721 files changed, 401740 insertions, 197287 deletions

diff --git a/xlators/Makefile.am b/xlators/Makefile.am
index f60fa85ce06..ef20cbb64fa 100644
--- a/xlators/Makefile.am
+++ b/xlators/Makefile.am
@@ -1,4 +1,13 @@
-SUBDIRS = cluster storage protocol performance debug features encryption mount nfs mgmt system \
-          playground
+if BUILD_GNFS
+  GNFS_DIR = nfs
+endif
+
+DIST_SUBDIRS = cluster storage protocol performance debug features \
+          mount nfs mgmt system playground meta
+
+SUBDIRS = cluster storage protocol performance debug features \
+          mount ${GNFS_DIR} mgmt system playground meta
+
+EXTRA_DIST = xlator.sym
 
 CLEANFILES =
diff --git a/xlators/bindings/Makefile.am b/xlators/bindings/Makefile.am
deleted file mode 100644
index f7766580257..00000000000
--- a/xlators/bindings/Makefile.am
+++ /dev/null
@@ -1 +0,0 @@
-SUBDIRS = $(BINDINGS_SUBDIRS)
diff --git a/xlators/bindings/python/src/Makefile.am b/xlators/bindings/python/src/Makefile.am
deleted file mode 100644
index 90370d86175..00000000000
--- a/xlators/bindings/python/src/Makefile.am
+++ /dev/null
@@ -1,19 +0,0 @@
-
-xlator_PROGRAMS = python.so
-
-xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/bindings
-
-python_PYTHON = gluster.py glustertypes.py glusterstack.py
-
-pythondir = $(xlatordir)/python
-
-python_so_SOURCES = python.c
-
-AM_CFLAGS = -fPIC $(GF_CPPFLAGS) -Wall \
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles \
-	$(PYTHON_CPPLAGS) -DGLUSTER_PYTHON_PATH=\"$(pythondir)\"
-
-AM_LDFLAGS = $(PYTHON_LDFLAGS)
-
-CLEANFILES = 
-
diff --git a/xlators/bindings/python/src/gluster.py b/xlators/bindings/python/src/gluster.py
deleted file mode 100644
index 337c983ecb9..00000000000
--- a/xlators/bindings/python/src/gluster.py
+++ /dev/null
@@ -1,40 +0,0 @@
-
-#  Copyright (c) 2007-2012 Red Hat, Inc. <http://www.redhat.com>
-#  This file is part of GlusterFS.
-#
-#  This file is licensed to you under your choice of the GNU Lesser
-#  General Public License, version 3 or any later version (LGPLv3 or
-#  later), or the GNU General Public License, version 2 (GPLv2), in all
-#  cases as published by the Free Software Foundation.
-
-from ctypes import *
-from glustertypes import *
-from glusterstack import *
-import sys
-import inspect
-
-libglusterfs = CDLL("libglusterfs.so")
-_gf_log = libglusterfs._gf_log
-_gf_log.restype = c_int32
-_gf_log.argtypes = [c_char_p, c_char_p, c_char_p, c_int32, c_int, c_char_p]
-
-gf_log_loglevel = c_int.in_dll(libglusterfs, "gf_log_loglevel")
-
-GF_LOG_NONE = 0
-GF_LOG_CRITICAL = 1
-GF_LOG_ERROR = 2
-GF_LOG_WARNING = 3
-GF_LOG_DEBUG = 4
-
-def gf_log(module, level, fmt, *params):
-    if level <= gf_log_loglevel:
-        frame = sys._getframe(1)
-        _gf_log(module, frame.f_code.co_filename, frame.f_code.co_name,
-                frame.f_lineno, level, fmt, *params)
-
-class ComplexTranslator(object):
-    def __init__(self, xlator):
-        self.xlator = xlator_t.from_address(xlator)
-
-    def __getattr__(self, item):
-        return getattr(self.xlator, item)
diff --git a/xlators/bindings/python/src/glusterstack.py b/xlators/bindings/python/src/glusterstack.py
deleted file mode 100644
index 0c071ae983a..00000000000
--- a/xlators/bindings/python/src/glusterstack.py
+++ /dev/null
@@ -1,48 +0,0 @@
-
-#  Copyright (c) 2007-2012 Red Hat, Inc. <http://www.redhat.com>
-#  This file is part of GlusterFS.
-#
-#  This file is licensed to you under your choice of the GNU Lesser
-#  General Public License, version 3 or any later version (LGPLv3 or
-#  later), or the GNU General Public License, version 2 (GPLv2), in all
-#  cases as published by the Free Software Foundation.
-
-from ctypes import *
-from glustertypes import *
-
-libc = CDLL("libc.so.6")
-calloc = libc.calloc
-calloc.argtypes = [c_int, c_int]
-calloc.restype = c_void_p
-
-# TODO: Can these be done in C somehow?
-def stack_wind(frame, rfn, obj, fn, *params):
-    """Frame is a frame object"""
-    _new = cast(calloc(1, sizeof(call_frame_t)), POINTER(call_frame_t))
-    _new[0].root = frame.root
-    _new[0].next = frame.root[0].frames.next
-    _new[0].prev = pointer(frame.root[0].frames)
-    if frame.root[0].frames.next:
-        frame.root[0].frames.next[0].prev = _new
-    frame.root[0].frames.next = _new
-    _new[0].this = obj
-    # TODO: Type checking like tmp_cbk?
-    _new[0].ret = rfn
-    _new[0].parent = pointer(frame)
-    _new[0].cookie = cast(_new, c_void_p)
-    # TODO: Initialize lock
-    #_new.lock.init()
-    frame.ref_count += 1
-    fn(_new, obj, *params)
-
-def stack_unwind(frame, *params):
-    """Frame is a frame object"""
-    fn = frame[0].ret
-    parent = frame[0].parent[0]
-    parent.ref_count -= 1
-
-    op_ret = params[0]
-    op_err = params[1]
-    params = params[2:]
-    fn(parent, call_frame_t.from_address(frame[0].cookie), parent.this,
-            op_ret, op_err, *params)
diff --git a/xlators/bindings/python/src/glustertypes.py b/xlators/bindings/python/src/glustertypes.py
deleted file mode 100644
index 98437d22ee9..00000000000
--- a/xlators/bindings/python/src/glustertypes.py
+++ /dev/null
@@ -1,160 +0,0 @@
-
-#  Copyright (c) 2007-2012 Red Hat, Inc. <http://www.redhat.com>
-#  This file is part of GlusterFS.
-#
-#  This file is licensed to you under your choice of the GNU Lesser
-#  General Public License, version 3 or any later version (LGPLv3 or
-#  later), or the GNU General Public License, version 2 (GPLv2), in all
-#  cases as published by the Free Software Foundation.
-
-from ctypes import *
-import collections
-
-#
-# Forward declaration of some gluster types
-#
-class call_frame_t(Structure):
-    pass
-
-class call_ctx_t(Structure):
-    pass
-
-class call_pool_t(Structure):
-    pass
-
-class xlator_t(Structure):
-    def _getFirstChild(self):
-        return self.children[0].xlator
-    firstChild = property(_getFirstChild)
-
-class xlator_list_t(Structure):
-    pass
-
-class xlator_fops(Structure):
-    pass
-
-class xlator_mops(Structure):
-    pass
-
-class glusterfs_ctx_t(Structure):
-    pass
-
-class list_head(Structure):
-    pass
-
-class dict_t(Structure):
-    pass
-
-class inode_table_t(Structure):
-    pass
-
-class fd_t(Structure):
-    pass
-
-class iovec(Structure):
-    _fields_ = [
-            ("iov_base", c_void_p),
-            ("iov_len", c_size_t),
-            ]
-
-    def __init__(self, s):
-        self.iov_base = cast(c_char_p(s), c_void_p)
-        self.iov_len = len(s)
-
-    def getBytes(self):
-        return string_at(self.iov_base, self.iov_len)
-
-# This is a pthread_spinlock_t
-# TODO: what happens to volatile-ness?
-gf_lock_t = c_int
-
-uid_t = c_uint32
-gid_t = c_uint32
-pid_t = c_int32
-
-off_t = c_int64
-
-#
-# Function pointer types
-#
-ret_fn_t = CFUNCTYPE(c_int32, POINTER(call_frame_t), POINTER(call_frame_t),
-                              POINTER(xlator_t), c_int32, c_int32)
-
-fini_fn_t = CFUNCTYPE(None, POINTER(xlator_t))
-init_fn_t = CFUNCTYPE(c_int32, POINTER(xlator_t))
-event_notify_fn_t = CFUNCTYPE(c_int32, POINTER(xlator_t), c_int32, c_void_p)
-
-list_head._fields_ = [
-        ("next", POINTER(list_head)),
-        ("prev", POINTER(list_head)),
-        ]
-
-call_frame_t._fields_ = [
-        ("root", POINTER(call_ctx_t)),
-        ("parent", POINTER(call_frame_t)),
-        ("next", POINTER(call_frame_t)),
-        ("prev", POINTER(call_frame_t)),
-        ("local", c_void_p),
-        ("this", POINTER(xlator_t)),
-        ("ret", ret_fn_t),
-        ("ref_count", c_int32),
-        ("lock", gf_lock_t),
-        ("cookie", c_void_p),
-        ("op", c_int32),
-        ("type", c_int8),
-        ]
-
-call_ctx_t._fields_ = [
-        ("all_frames", list_head),
-        ("trans", c_void_p),
-        ("pool", call_pool_t),
-        ("unique", c_uint64),
-        ("state", c_void_p),
-        ("uid", uid_t),
-        ("gid", gid_t),
-        ("pid", pid_t),
-        ("frames", call_frame_t),
-        ("req_refs", POINTER(dict_t)),
-        ("rsp_refs", POINTER(dict_t)),
-        ]
-
-xlator_t._fields_ = [
-        ("name", c_char_p),
-        ("type", c_char_p),
-        ("next", POINTER(xlator_t)),
-        ("prev", POINTER(xlator_t)),
-        ("parent", POINTER(xlator_t)),
-        ("children", POINTER(xlator_list_t)),
-        ("fops", POINTER(xlator_fops)),
-        ("mops", POINTER(xlator_mops)),
-        ("fini", fini_fn_t),
-        ("init", init_fn_t),
-        ("notify", event_notify_fn_t),
-        ("options", POINTER(dict_t)),
-        ("ctx", POINTER(glusterfs_ctx_t)),
-        ("itable", POINTER(inode_table_t)),
-        ("ready", c_char),
-        ("private", c_void_p),
-        ]
-
-xlator_list_t._fields_ = [
-        ("xlator", POINTER(xlator_t)),
-        ("next", POINTER(xlator_list_t)),
-        ]
-
-fop_functions = collections.defaultdict(lambda: c_void_p)
-fop_function_names = ['lookup', 'forget', 'stat', 'fstat', 'chmod', 'fchmod',
-        'chown', 'fchown', 'truncate', 'ftruncate', 'utimens', 'access',
-        'readlink', 'mknod', 'mkdir', 'unlink', 'rmdir', 'symlink',
-        'rename', 'link', 'create', 'open', 'readv', 'writev', 'flush',
-        'close', 'fsync', 'opendir', 'readdir', 'closedir', 'fsyncdir',
-        'statfs', 'setxattr', 'getxattr', 'removexattr', 'lk', 'writedir',
-        # TODO: Call backs?
-        ]
-
-fop_writev_t = CFUNCTYPE(c_int32, POINTER(call_frame_t), POINTER(xlator_t),
-                                  POINTER(fd_t), POINTER(iovec), c_int32,
-                                  off_t)
-
-fop_functions['writev'] = fop_writev_t
-xlator_fops._fields_ = [(f, fop_functions[f]) for f in fop_function_names]
diff --git a/xlators/bindings/python/src/python.c b/xlators/bindings/python/src/python.c
deleted file mode 100644
index 9b96790dec2..00000000000
--- a/xlators/bindings/python/src/python.c
+++ /dev/null
@@ -1,222 +0,0 @@
-/*
-   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
-   This file is part of GlusterFS.
-
-   This file is licensed to you under your choice of the GNU Lesser
-   General Public License, version 3 or any later version (LGPLv3 or
-   later), or the GNU General Public License, version 2 (GPLv2), in all
-   cases as published by the Free Software Foundation.
-*/
-#include <Python.h>
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "xlator.h"
-#include "logging.h"
-#include "defaults.h"
-
-typedef struct
-{
-    char        *scriptname;
-    PyObject    *pXlator;
-    PyObject    *pScriptModule;
-    PyObject    *pGlusterModule;
-    PyThreadState *pInterp;
-
-    PyObject    *pFrameType, *pVectorType, *pFdType;
-} python_private_t;
-
-int32_t
-python_writev (call_frame_t *frame,
-              xlator_t *this,
-              fd_t *fd,
-              struct iovec *vector,
-              int32_t count,
-              off_t offset)
-{
-  python_private_t *priv = (python_private_t *)this->private;
-  gf_log("python", GF_LOG_DEBUG, "In writev");
-  if (PyObject_HasAttrString(priv->pXlator, "writev"))
-  {
-
-      PyObject *retval = PyObject_CallMethod(priv->pXlator, "writev",
-              "O O O i l",
-              PyObject_CallMethod(priv->pFrameType, "from_address", "O&", PyLong_FromVoidPtr, frame),
-              PyObject_CallMethod(priv->pFdType, "from_address", "O&", PyLong_FromVoidPtr, fd),
-              PyObject_CallMethod(priv->pVectorType, "from_address", "O&", PyLong_FromVoidPtr, vector),
-              count,
-              offset);
-      if (PyErr_Occurred())
-      {
-          PyErr_Print();
-      }
-      Py_XDECREF(retval);
-  }
-  else
-  {
-      return default_writev(frame, this, fd, vector, count, offset);
-  }
-  return 0;
-}
-
-struct xlator_fops fops = {
-    .writev       = python_writev
-};
-
-static PyObject *
-AnonModule_FromFile (const char* fname)
-{
-    // Get the builtins
-    PyThreadState* pThread = PyThreadState_Get();
-    PyObject *pBuiltins = pThread->interp->builtins;
-
-    if (PyErr_Occurred())
-    {
-        PyErr_Print();
-        return NULL;
-    }
-
-    // Create a new dictionary for running code in
-    PyObject *pModuleDict = PyDict_New();
-    PyDict_SetItemString(pModuleDict, "__builtins__", pBuiltins);
-    Py_INCREF(pBuiltins);
-
-    // Run the file in the new context
-    FILE* fp = fopen(fname, "r");
-    PyRun_File(fp, fname, Py_file_input, pModuleDict, pModuleDict);
-    fclose(fp);
-    if (PyErr_Occurred())
-    {
-        PyErr_Print();
-        Py_DECREF(pModuleDict);
-        Py_DECREF(pBuiltins);
-        return NULL;
-    }
-
-    // Create an object to hold the new context
-    PyRun_String("class ModuleWrapper(object):\n\tpass\n", Py_single_input, pModuleDict, pModuleDict);
-    if (PyErr_Occurred())
-    {
-        PyErr_Print();
-        Py_DECREF(pModuleDict);
-        Py_DECREF(pBuiltins);
-        return NULL;
-    }
-    PyObject *pModule = PyRun_String("ModuleWrapper()", Py_eval_input, pModuleDict, pModuleDict);
-    if (PyErr_Occurred())
-    {
-        PyErr_Print();
-        Py_DECREF(pModuleDict);
-        Py_DECREF(pBuiltins);
-        Py_XDECREF(pModule);
-        return NULL;
-    }
-
-    // Set the new context's dictionary to the one we used to run the code
-    // inside
-    PyObject_SetAttrString(pModule, "__dict__", pModuleDict);
-    if (PyErr_Occurred())
-    {
-        PyErr_Print();
-        Py_DECREF(pModuleDict);
-        Py_DECREF(pBuiltins);
-        Py_DECREF(pModule);
-        return NULL;
-    }
-
-    return pModule;
-}
-
-int32_t
-init (xlator_t *this)
-{
-  // This is ok to call more than once per process
-  Py_InitializeEx(0);
-
-  if (!this->children) {
-    gf_log ("python", GF_LOG_ERROR,
-            "FATAL: python should have exactly one child");
-    return -1;
-  }
-
-  python_private_t *priv = CALLOC (sizeof (python_private_t), 1);
-  ERR_ABORT (priv);
-
-  data_t *scriptname = dict_get (this->options, "scriptname");
-  if (scriptname) {
-      priv->scriptname = data_to_str(scriptname);
-  } else {
-      gf_log("python", GF_LOG_ERROR,
-              "FATAL: python requires the scriptname parameter");
-      return -1;
-  }
-
-  priv->pInterp = Py_NewInterpreter();
-
-  // Adjust python's path
-  PyObject *syspath = PySys_GetObject("path");
-  PyObject *path = PyString_FromString(GLUSTER_PYTHON_PATH);
-  PyList_Append(syspath, path);
-  Py_DECREF(path);
-
-  gf_log("python", GF_LOG_DEBUG,
-          "Loading gluster module");
-
-  priv->pGlusterModule = PyImport_ImportModule("gluster");
-  if (PyErr_Occurred())
-  {
-      PyErr_Print();
-      return -1;
-  }
-
-  priv->pFrameType = PyObject_GetAttrString(priv->pGlusterModule, "call_frame_t");
-  priv->pFdType = PyObject_GetAttrString(priv->pGlusterModule, "fd_t");
-  priv->pVectorType = PyObject_GetAttrString(priv->pGlusterModule, "iovec");
-
-  gf_log("python", GF_LOG_DEBUG, "Loading script...%s", priv->scriptname);
-
-  priv->pScriptModule = AnonModule_FromFile(priv->scriptname);
-  if (!priv->pScriptModule || PyErr_Occurred())
-  {
-      gf_log("python", GF_LOG_ERROR, "Error loading %s", priv->scriptname);
-      PyErr_Print();
-      return -1;
-  }
-
-  if (!PyObject_HasAttrString(priv->pScriptModule, "xlator"))
-  {
-      gf_log("python", GF_LOG_ERROR, "%s does not have a xlator attribute", priv->scriptname);
-      return -1;
-  }
-  gf_log("python", GF_LOG_DEBUG, "Instantiating translator");
-  priv->pXlator = PyObject_CallMethod(priv->pScriptModule, "xlator", "O&",
-          PyLong_FromVoidPtr, this);
-  if (PyErr_Occurred() || !priv->pXlator)
-  {
-      PyErr_Print();
-      return -1;
-  }
-
-  this->private = priv;
-
-  gf_log ("python", GF_LOG_DEBUG, "python xlator loaded");
-  return 0;
-}
-
-void
-fini (xlator_t *this)
-{
-  python_private_t *priv = (python_private_t*)(this->private);
-  Py_DECREF(priv->pXlator);
-  Py_DECREF(priv->pScriptModule);
-  Py_DECREF(priv->pGlusterModule);
-  Py_DECREF(priv->pFrameType);
-  Py_DECREF(priv->pFdType);
-  Py_DECREF(priv->pVectorType);
-  Py_EndInterpreter(priv->pInterp);
-  return;
-}
diff --git a/xlators/bindings/python/src/testxlator.py b/xlators/bindings/python/src/testxlator.py
deleted file mode 100644
index 59a991dcab9..00000000000
--- a/xlators/bindings/python/src/testxlator.py
+++ /dev/null
@@ -1,49 +0,0 @@
-"""
-  Copyright (c) 2007-2012 Red Hat, Inc. <http://www.redhat.com>
-  This file is part of GlusterFS.
-
-  This file is licensed to you under your choice of the GNU Lesser
-  General Public License, version 3 or any later version (LGPLv3 or
-  later), or the GNU General Public License, version 2 (GPLv2), in all
-  cases as published by the Free Software Foundation.
-"""
-
-"""
-This is a test translator written in python.
-
-Important things to note:
-    This file must be import-able from glusterfsd.  This probably means
-    setting PYTHONPATH to where this file is located.
-
-    This file must have a top-level xlator class object that will be
-    used to instantiate individual translators.
-"""
-from gluster import *
-
-class MyXlator(ComplexTranslator):
-    name = "MyXlator"
-    def writev_cbk(self, frame, cookie, op_ret, op_errno, buf):
-        stack_unwind(frame, op_ret, op_errno, buf)
-        return 0
-
-    def writev(self, frame, fd, vector, count, offset):
-        gf_log(self.name, GF_LOG_WARNING, "writev %i bytes", vector.iov_len)
-        # TODO: Use cookie to pass this to writev_cbk
-        old_count = vector.iov_len
-
-        data = vector.getBytes().encode("zlib")
-
-        vector = iovec(data)
-        gf_log(self.name, GF_LOG_WARNING, "writev %i bytes", vector.iov_len)
-
-        @ret_fn_t
-        def rfn(frame, prev, this, op_ret, op_errno, *params):
-            if len(params) == 0:
-                params = [0]
-            return self.writev_cbk(frame, prev, old_count, op_errno, *params)
-
-        stack_wind(frame, rfn, self.firstChild,
-                self.firstChild[0].fops[0].writev, fd, vector, count, offset)
-        return 0
-
-xlator = MyXlator
diff --git a/xlators/cluster/Makefile.am b/xlators/cluster/Makefile.am
index 0990822a7d3..8e067d5ab58 100644
--- a/xlators/cluster/Makefile.am
+++ b/xlators/cluster/Makefile.am
@@ -1,3 +1,3 @@
-SUBDIRS = stripe afr dht
+SUBDIRS = afr dht ec
 
 CLEANFILES = 
diff --git a/xlators/cluster/afr/src/Makefile.am b/xlators/cluster/afr/src/Makefile.am
index 35d18a6c0da..610819b28fc 100644
--- a/xlators/cluster/afr/src/Makefile.am
+++ b/xlators/cluster/afr/src/Makefile.am
@@ -1,29 +1,28 @@
-xlator_LTLIBRARIES = afr.la pump.la
+xlator_LTLIBRARIES = afr.la
 xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/cluster
 
 afr_common_source = afr-dir-read.c afr-dir-write.c afr-inode-read.c \
-	afr-inode-write.c afr-open.c afr-transaction.c afr-self-heal-data.c \
-	afr-self-heal-common.c afr-self-heal-metadata.c afr-self-heal-entry.c \
-	afr-self-heal-algorithm.c afr-lk-common.c afr-self-heald.c \
+	afr-inode-write.c afr-open.c afr-transaction.c afr-lk-common.c \
+	afr-read-txn.c \
 	$(top_builddir)/xlators/lib/src/libxlator.c
 
-afr_la_LDFLAGS = -module -avoid-version
-afr_la_SOURCES = $(afr_common_source) afr.c
-afr_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+AFR_SELFHEAL_SOURCES = afr-self-heal-common.c afr-self-heal-data.c \
+	afr-self-heal-entry.c afr-self-heal-metadata.c afr-self-heald.c \
+	afr-self-heal-name.c
 
-pump_la_LDFLAGS = -module -avoid-version
-pump_la_SOURCES =  $(afr_common_source) pump.c
-pump_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+afr_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+afr_la_SOURCES = $(afr_common_source) $(AFR_SELFHEAL_SOURCES) afr.c
+afr_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
 noinst_HEADERS = afr.h afr-transaction.h afr-inode-write.h afr-inode-read.h \
-	afr-dir-read.h afr-dir-write.h afr-self-heal.h afr-self-heal-common.h \
-	afr-self-heal-algorithm.h pump.h afr-mem-types.h afr-common.c \
-	afr-self-heald.h $(top_builddir)/xlators/lib/src/libxlator.h \
-	$(top_builddir)/glusterfsd/src/glusterfsd.h
+	afr-dir-read.h afr-dir-write.h afr-self-heal.h afr-mem-types.h \
+	afr-common.c afr-self-heald.h \
+	$(top_builddir)/xlators/lib/src/libxlator.h afr-messages.h
 
 AM_CPPFLAGS = $(GF_CPPFLAGS) \
 	-I$(top_srcdir)/libglusterfs/src -I$(top_srcdir)/xlators/lib/src \
-	-I$(top_srcdir)/rpc/rpc-lib/src
+	-I$(top_srcdir)/rpc/rpc-lib/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
 
 AM_CFLAGS = -Wall $(GF_CFLAGS)
 
@@ -31,7 +30,6 @@ CLEANFILES =
 
 uninstall-local:
 	rm -f $(DESTDIR)$(xlatordir)/replicate.so
-	rm -f $(DESTDIR)$(xlatordir)/pump.so
 
 install-data-hook:
 	ln -sf afr.so $(DESTDIR)$(xlatordir)/replicate.so
diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c
index 8b6ab9a7f69..032ab5c8001 100644
--- a/xlators/cluster/afr/src/afr-common.c
+++ b/xlators/cluster/afr/src/afr-common.c
@@ -15,29 +15,20 @@
 #include <stdlib.h>
 #include <signal.h>
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
+#include <glusterfs/glusterfs.h>
 #include "afr.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
-#include "byte-order.h"
-#include "statedump.h"
-#include "inode.h"
-
-#include "fd.h"
+#include <glusterfs/dict.h>
+#include <glusterfs/hashfn.h>
+#include <glusterfs/list.h>
+#include <glusterfs/call-stub.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/byte-order.h>
+#include <glusterfs/statedump.h>
+#include <glusterfs/events.h>
+#include <glusterfs/upcall-utils.h>
 
 #include "afr-inode-read.h"
 #include "afr-inode-write.h"
@@ -45,4399 +36,7843 @@
 #include "afr-dir-write.h"
 #include "afr-transaction.h"
 #include "afr-self-heal.h"
-#include "afr-self-heal-common.h"
 #include "afr-self-heald.h"
-#include "pump.h"
+#include "afr-messages.h"
 
-#define AFR_ICTX_OPENDIR_DONE_MASK     0x0000000100000000ULL
-#define AFR_ICTX_READ_CHILD_MASK       0x00000000FFFFFFFFULL
+int32_t
+afr_quorum_errno(afr_private_t *priv)
+{
+    return ENOTCONN;
+}
 
-int
-afr_lookup_done_success_action (call_frame_t *frame, xlator_t *this,
-                                gf_boolean_t fail_conflict);
-void
-afr_children_copy (int32_t *dst, int32_t *src, unsigned int child_count)
+gf_boolean_t
+afr_is_private_directory(afr_private_t *priv, uuid_t pargfid, const char *name,
+                         pid_t pid)
 {
-        int     i = 0;
+    if (!__is_root_gfid(pargfid)) {
+        return _gf_false;
+    }
 
-        for (i = 0; i < child_count; i++)
-                dst[i] = src[i];
+    if (strcmp(name, GF_REPLICATE_TRASH_DIR) == 0) {
+        /*For backward compatibility /.landfill is private*/
+        return _gf_true;
+    }
+
+    if (pid == GF_CLIENT_PID_GSYNCD) {
+        /*geo-rep needs to create/sync private directory on slave because
+         * it appears in changelog*/
+        return _gf_false;
+    }
+
+    if (pid == GF_CLIENT_PID_GLFS_HEAL || pid == GF_CLIENT_PID_SELF_HEALD) {
+        if (strcmp(name, priv->anon_inode_name) == 0) {
+            /* anonymous-inode dir is private*/
+            return _gf_true;
+        }
+    } else {
+        if (strncmp(name, AFR_ANON_DIR_PREFIX, strlen(AFR_ANON_DIR_PREFIX)) ==
+            0) {
+            /* anonymous-inode dir prefix is private for geo-rep to work*/
+            return _gf_true;
+        }
+    }
+
+    return _gf_false;
 }
 
 void
-afr_xattr_req_prepare (xlator_t *this, dict_t *xattr_req, const char *path)
+afr_fill_success_replies(afr_local_t *local, afr_private_t *priv,
+                         unsigned char *replies)
 {
-        int             i           = 0;
-        afr_private_t   *priv       = NULL;
-        int             ret         = 0;
-
-        priv   = this->private;
+    int i = 0;
 
-        for (i = 0; i < priv->child_count; i++) {
-                ret = dict_set_uint64 (xattr_req, priv->pending_key[i],
-                                       3 * sizeof(int32_t));
-                if (ret < 0)
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "%s: Unable to set dict value for %s",
-                                path, priv->pending_key[i]);
-                /* 3 = data+metadata+entry */
-        }
-        ret = dict_set_int32 (xattr_req, GF_GFIDLESS_LOOKUP, 1);
-        if (ret) {
-                gf_log (this->name, GF_LOG_DEBUG, "%s: failed to set gfidless "
-                        "lookup", path);
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->replies[i].valid && local->replies[i].op_ret == 0) {
+            replies[i] = 1;
+        } else {
+            replies[i] = 0;
         }
+    }
 }
 
 int
-afr_lookup_xattr_req_prepare (afr_local_t *local, xlator_t *this,
-                              dict_t *xattr_req, loc_t *loc, void **gfid_req)
-{
-        int     ret = -ENOMEM;
+afr_fav_child_reset_sink_xattrs(void *opaque);
 
-        GF_ASSERT (gfid_req);
-
-        *gfid_req = NULL;
-        local->xattr_req = dict_new ();
-        if (!local->xattr_req)
-                goto out;
-        if (xattr_req)
-                dict_copy (xattr_req, local->xattr_req);
+int
+afr_fav_child_reset_sink_xattrs_cbk(int ret, call_frame_t *frame, void *opaque);
 
-        afr_xattr_req_prepare (this, local->xattr_req, loc->path);
-        ret = dict_set_uint64 (local->xattr_req, GLUSTERFS_INODELK_COUNT, 0);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: Unable to set dict value for %s",
-                        loc->path, GLUSTERFS_INODELK_COUNT);
-        }
-        ret = dict_set_uint64 (local->xattr_req, GLUSTERFS_ENTRYLK_COUNT, 0);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: Unable to set dict value for %s",
-                        loc->path, GLUSTERFS_ENTRYLK_COUNT);
-        }
+static void
+afr_discover_done(call_frame_t *frame, xlator_t *this);
 
-        ret = dict_set_uint32 (local->xattr_req, GLUSTERFS_PARENT_ENTRYLK, 0);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: Unable to set dict value for %s",
-                        loc->path, GLUSTERFS_PARENT_ENTRYLK);
-        }
+int
+afr_dom_lock_acquire_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int op_ret, int op_errno, dict_t *xdata)
+{
+    afr_local_t *local = frame->local;
+    afr_private_t *priv = this->private;
+    int i = (long)cookie;
 
-        ret = dict_get_ptr (local->xattr_req, "gfid-req", gfid_req);
-        if (ret) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "%s: failed to get the gfid from dict", loc->path);
-                *gfid_req = NULL;
-        } else {
-                if (loc->parent != NULL)
-                        dict_del (local->xattr_req, "gfid-req");
-        }
-        ret = 0;
-out:
-        return ret;
-}
+    local->cont.lk.dom_lock_op_ret[i] = op_ret;
+    local->cont.lk.dom_lock_op_errno[i] = op_errno;
+    if (op_ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, AFR_MSG_LK_HEAL_DOM,
+               "%s: Failed to acquire %s on %s",
+               uuid_utoa(local->fd->inode->gfid), AFR_LK_HEAL_DOM,
+               priv->children[i]->name);
+    } else {
+        local->cont.lk.dom_locked_nodes[i] = 1;
+    }
 
-void
-afr_lookup_save_gfid (uuid_t dst, void* new, const loc_t *loc)
-{
-        inode_t  *inode = NULL;
+    syncbarrier_wake(&local->barrier);
 
-        inode = loc->inode;
-        if (inode && !uuid_is_null (inode->gfid))
-                uuid_copy (dst, inode->gfid);
-        else if (!uuid_is_null (loc->gfid))
-                uuid_copy (dst, loc->gfid);
-        else if (new && !uuid_is_null (new))
-                uuid_copy (dst, new);
+    return 0;
 }
 
 int
-afr_errno_count (int32_t *children, int *child_errno,
-                 unsigned int child_count, int32_t op_errno)
-{
-        int i = 0;
-        int errno_count = 0;
-        int child = 0;
-
-        for (i = 0; i < child_count; i++) {
-                if (children) {
-                        child = children[i];
-                        if (child == -1)
-                                break;
-                } else {
-                        child = i;
-                }
-                if (child_errno[child] == op_errno)
-                        errno_count++;
+afr_dom_lock_acquire(call_frame_t *frame)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    struct gf_flock flock = {
+        0,
+    };
+    int i = 0;
+
+    priv = frame->this->private;
+    local = frame->local;
+    local->cont.lk.dom_locked_nodes = GF_CALLOC(
+        priv->child_count, sizeof(*local->cont.lk.locked_nodes),
+        gf_afr_mt_char);
+    if (!local->cont.lk.dom_locked_nodes) {
+        return -ENOMEM;
+    }
+    local->cont.lk.dom_lock_op_ret = GF_CALLOC(
+        priv->child_count, sizeof(*local->cont.lk.dom_lock_op_ret),
+        gf_afr_mt_int32_t);
+    if (!local->cont.lk.dom_lock_op_ret) {
+        return -ENOMEM; /* CALLOC'd members are freed in afr_local_cleanup. */
+    }
+    local->cont.lk.dom_lock_op_errno = GF_CALLOC(
+        priv->child_count, sizeof(*local->cont.lk.dom_lock_op_errno),
+        gf_afr_mt_int32_t);
+    if (!local->cont.lk.dom_lock_op_errno) {
+        return -ENOMEM; /* CALLOC'd members are freed in afr_local_cleanup. */
+    }
+    flock.l_type = F_WRLCK;
+
+    AFR_ONALL(frame, afr_dom_lock_acquire_cbk, finodelk, AFR_LK_HEAL_DOM,
+              local->fd, F_SETLK, &flock, NULL);
+
+    if (!afr_has_quorum(local->cont.lk.dom_locked_nodes, frame->this, NULL))
+        goto blocking_lock;
+
+    /*If any of the bricks returned EAGAIN, we still need blocking locks.*/
+    if (AFR_COUNT(local->cont.lk.dom_locked_nodes, priv->child_count) !=
+        priv->child_count) {
+        for (i = 0; i < priv->child_count; i++) {
+            if (local->cont.lk.dom_lock_op_ret[i] == -1 &&
+                local->cont.lk.dom_lock_op_errno[i] == EAGAIN)
+                goto blocking_lock;
         }
-        return errno_count;
-}
+    }
 
-int32_t
-afr_set_dict_gfid (dict_t *dict, uuid_t gfid)
-{
-        int ret       = 0;
-        uuid_t *pgfid = NULL;
+    return 0;
 
-        GF_ASSERT (gfid);
+blocking_lock:
+    afr_dom_lock_release(frame);
+    AFR_ONALL(frame, afr_dom_lock_acquire_cbk, finodelk, AFR_LK_HEAL_DOM,
+              local->fd, F_SETLKW, &flock, NULL);
+    if (!afr_has_quorum(local->cont.lk.dom_locked_nodes, frame->this, NULL)) {
+        afr_dom_lock_release(frame);
+        return -afr_quorum_errno(priv);
+    }
 
-        pgfid = GF_CALLOC (1, sizeof (uuid_t), gf_common_mt_char);
-        if (!pgfid) {
-                ret = -1;
-                goto out;
-        }
+    return 0;
+}
 
-        uuid_copy (*pgfid, gfid);
+int
+afr_dom_lock_release_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int op_ret, int op_errno, dict_t *xdata)
+{
+    afr_local_t *local = frame->local;
+    afr_private_t *priv = this->private;
+    int i = (long)cookie;
 
-        ret = dict_set_dynptr (dict, "gfid-req", pgfid, sizeof (uuid_t));
-        if (ret)
-                gf_log (THIS->name, GF_LOG_ERROR, "gfid set failed");
+    if (op_ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, AFR_MSG_LK_HEAL_DOM,
+               "%s: Failed to release %s on %s", local->loc.path,
+               AFR_LK_HEAL_DOM, priv->children[i]->name);
+    }
+    local->cont.lk.dom_locked_nodes[i] = 0;
 
-out:
-        if (ret && pgfid)
-                GF_FREE (pgfid);
+    syncbarrier_wake(&local->barrier);
 
-        return ret;
+    return 0;
 }
 
 void
-afr_inode_ctx_destroy (afr_inode_ctx_t *ctx)
-{
-        if (!ctx)
-                return;
-        GF_FREE (ctx->fresh_children);
-        GF_FREE (ctx);
+afr_dom_lock_release(call_frame_t *frame)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    unsigned char *locked_on = NULL;
+    struct gf_flock flock = {
+        0,
+    };
+
+    local = frame->local;
+    priv = frame->this->private;
+    locked_on = local->cont.lk.dom_locked_nodes;
+    if (AFR_COUNT(locked_on, priv->child_count) == 0)
+        return;
+    flock.l_type = F_UNLCK;
+
+    AFR_ONLIST(locked_on, frame, afr_dom_lock_release_cbk, finodelk,
+               AFR_LK_HEAL_DOM, local->fd, F_SETLK, &flock, NULL);
+
+    return;
 }
 
-afr_inode_ctx_t*
-__afr_inode_ctx_get (inode_t *inode, xlator_t *this)
+static void
+afr_lk_heal_info_cleanup(afr_lk_heal_info_t *info)
 {
-        int             ret      = 0;
-        uint64_t        ctx_addr = 0;
-        afr_inode_ctx_t *ctx     = NULL;
-        afr_private_t   *priv    = NULL;
+    if (!info)
+        return;
+    if (info->xdata_req)
+        dict_unref(info->xdata_req);
+    if (info->fd)
+        fd_unref(info->fd);
+    GF_FREE(info->locked_nodes);
+    GF_FREE(info->child_up_event_gen);
+    GF_FREE(info->child_down_event_gen);
+    GF_FREE(info);
+}
 
-        priv = this->private;
-        ret = __inode_ctx_get (inode, this, &ctx_addr);
-        if (ret < 0)
-                ctx_addr = 0;
-        if (ctx_addr != 0) {
-                ctx = (afr_inode_ctx_t*) (long) ctx_addr;
-                goto out;
-        }
-        ctx = GF_CALLOC (1, sizeof (*ctx),
-                         gf_afr_mt_inode_ctx_t);
-        if (!ctx)
-                goto fail;
-        ctx->fresh_children = GF_CALLOC (priv->child_count,
-                                         sizeof (*ctx->fresh_children),
-                                         gf_afr_mt_int32_t);
-        if (!ctx->fresh_children)
-                goto fail;
-        ret = __inode_ctx_put (inode, this, (uint64_t)ctx);
-        if (ret) {
-                gf_log_callingfn (this->name, GF_LOG_ERROR, "failed to "
-                                  "set the inode ctx (%s)",
-                                  uuid_utoa (inode->gfid));
-                goto fail;
+static int
+afr_add_lock_to_saved_locks(call_frame_t *frame, xlator_t *this)
+{
+    afr_private_t *priv = this->private;
+    afr_local_t *local = frame->local;
+    afr_lk_heal_info_t *info = NULL;
+    afr_fd_ctx_t *fd_ctx = NULL;
+    int ret = -ENOMEM;
+
+    info = GF_CALLOC(sizeof(*info), 1, gf_afr_mt_lk_heal_info_t);
+    if (!info) {
+        goto cleanup;
+    }
+    INIT_LIST_HEAD(&info->pos);
+    info->fd = fd_ref(local->fd);
+    info->cmd = local->cont.lk.cmd;
+    info->pid = frame->root->pid;
+    info->flock = local->cont.lk.user_flock;
+    info->xdata_req = dict_copy_with_ref(local->xdata_req, NULL);
+    if (!info->xdata_req) {
+        goto cleanup;
+    }
+    info->lk_owner = frame->root->lk_owner;
+    info->locked_nodes = GF_MALLOC(
+        sizeof(*info->locked_nodes) * priv->child_count, gf_afr_mt_char);
+    if (!info->locked_nodes) {
+        goto cleanup;
+    }
+    memcpy(info->locked_nodes, local->cont.lk.locked_nodes,
+           sizeof(*info->locked_nodes) * priv->child_count);
+    info->child_up_event_gen = GF_CALLOC(sizeof(*info->child_up_event_gen),
+                                         priv->child_count, gf_afr_mt_int32_t);
+    if (!info->child_up_event_gen) {
+        goto cleanup;
+    }
+    info->child_down_event_gen = GF_CALLOC(sizeof(*info->child_down_event_gen),
+                                           priv->child_count,
+                                           gf_afr_mt_int32_t);
+    if (!info->child_down_event_gen) {
+        goto cleanup;
+    }
+
+    LOCK(&local->fd->lock);
+    {
+        fd_ctx = __afr_fd_ctx_get(local->fd, this);
+        if (fd_ctx)
+            fd_ctx->lk_heal_info = info;
+    }
+    UNLOCK(&local->fd->lock);
+    if (!fd_ctx) {
+        goto cleanup;
+    }
+
+    LOCK(&priv->lock);
+    {
+        list_add_tail(&info->pos, &priv->saved_locks);
+    }
+    UNLOCK(&priv->lock);
+
+    return 0;
+cleanup:
+    gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_LK_HEAL_DOM,
+           "%s: Failed to add lock to healq",
+           uuid_utoa(local->fd->inode->gfid));
+    if (info) {
+        afr_lk_heal_info_cleanup(info);
+        if (fd_ctx) {
+            LOCK(&local->fd->lock);
+            {
+                fd_ctx->lk_heal_info = NULL;
+            }
+            UNLOCK(&local->fd->lock);
         }
+    }
+    return ret;
+}
 
+static int
+afr_remove_lock_from_saved_locks(afr_local_t *local, xlator_t *this)
+{
+    afr_private_t *priv = this->private;
+    struct gf_flock flock = local->cont.lk.user_flock;
+    afr_lk_heal_info_t *info = NULL;
+    afr_fd_ctx_t *fd_ctx = NULL;
+    int ret = -EINVAL;
+
+    fd_ctx = afr_fd_ctx_get(local->fd, this);
+    if (!fd_ctx || !fd_ctx->lk_heal_info) {
+        goto out;
+    }
+
+    info = fd_ctx->lk_heal_info;
+    if ((info->flock.l_start != flock.l_start) ||
+        (info->flock.l_whence != flock.l_whence) ||
+        (info->flock.l_len != flock.l_len)) {
+        /*TODO: Compare lkowners too.*/
+        goto out;
+    }
+
+    LOCK(&priv->lock);
+    {
+        list_del(&fd_ctx->lk_heal_info->pos);
+    }
+    UNLOCK(&priv->lock);
+
+    afr_lk_heal_info_cleanup(info);
+    fd_ctx->lk_heal_info = NULL;
+    ret = 0;
 out:
-        return ctx;
-
-fail:
-        afr_inode_ctx_destroy (ctx);
-        return NULL;
+    if (ret)
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_LK_HEAL_DOM,
+               "%s: Failed to remove lock from healq",
+               uuid_utoa(local->fd->inode->gfid));
+    return ret;
 }
 
-afr_inode_ctx_t*
-afr_inode_ctx_get (inode_t *inode, xlator_t *this)
+int
+afr_lock_heal_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct gf_flock *lock,
+                  dict_t *xdata)
 {
-        afr_inode_ctx_t *ctx = NULL;
+    afr_local_t *local = frame->local;
+    int i = (long)cookie;
 
-        LOCK (&inode->lock);
-        {
-                ctx = __afr_inode_ctx_get (inode, this);
-        }
-        UNLOCK (&inode->lock);
-        return ctx;
+    local->replies[i].valid = 1;
+    local->replies[i].op_ret = op_ret;
+    local->replies[i].op_errno = op_errno;
+    if (op_ret != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, AFR_MSG_LK_HEAL_DOM,
+               "Failed to heal lock on child %d for %s", i,
+               uuid_utoa(local->fd->inode->gfid));
+    }
+    syncbarrier_wake(&local->barrier);
+    return 0;
 }
 
-void
-afr_inode_get_ctx_params (xlator_t *this, inode_t *inode,
-                          afr_inode_params_t *params)
+int
+afr_getlk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct gf_flock *lock, dict_t *xdata)
 {
-        GF_ASSERT (inode);
-        GF_ASSERT (params);
+    afr_local_t *local = frame->local;
+    int i = (long)cookie;
 
-        afr_inode_ctx_t *ctx = NULL;
-        afr_private_t   *priv = NULL;
-        int             i = 0;
-        int32_t         read_child = -1;
-        int32_t         *fresh_children = NULL;
+    local->replies[i].valid = 1;
+    local->replies[i].op_ret = op_ret;
+    local->replies[i].op_errno = op_errno;
+    if (op_ret != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, AFR_MSG_LK_HEAL_DOM,
+               "Failed getlk for %s", uuid_utoa(local->fd->inode->gfid));
+    } else {
+        local->cont.lk.getlk_rsp[i] = *lock;
+    }
 
-        priv = this->private;
-        LOCK (&inode->lock);
-        {
-                ctx = __afr_inode_ctx_get (inode, this);
-                if (!ctx)
-                        goto unlock;
-                switch (params->op) {
-                case AFR_INODE_GET_READ_CTX:
-                        fresh_children = params->u.read_ctx.children;
-                        read_child = (int32_t)(ctx->masks &
-                                               AFR_ICTX_READ_CHILD_MASK);
-                        params->u.read_ctx.read_child = read_child;
-                        if (!fresh_children)
-                                goto unlock;
-                        for (i = 0; i < priv->child_count; i++)
-                                fresh_children[i] = ctx->fresh_children[i];
-                        break;
-                case AFR_INODE_GET_OPENDIR_DONE:
-                        params->u.value = _gf_false;
-                        if (ctx->masks & AFR_ICTX_OPENDIR_DONE_MASK)
-                                params->u.value = _gf_true;
-                        break;
-                default:
-                        GF_ASSERT (0);
-                        break;
-                }
-        }
-unlock:
-        UNLOCK (&inode->lock);
+    syncbarrier_wake(&local->barrier);
+    return 0;
 }
 
-gf_boolean_t
-afr_is_split_brain (xlator_t *this, inode_t *inode)
-{
-        afr_inode_ctx_t *ctx = NULL;
-        gf_boolean_t    spb  = _gf_false;
-
-        ctx = afr_inode_ctx_get (inode, this);
-        if (!ctx)
-                goto out;
-        if ((ctx->mdata_spb == SPB) || (ctx->data_spb == SPB))
-                spb = _gf_true;
+static gf_boolean_t
+afr_does_lk_owner_match(call_frame_t *frame, afr_private_t *priv,
+                        afr_lk_heal_info_t *info)
+{
+    int i = 0;
+    afr_local_t *local = frame->local;
+    struct gf_flock flock = {
+        0,
+    };
+    gf_boolean_t ret = _gf_true;
+    char *wind_on = alloca0(priv->child_count);
+    unsigned char *success_replies = alloca0(priv->child_count);
+    local->cont.lk.getlk_rsp = GF_CALLOC(sizeof(*local->cont.lk.getlk_rsp),
+                                         priv->child_count, gf_afr_mt_gf_lock);
+
+    flock = info->flock;
+    for (i = 0; i < priv->child_count; i++) {
+        if (info->locked_nodes[i])
+            wind_on[i] = 1;
+    }
+
+    AFR_ONLIST(wind_on, frame, afr_getlk_cbk, lk, info->fd, F_GETLK, &flock,
+               info->xdata_req);
+
+    afr_fill_success_replies(local, priv, success_replies);
+    if (AFR_COUNT(success_replies, priv->child_count) == 0) {
+        ret = _gf_false;
+        goto out;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid || local->replies[i].op_ret != 0)
+            continue;
+        if (local->cont.lk.getlk_rsp[i].l_type == F_UNLCK)
+            continue;
+        /*TODO: Do we really need to compare lkowner if F_UNLCK is true?*/
+        if (!is_same_lkowner(&local->cont.lk.getlk_rsp[i].l_owner,
+                             &info->lk_owner)) {
+            ret = _gf_false;
+            break;
+        }
+    }
 out:
-        return spb;
+    afr_local_replies_wipe(local, priv);
+    GF_FREE(local->cont.lk.getlk_rsp);
+    local->cont.lk.getlk_rsp = NULL;
+    return ret;
 }
 
-gf_boolean_t
-afr_is_opendir_done (xlator_t *this, inode_t *inode)
+static void
+afr_mark_fd_bad(fd_t *fd, xlator_t *this)
 {
-        afr_inode_params_t params = {0};
+    afr_fd_ctx_t *fd_ctx = NULL;
 
-        params.op = AFR_INODE_GET_OPENDIR_DONE;
-        afr_inode_get_ctx_params (this, inode, &params);
-        return params.u.value;
+    if (!fd)
+        return;
+    LOCK(&fd->lock);
+    {
+        fd_ctx = __afr_fd_ctx_get(fd, this);
+        if (fd_ctx) {
+            fd_ctx->is_fd_bad = _gf_true;
+            fd_ctx->lk_heal_info = NULL;
+        }
+    }
+    UNLOCK(&fd->lock);
 }
 
-int32_t
-afr_inode_get_read_ctx (xlator_t *this, inode_t *inode, int32_t *fresh_children)
+static void
+afr_add_lock_to_lkhealq(afr_private_t *priv, afr_lk_heal_info_t *info)
 {
-        afr_inode_params_t      params = {0};
+    LOCK(&priv->lock);
+    {
+        list_del(&info->pos);
+        list_add_tail(&info->pos, &priv->lk_healq);
+    }
+    UNLOCK(&priv->lock);
+}
 
-        params.op = AFR_INODE_GET_READ_CTX;
-        params.u.read_ctx.children = fresh_children;
-        afr_inode_get_ctx_params (this, inode, &params);
-        return params.u.read_ctx.read_child;
+static void
+afr_lock_heal_do(call_frame_t *frame, afr_private_t *priv,
+                 afr_lk_heal_info_t *info)
+{
+    int i = 0;
+    int op_errno = 0;
+    int32_t *current_event_gen = NULL;
+    afr_local_t *local = frame->local;
+    xlator_t *this = frame->this;
+    char *wind_on = alloca0(priv->child_count);
+    gf_boolean_t retry = _gf_true;
+
+    frame->root->pid = info->pid;
+    lk_owner_copy(&frame->root->lk_owner, &info->lk_owner);
+
+    op_errno = -afr_dom_lock_acquire(frame);
+    if ((op_errno != 0)) {
+        goto release;
+    }
+
+    if (!afr_does_lk_owner_match(frame, priv, info)) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_LK_HEAL_DOM,
+               "Ignoring lock heal for %s since lk-onwers mismatch. "
+               "Lock possibly pre-empted by another client.",
+               uuid_utoa(info->fd->inode->gfid));
+        goto release;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (info->locked_nodes[i])
+            continue;
+        wind_on[i] = 1;
+    }
+
+    current_event_gen = alloca(priv->child_count);
+    memcpy(current_event_gen, info->child_up_event_gen,
+           priv->child_count * sizeof *current_event_gen);
+    AFR_ONLIST(wind_on, frame, afr_lock_heal_cbk, lk, info->fd, info->cmd,
+               &info->flock, info->xdata_req);
+
+    LOCK(&priv->lock);
+    {
+        for (i = 0; i < priv->child_count; i++) {
+            if (!wind_on[i])
+                continue;
+            if ((!local->replies[i].valid) || (local->replies[i].op_ret != 0)) {
+                continue;
+            }
+
+            if ((current_event_gen[i] == info->child_up_event_gen[i]) &&
+                (current_event_gen[i] > info->child_down_event_gen[i])) {
+                info->locked_nodes[i] = 1;
+                retry = _gf_false;
+                list_del_init(&info->pos);
+                list_add_tail(&info->pos, &priv->saved_locks);
+            } else {
+                /*We received subsequent child up/down events while heal was in
+                 * progress; don't mark child as healed. Attempt again on the
+                 * new child up*/
+                gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_LK_HEAL_DOM,
+                       "Event gen mismatch: skipped healing lock on child %d "
+                       "for %s.",
+                       i, uuid_utoa(info->fd->inode->gfid));
+            }
+        }
+    }
+    UNLOCK(&priv->lock);
+
+release:
+    afr_dom_lock_release(frame);
+    if (retry)
+        afr_add_lock_to_lkhealq(priv, info);
+    return;
 }
 
-void
-afr_inode_ctx_set_read_child (afr_inode_ctx_t *ctx, int32_t read_child)
+static int
+afr_lock_heal_done(int ret, call_frame_t *frame, void *opaque)
 {
-        uint64_t        remaining_mask = 0;
-        uint64_t        mask         = 0;
+    STACK_DESTROY(frame->root);
+    return 0;
+}
 
-        remaining_mask = (~AFR_ICTX_READ_CHILD_MASK & ctx->masks);
-        mask = (AFR_ICTX_READ_CHILD_MASK & read_child);
-        ctx->masks = remaining_mask | mask;
+static int
+afr_lock_heal(void *opaque)
+{
+    call_frame_t *frame = (call_frame_t *)opaque;
+    call_frame_t *iter_frame = NULL;
+    xlator_t *this = frame->this;
+    afr_private_t *priv = this->private;
+    afr_lk_heal_info_t *info = NULL;
+    afr_lk_heal_info_t *tmp = NULL;
+    struct list_head healq = {
+        0,
+    };
+    int ret = 0;
+
+    iter_frame = afr_copy_frame(frame);
+    if (!iter_frame) {
+        return ENOMEM;
+    }
+
+    INIT_LIST_HEAD(&healq);
+    LOCK(&priv->lock);
+    {
+        list_splice_init(&priv->lk_healq, &healq);
+    }
+    UNLOCK(&priv->lock);
+
+    list_for_each_entry_safe(info, tmp, &healq, pos)
+    {
+        GF_ASSERT((AFR_COUNT(info->locked_nodes, priv->child_count) <
+                   priv->child_count));
+        ((afr_local_t *)(iter_frame->local))->fd = fd_ref(info->fd);
+        afr_lock_heal_do(iter_frame, priv, info);
+        AFR_STACK_RESET(iter_frame);
+        if (iter_frame->local == NULL) {
+            ret = ENOTCONN;
+            gf_msg(frame->this->name, GF_LOG_ERROR, ENOTCONN,
+                   AFR_MSG_LK_HEAL_DOM,
+                   "Aborting processing of lk_healq."
+                   "Healing will be reattempted on next child up for locks "
+                   "that are still in quorum.");
+            LOCK(&priv->lock);
+            {
+                list_add_tail(&healq, &priv->lk_healq);
+            }
+            UNLOCK(&priv->lock);
+            break;
+        }
+    }
+
+    AFR_STACK_DESTROY(iter_frame);
+    return ret;
 }
 
-void
-afr_inode_ctx_set_read_ctx (afr_inode_ctx_t *ctx, int32_t read_child,
-                            int32_t *fresh_children, int32_t child_count)
+static int
+__afr_lock_heal_synctask(xlator_t *this, afr_private_t *priv, int child)
 {
-        int             i            = 0;
+    int ret = 0;
+    call_frame_t *frame = NULL;
+    afr_lk_heal_info_t *info = NULL;
+    afr_lk_heal_info_t *tmp = NULL;
 
-        afr_inode_ctx_set_read_child (ctx, read_child);
-        for (i = 0; i < child_count; i++) {
-                if (fresh_children)
-                        ctx->fresh_children[i] = fresh_children[i];
-                else
-                        ctx->fresh_children[i] = -1;
-        }
-}
+    if (priv->shd.iamshd)
+        return 0;
 
-void
-afr_inode_ctx_rm_stale_children (afr_inode_ctx_t *ctx, int32_t *stale_children,
-                                 int32_t child_count)
-{
-        int             i            = 0;
-        int32_t         read_child   = -1;
+    list_for_each_entry_safe(info, tmp, &priv->saved_locks, pos)
+    {
+        info->child_up_event_gen[child] = priv->event_generation;
+        list_del_init(&info->pos);
+        list_add_tail(&info->pos, &priv->lk_healq);
+    }
 
-        GF_ASSERT (stale_children);
-        for (i = 0; i < child_count; i++) {
-                if (stale_children[i] == -1)
-                        break;
-                afr_children_rm_child (ctx->fresh_children,
-                                       stale_children[i], child_count);
-        }
-        read_child = (int32_t)(ctx->masks & AFR_ICTX_READ_CHILD_MASK);
-        if (!afr_is_child_present (ctx->fresh_children, child_count,
-                                   read_child))
-                afr_inode_ctx_set_read_child (ctx, ctx->fresh_children[0]);
+    frame = create_frame(this, this->ctx->pool);
+    if (!frame)
+        return -1;
+
+    ret = synctask_new(this->ctx->env, afr_lock_heal, afr_lock_heal_done, frame,
+                       frame);
+    if (ret)
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_LK_HEAL_DOM,
+               "Failed to launch lock heal synctask");
+
+    return ret;
 }
 
-void
-afr_inode_ctx_set_opendir_done (afr_inode_ctx_t *ctx)
+static int
+__afr_mark_pending_lk_heal(xlator_t *this, afr_private_t *priv, int child)
 {
-        uint64_t        remaining_mask = 0;
-        uint64_t        mask = 0;
+    afr_lk_heal_info_t *info = NULL;
+    afr_lk_heal_info_t *tmp = NULL;
+
+    if (priv->shd.iamshd)
+        return 0;
+    list_for_each_entry_safe(info, tmp, &priv->saved_locks, pos)
+    {
+        info->child_down_event_gen[child] = priv->event_generation;
+        if (info->locked_nodes[child] == 1)
+            info->locked_nodes[child] = 0;
+        if (!afr_has_quorum(info->locked_nodes, this, NULL)) {
+            /* Since the lock was lost on quorum no. of nodes, we should
+             * not attempt to heal it anymore. Some other client could have
+             * acquired the lock, modified data and released it and this
+             * client wouldn't know about it if we heal it.*/
+            afr_mark_fd_bad(info->fd, this);
+            list_del(&info->pos);
+            afr_lk_heal_info_cleanup(info);
+            /* We're not winding an unlock on the node where the lock is still
+             * present because when fencing logic switches over to the new
+             * client (since we marked the fd bad), it should preempt any
+             * existing lock. */
+        }
+    }
+    return 0;
+}
 
-        remaining_mask = (~AFR_ICTX_OPENDIR_DONE_MASK & ctx->masks);
-        mask = (0xFFFFFFFFFFFFFFFFULL & AFR_ICTX_OPENDIR_DONE_MASK);
-        ctx->masks = remaining_mask | mask;
+gf_boolean_t
+afr_is_consistent_io_possible(afr_local_t *local, afr_private_t *priv,
+                              int32_t *op_errno)
+{
+    if (priv->consistent_io && local->call_count != priv->child_count) {
+        gf_msg(THIS->name, GF_LOG_INFO, 0, AFR_MSG_SUBVOLS_DOWN,
+               "All subvolumes are not up");
+        if (op_errno)
+            *op_errno = ENOTCONN;
+        return _gf_false;
+    }
+    return _gf_true;
 }
 
-void
-afr_inode_set_ctx_params (xlator_t *this, inode_t *inode,
-                          afr_inode_params_t *params)
+gf_boolean_t
+afr_is_lock_mode_mandatory(dict_t *xdata)
 {
-        GF_ASSERT (inode);
-        GF_ASSERT (params);
+    int ret = 0;
+    uint32_t lk_mode = GF_LK_ADVISORY;
 
-        afr_inode_ctx_t *ctx            = NULL;
-        afr_private_t   *priv           = NULL;
-        int32_t         read_child      = -1;
-        int32_t         *fresh_children = NULL;
-        int32_t         *stale_children = NULL;
+    ret = dict_get_uint32(xdata, GF_LOCK_MODE, &lk_mode);
+    if (!ret && lk_mode == GF_LK_MANDATORY)
+        return _gf_true;
 
-        priv = this->private;
-        LOCK (&inode->lock);
-        {
-                ctx = __afr_inode_ctx_get (inode, this);
-                if (!ctx)
-                        goto unlock;
-                switch (params->op) {
-                case AFR_INODE_SET_READ_CTX:
-                        read_child = params->u.read_ctx.read_child;
-                        fresh_children = params->u.read_ctx.children;
-                        afr_inode_ctx_set_read_ctx (ctx, read_child,
-                                                    fresh_children,
-                                                    priv->child_count);
-                        break;
-                case AFR_INODE_RM_STALE_CHILDREN:
-                        stale_children = params->u.read_ctx.children;
-                        afr_inode_ctx_rm_stale_children (ctx,
-                                                         stale_children,
-                                                         priv->child_count);
-                        break;
-                case AFR_INODE_SET_OPENDIR_DONE:
-                        afr_inode_ctx_set_opendir_done (ctx);
-                        break;
-                default:
-                        GF_ASSERT (0);
-                        break;
-                }
-        }
-unlock:
-        UNLOCK (&inode->lock);
+    return _gf_false;
 }
 
-void
-afr_set_split_brain (xlator_t *this, inode_t *inode, afr_spb_state_t mdata_spb,
-                     afr_spb_state_t data_spb)
+call_frame_t *
+afr_copy_frame(call_frame_t *base)
 {
-        afr_inode_ctx_t *ctx = NULL;
+    afr_local_t *local = NULL;
+    call_frame_t *frame = NULL;
+    int op_errno = 0;
 
-        ctx = afr_inode_ctx_get (inode, this);
-        if (mdata_spb != DONT_KNOW)
-                ctx->mdata_spb = mdata_spb;
-        if (data_spb != DONT_KNOW)
-                ctx->data_spb = data_spb;
+    frame = copy_frame(base);
+    if (!frame)
+        return NULL;
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local) {
+        AFR_STACK_DESTROY(frame);
+        return NULL;
+    }
+
+    return frame;
 }
 
-void
-afr_set_opendir_done (xlator_t *this, inode_t *inode)
-{
-        afr_inode_params_t params = {0};
+/* Check if an entry or inode could be undergoing a transaction. */
+gf_boolean_t
+afr_is_possibly_under_txn(afr_transaction_type type, afr_local_t *local,
+                          xlator_t *this)
+{
+    int i = 0;
+    int tmp = 0;
+    afr_private_t *priv = NULL;
+    GF_UNUSED char *key = NULL;
+    int keylen = 0;
+
+    priv = this->private;
+
+    if (type == AFR_ENTRY_TRANSACTION) {
+        key = GLUSTERFS_PARENT_ENTRYLK;
+        keylen = SLEN(GLUSTERFS_PARENT_ENTRYLK);
+    } else if (type == AFR_DATA_TRANSACTION) {
+        /*FIXME: Use GLUSTERFS_INODELK_DOM_COUNT etc. once
+         * pl_inodelk_xattr_fill supports separate keys for different
+         * domains.*/
+        key = GLUSTERFS_INODELK_COUNT;
+        keylen = SLEN(GLUSTERFS_INODELK_COUNT);
+    }
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].xdata)
+            continue;
+        if (dict_get_int32n(local->replies[i].xdata, key, keylen, &tmp) == 0)
+            if (tmp)
+                return _gf_true;
+    }
 
-        params.op = AFR_INODE_SET_OPENDIR_DONE;
-        afr_inode_set_ctx_params (this, inode, &params);
+    return _gf_false;
 }
 
-void
-afr_inode_set_read_ctx (xlator_t *this, inode_t *inode, int32_t read_child,
-                        int32_t *fresh_children)
+static void
+afr_inode_ctx_destroy(afr_inode_ctx_t *ctx)
 {
-        afr_inode_params_t params = {0};
-        afr_private_t      *priv  = NULL;
+    int i = 0;
 
-        priv = this->private;
-        GF_ASSERT (read_child >= 0);
-        GF_ASSERT (fresh_children);
-        GF_ASSERT (afr_is_child_present (fresh_children, priv->child_count,
-                                         read_child));
+    if (!ctx)
+        return;
+
+    for (i = 0; i < AFR_NUM_CHANGE_LOGS; i++) {
+        GF_FREE(ctx->pre_op_done[i]);
+    }
 
-        params.op = AFR_INODE_SET_READ_CTX;
-        params.u.read_ctx.read_child     = read_child;
-        params.u.read_ctx.children = fresh_children;
-        afr_inode_set_ctx_params (this, inode, &params);
+    GF_FREE(ctx);
 }
 
-void
-afr_inode_rm_stale_children (xlator_t *this, inode_t *inode,
-                             int32_t *stale_children)
-{
-        afr_inode_params_t params = {0};
+int
+__afr_inode_ctx_get(xlator_t *this, inode_t *inode, afr_inode_ctx_t **ctx)
+{
+    uint64_t ctx_int = 0;
+    int ret = -1;
+    int i = -1;
+    int num_locks = -1;
+    afr_inode_ctx_t *ictx = NULL;
+    afr_lock_t *lock = NULL;
+    afr_private_t *priv = this->private;
+
+    ret = __inode_ctx_get(inode, this, &ctx_int);
+    if (ret == 0) {
+        *ctx = (afr_inode_ctx_t *)(uintptr_t)ctx_int;
+        return 0;
+    }
+
+    ictx = GF_CALLOC(1, sizeof(afr_inode_ctx_t), gf_afr_mt_inode_ctx_t);
+    if (!ictx)
+        goto out;
+
+    for (i = 0; i < AFR_NUM_CHANGE_LOGS; i++) {
+        ictx->pre_op_done[i] = GF_CALLOC(sizeof *ictx->pre_op_done[i],
+                                         priv->child_count, gf_afr_mt_int32_t);
+        if (!ictx->pre_op_done[i]) {
+            ret = -ENOMEM;
+            goto out;
+        }
+    }
+
+    num_locks = sizeof(ictx->lock) / sizeof(afr_lock_t);
+    for (i = 0; i < num_locks; i++) {
+        lock = &ictx->lock[i];
+        INIT_LIST_HEAD(&lock->post_op);
+        INIT_LIST_HEAD(&lock->frozen);
+        INIT_LIST_HEAD(&lock->waiting);
+        INIT_LIST_HEAD(&lock->owners);
+    }
+
+    ctx_int = (uint64_t)(uintptr_t)ictx;
+    ret = __inode_ctx_set(inode, this, &ctx_int);
+    if (ret) {
+        goto out;
+    }
+
+    ictx->spb_choice = -1;
+    ictx->read_subvol = 0;
+    ictx->write_subvol = 0;
+    ictx->lock_count = 0;
+    ret = 0;
+    *ctx = ictx;
+out:
+    if (ret) {
+        afr_inode_ctx_destroy(ictx);
+    }
+    return ret;
+}
+
+/*
+ * INODE CTX 64-bit VALUE FORMAT FOR SMALL (<= 16) SUBVOL COUNTS:
+ *
+ * |<----------   64bit   ------------>|
+ *  63           32 31    16 15       0
+ * |   EVENT_GEN   |  DATA  | METADATA |
+ *
+ *
+ *  METADATA (bit-0 .. bit-15): bitmap representing subvolumes from which
+ *                              metadata can be attempted to be read.
+ *
+ *                              bit-0 => priv->subvolumes[0]
+ *                              bit-1 => priv->subvolumes[1]
+ *                              ... etc. till bit-15
+ *
+ *  DATA (bit-16 .. bit-31): bitmap representing subvolumes from which data
+ *                           can be attempted to be read.
+ *
+ *                           bit-16 => priv->subvolumes[0]
+ *                           bit-17 => priv->subvolumes[1]
+ *                           ... etc. till bit-31
+ *
+ *  EVENT_GEN (bit-32 .. bit-63): event generation (i.e priv->event_generation)
+ *                                when DATA and METADATA was last updated.
+ *
+ *                                If EVENT_GEN is < priv->event_generation,
+ *                                or is 0, it means afr_inode_refresh() needs
+ *                                to be called to recalculate the bitmaps.
+ */
 
-        GF_ASSERT (stale_children);
+int
+__afr_set_in_flight_sb_status(xlator_t *this, afr_local_t *local,
+                              inode_t *inode)
+{
+    int i = 0;
+    int txn_type = 0;
+    int count = 0;
+    int index = -1;
+    uint16_t datamap_old = 0;
+    uint16_t metadatamap_old = 0;
+    uint16_t datamap = 0;
+    uint16_t metadatamap = 0;
+    uint16_t tmp_map = 0;
+    uint16_t mask = 0;
+    uint32_t event = 0;
+    uint64_t val = 0;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+    txn_type = local->transaction.type;
+
+    if (txn_type == AFR_DATA_TRANSACTION)
+        val = local->inode_ctx->write_subvol;
+    else
+        val = local->inode_ctx->read_subvol;
+
+    metadatamap_old = metadatamap = (val & 0x000000000000ffff);
+    datamap_old = datamap = (val & 0x00000000ffff0000) >> 16;
+    event = (val & 0xffffffff00000000) >> 32;
+
+    if (txn_type == AFR_DATA_TRANSACTION)
+        tmp_map = datamap;
+    else if (txn_type == AFR_METADATA_TRANSACTION)
+        tmp_map = metadatamap;
+
+    count = gf_bits_count(tmp_map);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->transaction.failed_subvols[i])
+            continue;
+
+        mask = 1 << i;
+        if (txn_type == AFR_METADATA_TRANSACTION)
+            metadatamap &= ~mask;
+        else if (txn_type == AFR_DATA_TRANSACTION)
+            datamap &= ~mask;
+    }
+
+    switch (txn_type) {
+        case AFR_METADATA_TRANSACTION:
+            if ((metadatamap_old != 0) && (metadatamap == 0) && (count == 1)) {
+                index = gf_bits_index(tmp_map);
+                local->transaction.in_flight_sb_errno = local->replies[index]
+                                                            .op_errno;
+                local->transaction.in_flight_sb = _gf_true;
+                metadatamap |= (1 << index);
+            }
+            if (metadatamap_old != metadatamap) {
+                __afr_inode_need_refresh_set(inode, this);
+            }
+            break;
+
+        case AFR_DATA_TRANSACTION:
+            if ((datamap_old != 0) && (datamap == 0) && (count == 1)) {
+                index = gf_bits_index(tmp_map);
+                local->transaction.in_flight_sb_errno = local->replies[index]
+                                                            .op_errno;
+                local->transaction.in_flight_sb = _gf_true;
+                datamap |= (1 << index);
+            }
+            if (datamap_old != datamap)
+                __afr_inode_need_refresh_set(inode, this);
+            break;
+
+        default:
+            break;
+    }
 
-        params.op = AFR_INODE_RM_STALE_CHILDREN;
-        params.u.read_ctx.children = stale_children;
-        afr_inode_set_ctx_params (this, inode, &params);
+    val = ((uint64_t)metadatamap) | (((uint64_t)datamap) << 16) |
+          (((uint64_t)event) << 32);
+
+    if (txn_type == AFR_DATA_TRANSACTION)
+        local->inode_ctx->write_subvol = val;
+    local->inode_ctx->read_subvol = val;
+
+    return 0;
 }
 
 gf_boolean_t
-afr_is_source_child (int32_t *sources, int32_t child_count, int32_t child)
+afr_is_symmetric_error(call_frame_t *frame, xlator_t *this)
 {
-        gf_boolean_t             source_xattrs = _gf_false;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int op_errno = 0;
+    int i_errno = 0;
+    gf_boolean_t matching_errors = _gf_true;
+    int i = 0;
+
+    priv = this->private;
+    local = frame->local;
 
-        GF_ASSERT (child < child_count);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
+        if (local->replies[i].op_ret != -1) {
+            /* Operation succeeded on at least one subvol,
+               so it is not a failed-everywhere situation.
+            */
+            matching_errors = _gf_false;
+            break;
+        }
+        i_errno = local->replies[i].op_errno;
+
+        if (i_errno == ENOTCONN) {
+            /* ENOTCONN is not a symmetric error. We do not
+               know if the operation was performed on the
+               backend or not.
+            */
+            matching_errors = _gf_false;
+            break;
+        }
 
-        if ((child >= 0) && (child < child_count) &&
-             sources[child]) {
-                source_xattrs = _gf_true;
+        if (!op_errno) {
+            op_errno = i_errno;
+        } else if (op_errno != i_errno) {
+            /* Mismatching op_errno's */
+            matching_errors = _gf_false;
+            break;
         }
-        return source_xattrs;
+    }
+
+    return matching_errors;
 }
 
-gf_boolean_t
-afr_is_child_present (int32_t *success_children, int32_t child_count,
-                      int32_t child)
+int
+afr_set_in_flight_sb_status(xlator_t *this, call_frame_t *frame, inode_t *inode)
 {
-        gf_boolean_t             success_child = _gf_false;
-        int                      i = 0;
+    int ret = -1;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
 
-        GF_ASSERT (child < child_count);
+    priv = this->private;
+    local = frame->local;
 
-        for (i = 0; i < child_count; i++) {
-                if (success_children[i] == -1)
-                        break;
-                if (child == success_children[i]) {
-                        success_child = _gf_true;
-                        break;
-                }
-        }
-        return success_child;
+    /* If this transaction saw no failures, then exit. */
+    if (AFR_COUNT(local->transaction.failed_subvols, priv->child_count) == 0)
+        return 0;
+
+    if (afr_is_symmetric_error(frame, this))
+        return 0;
+
+    LOCK(&inode->lock);
+    {
+        ret = __afr_set_in_flight_sb_status(this, local, inode);
+    }
+    UNLOCK(&inode->lock);
+
+    return ret;
 }
 
-gf_boolean_t
-afr_is_read_child (int32_t *success_children, int32_t *sources,
-                   int32_t child_count, int32_t child)
+int
+__afr_inode_read_subvol_get_small(inode_t *inode, xlator_t *this,
+                                  unsigned char *data, unsigned char *metadata,
+                                  int *event_p)
+{
+    afr_private_t *priv = NULL;
+    int ret = -1;
+    uint16_t datamap = 0;
+    uint16_t metadatamap = 0;
+    uint32_t event = 0;
+    uint64_t val = 0;
+    int i = 0;
+    afr_inode_ctx_t *ctx = NULL;
+
+    priv = this->private;
+
+    ret = __afr_inode_ctx_get(this, inode, &ctx);
+    if (ret < 0)
+        return ret;
+
+    val = ctx->read_subvol;
+
+    metadatamap = (val & 0x000000000000ffff);
+    datamap = (val & 0x00000000ffff0000) >> 16;
+    event = (val & 0xffffffff00000000) >> 32;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (metadata)
+            metadata[i] = (metadatamap >> i) & 1;
+        if (data)
+            data[i] = (datamap >> i) & 1;
+    }
+
+    if (event_p)
+        *event_p = event;
+    return ret;
+}
+
+int
+__afr_inode_read_subvol_set_small(inode_t *inode, xlator_t *this,
+                                  unsigned char *data, unsigned char *metadata,
+                                  int event)
 {
-        gf_boolean_t             success_child = _gf_false;
-        gf_boolean_t             source        = _gf_false;
+    afr_private_t *priv = NULL;
+    uint16_t datamap = 0;
+    uint16_t metadatamap = 0;
+    uint64_t val = 0;
+    int i = 0;
+    int ret = -1;
+    afr_inode_ctx_t *ctx = NULL;
 
-        if (child < 0) {
-                return _gf_false;
-        }
+    priv = this->private;
 
-        GF_ASSERT (success_children);
-        GF_ASSERT (child_count > 0);
+    ret = __afr_inode_ctx_get(this, inode, &ctx);
+    if (ret)
+        goto out;
 
-        success_child = afr_is_child_present (success_children, child_count,
-                                              child);
-        if (!success_child)
-                goto out;
-        if (NULL == sources) {
-                source = _gf_true;
-                goto out;
-        }
-        source = afr_is_source_child (sources, child_count, child);
+    for (i = 0; i < priv->child_count; i++) {
+        if (data[i])
+            datamap |= (1 << i);
+        if (metadata[i])
+            metadatamap |= (1 << i);
+    }
+
+    val = ((uint64_t)metadatamap) | (((uint64_t)datamap) << 16) |
+          (((uint64_t)event) << 32);
+
+    ctx->read_subvol = val;
+
+    ret = 0;
 out:
-        return (success_child && source);
+    return ret;
 }
 
-int32_t
-afr_hash_child (int32_t *success_children, int32_t child_count,
-                unsigned int hmode, uuid_t gfid)
+int
+__afr_inode_read_subvol_get(inode_t *inode, xlator_t *this, unsigned char *data,
+                            unsigned char *metadata, int *event_p)
 {
-        uuid_t  gfid_copy = {0,};
-        pid_t pid;
+    afr_private_t *priv = NULL;
+    int ret = -1;
 
-        if (!hmode) {
-                return -1;
-        }
+    priv = this->private;
 
-        if (gfid) {
-               uuid_copy(gfid_copy,gfid);
-        }
-        if (hmode > 1) {
-                /*
-                 * Why getpid?  Because it's one of the cheapest calls
-                 * available - faster than gethostname etc. - and returns a
-                 * constant-length value that's sure to be shorter than a UUID.
-                 * It's still very unlikely to be the same across clients, so
-                 * it still provides good mixing.  We're not trying for
-                 * perfection here.  All we need is a low probability that
-                 * multiple clients won't converge on the same subvolume.
-                 */
-                pid = getpid();
-                memcpy (gfid_copy, &pid, sizeof(pid));
-        }
+    if (priv->child_count <= 16)
+        ret = __afr_inode_read_subvol_get_small(inode, this, data, metadata,
+                                                event_p);
+    else
+        /* TBD: allocate structure with array and read from it */
+        ret = -1;
 
-        return SuperFastHash((char *)gfid_copy,
-                             sizeof(gfid_copy)) % child_count;
+    return ret;
 }
 
-/* If sources is NULL the xattrs are assumed to be of source for all
- * success_children.
- */
 int
-afr_select_read_child_from_policy (int32_t *success_children,
-                                   int32_t child_count, int32_t prev_read_child,
-                                   int32_t config_read_child, int32_t *sources,
-                                   unsigned int hmode, uuid_t gfid)
+__afr_inode_split_brain_choice_get(inode_t *inode, xlator_t *this,
+                                   int *spb_choice)
 {
-        int32_t                  read_child   = -1;
-        int                      i            = 0;
+    afr_inode_ctx_t *ctx = NULL;
+    int ret = -1;
 
-        GF_ASSERT (success_children);
+    ret = __afr_inode_ctx_get(this, inode, &ctx);
+    if (ret < 0)
+        return ret;
 
-        read_child = config_read_child;
-        if (afr_is_read_child (success_children, sources, child_count,
-                               read_child))
-                goto out;
+    *spb_choice = ctx->spb_choice;
+    return 0;
+}
 
-        read_child = prev_read_child;
-        if (afr_is_read_child (success_children, sources, child_count,
-                               read_child))
-                goto out;
+int
+__afr_inode_read_subvol_set(inode_t *inode, xlator_t *this, unsigned char *data,
+                            unsigned char *metadata, int event)
+{
+    afr_private_t *priv = NULL;
+    int ret = -1;
 
-        read_child = afr_hash_child (success_children, child_count,
-                                     hmode, gfid);
-        if (afr_is_read_child (success_children, sources, child_count,
-                               read_child)) {
-                goto out;
-        }
+    priv = this->private;
 
-        for (i = 0; i < child_count; i++) {
-                read_child = success_children[i];
-                if (read_child < 0)
-                        break;
-                if (afr_is_read_child (success_children, sources, child_count,
-                                       read_child))
-                        goto out;
-        }
-        read_child = -1;
+    if (priv->child_count <= 16)
+        ret = __afr_inode_read_subvol_set_small(inode, this, data, metadata,
+                                                event);
+    else
+        ret = -1;
 
-out:
-        return read_child;
+    return ret;
 }
 
-/* This function should be used when all the success_children are sources
- */
-void
-afr_set_read_ctx_from_policy (xlator_t *this, inode_t *inode,
-                              int32_t *fresh_children, int32_t prev_read_child,
-                              int32_t config_read_child, uuid_t gfid)
-{
-        int                      read_child = -1;
-        afr_private_t            *priv = NULL;
-
-        priv = this->private;
-        read_child = afr_select_read_child_from_policy (fresh_children,
-                                                        priv->child_count,
-                                                        prev_read_child,
-                                                        config_read_child,
-                                                        NULL,
-                                                        priv->hash_mode, gfid);
-        if (read_child >= 0)
-                afr_inode_set_read_ctx (this, inode, read_child,
-                                        fresh_children);
-}
-
-/* afr_next_call_child ()
- * This is a common function used by all the read-type fops
- * This function should not be called with the inode's read_children array.
- * The fop's handler should make a copy of the inode's read_children,
- * preferred read_child into the local vars, because while this function is
- * in execution there is a chance for inode's read_ctx to change.
- */
-int32_t
-afr_next_call_child (int32_t *fresh_children, unsigned char *child_up,
-                     size_t child_count, int32_t *last_index,
-                     int32_t read_child)
+int
+__afr_inode_split_brain_choice_set(inode_t *inode, xlator_t *this,
+                                   int spb_choice)
 {
-        int             next_index      = 0;
-        int32_t         next_call_child = -1;
+    afr_inode_ctx_t *ctx = NULL;
+    int ret = -1;
 
-        GF_ASSERT (last_index);
+    ret = __afr_inode_ctx_get(this, inode, &ctx);
+    if (ret)
+        goto out;
 
-        next_index = *last_index;
-retry:
-        next_index++;
-        if ((next_index >= child_count) ||
-           (fresh_children[next_index] == -1))
-                goto out;
-        if ((fresh_children[next_index] == read_child) ||
-           (!child_up[fresh_children[next_index]]))
-                goto retry;
-        *last_index = next_index;
-        next_call_child = fresh_children[next_index];
+    ctx->spb_choice = spb_choice;
+
+    ret = 0;
 out:
-        return next_call_child;
+    return ret;
 }
 
- /* This function should not be called with the inode's read_children array.
- * The fop's handler should make a copy of the inode's read_children,
- * preferred read_child into the local vars, because while this function is
- * in execution there is a chance for inode's read_ctx to change.
- */
-int32_t
-afr_get_call_child (xlator_t *this, unsigned char *child_up, int32_t read_child,
-                    int32_t *fresh_children,
-                    int32_t *call_child, int32_t *last_index)
+int
+afr_inode_read_subvol_get(inode_t *inode, xlator_t *this, unsigned char *data,
+                          unsigned char *metadata, int *event_p)
 {
-        int             ret   = 0;
-        afr_private_t   *priv = NULL;
-        int             i     = 0;
+    int ret = -1;
 
-        GF_ASSERT (child_up);
-        GF_ASSERT (call_child);
-        GF_ASSERT (last_index);
-        GF_ASSERT (fresh_children);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
 
-        if (read_child < 0) {
-                ret = -EIO;
-                goto out;
-        }
-        priv = this->private;
-        *call_child = -1;
-        *last_index = -1;
+    LOCK(&inode->lock);
+    {
+        ret = __afr_inode_read_subvol_get(inode, this, data, metadata, event_p);
+    }
+    UNLOCK(&inode->lock);
+out:
+    return ret;
+}
 
-        if (child_up[read_child]) {
-                *call_child = read_child;
-        } else {
-                for (i = 0; i < priv->child_count; i++) {
-                        if (fresh_children[i] == -1)
-                                break;
-                        if (child_up[fresh_children[i]]) {
-                                *call_child = fresh_children[i];
-                                ret = 0;
-                                break;
-                        }
-                }
+int
+afr_inode_get_readable(call_frame_t *frame, inode_t *inode, xlator_t *this,
+                       unsigned char *readable, int *event_p, int type)
+{
+    afr_private_t *priv = this->private;
+    afr_local_t *local = frame->local;
+    unsigned char *data = alloca0(priv->child_count);
+    unsigned char *metadata = alloca0(priv->child_count);
+    int data_count = 0;
+    int metadata_count = 0;
+    int event_generation = 0;
+    int ret = 0;
+
+    ret = afr_inode_read_subvol_get(inode, this, data, metadata,
+                                    &event_generation);
+    if (ret == -1)
+        return -EIO;
+
+    data_count = AFR_COUNT(data, priv->child_count);
+    metadata_count = AFR_COUNT(metadata, priv->child_count);
+
+    if (inode->ia_type == IA_IFDIR) {
+        /* For directories, allow even if it is in data split-brain. */
+        if (type == AFR_METADATA_TRANSACTION || local->op == GF_FOP_STAT ||
+            local->op == GF_FOP_FSTAT) {
+            if (!metadata_count)
+                return -EIO;
+        }
+    } else {
+        /* For files, abort in case of data/metadata split-brain. */
+        if (!data_count || !metadata_count) {
+            return -EIO;
+        }
+    }
+
+    if (type == AFR_METADATA_TRANSACTION && readable)
+        memcpy(readable, metadata, priv->child_count * sizeof *metadata);
+    if (type == AFR_DATA_TRANSACTION && readable) {
+        if (!data_count)
+            memcpy(readable, local->child_up,
+                   priv->child_count * sizeof *readable);
+        else
+            memcpy(readable, data, priv->child_count * sizeof *data);
+    }
+    if (event_p)
+        *event_p = event_generation;
+    return 0;
+}
 
-                if (*call_child == -1) {
-                        ret = -ENOTCONN;
-                        goto out;
-                }
+static int
+afr_inode_split_brain_choice_get(inode_t *inode, xlator_t *this,
+                                 int *spb_choice)
+{
+    int ret = -1;
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
 
-                *last_index = i;
-        }
+    LOCK(&inode->lock);
+    {
+        ret = __afr_inode_split_brain_choice_get(inode, this, spb_choice);
+    }
+    UNLOCK(&inode->lock);
 out:
-        gf_log (this->name, GF_LOG_DEBUG, "Returning %d, call_child: %d, "
-                "last_index: %d", ret, *call_child, *last_index);
-        return ret;
+    return ret;
 }
 
-void
-afr_reset_xattr (dict_t **xattr, unsigned int child_count)
+/*
+ * frame is used to get the favourite policy. Since
+ * afr_inode_split_brain_choice_get was called with afr_open, it is possible to
+ * have a frame with out local->replies. So in that case, frame is passed as
+ * null, hence this function will handle the frame NULL case.
+ */
+int
+afr_split_brain_read_subvol_get(inode_t *inode, xlator_t *this,
+                                call_frame_t *frame, int *spb_subvol)
 {
-        unsigned int i = 0;
+    int ret = -1;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        if (!xattr)
-                goto out;
-        for (i = 0; i < child_count; i++) {
-                if (xattr[i]) {
-                        dict_unref (xattr[i]);
-                        xattr[i] = NULL;
-                }
+    GF_VALIDATE_OR_GOTO("afr", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+    GF_VALIDATE_OR_GOTO(this->name, spb_subvol, out);
+
+    priv = this->private;
+
+    ret = afr_inode_split_brain_choice_get(inode, this, spb_subvol);
+    if (*spb_subvol < 0 && priv->fav_child_policy && frame && frame->local) {
+        local = frame->local;
+        *spb_subvol = afr_sh_get_fav_by_policy(this, local->replies, inode,
+                                               NULL);
+        if (*spb_subvol >= 0) {
+            ret = 0;
         }
+    }
+
 out:
-        return;
+    return ret;
 }
+int
+afr_inode_read_subvol_set(inode_t *inode, xlator_t *this, unsigned char *data,
+                          unsigned char *metadata, int event)
+{
+    int ret = -1;
 
-void
-afr_xattr_array_destroy (dict_t **xattr, unsigned int child_count)
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+
+    LOCK(&inode->lock);
+    {
+        ret = __afr_inode_read_subvol_set(inode, this, data, metadata, event);
+    }
+    UNLOCK(&inode->lock);
+out:
+    return ret;
+}
+
+int
+afr_inode_split_brain_choice_set(inode_t *inode, xlator_t *this, int spb_choice)
 {
-        afr_reset_xattr (xattr, child_count);
-        GF_FREE (xattr);
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+
+    LOCK(&inode->lock);
+    {
+        ret = __afr_inode_split_brain_choice_set(inode, this, spb_choice);
+    }
+    UNLOCK(&inode->lock);
+out:
+    return ret;
 }
 
-void
-afr_local_sh_cleanup (afr_local_t *local, xlator_t *this)
+/* The caller of this should perform afr_inode_refresh, if this function
+ * returns _gf_true
+ */
+gf_boolean_t
+afr_is_inode_refresh_reqd(inode_t *inode, xlator_t *this, int event_gen1,
+                          int event_gen2)
 {
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
+    gf_boolean_t need_refresh = _gf_false;
+    afr_inode_ctx_t *ctx = NULL;
+    int ret = -1;
 
-        sh = &local->self_heal;
-        priv = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
 
-        if (sh->data_sh_info && strcmp (sh->data_sh_info, ""))
-                GF_FREE (sh->data_sh_info);
+    LOCK(&inode->lock);
+    {
+        ret = __afr_inode_ctx_get(this, inode, &ctx);
+        if (ret)
+            goto unlock;
 
-        if (sh->metadata_sh_info && strcmp (sh->metadata_sh_info, ""))
-                GF_FREE (sh->metadata_sh_info);
+        need_refresh = ctx->need_refresh;
+        /* Hoping that the caller will do inode_refresh followed by
+         * this, hence setting the need_refresh to false */
+        ctx->need_refresh = _gf_false;
+    }
+unlock:
+    UNLOCK(&inode->lock);
 
-        GF_FREE (sh->buf);
+    if (event_gen1 != event_gen2)
+        need_refresh = _gf_true;
+out:
+    return need_refresh;
+}
 
-        GF_FREE (sh->parentbufs);
+int
+__afr_inode_need_refresh_set(inode_t *inode, xlator_t *this)
+{
+    int ret = -1;
+    afr_inode_ctx_t *ctx = NULL;
 
-        if (sh->inode)
-                inode_unref (sh->inode);
+    ret = __afr_inode_ctx_get(this, inode, &ctx);
+    if (ret == 0) {
+        ctx->need_refresh = _gf_true;
+    }
 
-        afr_xattr_array_destroy (sh->xattr, priv->child_count);
+    return ret;
+}
 
-        GF_FREE (sh->child_errno);
+int
+afr_inode_need_refresh_set(inode_t *inode, xlator_t *this)
+{
+    int ret = -1;
 
-        afr_matrix_cleanup (sh->pending_matrix, priv->child_count);
-        afr_matrix_cleanup (sh->delta_matrix, priv->child_count);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
 
-        GF_FREE (sh->sources);
+    LOCK(&inode->lock);
+    {
+        ret = __afr_inode_need_refresh_set(inode, this);
+    }
+    UNLOCK(&inode->lock);
+out:
+    return ret;
+}
 
-        GF_FREE (sh->success);
+int
+afr_spb_choice_timeout_cancel(xlator_t *this, inode_t *inode)
+{
+    afr_inode_ctx_t *ctx = NULL;
+    int ret = -1;
 
-        GF_FREE (sh->locked_nodes);
+    if (!inode)
+        return ret;
 
-        if (sh->healing_fd) {
-                fd_unref (sh->healing_fd);
-                sh->healing_fd = NULL;
+    LOCK(&inode->lock);
+    {
+        ret = __afr_inode_ctx_get(this, inode, &ctx);
+        if (ret < 0 || !ctx) {
+            UNLOCK(&inode->lock);
+            gf_msg(this->name, GF_LOG_WARNING, 0,
+                   AFR_MSG_SPLIT_BRAIN_CHOICE_ERROR,
+                   "Failed to cancel split-brain choice timer.");
+            goto out;
+        }
+        ctx->spb_choice = -1;
+        if (ctx->timer) {
+            gf_timer_call_cancel(this->ctx, ctx->timer);
+            ctx->timer = NULL;
         }
+        ret = 0;
+    }
+    UNLOCK(&inode->lock);
+out:
+    return ret;
+}
 
-        GF_FREE ((char *)sh->linkname);
+void
+afr_set_split_brain_choice_cbk(void *data)
+{
+    inode_t *inode = data;
+    xlator_t *this = THIS;
 
-        GF_FREE (sh->success_children);
+    afr_spb_choice_timeout_cancel(this, inode);
+    inode_invalidate(inode);
+    inode_unref(inode);
+    return;
+}
 
-        GF_FREE (sh->fresh_children);
+int
+afr_set_split_brain_choice(int ret, call_frame_t *frame, void *opaque)
+{
+    int op_errno = ENOMEM;
+    afr_private_t *priv = NULL;
+    afr_inode_ctx_t *ctx = NULL;
+    inode_t *inode = NULL;
+    loc_t *loc = NULL;
+    xlator_t *this = NULL;
+    afr_spbc_timeout_t *data = opaque;
+    struct timespec delta = {
+        0,
+    };
+    gf_boolean_t timer_set = _gf_false;
+    gf_boolean_t timer_cancelled = _gf_false;
+    gf_boolean_t timer_reset = _gf_false;
+    int old_spb_choice = -1;
+
+    frame = data->frame;
+    loc = data->loc;
+    this = frame->this;
+    priv = this->private;
+
+    if (ret) {
+        op_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    delta.tv_sec = priv->spb_choice_timeout;
+    delta.tv_nsec = 0;
+
+    if (!loc->inode) {
+        ret = -1;
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    if (!(data->d_spb || data->m_spb)) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_SPLIT_BRAIN_CHOICE_ERROR,
+               "Cannot set "
+               "replica.split-brain-choice on %s. File is"
+               " not in data/metadata split-brain.",
+               uuid_utoa(loc->gfid));
+        ret = -1;
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    /*
+     * we're ref'ing the inode before LOCK like it is done elsewhere in the
+     * code. If we ref after LOCK, coverity complains of possible deadlocks.
+     */
+    inode = inode_ref(loc->inode);
+
+    LOCK(&inode->lock);
+    {
+        ret = __afr_inode_ctx_get(this, inode, &ctx);
+        if (ret) {
+            UNLOCK(&inode->lock);
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   AFR_MSG_SPLIT_BRAIN_CHOICE_ERROR,
+                   "Failed to get inode_ctx for %s", loc->name);
+            goto post_unlock;
+        }
 
-        GF_FREE (sh->fresh_parent_dirs);
+        old_spb_choice = ctx->spb_choice;
+        ctx->spb_choice = data->spb_child_index;
 
-        loc_wipe (&sh->parent_loc);
-        loc_wipe (&sh->lookup_loc);
+        /* Possible changes in spb-choice :
+         *         valid to -1    : cancel timer and unref
+         *         valid to valid : cancel timer and inject new one
+         *         -1    to -1    : unref and do not do anything
+         *         -1 to valid    : inject timer
+         */
 
-        GF_FREE (sh->checksum);
+        /* ctx->timer is NULL iff previous value of
+         * ctx->spb_choice is -1
+         */
+        if (ctx->timer) {
+            if (ctx->spb_choice == -1) {
+                if (!gf_timer_call_cancel(this->ctx, ctx->timer)) {
+                    ctx->timer = NULL;
+                    timer_cancelled = _gf_true;
+                }
+                /* If timer cancel failed here it means that the
+                 *  previous cbk will be executed which will set
+                 *  spb_choice to -1. So we can consider the
+                 *  'valid to -1' case to be a success
+                 *  (i.e. ret = 0) and goto unlock.
+                 */
+                goto unlock;
+            }
+            goto reset_timer;
+        } else {
+            if (ctx->spb_choice == -1)
+                goto unlock;
+            goto set_timer;
+        }
 
-        GF_FREE (sh->write_needed);
-        if (sh->healing_fd)
-                fd_unref (sh->healing_fd);
+    reset_timer:
+        ret = gf_timer_call_cancel(this->ctx, ctx->timer);
+        if (ret != 0) {
+            /* We need to bail out now instead of launching a new
+             * timer. Otherwise the cbk of the previous timer event
+             * will cancel the new ctx->timer.
+             */
+            ctx->spb_choice = old_spb_choice;
+            ret = -1;
+            op_errno = EAGAIN;
+            goto unlock;
+        }
+        ctx->timer = NULL;
+        timer_reset = _gf_true;
+
+    set_timer:
+        ctx->timer = gf_timer_call_after(this->ctx, delta,
+                                         afr_set_split_brain_choice_cbk, inode);
+        if (!ctx->timer) {
+            ctx->spb_choice = old_spb_choice;
+            ret = -1;
+            op_errno = ENOMEM;
+        }
+        if (!timer_reset && ctx->timer)
+            timer_set = _gf_true;
+        if (timer_reset && !ctx->timer)
+            timer_cancelled = _gf_true;
+    }
+unlock:
+    UNLOCK(&inode->lock);
+post_unlock:
+    if (!timer_set)
+        inode_unref(inode);
+    if (timer_cancelled)
+        inode_unref(inode);
+    /*
+     * We need to invalidate the inode to prevent the kernel from serving
+     * reads from an older cached value despite a change in spb_choice to
+     * a new value.
+     */
+    inode_invalidate(inode);
+out:
+    GF_FREE(data);
+    AFR_STACK_UNWIND(setxattr, frame, ret, op_errno, NULL);
+    return 0;
 }
 
-
-void
-afr_local_transaction_cleanup (afr_local_t *local, xlator_t *this)
+int
+afr_accused_fill(xlator_t *this, dict_t *xdata, unsigned char *accused,
+                 afr_transaction_type type)
 {
-        afr_private_t *priv    = NULL;
-        int           i        = 0;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int idx = afr_index_for_transaction_type(type);
+    void *pending_raw = NULL;
+    int pending[3];
+    int ret = 0;
 
-        priv = this->private;
+    priv = this->private;
 
-        afr_matrix_cleanup (local->pending, priv->child_count);
-        afr_matrix_cleanup (local->transaction.txn_changelog,
-                            priv->child_count);
+    for (i = 0; i < priv->child_count; i++) {
+        ret = dict_get_ptr(xdata, priv->pending_key[i], &pending_raw);
+        if (ret) /* no pending flags */
+            continue;
+        memcpy(pending, pending_raw, sizeof(pending));
 
-        GF_FREE (local->internal_lock.locked_nodes);
+        if (ntoh32(pending[idx]))
+            accused[i] = 1;
+    }
 
-        for (i = 0; local->internal_lock.inodelk[i].domain; i++) {
-                GF_FREE (local->internal_lock.inodelk[i].locked_nodes);
-        }
+    return 0;
+}
 
-        GF_FREE (local->internal_lock.lower_locked_nodes);
+int
+afr_accuse_smallfiles(xlator_t *this, struct afr_reply *replies,
+                      unsigned char *data_accused)
+{
+    int i = 0;
+    afr_private_t *priv = NULL;
+    uint64_t maxsize = 0;
 
-        afr_entry_lockee_cleanup (&local->internal_lock);
+    priv = this->private;
 
-        GF_FREE (local->transaction.pre_op);
-        GF_FREE (local->transaction.eager_lock);
+    for (i = 0; i < priv->child_count; i++) {
+        if (replies[i].valid && replies[i].xdata &&
+            dict_get_sizen(replies[i].xdata, GLUSTERFS_BAD_INODE))
+            continue;
+        if (data_accused[i])
+            continue;
+        if (replies[i].poststat.ia_size > maxsize)
+            maxsize = replies[i].poststat.ia_size;
+    }
 
-        GF_FREE (local->transaction.basename);
-        GF_FREE (local->transaction.new_basename);
+    for (i = 0; i < priv->child_count; i++) {
+        if (data_accused[i])
+            continue;
+        if (AFR_IS_ARBITER_BRICK(priv, i))
+            continue;
+        if (replies[i].poststat.ia_size < maxsize)
+            data_accused[i] = 1;
+    }
 
-        loc_wipe (&local->transaction.parent_loc);
-        loc_wipe (&local->transaction.new_parent_loc);
+    return 0;
+}
 
-        GF_FREE (local->transaction.postop_piggybacked);
+int
+afr_readables_fill(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                   unsigned char *data_accused, unsigned char *metadata_accused,
+                   unsigned char *data_readable,
+                   unsigned char *metadata_readable, struct afr_reply *replies)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    dict_t *xdata = NULL;
+    int i = 0;
+    int ret = 0;
+    ia_type_t ia_type = IA_INVAL;
+
+    local = frame->local;
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        data_readable[i] = 1;
+        metadata_readable[i] = 1;
+    }
+    if (AFR_IS_ARBITER_BRICK(priv, ARBITER_BRICK_INDEX)) {
+        data_readable[ARBITER_BRICK_INDEX] = 0;
+        metadata_readable[ARBITER_BRICK_INDEX] = 0;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (replies) { /* Lookup */
+            if (!replies[i].valid || replies[i].op_ret == -1 ||
+                (replies[i].xdata &&
+                 dict_get_sizen(replies[i].xdata, GLUSTERFS_BAD_INODE))) {
+                data_readable[i] = 0;
+                metadata_readable[i] = 0;
+                continue;
+            }
+
+            xdata = replies[i].xdata;
+            ia_type = replies[i].poststat.ia_type;
+        } else { /* pre-op xattrop */
+            xdata = local->transaction.changelog_xdata[i];
+            ia_type = inode->ia_type;
+        }
+
+        if (!xdata)
+            continue; /* mkdir_cbk sends NULL xdata_rsp. */
+        afr_accused_fill(this, xdata, data_accused,
+                         (ia_type == IA_IFDIR) ? AFR_ENTRY_TRANSACTION
+                                               : AFR_DATA_TRANSACTION);
+
+        afr_accused_fill(this, xdata, metadata_accused,
+                         AFR_METADATA_TRANSACTION);
+    }
+
+    if (replies && ia_type != IA_INVAL && ia_type != IA_IFDIR &&
+        /* We want to accuse small files only when we know for
+         * sure that there is no IO happening. Otherwise, the
+         * ia_sizes obtained in post-refresh replies may
+         * mismatch due to a race between inode-refresh and
+         * ongoing writes, causing spurious heal launches*/
+        !afr_is_possibly_under_txn(AFR_DATA_TRANSACTION, local, this)) {
+        afr_accuse_smallfiles(this, replies, data_accused);
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (data_accused[i]) {
+            data_readable[i] = 0;
+            ret = 1;
+        }
+        if (metadata_accused[i]) {
+            metadata_readable[i] = 0;
+            ret = 1;
+        }
+    }
+    return ret;
 }
 
+int
+afr_replies_interpret(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                      gf_boolean_t *start_heal)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    struct afr_reply *replies = NULL;
+    int event_generation = 0;
+    int i = 0;
+    unsigned char *data_accused = NULL;
+    unsigned char *metadata_accused = NULL;
+    unsigned char *data_readable = NULL;
+    unsigned char *metadata_readable = NULL;
+    int ret = 0;
+
+    local = frame->local;
+    priv = this->private;
+    replies = local->replies;
+    event_generation = local->event_generation;
+
+    data_accused = alloca0(priv->child_count);
+    data_readable = alloca0(priv->child_count);
+    metadata_accused = alloca0(priv->child_count);
+    metadata_readable = alloca0(priv->child_count);
+
+    ret = afr_readables_fill(frame, this, inode, data_accused, metadata_accused,
+                             data_readable, metadata_readable, replies);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (start_heal && priv->child_up[i] &&
+            (data_accused[i] || metadata_accused[i])) {
+            *start_heal = _gf_true;
+            break;
+        }
+    }
+    afr_inode_read_subvol_set(inode, this, data_readable, metadata_readable,
+                              event_generation);
+    return ret;
+}
 
-void
-afr_local_cleanup (afr_local_t *local, xlator_t *this)
+int
+afr_refresh_selfheal_done(int ret, call_frame_t *heal, void *opaque)
 {
-        afr_private_t * priv = NULL;
+    if (heal)
+        AFR_STACK_DESTROY(heal);
+    return 0;
+}
 
-        if (!local)
-                return;
+int
+afr_inode_refresh_err(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int err = 0;
 
-        afr_local_sh_cleanup (local, this);
+    local = frame->local;
+    priv = this->private;
 
-        afr_local_transaction_cleanup (local, this);
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->replies[i].valid && !local->replies[i].op_ret) {
+            err = 0;
+            goto ret;
+        }
+    }
 
-        priv = this->private;
+    err = afr_final_errno(local, priv);
+ret:
+    return err;
+}
 
-        loc_wipe (&local->loc);
-        loc_wipe (&local->newloc);
+gf_boolean_t
+afr_selfheal_enabled(const xlator_t *this)
+{
+    const afr_private_t *priv = this->private;
 
-        if (local->fd)
-                fd_unref (local->fd);
+    return priv->data_self_heal || priv->metadata_self_heal ||
+           priv->entry_self_heal;
+}
 
-        if (local->xattr_req)
-                dict_unref (local->xattr_req);
+int
+afr_txn_refresh_done(call_frame_t *frame, xlator_t *this, int err)
+{
+    call_frame_t *heal_frame = NULL;
+    afr_local_t *heal_local = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    inode_t *inode = NULL;
+    int event_generation = 0;
+    int read_subvol = -1;
+    int ret = 0;
+
+    local = frame->local;
+    inode = local->inode;
+    priv = this->private;
+
+    if (err)
+        goto refresh_done;
+
+    if (local->op == GF_FOP_LOOKUP)
+        goto refresh_done;
+
+    ret = afr_inode_get_readable(frame, inode, this, local->readable,
+                                 &event_generation, local->transaction.type);
+
+    if (ret == -EIO) {
+        /* No readable subvolume even after refresh ==> splitbrain.*/
+        if (!priv->fav_child_policy) {
+            err = EIO;
+            goto refresh_done;
+        }
+        read_subvol = afr_sh_get_fav_by_policy(this, local->replies, inode,
+                                               NULL);
+        if (read_subvol == -1) {
+            err = EIO;
+            goto refresh_done;
+        }
+
+        heal_frame = afr_frame_create(this, NULL);
+        if (!heal_frame) {
+            err = EIO;
+            goto refresh_done;
+        }
+        heal_local = heal_frame->local;
+        heal_local->xdata_req = dict_new();
+        if (!heal_local->xdata_req) {
+            err = EIO;
+            AFR_STACK_DESTROY(heal_frame);
+            goto refresh_done;
+        }
+        heal_local->heal_frame = frame;
+        ret = synctask_new(this->ctx->env, afr_fav_child_reset_sink_xattrs,
+                           afr_fav_child_reset_sink_xattrs_cbk, heal_frame,
+                           heal_frame);
+        return 0;
+    }
 
-        if (local->dict)
-                dict_unref (local->dict);
+refresh_done:
+    afr_local_replies_wipe(local, this->private);
+    local->refreshfn(frame, this, err);
 
-	GF_FREE(local->replies);
+    return 0;
+}
 
-        GF_FREE (local->child_up);
+int
+afr_inode_refresh_done(call_frame_t *frame, xlator_t *this, int error)
+{
+    call_frame_t *heal_frame = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    gf_boolean_t start_heal = _gf_false;
+    afr_local_t *heal_local = NULL;
+    unsigned char *success_replies = NULL;
+    int ret = 0;
+
+    if (error != 0) {
+        goto refresh_done;
+    }
+
+    local = frame->local;
+    priv = this->private;
+    success_replies = alloca0(priv->child_count);
+    afr_fill_success_replies(local, priv, success_replies);
+
+    if (priv->thin_arbiter_count && local->is_read_txn &&
+        AFR_COUNT(success_replies, priv->child_count) != priv->child_count) {
+        /* We need to query the good bricks and/or thin-arbiter.*/
+        if (success_replies[0]) {
+            local->read_txn_query_child = AFR_CHILD_ZERO;
+        } else if (success_replies[1]) {
+            local->read_txn_query_child = AFR_CHILD_ONE;
+        }
+        error = EINVAL;
+        goto refresh_done;
+    }
+
+    if (!afr_has_quorum(success_replies, this, frame)) {
+        error = afr_final_errno(frame->local, this->private);
+        if (!error)
+            error = afr_quorum_errno(priv);
+        goto refresh_done;
+    }
+
+    ret = afr_replies_interpret(frame, this, local->refreshinode, &start_heal);
+
+    if (ret && afr_selfheal_enabled(this) && start_heal) {
+        heal_frame = afr_frame_create(this, NULL);
+        if (!heal_frame)
+            goto refresh_done;
+        heal_local = heal_frame->local;
+        heal_local->refreshinode = inode_ref(local->refreshinode);
+        heal_local->heal_frame = heal_frame;
+        if (!afr_throttled_selfheal(heal_frame, this)) {
+            AFR_STACK_DESTROY(heal_frame);
+            goto refresh_done;
+        }
+    }
+
+refresh_done:
+    afr_txn_refresh_done(frame, this, error);
+
+    return 0;
+}
 
-        GF_FREE (local->child_errno);
+void
+afr_inode_refresh_subvol_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                             int op_ret, int op_errno, struct iatt *buf,
+                             dict_t *xdata, struct iatt *par)
+{
+    afr_local_t *local = NULL;
+    int call_child = (long)cookie;
+    int8_t need_heal = 1;
+    int call_count = 0;
+    int ret = 0;
+
+    local = frame->local;
+    local->replies[call_child].valid = 1;
+    local->replies[call_child].op_ret = op_ret;
+    local->replies[call_child].op_errno = op_errno;
+    if (op_ret != -1) {
+        local->replies[call_child].poststat = *buf;
+        if (par)
+            local->replies[call_child].postparent = *par;
+        if (xdata)
+            local->replies[call_child].xdata = dict_ref(xdata);
+    }
+
+    if (xdata) {
+        ret = dict_get_int8(xdata, "link-count", &need_heal);
+        if (ret) {
+            gf_msg_debug(this->name, -ret, "Unable to get link count");
+        }
+    }
 
-        GF_FREE (local->fresh_children);
+    local->replies[call_child].need_heal = need_heal;
+    call_count = afr_frame_return(frame);
+    if (call_count == 0) {
+        afr_set_need_heal(this, local);
+        ret = afr_inode_refresh_err(frame, this);
+        if (ret) {
+            gf_msg_debug(this->name, ret, "afr_inode_refresh_err failed");
+        }
+        afr_inode_refresh_done(frame, this, ret);
+    }
+}
 
-        { /* lookup */
-                if (local->cont.lookup.xattrs) {
-                        afr_reset_xattr (local->cont.lookup.xattrs,
-                                         priv->child_count);
-                        GF_FREE (local->cont.lookup.xattrs);
-                        local->cont.lookup.xattrs = NULL;
-                }
+int
+afr_inode_refresh_subvol_with_lookup_cbk(call_frame_t *frame, void *cookie,
+                                         xlator_t *this, int op_ret,
+                                         int op_errno, inode_t *inode,
+                                         struct iatt *buf, dict_t *xdata,
+                                         struct iatt *par)
+{
+    afr_inode_refresh_subvol_cbk(frame, cookie, this, op_ret, op_errno, buf,
+                                 xdata, par);
+    return 0;
+}
 
-                if (local->cont.lookup.xattr) {
-                        dict_unref (local->cont.lookup.xattr);
-                }
+int
+afr_inode_refresh_subvol_with_lookup(call_frame_t *frame, xlator_t *this, int i,
+                                     inode_t *inode, uuid_t gfid, dict_t *xdata)
+{
+    loc_t loc = {
+        0,
+    };
+    afr_private_t *priv = NULL;
 
-                if (local->cont.lookup.inode) {
-                        inode_unref (local->cont.lookup.inode);
-                }
+    priv = this->private;
 
-                GF_FREE (local->cont.lookup.postparents);
+    loc.inode = inode;
+    if (gf_uuid_is_null(inode->gfid) && gfid) {
+        /* To handle setattr/setxattr on yet to be linked inode from
+         * dht */
+        gf_uuid_copy(loc.gfid, gfid);
+    } else {
+        gf_uuid_copy(loc.gfid, inode->gfid);
+    }
 
-                GF_FREE (local->cont.lookup.bufs);
+    STACK_WIND_COOKIE(frame, afr_inode_refresh_subvol_with_lookup_cbk,
+                      (void *)(long)i, priv->children[i],
+                      priv->children[i]->fops->lookup, &loc, xdata);
+    return 0;
+}
 
-                GF_FREE (local->cont.lookup.success_children);
+int
+afr_inode_refresh_subvol_with_fstat_cbk(call_frame_t *frame, void *cookie,
+                                        xlator_t *this, int32_t op_ret,
+                                        int32_t op_errno, struct iatt *buf,
+                                        dict_t *xdata)
+{
+    afr_inode_refresh_subvol_cbk(frame, cookie, this, op_ret, op_errno, buf,
+                                 xdata, NULL);
+    return 0;
+}
 
-                GF_FREE (local->cont.lookup.sources);
-                afr_matrix_cleanup (local->cont.lookup.pending_matrix,
-                                    priv->child_count);
-        }
+int
+afr_inode_refresh_subvol_with_fstat(call_frame_t *frame, xlator_t *this, int i,
+                                    dict_t *xdata)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
 
-        { /* getxattr */
-                GF_FREE (local->cont.getxattr.name);
-        }
+    priv = this->private;
+    local = frame->local;
 
-        { /* lk */
-                GF_FREE (local->cont.lk.locked_nodes);
-        }
+    STACK_WIND_COOKIE(frame, afr_inode_refresh_subvol_with_fstat_cbk,
+                      (void *)(long)i, priv->children[i],
+                      priv->children[i]->fops->fstat, local->fd, xdata);
+    return 0;
+}
 
-        { /* create */
-                if (local->cont.create.fd)
-                        fd_unref (local->cont.create.fd);
-                if (local->cont.create.params)
-                        dict_unref (local->cont.create.params);
-        }
+int
+afr_inode_refresh_do(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int call_count = 0;
+    int i = 0;
+    int ret = 0;
+    dict_t *xdata = NULL;
+    afr_fd_ctx_t *fd_ctx = NULL;
+    unsigned char *wind_subvols = NULL;
 
-        { /* mknod */
-                if (local->cont.mknod.params)
-                        dict_unref (local->cont.mknod.params);
-        }
+    priv = this->private;
+    local = frame->local;
+    wind_subvols = alloca0(priv->child_count);
 
-        { /* mkdir */
-                if (local->cont.mkdir.params)
-                        dict_unref (local->cont.mkdir.params);
-        }
+    afr_local_replies_wipe(local, priv);
 
-        { /* symlink */
-                if (local->cont.symlink.params)
-                        dict_unref (local->cont.symlink.params);
+    if (local->fd) {
+        fd_ctx = afr_fd_ctx_get(local->fd, this);
+        if (!fd_ctx) {
+            afr_inode_refresh_done(frame, this, EINVAL);
+            return 0;
         }
+    }
 
-        { /* writev */
-                GF_FREE (local->cont.writev.vector);
-        }
+    xdata = dict_new();
+    if (!xdata) {
+        afr_inode_refresh_done(frame, this, ENOMEM);
+        return 0;
+    }
 
-        { /* setxattr */
-                if (local->cont.setxattr.dict)
-                        dict_unref (local->cont.setxattr.dict);
-        }
+    ret = afr_xattr_req_prepare(this, xdata);
+    if (ret != 0) {
+        dict_unref(xdata);
+        afr_inode_refresh_done(frame, this, -ret);
+        return 0;
+    }
 
-        { /* fsetxattr */
-                if (local->cont.fsetxattr.dict)
-                        dict_unref (local->cont.fsetxattr.dict);
-        }
+    ret = dict_set_sizen_str_sizen(xdata, "link-count", GF_XATTROP_INDEX_COUNT);
+    if (ret) {
+        gf_msg_debug(this->name, -ret, "Unable to set link-count in dict ");
+    }
 
-        { /* removexattr */
-                GF_FREE (local->cont.removexattr.name);
-        }
-        { /* xattrop */
-                if (local->cont.xattrop.xattr)
-                        dict_unref (local->cont.xattrop.xattr);
-        }
-        { /* fxattrop */
-                if (local->cont.fxattrop.xattr)
-                        dict_unref (local->cont.fxattrop.xattr);
-        }
-        { /* symlink */
-                GF_FREE (local->cont.symlink.linkpath);
-        }
+    ret = dict_set_str_sizen(xdata, GLUSTERFS_INODELK_DOM_COUNT, this->name);
+    if (ret) {
+        gf_msg_debug(this->name, -ret,
+                     "Unable to set inodelk-dom-count in dict ");
+    }
 
-        { /* opendir */
-                GF_FREE (local->cont.opendir.checksum);
+    if (local->fd) {
+        for (i = 0; i < priv->child_count; i++) {
+            if (local->child_up[i] && fd_ctx->opened_on[i] == AFR_FD_OPENED)
+                wind_subvols[i] = 1;
         }
+    } else {
+        memcpy(wind_subvols, local->child_up,
+               sizeof(*local->child_up) * priv->child_count);
+    }
 
-        { /* readdirp */
-                if (local->cont.readdir.dict)
-                        dict_unref (local->cont.readdir.dict);
-        }
+    local->call_count = AFR_COUNT(wind_subvols, priv->child_count);
 
-        if (local->xdata_req)
-                dict_unref (local->xdata_req);
+    call_count = local->call_count;
+    if (!call_count) {
+        dict_unref(xdata);
+        if (local->fd && AFR_COUNT(local->child_up, priv->child_count))
+            afr_inode_refresh_done(frame, this, EBADFD);
+        else
+            afr_inode_refresh_done(frame, this, ENOTCONN);
+        return 0;
+    }
+    for (i = 0; i < priv->child_count; i++) {
+        if (!wind_subvols[i])
+            continue;
 
-        if (local->xdata_rsp)
-                dict_unref (local->xdata_rsp);
-}
+        if (local->fd)
+            afr_inode_refresh_subvol_with_fstat(frame, this, i, xdata);
+        else
+            afr_inode_refresh_subvol_with_lookup(
+                frame, this, i, local->refreshinode, local->refreshgfid, xdata);
+
+        if (!--call_count)
+            break;
+    }
 
+    dict_unref(xdata);
+
+    return 0;
+}
 
 int
-afr_frame_return (call_frame_t *frame)
+afr_inode_refresh(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                  uuid_t gfid, afr_inode_refresh_cbk_t refreshfn)
 {
-        afr_local_t *local = NULL;
-        int          call_count = 0;
+    afr_local_t *local = NULL;
 
-        local = frame->local;
+    local = frame->local;
 
-        LOCK (&frame->lock);
-        {
-                call_count = --local->call_count;
-        }
-        UNLOCK (&frame->lock);
+    local->refreshfn = refreshfn;
+
+    if (local->refreshinode) {
+        inode_unref(local->refreshinode);
+        local->refreshinode = NULL;
+    }
 
-        return call_count;
+    local->refreshinode = inode_ref(inode);
+
+    if (gfid)
+        gf_uuid_copy(local->refreshgfid, gfid);
+    else
+        gf_uuid_clear(local->refreshgfid);
+
+    afr_inode_refresh_do(frame, this);
+
+    return 0;
 }
 
 int
-afr_set_elem_count_get (unsigned char *elems, int child_count)
+afr_xattr_req_prepare(xlator_t *this, dict_t *xattr_req)
 {
-        int i   = 0;
-        int ret = 0;
-
-        for (i = 0; i < child_count; i++)
-                if (elems[i])
-                        ret++;
-        return ret;
-}
+    int i = 0;
+    afr_private_t *priv = NULL;
+    int ret = 0;
 
-/**
- * up_children_count - return the number of children that are up
- */
+    priv = this->private;
 
-unsigned int
-afr_up_children_count (unsigned char *child_up, unsigned int child_count)
-{
-        return afr_set_elem_count_get (child_up, child_count);
+    for (i = 0; i < priv->child_count; i++) {
+        ret = dict_set_uint64(xattr_req, priv->pending_key[i],
+                              AFR_NUM_CHANGE_LOGS * sizeof(int));
+        if (ret < 0)
+            gf_msg(this->name, GF_LOG_WARNING, -ret, AFR_MSG_DICT_SET_FAILED,
+                   "Unable to set dict value for %s", priv->pending_key[i]);
+        /* 3 = data+metadata+entry */
+    }
+    ret = dict_set_uint64(xattr_req, AFR_DIRTY,
+                          AFR_NUM_CHANGE_LOGS * sizeof(int));
+    if (ret) {
+        gf_msg_debug(this->name, -ret,
+                     "failed to set dirty "
+                     "query flag");
+    }
+
+    ret = dict_set_int32_sizen(xattr_req, "list-xattr", 1);
+    if (ret) {
+        gf_msg_debug(this->name, -ret, "Unable to set list-xattr in dict ");
+    }
+
+    return ret;
 }
 
-unsigned int
-afr_locked_children_count (unsigned char *children, unsigned int child_count)
-{
-        return afr_set_elem_count_get (children, child_count);
+int
+afr_lookup_xattr_req_prepare(afr_local_t *local, xlator_t *this,
+                             dict_t *xattr_req, loc_t *loc)
+{
+    int ret = -ENOMEM;
+
+    if (!local->xattr_req)
+        local->xattr_req = dict_new();
+
+    if (!local->xattr_req)
+        goto out;
+
+    if (xattr_req && (xattr_req != local->xattr_req))
+        dict_copy(xattr_req, local->xattr_req);
+
+    ret = afr_xattr_req_prepare(this, local->xattr_req);
+
+    ret = dict_set_uint64(local->xattr_req, GLUSTERFS_INODELK_COUNT, 0);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, AFR_MSG_DICT_SET_FAILED,
+               "%s: Unable to set dict value for %s", loc->path,
+               GLUSTERFS_INODELK_COUNT);
+    }
+    ret = dict_set_uint64(local->xattr_req, GLUSTERFS_ENTRYLK_COUNT, 0);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, AFR_MSG_DICT_SET_FAILED,
+               "%s: Unable to set dict value for %s", loc->path,
+               GLUSTERFS_ENTRYLK_COUNT);
+    }
+
+    ret = dict_set_uint32(local->xattr_req, GLUSTERFS_PARENT_ENTRYLK, 0);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, AFR_MSG_DICT_SET_FAILED,
+               "%s: Unable to set dict value for %s", loc->path,
+               GLUSTERFS_PARENT_ENTRYLK);
+    }
+
+    ret = dict_set_sizen_str_sizen(local->xattr_req, "link-count",
+                                   GF_XATTROP_INDEX_COUNT);
+    if (ret) {
+        gf_msg_debug(this->name, -ret, "Unable to set link-count in dict ");
+    }
+
+    ret = 0;
+out:
+    return ret;
 }
 
-unsigned int
-afr_pre_op_done_children_count (unsigned char *pre_op,
-                                unsigned int child_count)
+int
+afr_least_pending_reads_child(afr_private_t *priv, unsigned char *readable)
 {
-        return afr_set_elem_count_get (pre_op, child_count);
+    int i = 0;
+    int child = -1;
+    int64_t read_iter = -1;
+    int64_t pending_read = -1;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (AFR_IS_ARBITER_BRICK(priv, i) || !readable[i])
+            continue;
+        read_iter = GF_ATOMIC_GET(priv->pending_reads[i]);
+        if (child == -1 || read_iter < pending_read) {
+            pending_read = read_iter;
+            child = i;
+        }
+    }
+
+    return child;
 }
 
-gf_boolean_t
-afr_is_fresh_lookup (loc_t *loc, xlator_t *this)
+static int32_t
+afr_least_latency_child(afr_private_t *priv, unsigned char *readable)
 {
-        uint64_t          ctx = 0;
-        int32_t           ret = 0;
+    int32_t i = 0;
+    int child = -1;
 
-        GF_ASSERT (loc);
-        GF_ASSERT (this);
-        GF_ASSERT (loc->inode);
+    for (i = 0; i < priv->child_count; i++) {
+        if (AFR_IS_ARBITER_BRICK(priv, i) || !readable[i] ||
+            priv->child_latency[i] < 0)
+            continue;
 
-        ret = inode_ctx_get (loc->inode, this, &ctx);
-        if (0 == ret)
-                return _gf_false;
-        return _gf_true;
+        if (child == -1 ||
+            priv->child_latency[i] < priv->child_latency[child]) {
+            child = i;
+        }
+    }
+    return child;
 }
 
-void
-afr_update_loc_gfids (loc_t *loc, struct iatt *buf, struct iatt *postparent)
+static int32_t
+afr_least_latency_times_pending_reads_child(afr_private_t *priv,
+                                            unsigned char *readable)
 {
-        GF_ASSERT (loc);
-        GF_ASSERT (buf);
+    int32_t i = 0;
+    int child = -1;
+    int64_t pending_read = 0;
+    int64_t latency = -1;
+    int64_t least_latency = -1;
 
-        uuid_copy (loc->gfid, buf->ia_gfid);
-        if (postparent)
-                uuid_copy (loc->pargfid, postparent->ia_gfid);
-}
+    for (i = 0; i < priv->child_count; i++) {
+        if (AFR_IS_ARBITER_BRICK(priv, i) || !readable[i] ||
+            priv->child_latency[i] < 0)
+            continue;
 
-int
-afr_lookup_build_response_params (afr_local_t *local, xlator_t *this)
-{
-        struct iatt     *buf = NULL;
-        struct iatt     *postparent = NULL;
-        dict_t          **xattr = NULL;
-        int32_t         *success_children = NULL;
-        int32_t         *sources = NULL;
-        afr_private_t   *priv = NULL;
-        int32_t         read_child = -1;
-        int             ret = 0;
-        int             i = 0;
-
-        GF_ASSERT (local);
-
-        buf = &local->cont.lookup.buf;
-        postparent = &local->cont.lookup.postparent;
-        xattr = &local->cont.lookup.xattr;
-        priv = this->private;
-
-        read_child = afr_inode_get_read_ctx (this, local->cont.lookup.inode,
-                                             local->fresh_children);
-        if (read_child < 0) {
-                ret = -1;
-                goto out;
-        }
-        success_children = local->cont.lookup.success_children;
-        sources = local->cont.lookup.sources;
-        memset (sources, 0, sizeof (*sources) * priv->child_count);
-        afr_children_intersection_get (local->fresh_children, success_children,
-                                       sources, priv->child_count);
-        if (!sources[read_child]) {
-                read_child = -1;
-                for (i = 0; i < priv->child_count; i++) {
-                        if (sources[i]) {
-                                read_child = i;
-                                break;
-                        }
-                }
-        }
-        if (read_child < 0) {
-                ret = -1;
-                goto out;
-        }
-        gf_log (this->name, GF_LOG_DEBUG, "Building lookup response from %d",
-                read_child);
-        if (!*xattr)
-                *xattr = dict_ref (local->cont.lookup.xattrs[read_child]);
-        *buf = local->cont.lookup.bufs[read_child];
-        *postparent = local->cont.lookup.postparents[read_child];
+        pending_read = GF_ATOMIC_GET(priv->pending_reads[i]);
+        latency = (pending_read + 1) * priv->child_latency[i];
 
-        if (IA_INVAL == local->cont.lookup.inode->ia_type) {
-                /* fix for RT #602 */
-                local->cont.lookup.inode->ia_type = buf->ia_type;
+        if (child == -1 || latency < least_latency) {
+            least_latency = latency;
+            child = i;
         }
-out:
-        return ret;
+    }
+    return child;
 }
 
-static void
-afr_lookup_update_lk_counts (afr_local_t *local, xlator_t *this,
-                            int child_index, dict_t *xattr)
-{
-        uint32_t inodelk_count = 0;
-        uint32_t entrylk_count = 0;
-        int      ret           = -1;
-        uint32_t parent_entrylk = 0;
-
-        GF_ASSERT (local);
-        GF_ASSERT (this);
-        GF_ASSERT (xattr);
-        GF_ASSERT (child_index >= 0);
-
-        ret = dict_get_uint32 (xattr, GLUSTERFS_INODELK_COUNT,
-                               &inodelk_count);
-        if (ret == 0)
-                local->inodelk_count += inodelk_count;
-
-        ret = dict_get_uint32 (xattr, GLUSTERFS_ENTRYLK_COUNT,
-                               &entrylk_count);
-        if (ret == 0)
-                local->entrylk_count += entrylk_count;
-        ret = dict_get_uint32 (xattr, GLUSTERFS_PARENT_ENTRYLK,
-                               &parent_entrylk);
-        if (!ret)
-                local->cont.lookup.parent_entrylk += parent_entrylk;
+int
+afr_hash_child(afr_read_subvol_args_t *args, afr_private_t *priv,
+               unsigned char *readable)
+{
+    uuid_t gfid_copy = {
+        0,
+    };
+    pid_t pid;
+    int child = -1;
+
+    switch (priv->hash_mode) {
+        case AFR_READ_POLICY_FIRST_UP:
+            break;
+        case AFR_READ_POLICY_GFID_HASH:
+            gf_uuid_copy(gfid_copy, args->gfid);
+            child = SuperFastHash((char *)gfid_copy, sizeof(gfid_copy)) %
+                    priv->child_count;
+            break;
+        case AFR_READ_POLICY_GFID_PID_HASH:
+            if (args->ia_type != IA_IFDIR) {
+                /*
+                 * Why getpid?  Because it's one of the cheapest calls
+                 * available - faster than gethostname etc. - and
+                 * returns a constant-length value that's sure to be
+                 * shorter than a UUID. It's still very unlikely to be
+                 * the same across clients, so it still provides good
+                 * mixing.  We're not trying for perfection here. All we
+                 * need is a low probability that multiple clients
+                 * won't converge on the same subvolume.
+                 */
+                gf_uuid_copy(gfid_copy, args->gfid);
+                pid = getpid();
+                *(pid_t *)gfid_copy ^= pid;
+            }
+            child = SuperFastHash((char *)gfid_copy, sizeof(gfid_copy)) %
+                    priv->child_count;
+            break;
+        case AFR_READ_POLICY_LESS_LOAD:
+            child = afr_least_pending_reads_child(priv, readable);
+            break;
+        case AFR_READ_POLICY_LEAST_LATENCY:
+            child = afr_least_latency_child(priv, readable);
+            break;
+        case AFR_READ_POLICY_LOAD_LATENCY_HYBRID:
+            child = afr_least_latency_times_pending_reads_child(priv, readable);
+            break;
+    }
+
+    return child;
 }
 
-/*
- * It's important to maintain a commutative property on do_*_self_heal and
- * found*; once set, they must not be cleared by a subsequent iteration or
- * call, so that they represent a logical OR of all iterations and calls
- * regardless of child/key order.  That allows the caller to call us multiple
- * times without having to use a separate variable as a "reduce" accumulator.
- */
-static void
-afr_lookup_set_self_heal_params_by_xattr (afr_local_t *local, xlator_t *this,
-                                          dict_t *xattr)
+int
+afr_read_subvol_select_by_policy(inode_t *inode, xlator_t *this,
+                                 unsigned char *readable,
+                                 afr_read_subvol_args_t *args)
 {
-        afr_private_t *priv        = NULL;
-        int            i           = 0;
-        int            ret         = -1;
-        void          *pending_raw = NULL;
-        int32_t       *pending     = NULL;
+    int i = 0;
+    int read_subvol = -1;
+    afr_private_t *priv = NULL;
+    afr_read_subvol_args_t local_args = {
+        0,
+    };
 
-        GF_ASSERT (local);
-        GF_ASSERT (this);
-        GF_ASSERT (xattr);
+    priv = this->private;
 
-        priv = this->private;
+    /* first preference - explicitly specified or local subvolume */
+    if (priv->read_child >= 0 && readable[priv->read_child])
+        return priv->read_child;
 
-        for (i = 0; i < priv->child_count; i++) {
-                ret = dict_get_ptr (xattr, priv->pending_key[i],
-                                    &pending_raw);
-                if (ret != 0) {
-                        continue;
-                }
-                pending = pending_raw;
+    if (inode_is_linked(inode)) {
+        gf_uuid_copy(local_args.gfid, inode->gfid);
+        local_args.ia_type = inode->ia_type;
+    } else if (args) {
+        local_args = *args;
+    }
 
-                if (pending[AFR_METADATA_TRANSACTION]) {
-                        gf_log(this->name, GF_LOG_DEBUG,
-                               "metadata self-heal is pending for %s.",
-                               local->loc.path);
-                        local->self_heal.do_metadata_self_heal = _gf_true;
-                }
+    /* second preference - use hashed mode */
+    read_subvol = afr_hash_child(&local_args, priv, readable);
+    if (read_subvol >= 0 && readable[read_subvol])
+        return read_subvol;
 
-                if (pending[AFR_ENTRY_TRANSACTION]) {
-                        gf_log(this->name, GF_LOG_DEBUG,
-                               "entry self-heal is pending for %s.",
-                               local->loc.path);
-                        local->self_heal.do_entry_self_heal = _gf_true;
-                }
+    for (i = 0; i < priv->child_count; i++) {
+        if (readable[i])
+            return i;
+    }
 
-                if (pending[AFR_DATA_TRANSACTION]) {
-                        gf_log(this->name, GF_LOG_DEBUG,
-                               "data self-heal is pending for %s.",
-                               local->loc.path);
-                        local->self_heal.do_data_self_heal = _gf_true;
-                }
-        }
+    /* no readable subvolumes, either split brain or all subvols down */
+
+    return -1;
 }
 
-void
-afr_lookup_check_set_metadata_split_brain (afr_local_t *local, xlator_t *this)
+int
+afr_inode_read_subvol_type_get(inode_t *inode, xlator_t *this,
+                               unsigned char *readable, int *event_p, int type)
 {
-        int32_t                  *sources = NULL;
-        afr_private_t            *priv = NULL;
-        int32_t                  subvol_status = 0;
-        int32_t                  *success_children   = NULL;
-        dict_t                   **xattrs = NULL;
-        struct iatt              *bufs = NULL;
-        int32_t                  **pending_matrix = NULL;
+    int ret = -1;
 
-        priv = this->private;
-
-        sources = GF_CALLOC (priv->child_count, sizeof (*sources),
-                             gf_afr_mt_int32_t);
-        if (NULL == sources)
-                goto out;
-        success_children = local->cont.lookup.success_children;
-        xattrs = local->cont.lookup.xattrs;
-        bufs = local->cont.lookup.bufs;
-        pending_matrix = local->cont.lookup.pending_matrix;
-        afr_build_sources (this, xattrs, bufs, pending_matrix,
-                           sources, success_children, AFR_METADATA_TRANSACTION,
-                           &subvol_status, _gf_false);
-        if (subvol_status & SPLIT_BRAIN)
-                local->cont.lookup.possible_spb = _gf_true;
-out:
-        GF_FREE (sources);
+    if (type == AFR_METADATA_TRANSACTION)
+        ret = afr_inode_read_subvol_get(inode, this, 0, readable, event_p);
+    else
+        ret = afr_inode_read_subvol_get(inode, this, readable, 0, event_p);
+    return ret;
 }
 
-static void
-afr_detect_self_heal_by_iatt (afr_local_t *local, xlator_t *this,
-                            struct iatt *buf, struct iatt *lookup_buf)
+void
+afr_readables_intersect_get(inode_t *inode, xlator_t *this, int *event,
+                            unsigned char *intersection)
 {
-        if (PERMISSION_DIFFERS (buf, lookup_buf)) {
-                /* mismatching permissions */
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "permissions differ for %s ", local->loc.path);
-                local->self_heal.do_metadata_self_heal = _gf_true;
-        }
+    afr_private_t *priv = NULL;
+    unsigned char *data_readable = NULL;
+    unsigned char *metadata_readable = NULL;
+    unsigned char *intersect = NULL;
 
-        if (OWNERSHIP_DIFFERS (buf, lookup_buf)) {
-                /* mismatching permissions */
-                local->self_heal.do_metadata_self_heal = _gf_true;
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "ownership differs for %s ", local->loc.path);
-        }
+    priv = this->private;
+    data_readable = alloca0(priv->child_count);
+    metadata_readable = alloca0(priv->child_count);
+    intersect = alloca0(priv->child_count);
 
-        if (SIZE_DIFFERS (buf, lookup_buf)
-            && IA_ISREG (buf->ia_type)) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "size differs for %s ", local->loc.path);
-                local->self_heal.do_data_self_heal = _gf_true;
-        }
+    afr_inode_read_subvol_get(inode, this, data_readable, metadata_readable,
+                              event);
 
-        if (uuid_compare (buf->ia_gfid, lookup_buf->ia_gfid)) {
-                /* mismatching gfid */
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "%s: gfid different on subvolume", local->loc.path);
-        }
+    AFR_INTERSECT(intersect, data_readable, metadata_readable,
+                  priv->child_count);
+    if (intersection)
+        memcpy(intersection, intersect,
+               sizeof(*intersection) * priv->child_count);
 }
 
-static void
-afr_detect_self_heal_by_split_brain_status (afr_local_t *local, xlator_t *this)
+int
+afr_read_subvol_get(inode_t *inode, xlator_t *this, int *subvol_p,
+                    unsigned char *readables, int *event_p,
+                    afr_transaction_type type, afr_read_subvol_args_t *args)
 {
-        gf_boolean_t split_brain = _gf_false;
-        afr_self_heal_t *sh = NULL;
+    afr_private_t *priv = NULL;
+    unsigned char *readable = NULL;
+    unsigned char *intersection = NULL;
+    int subvol = -1;
+    int event = 0;
 
-        sh = &local->self_heal;
+    priv = this->private;
 
-        split_brain = afr_is_split_brain (this, local->cont.lookup.inode);
-        split_brain = split_brain || local->cont.lookup.possible_spb;
-        if ((local->success_count > 0) && split_brain &&
-            IA_ISREG (local->cont.lookup.inode->ia_type)) {
-                sh->force_confirm_spb = _gf_true;
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "split brain detected during lookup of %s.",
-                        local->loc.path);
-        }
-}
+    readable = alloca0(priv->child_count);
+    intersection = alloca0(priv->child_count);
 
-static void
-afr_detect_self_heal_by_lookup_status (afr_local_t *local, xlator_t *this)
-{
-        GF_ASSERT (local);
-        GF_ASSERT (this);
+    afr_inode_read_subvol_type_get(inode, this, readable, &event, type);
 
-        if ((local->success_count > 0) && (local->enoent_count > 0)) {
-                local->self_heal.do_metadata_self_heal = _gf_true;
-                local->self_heal.do_data_self_heal     = _gf_true;
-                local->self_heal.do_entry_self_heal    = _gf_true;
-                local->self_heal.do_gfid_self_heal    = _gf_true;
-                local->self_heal.do_missing_entry_self_heal    = _gf_true;
-                gf_log(this->name, GF_LOG_DEBUG,
-                       "entries are missing in lookup of %s.",
-                       local->loc.path);
-        }
+    afr_readables_intersect_get(inode, this, &event, intersection);
 
-        return;
+    if (AFR_COUNT(intersection, priv->child_count) > 0)
+        subvol = afr_read_subvol_select_by_policy(inode, this, intersection,
+                                                  args);
+    else
+        subvol = afr_read_subvol_select_by_policy(inode, this, readable, args);
+    if (subvol_p)
+        *subvol_p = subvol;
+    if (event_p)
+        *event_p = event;
+    if (readables)
+        memcpy(readables, readable, sizeof(*readables) * priv->child_count);
+    return subvol;
 }
 
-gf_boolean_t
-afr_can_self_heal_proceed (afr_self_heal_t *sh, afr_private_t *priv)
+void
+afr_local_transaction_cleanup(afr_local_t *local, xlator_t *this)
 {
-        GF_ASSERT (sh);
-        GF_ASSERT (priv);
+    afr_private_t *priv = NULL;
+    int i = 0;
 
-        if (sh->force_confirm_spb)
-                return _gf_true;
-        return (sh->do_gfid_self_heal
-                || sh->do_missing_entry_self_heal
-                || (afr_data_self_heal_enabled (priv->data_self_heal) &&
-                    sh->do_data_self_heal)
-                || (priv->metadata_self_heal && sh->do_metadata_self_heal)
-                || (priv->entry_self_heal && sh->do_entry_self_heal));
-}
+    priv = this->private;
 
-afr_transaction_type
-afr_transaction_type_get (ia_type_t ia_type)
-{
-        afr_transaction_type    type = AFR_METADATA_TRANSACTION;
+    afr_matrix_cleanup(local->pending, priv->child_count);
 
-        GF_ASSERT (ia_type != IA_INVAL);
+    GF_FREE(local->internal_lock.lower_locked_nodes);
 
-        if (IA_ISDIR (ia_type)) {
-                type = AFR_ENTRY_TRANSACTION;
-        } else if (IA_ISREG (ia_type)) {
-                type = AFR_DATA_TRANSACTION;
-        }
-        return type;
-}
+    afr_lockees_cleanup(&local->internal_lock);
 
-int
-afr_lookup_select_read_child (afr_local_t *local, xlator_t *this,
-                              int32_t *read_child)
-{
-        ia_type_t               ia_type        = IA_INVAL;
-        int32_t                 source         = -1;
-        int                     ret            = -1;
-        dict_t                  **xattrs       = NULL;
-        int32_t                 *success_children = NULL;
-        afr_transaction_type    type           = AFR_METADATA_TRANSACTION;
-        uuid_t                  *gfid          = NULL;
-
-        GF_ASSERT (local);
-        GF_ASSERT (this);
-        GF_ASSERT (local->success_count > 0);
-
-        success_children = local->cont.lookup.success_children;
-        /*We can take the success_children[0] only because we already
-         *handle the conflicting children other wise, we could select the
-         *read_child based on wrong file type
-         */
-        ia_type = local->cont.lookup.bufs[success_children[0]].ia_type;
-        type = afr_transaction_type_get (ia_type);
-        xattrs = local->cont.lookup.xattrs;
-        gfid = &local->cont.lookup.buf.ia_gfid;
-        source = afr_lookup_select_read_child_by_txn_type (this, local, xattrs,
-                                                           type, *gfid);
-        if (source < 0) {
-                gf_log (this->name, GF_LOG_DEBUG, "failed to select source "
-                        "for %s", local->loc.path);
-                goto out;
+    GF_FREE(local->transaction.pre_op);
+
+    GF_FREE(local->transaction.pre_op_sources);
+    if (local->transaction.changelog_xdata) {
+        for (i = 0; i < priv->child_count; i++) {
+            if (!local->transaction.changelog_xdata[i])
+                continue;
+            dict_unref(local->transaction.changelog_xdata[i]);
         }
+        GF_FREE(local->transaction.changelog_xdata);
+    }
 
-        gf_log (this->name, GF_LOG_DEBUG, "Source selected as %d for %s",
-                source, local->loc.path);
-        *read_child = source;
-        ret = 0;
-out:
-        return ret;
-}
+    GF_FREE(local->transaction.failed_subvols);
 
-static inline gf_boolean_t
-afr_is_transaction_running (afr_local_t *local)
-{
-        GF_ASSERT (local->fop == GF_FOP_LOOKUP);
-        return ((local->inodelk_count > 0) || (local->entrylk_count > 0));
+    GF_FREE(local->transaction.basename);
+    GF_FREE(local->transaction.new_basename);
+
+    loc_wipe(&local->transaction.parent_loc);
+    loc_wipe(&local->transaction.new_parent_loc);
 }
 
 void
-afr_launch_self_heal (call_frame_t *frame, xlator_t *this, inode_t *inode,
-                      gf_boolean_t background, ia_type_t ia_type, char *reason,
-                      void (*gfid_sh_success_cbk) (call_frame_t *sh_frame,
-                                                   xlator_t *this),
-                      int (*unwind) (call_frame_t *frame, xlator_t *this,
-                                     int32_t op_ret, int32_t op_errno,
-                                     int32_t sh_failed))
-{
-        afr_local_t             *local = NULL;
-        char                    sh_type_str[256] = {0,};
-        char                    *bg = "";
-
-        GF_ASSERT (frame);
-        GF_ASSERT (this);
-        GF_ASSERT (inode);
-        GF_ASSERT (ia_type != IA_INVAL);
-
-        local = frame->local;
-        local->self_heal.background = background;
-        local->self_heal.type       = ia_type;
-        local->self_heal.unwind     = unwind;
-        local->self_heal.gfid_sh_success_cbk     = gfid_sh_success_cbk;
-
-        afr_self_heal_type_str_get (&local->self_heal,
-                                    sh_type_str,
-                                    sizeof (sh_type_str));
-
-        if (background)
-                bg = "background";
-        gf_log (this->name, GF_LOG_DEBUG,
-                "%s %s self-heal triggered. path: %s, reason: %s", bg,
-                sh_type_str, local->loc.path, reason);
-
-        afr_self_heal (frame, this, inode);
-}
-
-unsigned int
-afr_gfid_missing_count (const char *xlator_name, int32_t *success_children,
-                        struct iatt *bufs, unsigned int child_count,
-                        const char *path)
-{
-        unsigned int    gfid_miss_count   = 0;
-        int             i              = 0;
-        struct iatt     *child1        = NULL;
-
-        for (i = 0; i < child_count; i++) {
-                if (success_children[i] == -1)
-                        break;
-                child1 = &bufs[success_children[i]];
-                if (uuid_is_null (child1->ia_gfid)) {
-                        gf_log (xlator_name, GF_LOG_DEBUG, "%s: gfid is null"
-                                " on subvolume %d", path, success_children[i]);
-                        gfid_miss_count++;
-                }
-        }
+afr_reply_wipe(struct afr_reply *reply)
+{
+    if (reply->xdata) {
+        dict_unref(reply->xdata);
+        reply->xdata = NULL;
+    }
 
-        return gfid_miss_count;
+    if (reply->xattr) {
+        dict_unref(reply->xattr);
+        reply->xattr = NULL;
+    }
 }
 
-static int
-afr_lookup_gfid_missing_count (afr_local_t *local, xlator_t *this)
+void
+afr_replies_wipe(struct afr_reply *replies, int count)
 {
-        int32_t         *success_children = NULL;
-        afr_private_t   *priv          = NULL;
-        struct iatt     *bufs          = NULL;
-        int             miss_count     = 0;
-
-        priv = this->private;
-        bufs = local->cont.lookup.bufs;
-        success_children = local->cont.lookup.success_children;
+    int i = 0;
 
-        miss_count =  afr_gfid_missing_count (this->name, success_children,
-                                              bufs, priv->child_count,
-                                              local->loc.path);
-        return miss_count;
-}
-
-gf_boolean_t
-afr_conflicting_iattrs (struct iatt *bufs, int32_t *success_children,
-                        unsigned int child_count, const char *path,
-                        const char *xlator_name)
-{
-        gf_boolean_t    conflicting    = _gf_false;
-        int             i              = 0;
-        struct iatt     *child1        = NULL;
-        struct iatt     *child2        = NULL;
-        uuid_t          *gfid          = NULL;
-
-        for (i = 0; i < child_count; i++) {
-                if (success_children[i] == -1)
-                        break;
-                child1 = &bufs[success_children[i]];
-                if ((!gfid) && (!uuid_is_null (child1->ia_gfid)))
-                        gfid = &child1->ia_gfid;
-
-                if (i == 0)
-                        continue;
-
-                child2 = &bufs[success_children[i-1]];
-                if (FILETYPE_DIFFERS (child1, child2)) {
-                        gf_log (xlator_name, GF_LOG_DEBUG, "%s: filetype "
-                                "differs on subvolumes (%d, %d)", path,
-                                success_children[i-1], success_children[i]);
-                        conflicting = _gf_true;
-                        goto out;
-                }
-                if (!gfid || uuid_is_null (child1->ia_gfid))
-                        continue;
-                if (uuid_compare (*gfid, child1->ia_gfid)) {
-                       gf_log (xlator_name, GF_LOG_DEBUG, "%s: gfid differs"
-                               " on subvolume %d", path, success_children[i]);
-                       conflicting = _gf_true;
-                       goto out;
-                }
-        }
-out:
-        return conflicting;
+    for (i = 0; i < count; i++) {
+        afr_reply_wipe(&replies[i]);
+    }
 }
 
-/* afr_update_gfid_from_iatts: This function should be called only if the
- * iatts are not conflicting.
- */
 void
-afr_update_gfid_from_iatts (uuid_t uuid, struct iatt *bufs,
-                            int32_t *success_children, unsigned int child_count)
-{
-        uuid_t          *gfid = NULL;
-        int             i = 0;
-        int             child = 0;
-
-        for (i = 0; i < child_count; i++) {
-                child = success_children[i];
-                if (child == -1)
-                        break;
-                if ((!gfid) && (!uuid_is_null (bufs[child].ia_gfid))) {
-                        gfid = &bufs[child].ia_gfid;
-                } else if (gfid && (!uuid_is_null (bufs[child].ia_gfid))) {
-                        if (uuid_compare (*gfid, bufs[child].ia_gfid)) {
-                                GF_ASSERT (0);
-                                goto out;
-                        }
-                }
-        }
-        if (gfid && (!uuid_is_null (*gfid)))
-                uuid_copy (uuid, *gfid);
-out:
+afr_local_replies_wipe(afr_local_t *local, afr_private_t *priv)
+{
+    if (!local->replies)
         return;
-}
 
-static gf_boolean_t
-afr_lookup_conflicting_entries (afr_local_t *local, xlator_t *this)
-{
-        afr_private_t           *priv = NULL;
-        gf_boolean_t            conflict = _gf_false;
+    afr_replies_wipe(local->replies, priv->child_count);
 
-        priv = this->private;
-        conflict =  afr_conflicting_iattrs (local->cont.lookup.bufs,
-                                            local->cont.lookup.success_children,
-                                            priv->child_count, local->loc.path,
-                                            this->name);
-        return conflict;
+    memset(local->replies, 0, sizeof(*local->replies) * priv->child_count);
 }
 
-gf_boolean_t
-afr_open_only_data_self_heal (char *data_self_heal)
-{
-        return !strcmp (data_self_heal, "open");
+static gf_boolean_t
+afr_fop_lock_is_unlock(call_frame_t *frame)
+{
+    afr_local_t *local = frame->local;
+    switch (local->op) {
+        case GF_FOP_INODELK:
+        case GF_FOP_FINODELK:
+            if ((F_UNLCK == local->cont.inodelk.in_flock.l_type) &&
+                (local->cont.inodelk.in_cmd == F_SETLKW ||
+                 local->cont.inodelk.in_cmd == F_SETLK))
+                return _gf_true;
+            break;
+        case GF_FOP_ENTRYLK:
+        case GF_FOP_FENTRYLK:
+            if (ENTRYLK_UNLOCK == local->cont.entrylk.in_cmd)
+                return _gf_true;
+            break;
+        default:
+            return _gf_false;
+    }
+    return _gf_false;
 }
 
-gf_boolean_t
-afr_data_self_heal_enabled (char *data_self_heal)
+static gf_boolean_t
+afr_lk_is_unlock(int32_t cmd, struct gf_flock *flock)
 {
-        gf_boolean_t    enabled = _gf_false;
+    switch (cmd) {
+        case F_RESLK_UNLCK:
+            return _gf_true;
+            break;
 
-        if (gf_string2boolean (data_self_heal, &enabled) == -1) {
-                enabled = !strcmp (data_self_heal, "open");
-                GF_ASSERT (enabled);
-        }
+#if F_SETLKW != F_SETLKW64
+        case F_SETLKW64:
+#endif
+        case F_SETLKW:
 
-        return enabled;
+#if F_SETLK != F_SETLK64
+        case F_SETLK64:
+#endif
+        case F_SETLK:
+            if (F_UNLCK == flock->l_type)
+                return _gf_true;
+            break;
+        default:
+            return _gf_false;
+    }
+    return _gf_false;
 }
 
-static void
-afr_lookup_set_self_heal_params (afr_local_t *local, xlator_t *this)
+void
+afr_handle_inconsistent_fop(call_frame_t *frame, int32_t *op_ret,
+                            int32_t *op_errno)
 {
-        int                     i = 0;
-        struct iatt             *bufs = NULL;
-        dict_t                  **xattr = NULL;
-        afr_private_t           *priv = NULL;
-        int32_t                 child1 = -1;
-        int32_t                 child2 = -1;
-        afr_self_heal_t         *sh = NULL;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
 
-        priv  = this->private;
-        sh = &local->self_heal;
+    if (!frame || !frame->this || !frame->local || !frame->this->private)
+        return;
 
-        afr_detect_self_heal_by_lookup_status (local, this);
+    if (*op_ret < 0)
+        return;
 
-        if (afr_lookup_gfid_missing_count (local, this))
-                local->self_heal.do_gfid_self_heal    = _gf_true;
+    /* Failing inodelk/entrylk/lk here is not a good idea because we
+     * need to cleanup the locks on the other bricks if we choose to fail
+     * the fop here. The brick may go down just after unwind happens as well
+     * so anyways the fop will fail when the next fop is sent so leaving
+     * it like this for now.*/
+    local = frame->local;
+    switch (local->op) {
+        case GF_FOP_LOOKUP:
+        case GF_FOP_INODELK:
+        case GF_FOP_FINODELK:
+        case GF_FOP_ENTRYLK:
+        case GF_FOP_FENTRYLK:
+        case GF_FOP_LK:
+            return;
+        default:
+            break;
+    }
 
-        if (_gf_true == afr_lookup_conflicting_entries (local, this))
-                local->self_heal.do_missing_entry_self_heal    = _gf_true;
-        else
-                afr_update_gfid_from_iatts (local->self_heal.sh_gfid_req,
-                                            local->cont.lookup.bufs,
-                                            local->cont.lookup.success_children,
-                                            priv->child_count);
+    priv = frame->this->private;
+    if (!priv->consistent_io)
+        return;
 
-        bufs = local->cont.lookup.bufs;
-        for (i = 1; i < local->success_count; i++) {
-                child1 = local->cont.lookup.success_children[i-1];
-                child2 = local->cont.lookup.success_children[i];
-                afr_detect_self_heal_by_iatt (local, this,
-                                              &bufs[child1], &bufs[child2]);
-        }
+    if (local->event_generation &&
+        (local->event_generation != priv->event_generation))
+        goto inconsistent;
 
-        xattr = local->cont.lookup.xattrs;
-        for (i = 0; i < local->success_count; i++) {
-                child1 = local->cont.lookup.success_children[i];
-                afr_lookup_set_self_heal_params_by_xattr (local, this,
-                                                          xattr[child1]);
-        }
-        if (afr_open_only_data_self_heal (priv->data_self_heal))
-                sh->do_data_self_heal = _gf_false;
-        if (sh->do_metadata_self_heal)
-                afr_lookup_check_set_metadata_split_brain (local, this);
-        afr_detect_self_heal_by_split_brain_status (local, this);
+    return;
+inconsistent:
+    *op_ret = -1;
+    *op_errno = ENOTCONN;
 }
 
-int
-afr_self_heal_lookup_unwind (call_frame_t *frame, xlator_t *this,
-                             int32_t op_ret, int32_t op_errno,
-                             int32_t sh_failed)
+void
+afr_local_cleanup(afr_local_t *local, xlator_t *this)
 {
-        afr_local_t *local = NULL;
-        int         ret    = -1;
-        dict_t      *xattr = NULL;
+    afr_private_t *priv = NULL;
 
-        local = frame->local;
+    if (!local)
+        return;
 
-        if (op_ret == -1) {
-                local->op_ret = -1;
-		local->op_errno = afr_most_important_error(local->op_errno,
-							   op_errno, _gf_true);
+    syncbarrier_destroy(&local->barrier);
 
-                goto out;
-        } else {
-                local->op_ret = 0;
-        }
-
-        afr_lookup_done_success_action (frame, this, _gf_true);
-        xattr = local->cont.lookup.xattr;
-        if (xattr) {
-                ret = dict_set_int32 (xattr, "sh-failed", sh_failed);
-                if (ret)
-                        gf_log (this->name, GF_LOG_ERROR, "%s: Failed to set "
-                                "sh-failed to %d", local->loc.path, sh_failed);
-
-                if (local->self_heal.actual_sh_started == _gf_true &&
-                    sh_failed == 0) {
-                        ret = dict_set_int32 (xattr, "actual-sh-done", 1);
-                        if (ret)
-                                gf_log(this->name, GF_LOG_ERROR, "%s: Failed to"
-                                       " set actual-sh-done to %d",
-                                       local->loc.path,
-                                       local->self_heal.actual_sh_started);
-                }
-        }
-out:
-        AFR_STACK_UNWIND (lookup, frame, local->op_ret, local->op_errno,
-                          local->cont.lookup.inode, &local->cont.lookup.buf,
-                          local->cont.lookup.xattr,
-                          &local->cont.lookup.postparent);
+    afr_local_transaction_cleanup(local, this);
 
-        return 0;
-}
+    priv = this->private;
 
-//TODO: At the moment only lookup needs this, so not doing any checks, in the
-// future we will have to do fop specific operations
-void
-afr_post_gfid_sh_success (call_frame_t *sh_frame, xlator_t *this)
-{
-        afr_local_t             *local = NULL;
-        afr_local_t             *sh_local = NULL;
-        afr_private_t           *priv = NULL;
-        afr_self_heal_t         *sh = NULL;
-        int                     i = 0;
-        struct iatt             *lookup_bufs = NULL;
-        struct iatt             *lookup_parentbufs = NULL;
+    loc_wipe(&local->loc);
+    loc_wipe(&local->newloc);
 
-        sh_local = sh_frame->local;
-        sh       = &sh_local->self_heal;
-        local = sh->orig_frame->local;
-        lookup_bufs = local->cont.lookup.bufs;
-        lookup_parentbufs = local->cont.lookup.postparents;
-        priv = this->private;
+    if (local->fd)
+        fd_unref(local->fd);
 
-        memcpy (lookup_bufs, sh->buf, priv->child_count * sizeof (*sh->buf));
-        memcpy (lookup_parentbufs, sh->parentbufs,
-                priv->child_count * sizeof (*sh->parentbufs));
+    if (local->xattr_req)
+        dict_unref(local->xattr_req);
 
-        afr_reset_xattr (local->cont.lookup.xattrs, priv->child_count);
-        if (local->cont.lookup.xattr) {
-                dict_unref (local->cont.lookup.xattr);
-                local->cont.lookup.xattr = NULL;
-        }
+    if (local->xattr_rsp)
+        dict_unref(local->xattr_rsp);
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (sh->xattr[i])
-                        local->cont.lookup.xattrs[i] = dict_ref (sh->xattr[i]);
-        }
+    if (local->dict)
+        dict_unref(local->dict);
 
-        afr_reset_children (local->cont.lookup.success_children,
-                            priv->child_count);
-        afr_children_copy (local->cont.lookup.success_children,
-                           sh->fresh_children, priv->child_count);
-}
+    afr_local_replies_wipe(local, priv);
+    GF_FREE(local->replies);
 
-static void
-afr_lookup_perform_self_heal (call_frame_t *frame, xlator_t *this,
-                              gf_boolean_t *sh_launched)
-{
-        unsigned int         up_count = 0;
-        afr_private_t       *priv    = NULL;
-        afr_local_t         *local   = NULL;
-        char                *reason  = NULL;
-
-        GF_ASSERT (sh_launched);
-        *sh_launched = _gf_false;
-        priv         = this->private;
-        local        = frame->local;
-
-        up_count  = afr_up_children_count (local->child_up, priv->child_count);
-        if (up_count == 1) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Only 1 child up - do not attempt to detect self heal");
-                goto out;
-        }
+    GF_FREE(local->child_up);
 
-        afr_lookup_set_self_heal_params (local, this);
-        if (afr_can_self_heal_proceed (&local->self_heal, priv)) {
-                if  (afr_is_transaction_running (local))
-                        goto out;
+    GF_FREE(local->read_attempted);
+
+    GF_FREE(local->readable);
+    GF_FREE(local->readable2);
+
+    if (local->inode)
+        inode_unref(local->inode);
 
-                reason = "lookup detected pending operations";
-                afr_launch_self_heal (frame, this, local->cont.lookup.inode,
-                                      _gf_true, local->cont.lookup.buf.ia_type,
-                                      reason, afr_post_gfid_sh_success,
-                                      afr_self_heal_lookup_unwind);
-                *sh_launched = _gf_true;
-        }
-out:
-        return;
-}
+    if (local->parent)
+        inode_unref(local->parent);
 
-void
-afr_get_fresh_children (int32_t *success_children, int32_t *sources,
-                        int32_t *fresh_children, unsigned int child_count)
-{
-        unsigned int i = 0;
-        unsigned int j = 0;
-
-        GF_ASSERT (success_children);
-        GF_ASSERT (sources);
-        GF_ASSERT (fresh_children);
-
-        afr_reset_children (fresh_children, child_count);
-        for (i = 0; i < child_count; i++) {
-                if (success_children[i] == -1)
-                        break;
-                if (afr_is_read_child (success_children, sources, child_count,
-                                       success_children[i])) {
-                        fresh_children[j] = success_children[i];
-                        j++;
-                }
-        }
-}
+    if (local->parent2)
+        inode_unref(local->parent2);
 
-static int
-afr_lookup_set_read_ctx (afr_local_t *local, xlator_t *this, int32_t read_child)
-{
-        afr_private_t           *priv = NULL;
+    if (local->refreshinode)
+        inode_unref(local->refreshinode);
 
-        GF_ASSERT (read_child >= 0);
+    { /* getxattr */
+        GF_FREE(local->cont.getxattr.name);
+    }
 
-        priv = this->private;
-        afr_get_fresh_children (local->cont.lookup.success_children,
-                                local->cont.lookup.sources,
-                                local->fresh_children, priv->child_count);
-        afr_inode_set_read_ctx (this, local->cont.lookup.inode, read_child,
-                                local->fresh_children);
+    { /* lk */
+        GF_FREE(local->cont.lk.locked_nodes);
+        GF_FREE(local->cont.lk.dom_locked_nodes);
+        GF_FREE(local->cont.lk.dom_lock_op_ret);
+        GF_FREE(local->cont.lk.dom_lock_op_errno);
+    }
 
-        return 0;
-}
+    { /* create */
+        if (local->cont.create.fd)
+            fd_unref(local->cont.create.fd);
+        if (local->cont.create.params)
+            dict_unref(local->cont.create.params);
+    }
 
-int
-afr_lookup_done_success_action (call_frame_t *frame, xlator_t *this,
-                                gf_boolean_t fail_conflict)
-{
-        int32_t             read_child = -1;
-        int32_t             ret        = -1;
-        afr_local_t         *local     = NULL;
-        gf_boolean_t        fresh_lookup = _gf_false;
+    { /* mknod */
+        if (local->cont.mknod.params)
+            dict_unref(local->cont.mknod.params);
+    }
 
-        local   = frame->local;
-        fresh_lookup = local->cont.lookup.fresh_lookup;
+    { /* mkdir */
+        if (local->cont.mkdir.params)
+            dict_unref(local->cont.mkdir.params);
+    }
 
-        if (local->loc.parent == NULL)
-                fail_conflict = _gf_true;
+    { /* symlink */
+        if (local->cont.symlink.params)
+            dict_unref(local->cont.symlink.params);
+    }
 
-        if (afr_lookup_conflicting_entries (local, this)) {
-                if (fail_conflict == _gf_false)
-                        ret = 0;
-                goto out;
-        }
+    { /* writev */
+        GF_FREE(local->cont.writev.vector);
+        if (local->cont.writev.iobref)
+            iobref_unref(local->cont.writev.iobref);
+    }
 
-        ret = afr_lookup_select_read_child (local, this, &read_child);
-        if (!afr_is_transaction_running (local) || fresh_lookup) {
-                if (read_child < 0)
-                        goto out;
+    { /* setxattr */
+        if (local->cont.setxattr.dict)
+            dict_unref(local->cont.setxattr.dict);
+    }
 
-                ret = afr_lookup_set_read_ctx (local, this, read_child);
-                if (ret)
-                        goto out;
-        }
+    { /* fsetxattr */
+        if (local->cont.fsetxattr.dict)
+            dict_unref(local->cont.fsetxattr.dict);
+    }
 
-        ret = afr_lookup_build_response_params (local, this);
-        if (ret)
-                goto out;
-        afr_update_loc_gfids (&local->loc,
-                              &local->cont.lookup.buf,
-                              &local->cont.lookup.postparent);
+    { /* removexattr */
+        GF_FREE(local->cont.removexattr.name);
+    }
+    { /* xattrop */
+        if (local->cont.xattrop.xattr)
+            dict_unref(local->cont.xattrop.xattr);
+    }
+    { /* symlink */
+        GF_FREE(local->cont.symlink.linkpath);
+    }
 
-        ret = 0;
-out:
-        if (ret) {
-                local->op_ret = -1;
-                local->op_errno = EIO;
-        }
-        return ret;
+    { /* opendir */
+        GF_FREE(local->cont.opendir.checksum);
+    }
+
+    { /* open */
+        if (local->cont.open.fd)
+            fd_unref(local->cont.open.fd);
+    }
+
+    { /* readdirp */
+        if (local->cont.readdir.dict)
+            dict_unref(local->cont.readdir.dict);
+    }
+
+    { /* inodelk */
+        GF_FREE(local->cont.inodelk.volume);
+        if (local->cont.inodelk.xdata)
+            dict_unref(local->cont.inodelk.xdata);
+    }
+
+    { /* entrylk */
+        GF_FREE(local->cont.entrylk.volume);
+        GF_FREE(local->cont.entrylk.basename);
+        if (local->cont.entrylk.xdata)
+            dict_unref(local->cont.entrylk.xdata);
+    }
+
+    if (local->xdata_req)
+        dict_unref(local->xdata_req);
+
+    if (local->xdata_rsp)
+        dict_unref(local->xdata_rsp);
 }
 
 int
-afr_lookup_get_latest_subvol (afr_local_t *local, xlator_t *this)
-{
-        afr_private_t *priv = NULL;
-        int32_t       *success_children = NULL;
-        struct iatt   *bufs = NULL;
-        int           i = 0;
-        int           child = 0;
-        int           lsubvol = -1;
-
-        priv = this->private;
-        success_children = local->cont.lookup.success_children;
-        bufs = local->cont.lookup.bufs;
-        for (i = 0; i < priv->child_count; i++) {
-                child = success_children[i];
-                if (child == -1)
-                        break;
-                if (uuid_is_null (bufs[child].ia_gfid))
-                        continue;
-                if (lsubvol < 0) {
-                        lsubvol = child;
-                } else if (bufs[lsubvol].ia_ctime < bufs[child].ia_ctime) {
-                        lsubvol = child;
-                } else if ((bufs[lsubvol].ia_ctime == bufs[child].ia_ctime) &&
-                  (bufs[lsubvol].ia_ctime_nsec < bufs[child].ia_ctime_nsec)) {
-                        lsubvol = child;
-                }
-        }
-        return lsubvol;
+afr_frame_return(call_frame_t *frame)
+{
+    afr_local_t *local = NULL;
+    int call_count = 0;
+
+    local = frame->local;
+
+    LOCK(&frame->lock);
+    {
+        call_count = --local->call_count;
+    }
+    UNLOCK(&frame->lock);
+
+    return call_count;
 }
 
-void
-afr_lookup_mark_other_entries_stale (afr_local_t *local, xlator_t *this,
-                                     int subvol)
-{
-        afr_private_t *priv = NULL;
-        int32_t       *success_children = NULL;
-        struct iatt   *bufs = NULL;
-        int           i = 0;
-        int           child = 0;
-
-        priv = this->private;
-        success_children = local->cont.lookup.success_children;
-        bufs = local->cont.lookup.bufs;
-        memcpy (local->fresh_children, success_children,
-                sizeof (*success_children) * priv->child_count);
-        for (i = 0; i < priv->child_count; i++) {
-                child = local->fresh_children[i];
-                if (child == -1)
-                        break;
-                if (child == subvol)
-                        continue;
-                if (uuid_is_null (bufs[child].ia_gfid) &&
-                    (bufs[child].ia_type == bufs[subvol].ia_type))
-                        continue;
-                afr_children_rm_child (success_children, child,
-                                       priv->child_count);
-                local->success_count--;
-        }
-        afr_reset_children (local->fresh_children, priv->child_count);
+static char *afr_ignore_xattrs[] = {GF_SELINUX_XATTR_KEY, QUOTA_SIZE_KEY, NULL};
+
+gf_boolean_t
+afr_is_xattr_ignorable(char *key)
+{
+    int i = 0;
+
+    if (!strncmp(key, AFR_XATTR_PREFIX, SLEN(AFR_XATTR_PREFIX)))
+        return _gf_true;
+    for (i = 0; afr_ignore_xattrs[i]; i++) {
+        if (!strcmp(key, afr_ignore_xattrs[i]))
+            return _gf_true;
+    }
+    return _gf_false;
 }
 
-void
-afr_succeed_lookup_on_latest_iatt (afr_local_t *local, xlator_t *this)
+static gf_boolean_t
+afr_xattr_match_needed(dict_t *this, char *key1, data_t *value1, void *data)
 {
-        int    lsubvol = 0;
+    /* Ignore all non-disk (i.e. virtual) xattrs right away. */
+    if (!gf_is_valid_xattr_namespace(key1))
+        return _gf_false;
 
-        if (!afr_lookup_conflicting_entries (local, this))
-                goto out;
+    /* Ignore on-disk xattrs that AFR doesn't need to heal. */
+    if (!afr_is_xattr_ignorable(key1))
+        return _gf_true;
 
-        lsubvol = afr_lookup_get_latest_subvol (local, this);
-        if (lsubvol < 0)
-                goto out;
-        afr_lookup_mark_other_entries_stale (local, this, lsubvol);
-out:
-        return;
+    return _gf_false;
 }
 
 gf_boolean_t
-afr_is_entry_possibly_under_creation (afr_local_t *local, xlator_t *this)
-{
-        /*
-         * We need to perform this test in lookup done and treat on going
-         * create/DELETE as ENOENT.
-         * Reason:
-        Multiple clients A, B and C are attempting 'mkdir -p /mnt/a/b/c'
-
-        1 Client A is in the middle of mkdir(/a). It has acquired lock.
-          It has performed mkdir(/a) on one subvol, and second one is still
-          in progress
-        2 Client B performs a lookup, sees directory /a on one,
-          ENOENT on the other, succeeds lookup.
-        3 Client B performs lookup on /a/b on both subvols, both return ENOENT
-          (one subvol because /a/b does not exist, another because /a
-          itself does not exist)
-        4 Client B proceeds to mkdir /a/b. It obtains entrylk on inode=/a with
-          basename=b on one subvol, but fails on other subvol as /a is yet to
-          be created by Client A.
-        5 Client A finishes mkdir of /a on other subvol
-        6 Client C also attempts to create /a/b, lookup returns ENOENT on
-          both subvols.
-        7 Client C tries to obtain entrylk on on inode=/a with basename=b,
-          obtains on one subvol (where B had failed), and waits for B to unlock
-          on other subvol.
-        8 Client B finishes mkdir() on one subvol with GFID-1 and completes
-          transaction and unlocks
-        9 Client C gets the lock on the second subvol, At this stage second
-          subvol already has /a/b created from Client B, but Client C does not
-          check that in the middle of mkdir transaction
-        10 Client C attempts mkdir /a/b on both subvols. It succeeds on
-           ONLY ONE (where Client B could not get lock because of
-           missing parent /a dir) with GFID-2, and gets EEXIST from ONE subvol.
-        This way we have /a/b in GFID mismatch. One subvol got GFID-1 because
-        Client B performed transaction on only one subvol (because entrylk()
-        could not be obtained on second subvol because of missing parent dir --
-        caused by premature/speculative succeeding of lookup() on /a when locks
-        are detected). Other subvol gets GFID-2 from Client C because while
-        it was waiting for entrylk() on both subvols, Client B was in the
-        middle of creating mkdir() on only one subvol, and Client C does not
-        "expect" this when it is between lock() and pre-op()/op() phase of the
-        transaction.
-         */
-	if (local->cont.lookup.parent_entrylk && local->enoent_count)
-		return _gf_true;
-
-	return _gf_false;
+afr_xattrs_are_equal(dict_t *dict1, dict_t *dict2)
+{
+    return are_dicts_equal(dict1, dict2, afr_xattr_match_needed, NULL);
 }
 
+static int
+afr_get_parent_read_subvol(xlator_t *this, inode_t *parent,
+                           struct afr_reply *replies, unsigned char *readable)
+{
+    int i = 0;
+    int par_read_subvol = -1;
+    int par_read_subvol_iter = -1;
+    afr_private_t *priv = NULL;
 
-static void
-afr_lookup_done (call_frame_t *frame, xlator_t *this)
-{
-        int                 unwind = 1;
-        afr_private_t       *priv  = NULL;
-        afr_local_t         *local = NULL;
-        int                 ret = -1;
-        gf_boolean_t        sh_launched = _gf_false;
-        gf_boolean_t        fail_conflict = _gf_false;
-        int                 gfid_miss_count = 0;
-        int                 enotconn_count = 0;
-        int                 up_children_count = 0;
-
-        priv  = this->private;
-        local = frame->local;
+    priv = this->private;
 
-	if (afr_is_entry_possibly_under_creation (local, this)) {
-		local->op_ret = -1;
-		local->op_errno = ENOENT;
-		goto unwind;
-	}
+    if (parent)
+        par_read_subvol = afr_data_subvol_get(parent, this, NULL, NULL, NULL,
+                                              NULL);
 
-        if (local->op_ret < 0)
-                goto unwind;
-
-        if (local->cont.lookup.parent_entrylk && local->success_count > 1)
-                afr_succeed_lookup_on_latest_iatt (local, this);
-
-        gfid_miss_count = afr_lookup_gfid_missing_count (local, this);
-        up_children_count = afr_up_children_count (local->child_up,
-                                                   priv->child_count);
-        enotconn_count = priv->child_count - up_children_count;
-        if ((gfid_miss_count == local->success_count) &&
-            (enotconn_count > 0)) {
-                local->op_ret = -1;
-                local->op_errno = EIO;
-                gf_log (this->name, GF_LOG_ERROR, "Failing lookup for %s, "
-                        "LOOKUP on a file without gfid is not allowed when "
-                        "some of the children are down", local->loc.path);
-                goto unwind;
-        }
-
-        if ((gfid_miss_count == local->success_count) &&
-            uuid_is_null (local->cont.lookup.gfid_req)) {
-                local->op_ret = -1;
-                local->op_errno = ENODATA;
-                gf_log (this->name, GF_LOG_ERROR, "%s: No gfid present",
-                        local->loc.path);
-                goto unwind;
-        }
-
-        if (gfid_miss_count && uuid_is_null (local->cont.lookup.gfid_req))
-                fail_conflict = _gf_true;
-        ret = afr_lookup_done_success_action (frame, this, fail_conflict);
-        if (ret)
-                goto unwind;
-        uuid_copy (local->self_heal.sh_gfid_req, local->cont.lookup.gfid_req);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid)
+            continue;
 
-        afr_lookup_perform_self_heal (frame, this, &sh_launched);
-        if (sh_launched) {
-                unwind = 0;
-                goto unwind;
-        }
+        if (replies[i].op_ret < 0)
+            continue;
 
- unwind:
-         if (unwind) {
-                 AFR_STACK_UNWIND (lookup, frame, local->op_ret,
-                                   local->op_errno, local->cont.lookup.inode,
-                                   &local->cont.lookup.buf,
-                                   local->cont.lookup.xattr,
-                                   &local->cont.lookup.postparent);
+        if (par_read_subvol_iter == -1) {
+            par_read_subvol_iter = i;
+            continue;
         }
+
+        if ((par_read_subvol_iter != par_read_subvol) && readable[i])
+            par_read_subvol_iter = i;
+
+        if (i == par_read_subvol)
+            par_read_subvol_iter = i;
+    }
+    /* At the end of the for-loop, the only reason why @par_read_subvol_iter
+     * could be -1 is when this LOOKUP has failed on all sub-volumes.
+     * So it is okay to send an arbitrary subvolume (0 in this case)
+     * as parent read subvol.
+     */
+    if (par_read_subvol_iter == -1)
+        par_read_subvol_iter = 0;
+
+    return par_read_subvol_iter;
 }
 
-/*
- * During a lookup, some errors are more "important" than
- * others in that they must be given higher priority while
- * returning to the user.
- *
- * The hierarchy is ESTALE > EIO > ENOENT > others
- */
-int32_t
-afr_most_important_error(int32_t old_errno, int32_t new_errno,
-			 gf_boolean_t eio)
+int
+afr_read_subvol_decide(inode_t *inode, xlator_t *this,
+                       afr_read_subvol_args_t *args, unsigned char *readable)
 {
-	if (old_errno == ESTALE || new_errno == ESTALE)
-		return ESTALE;
-	if (eio && (old_errno == EIO || new_errno == EIO))
-		return EIO;
-	if (old_errno == ENOENT || new_errno == ENOENT)
-		return ENOENT;
+    int event = 0;
+    afr_private_t *priv = NULL;
+    unsigned char *intersection = NULL;
 
-	return new_errno;
-}
+    priv = this->private;
+    intersection = alloca0(priv->child_count);
 
-int32_t
-afr_resultant_errno_get (int32_t *children,
-                         int *child_errno, unsigned int child_count)
-{
-        int     i = 0;
-        int32_t op_errno = 0;
-        int     child = 0;
-
-        for (i = 0; i < child_count; i++) {
-                if (children) {
-                        child = children[i];
-                        if (child == -1)
-                                break;
-                } else {
-                        child = i;
-                }
-		op_errno = afr_most_important_error(op_errno,
-						    child_errno[child],
-						    _gf_false);
-        }
-        return op_errno;
+    afr_readables_intersect_get(inode, this, &event, intersection);
+
+    if (AFR_COUNT(intersection, priv->child_count) <= 0) {
+        /* TODO: If we have one brick with valid data_readable and
+         * another with metadata_readable, try to send an iatt with
+         * valid bits from both.*/
+        return -1;
+    }
+
+    memcpy(readable, intersection, sizeof(*readable) * priv->child_count);
+
+    return afr_read_subvol_select_by_policy(inode, this, intersection, args);
 }
 
-static void
-afr_lookup_handle_error (afr_local_t *local, int32_t op_ret,  int32_t op_errno)
+static inline int
+afr_first_up_child(call_frame_t *frame, xlator_t *this)
 {
-        GF_ASSERT (local);
-        if (op_errno == ENOENT)
-                local->enoent_count++;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int i = 0;
 
-	local->op_errno = afr_most_important_error(local->op_errno, op_errno,
-						   _gf_false);
+    local = frame->local;
+    priv = this->private;
 
-        if (local->op_errno == ESTALE) {
-                local->op_ret = -1;
-        }
+    for (i = 0; i < priv->child_count; i++)
+        if (local->replies[i].valid && local->replies[i].op_ret == 0)
+            return i;
+    return -1;
 }
 
 static void
-afr_set_root_inode_on_first_lookup (afr_local_t *local, xlator_t *this,
-                                    inode_t *inode)
+afr_attempt_readsubvol_set(call_frame_t *frame, xlator_t *this,
+                           unsigned char *success_replies,
+                           unsigned char *data_readable, int *read_subvol)
 {
-        afr_private_t           *priv = NULL;
-        GF_ASSERT (inode);
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int spb_subvol = -1;
+    int child_count = -1;
 
-        if (!__is_root_gfid (inode->gfid))
-                goto out;
-        if (!afr_is_fresh_lookup (&local->loc, this))
-                goto out;
-        priv = this->private;
-        if ((priv->first_lookup)) {
-                gf_log (this->name, GF_LOG_INFO, "added root inode");
-                priv->root_inode = inode_ref (inode);
-                priv->first_lookup = 0;
-        }
-out:
+    if (*read_subvol != -1)
         return;
-}
 
-static void
-afr_lookup_cache_args (afr_local_t *local, int child_index, dict_t *xattr,
-                       struct iatt *buf, struct iatt *postparent)
-{
-        GF_ASSERT (child_index >= 0);
-        local->cont.lookup.xattrs[child_index] = dict_ref (xattr);
-        local->cont.lookup.postparents[child_index] = *postparent;
-        local->cont.lookup.bufs[child_index] = *buf;
+    priv = this->private;
+    local = frame->local;
+    child_count = priv->child_count;
+
+    afr_split_brain_read_subvol_get(local->inode, this, frame, &spb_subvol);
+    if ((spb_subvol >= 0) &&
+        (AFR_COUNT(success_replies, child_count) == child_count)) {
+        *read_subvol = spb_subvol;
+    } else if (!priv->quorum_count ||
+               frame->root->pid == GF_CLIENT_PID_GLFS_HEAL) {
+        *read_subvol = afr_first_up_child(frame, this);
+    } else if (priv->quorum_count &&
+               afr_has_quorum(data_readable, this, NULL)) {
+        /* read_subvol is guaranteed to be valid if we hit this path. */
+        *read_subvol = afr_first_up_child(frame, this);
+    } else {
+        /* If quorum is enabled and we do not have a
+           readable yet, it means all good copies are down.
+        */
+        local->op_ret = -1;
+        local->op_errno = ENOTCONN;
+        gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_READ_SUBVOL_ERROR,
+               "no read "
+               "subvols for %s",
+               local->loc.path);
+    }
+    if (*read_subvol >= 0)
+        dict_del_sizen(local->replies[*read_subvol].xdata, GF_CONTENT_KEY);
 }
 
 static void
-afr_lookup_handle_first_success (afr_local_t *local, xlator_t *this,
-                                 inode_t *inode, struct iatt *buf)
-{
-        local->cont.lookup.inode      = inode_ref (inode);
-        local->cont.lookup.buf        = *buf;
-        afr_set_root_inode_on_first_lookup (local, this, inode);
-}
+afr_lookup_done(call_frame_t *frame, xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int i = -1;
+    int op_errno = 0;
+    int read_subvol = 0;
+    int par_read_subvol = 0;
+    int ret = -1;
+    unsigned char *readable = NULL;
+    unsigned char *success_replies = NULL;
+    int event = 0;
+    struct afr_reply *replies = NULL;
+    uuid_t read_gfid = {
+        0,
+    };
+    gf_boolean_t locked_entry = _gf_false;
+    gf_boolean_t in_flight_create = _gf_false;
+    gf_boolean_t can_interpret = _gf_true;
+    inode_t *parent = NULL;
+    ia_type_t ia_type = IA_INVAL;
+    afr_read_subvol_args_t args = {
+        0,
+    };
+    char *gfid_heal_msg = NULL;
+
+    priv = this->private;
+    local = frame->local;
+    replies = local->replies;
+    parent = local->loc.parent;
+
+    locked_entry = afr_is_possibly_under_txn(AFR_ENTRY_TRANSACTION, local,
+                                             this);
+
+    readable = alloca0(priv->child_count);
+    success_replies = alloca0(priv->child_count);
+
+    afr_inode_read_subvol_get(parent, this, readable, NULL, &event);
+    par_read_subvol = afr_get_parent_read_subvol(this, parent, replies,
+                                                 readable);
+
+    /* First, check if we have a gfid-change from somewhere,
+       If so, propagate that so that a fresh lookup can be
+       issued
+    */
+    if (local->cont.lookup.needs_fresh_lookup) {
+        local->op_ret = -1;
+        local->op_errno = ESTALE;
+        goto error;
+    }
 
-static int32_t
-afr_discovery_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, dict_t *dict,
-                   dict_t *xdata)
-{
-        int              ret            = 0;
-        char            *pathinfo       = NULL;
-        gf_boolean_t     is_local        = _gf_false;
-        afr_private_t   *priv           = NULL;
-        int32_t          child_index    = -1;
+    op_errno = afr_final_errno(frame->local, this->private);
+    local->op_errno = op_errno;
 
-        if (op_ret != 0) {
-                goto out;
+    read_subvol = -1;
+    afr_fill_success_replies(local, priv, success_replies);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid)
+            continue;
+
+        if (replies[i].op_ret == -1) {
+            if (locked_entry && replies[i].op_errno == ENOENT) {
+                in_flight_create = _gf_true;
+            }
+            continue;
         }
 
-        ret = dict_get_str (dict, GF_XATTR_PATHINFO_KEY, &pathinfo);
-        if (ret != 0) {
-                goto out;
+        if (read_subvol == -1 || !readable[read_subvol]) {
+            read_subvol = i;
+            gf_uuid_copy(read_gfid, replies[i].poststat.ia_gfid);
+            ia_type = replies[i].poststat.ia_type;
+            local->op_ret = 0;
         }
+    }
 
-        ret = afr_local_pathinfo (pathinfo, &is_local);
+    if (in_flight_create && !afr_has_quorum(success_replies, this, NULL)) {
+        local->op_ret = -1;
+        local->op_errno = ENOENT;
+        goto error;
+    }
+
+    if (read_subvol == -1)
+        goto error;
+    /* We now have a read_subvol, which is readable[] (if there
+       were any). Next we look for GFID mismatches. We don't
+       consider a GFID mismatch as an error if read_subvol is
+       readable[] but the mismatching GFID subvol is not.
+    */
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret == -1) {
+            continue;
+        }
+
+        if (!gf_uuid_compare(replies[i].poststat.ia_gfid, read_gfid))
+            continue;
+
+        can_interpret = _gf_false;
+
+        if (locked_entry)
+            continue;
+
+        /* Now GFIDs mismatch. It's OK as long as this subvol
+           is not readable[] but read_subvol is */
+        if (readable[read_subvol] && !readable[i])
+            continue;
+
+        /* If we were called from glfsheal and there is still a gfid
+         * mismatch, succeed the lookup and let glfsheal print the
+         * response via gfid-heal-msg.*/
+        if (!dict_get_str_sizen(local->xattr_req, "gfid-heal-msg",
+                                &gfid_heal_msg))
+            goto cant_interpret;
+
+        /* LOG ERROR */
+        local->op_ret = -1;
+        local->op_errno = EIO;
+        goto error;
+    }
+
+    /* Forth, for the finalized GFID, pick the best subvolume
+       to return stats from.
+    */
+    read_subvol = -1;
+    memset(readable, 0, sizeof(*readable) * priv->child_count);
+    if (can_interpret) {
+        if (!afr_has_quorum(success_replies, this, NULL))
+            goto cant_interpret;
+        /* It is safe to call afr_replies_interpret() because we have
+           a response from all the UP subvolumes and all of them resolved
+           to the same GFID
+        */
+        gf_uuid_copy(args.gfid, read_gfid);
+        args.ia_type = ia_type;
+        ret = afr_replies_interpret(frame, this, local->inode, NULL);
+        read_subvol = afr_read_subvol_decide(local->inode, this, &args,
+                                             readable);
+        if (read_subvol == -1)
+            goto cant_interpret;
         if (ret) {
-                goto out;
+            afr_inode_need_refresh_set(local->inode, this);
+            dict_del_sizen(local->replies[read_subvol].xdata, GF_CONTENT_KEY);
+        }
+    } else {
+    cant_interpret:
+        afr_attempt_readsubvol_set(frame, this, success_replies, readable,
+                                   &read_subvol);
+        if (read_subvol == -1) {
+            goto error;
         }
+    }
 
-        priv = this->private;
-        /*
-         * Note that one local subvolume will override another here.  The only
-         * way to avoid that would be to retain extra information about whether
-         * the previous read_child is local, and it's just not worth it.  Even
-         * the slowest local subvolume is far preferable to a remote one.
-         */
-        if (is_local) {
-                child_index = (int32_t)(long)cookie;
-                gf_log (this->name, GF_LOG_INFO,
-                        "selecting local read_child %s",
-                        priv->children[child_index]->name);
-                priv->read_child = child_index;
+    afr_handle_quota_size(frame, this);
+
+    afr_set_need_heal(this, local);
+    if (AFR_IS_ARBITER_BRICK(priv, read_subvol) && local->op_ret == 0) {
+        local->op_ret = -1;
+        local->op_errno = ENOTCONN;
+        gf_msg_debug(this->name, 0,
+                     "Arbiter cannot be a read subvol "
+                     "for %s",
+                     local->loc.path);
+        goto error;
+    }
+
+    ret = dict_get_str_sizen(local->xattr_req, "gfid-heal-msg", &gfid_heal_msg);
+    if (!ret) {
+        ret = dict_set_str_sizen(local->replies[read_subvol].xdata,
+                                 "gfid-heal-msg", gfid_heal_msg);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_DICT_SET_FAILED,
+                   "Error setting gfid-heal-msg dict");
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
         }
+    }
 
-out:
-        STACK_DESTROY(frame->root);
-        return 0;
+    AFR_STACK_UNWIND(lookup, frame, local->op_ret, local->op_errno,
+                     local->inode, &local->replies[read_subvol].poststat,
+                     local->replies[read_subvol].xdata,
+                     &local->replies[par_read_subvol].postparent);
+    return;
+
+error:
+    AFR_STACK_UNWIND(lookup, frame, local->op_ret, local->op_errno, NULL, NULL,
+                     NULL, NULL);
 }
 
-static void
-afr_attempt_local_discovery (xlator_t *this, int32_t child_index)
+/*
+ * During a lookup, some errors are more "important" than
+ * others in that they must be given higher priority while
+ * returning to the user.
+ *
+ * The hierarchy is ENODATA > ENOENT > ESTALE > ENOSPC others
+ */
+
+int
+afr_higher_errno(int32_t old_errno, int32_t new_errno)
 {
-        call_frame_t    *newframe = NULL;
-        loc_t            tmploc = {0,};
-        afr_private_t   *priv = this->private;
+    if (old_errno == ENODATA || new_errno == ENODATA)
+        return ENODATA;
+    if (old_errno == ENOENT || new_errno == ENOENT)
+        return ENOENT;
+    if (old_errno == ESTALE || new_errno == ESTALE)
+        return ESTALE;
+    if (old_errno == ENOSPC || new_errno == ENOSPC)
+        return ENOSPC;
 
-        newframe = create_frame(this,this->ctx->pool);
-        if (!newframe) {
-                return;
-        }
+    return new_errno;
+}
 
-        tmploc.gfid[sizeof(tmploc.gfid)-1] = 1;
-        STACK_WIND_COOKIE (newframe, afr_discovery_cbk,
-                           (void *)(long)child_index,
-                           priv->children[child_index],
-                           priv->children[child_index]->fops->getxattr,
-                           &tmploc, GF_XATTR_PATHINFO_KEY, NULL);
+int
+afr_final_errno(afr_local_t *local, afr_private_t *priv)
+{
+    int i = 0;
+    int op_errno = 0;
+    int tmp_errno = 0;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
+        if (local->replies[i].op_ret >= 0)
+            continue;
+        tmp_errno = local->replies[i].op_errno;
+        op_errno = afr_higher_errno(op_errno, tmp_errno);
+    }
+
+    return op_errno;
+}
+
+static int32_t
+afr_local_discovery_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *dict,
+                        dict_t *xdata)
+{
+    int ret = 0;
+    char *pathinfo = NULL;
+    gf_boolean_t is_local = _gf_false;
+    afr_private_t *priv = NULL;
+    int32_t child_index = -1;
+
+    if (op_ret != 0) {
+        goto out;
+    }
+
+    priv = this->private;
+    child_index = (int32_t)(long)cookie;
+
+    ret = dict_get_str_sizen(dict, GF_XATTR_PATHINFO_KEY, &pathinfo);
+    if (ret != 0) {
+        goto out;
+    }
+
+    ret = glusterfs_is_local_pathinfo(pathinfo, &is_local);
+    if (ret) {
+        goto out;
+    }
+
+    /*
+     * Note that one local subvolume will override another here.  The only
+     * way to avoid that would be to retain extra information about whether
+     * the previous read_child is local, and it's just not worth it.  Even
+     * the slowest local subvolume is far preferable to a remote one.
+     */
+    if (is_local) {
+        priv->local[child_index] = 1;
+        /* Don't set arbiter as read child. */
+        if (AFR_IS_ARBITER_BRICK(priv, child_index))
+            goto out;
+        gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_LOCAL_CHILD,
+               "selecting local read_child %s",
+               priv->children[child_index]->name);
+
+        priv->read_child = child_index;
+    }
+out:
+    STACK_DESTROY(frame->root);
+    return 0;
 }
 
 static void
-afr_lookup_handle_success (afr_local_t *local, xlator_t *this, int32_t child_index,
-                           int32_t op_ret, int32_t op_errno, inode_t *inode,
-                           struct iatt *buf, dict_t *xattr,
-                           struct iatt *postparent)
+afr_attempt_local_discovery(xlator_t *this, int32_t child_index)
 {
-        afr_private_t   *priv   = this->private;
+    call_frame_t *newframe = NULL;
+    loc_t tmploc = {
+        0,
+    };
+    afr_private_t *priv = this->private;
 
-        if (local->success_count == 0) {
-                if (local->op_errno != ESTALE) {
-                        local->op_ret = op_ret;
-                        local->op_errno = 0;
-                }
-                afr_lookup_handle_first_success (local, this, inode, buf);
-        }
-        afr_lookup_update_lk_counts (local, this,
-                                     child_index, xattr);
+    newframe = create_frame(this, this->ctx->pool);
+    if (!newframe) {
+        return;
+    }
 
-        afr_lookup_cache_args (local, child_index, xattr,
-                               buf, postparent);
+    tmploc.gfid[sizeof(tmploc.gfid) - 1] = 1;
+    STACK_WIND_COOKIE(newframe, afr_local_discovery_cbk,
+                      (void *)(long)child_index, priv->children[child_index],
+                      priv->children[child_index]->fops->getxattr, &tmploc,
+                      GF_XATTR_PATHINFO_KEY, NULL);
+}
 
-        if (local->do_discovery && (priv->read_child == (-1))) {
-                afr_attempt_local_discovery(this,child_index);
-        }
+int
+afr_lookup_sh_metadata_wrap(void *opaque)
+{
+    call_frame_t *frame = opaque;
+    afr_local_t *local = NULL;
+    xlator_t *this = NULL;
+    inode_t *inode = NULL;
+    afr_private_t *priv = NULL;
+    struct afr_reply *replies = NULL;
+    int i = 0, first = -1;
+    int ret = -1;
+    dict_t *dict = NULL;
+
+    local = frame->local;
+    this = frame->this;
+    priv = this->private;
+    replies = local->replies;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret == -1)
+            continue;
+        first = i;
+        break;
+    }
+    if (first == -1)
+        goto out;
+
+    if (afr_selfheal_metadata_by_stbuf(this, &replies[first].poststat))
+        goto out;
+
+    afr_local_replies_wipe(local, this->private);
+
+    dict = dict_new();
+    if (!dict)
+        goto out;
+    if (local->xattr_req) {
+        dict_copy(local->xattr_req, dict);
+    }
+
+    ret = dict_set_sizen_str_sizen(dict, "link-count", GF_XATTROP_INDEX_COUNT);
+    if (ret) {
+        gf_msg_debug(this->name, -ret, "Unable to set link-count in dict ");
+    }
+
+    if (loc_is_nameless(&local->loc)) {
+        ret = afr_selfheal_unlocked_discover_on(frame, local->inode,
+                                                local->loc.gfid, local->replies,
+                                                local->child_up, dict);
+    } else {
+        inode = afr_selfheal_unlocked_lookup_on(frame, local->loc.parent,
+                                                local->loc.name, local->replies,
+                                                local->child_up, dict);
+    }
+    if (inode)
+        inode_unref(inode);
+out:
+    if (loc_is_nameless(&local->loc))
+        afr_discover_done(frame, this);
+    else
+        afr_lookup_done(frame, this);
 
-        local->cont.lookup.success_children[local->success_count] = child_index;
-        local->success_count++;
+    if (dict)
+        dict_unref(dict);
+
+    return 0;
 }
 
-int
-afr_lookup_cbk (call_frame_t *frame, void *cookie,
-                xlator_t *this,  int32_t op_ret,  int32_t op_errno,
-                inode_t *inode,   struct iatt *buf, dict_t *xattr,
-                struct iatt *postparent)
+gf_boolean_t
+afr_is_pending_set(xlator_t *this, dict_t *xdata, int type)
 {
-        afr_local_t *   local = NULL;
-        int             call_count      = -1;
-        int             child_index     = -1;
+    int idx = -1;
+    afr_private_t *priv = NULL;
+    void *pending_raw = NULL;
+    int *pending_int = NULL;
+    int i = 0;
 
-         child_index = (long) cookie;
+    priv = this->private;
+    idx = afr_index_for_transaction_type(type);
 
-        LOCK (&frame->lock);
-        {
-                local = frame->local;
+    if (dict_get_ptr(xdata, AFR_DIRTY, &pending_raw) == 0) {
+        if (pending_raw) {
+            pending_int = pending_raw;
 
-                if (op_ret == -1) {
-                        afr_lookup_handle_error (local, op_ret, op_errno);
-                        goto unlock;
-                }
-                afr_lookup_handle_success (local, this, child_index, op_ret,
-                                           op_errno, inode, buf, xattr,
-                                           postparent);
+            if (ntoh32(pending_int[idx]))
+                return _gf_true;
+        }
+    }
 
-         }
-unlock:
-        UNLOCK (&frame->lock);
+    for (i = 0; i < priv->child_count; i++) {
+        if (dict_get_ptr(xdata, priv->pending_key[i], &pending_raw))
+            continue;
+        if (!pending_raw)
+            continue;
+        pending_int = pending_raw;
 
-        call_count = afr_frame_return (frame);
-        if (call_count == 0) {
-               afr_lookup_done (frame, this);
-        }
+        if (ntoh32(pending_int[idx]))
+            return _gf_true;
+    }
 
-         return 0;
+    return _gf_false;
 }
 
-int
-afr_lookup_cont_init (afr_local_t *local, unsigned int child_count)
-{
-        int               ret            = -ENOMEM;
-        struct iatt       *iatts         = NULL;
-        int32_t           *success_children = NULL;
-        int32_t           *sources       = NULL;
-        int32_t           **pending_matrix = NULL;
-
-        GF_ASSERT (local);
-        local->cont.lookup.xattrs = GF_CALLOC (child_count,
-                                               sizeof (*local->cont.lookup.xattr),
-                                               gf_afr_mt_dict_t);
-        if (NULL == local->cont.lookup.xattrs)
-                goto out;
+static gf_boolean_t
+afr_can_start_metadata_self_heal(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    struct afr_reply *replies = NULL;
+    int i = 0, first = -1;
+    gf_boolean_t start = _gf_false;
+    struct iatt stbuf = {
+        0,
+    };
+
+    local = frame->local;
+    replies = local->replies;
+    priv = this->private;
+
+    if (!priv->metadata_self_heal)
+        return _gf_false;
 
-        iatts = GF_CALLOC (child_count, sizeof (*iatts), gf_afr_mt_iatt);
-        if (NULL == iatts)
-                goto out;
-        local->cont.lookup.postparents = iatts;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret == -1)
+            continue;
+        if (first == -1) {
+            first = i;
+            stbuf = replies[i].poststat;
+            continue;
+        }
 
-        iatts = GF_CALLOC (child_count, sizeof (*iatts), gf_afr_mt_iatt);
-        if (NULL == iatts)
-                goto out;
-        local->cont.lookup.bufs = iatts;
+        if (afr_is_pending_set(this, replies[i].xdata,
+                               AFR_METADATA_TRANSACTION)) {
+            /* Let shd do the heal so that lookup is not blocked
+             * on getting metadata lock/doing the heal */
+            start = _gf_false;
+            break;
+        }
 
-        success_children = afr_children_create (child_count);
-        if (NULL == success_children)
-                goto out;
-        local->cont.lookup.success_children = success_children;
+        if (gf_uuid_compare(stbuf.ia_gfid, replies[i].poststat.ia_gfid)) {
+            start = _gf_false;
+            break;
+        }
+        if (!IA_EQUAL(stbuf, replies[i].poststat, type)) {
+            start = _gf_false;
+            break;
+        }
 
-        local->fresh_children = afr_children_create (child_count);
-        if (NULL == local->fresh_children)
-                goto out;
+        /*Check if iattrs need heal*/
+        if ((!IA_EQUAL(stbuf, replies[i].poststat, uid)) ||
+            (!IA_EQUAL(stbuf, replies[i].poststat, gid)) ||
+            (!IA_EQUAL(stbuf, replies[i].poststat, prot))) {
+            start = _gf_true;
+            continue;
+        }
 
-        sources = GF_CALLOC (sizeof (*sources), child_count, gf_afr_mt_int32_t);
-        if (NULL == sources)
-                goto out;
-        local->cont.lookup.sources = sources;
+        /*Check if xattrs need heal*/
+        if (!afr_xattrs_are_equal(replies[first].xdata, replies[i].xdata))
+            start = _gf_true;
+    }
 
-        pending_matrix = afr_matrix_create (child_count, child_count);
-        if (NULL == pending_matrix)
-                goto out;
-        local->cont.lookup.pending_matrix = pending_matrix;
+    return start;
+}
 
-        ret = 0;
+int
+afr_lookup_metadata_heal_check(call_frame_t *frame, xlator_t *this)
+
+{
+    call_frame_t *heal = NULL;
+    afr_local_t *local = NULL;
+    int ret = 0;
+
+    local = frame->local;
+    if (!afr_can_start_metadata_self_heal(frame, this))
+        goto out;
+
+    heal = afr_frame_create(this, &ret);
+    if (!heal) {
+        ret = -ret;
+        goto out;
+    }
+
+    ret = synctask_new(this->ctx->env, afr_lookup_sh_metadata_wrap,
+                       afr_refresh_selfheal_done, heal, frame);
+    if (ret)
+        goto out;
+    return ret;
 out:
-        return ret;
+    if (loc_is_nameless(&local->loc))
+        afr_discover_done(frame, this);
+    else
+        afr_lookup_done(frame, this);
+    if (heal)
+        AFR_STACK_DESTROY(heal);
+    return ret;
 }
 
 int
-afr_lookup (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, dict_t *xattr_req)
+afr_lookup_selfheal_wrap(void *opaque)
 {
-        afr_private_t    *priv           = NULL;
-        afr_local_t      *local          = NULL;
-        void              *gfid_req      = NULL;
-        int               ret            = -1;
-        int               i              = 0;
-        int               call_count     = 0;
-        uint64_t          ctx            = 0;
-        int32_t           op_errno       = 0;
+    int ret = 0;
+    call_frame_t *frame = opaque;
+    afr_local_t *local = NULL;
+    xlator_t *this = NULL;
+    inode_t *inode = NULL;
+    uuid_t pargfid = {
+        0,
+    };
 
-        priv = this->private;
+    local = frame->local;
+    this = frame->this;
+    loc_pargfid(&local->loc, pargfid);
 
-        AFR_LOCAL_ALLOC_OR_GOTO (local, out);
+    ret = afr_selfheal_name(frame->this, pargfid, local->loc.name,
+                            &local->cont.lookup.gfid_req, local->xattr_req);
+    if (ret == -EIO)
+        goto unwind;
 
-        local->op_ret = -1;
+    afr_local_replies_wipe(local, this->private);
 
-        frame->local = local;
-        local->fop = GF_FOP_LOOKUP;
+    inode = afr_selfheal_unlocked_lookup_on(frame, local->loc.parent,
+                                            local->loc.name, local->replies,
+                                            local->child_up, local->xattr_req);
+    if (inode)
+        inode_unref(inode);
 
-        loc_copy (&local->loc, loc);
-        ret = loc_path (&local->loc, NULL);
-        if (ret < 0) {
-                op_errno = EINVAL;
-                goto out;
-        }
-
-        if (local->loc.path &&
-            (strcmp (local->loc.path, "/" GF_REPLICATE_TRASH_DIR) == 0)) {
-                op_errno = EPERM;
-                ret = -1;
-                goto out;
-        }
+    afr_lookup_metadata_heal_check(frame, this);
+    return 0;
 
-        ret = inode_ctx_get (local->loc.inode, this, &ctx);
-        if (ret == 0) {
-                /* lookup is a revalidate */
+unwind:
+    AFR_STACK_UNWIND(lookup, frame, -1, EIO, NULL, NULL, NULL, NULL);
+    return 0;
+}
 
-                local->read_child_index = afr_inode_get_read_ctx (this,
-                                                               local->loc.inode,
-                                                               NULL);
+int
+afr_lookup_entry_heal(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    call_frame_t *heal = NULL;
+    int i = 0, first = -1;
+    gf_boolean_t name_state_mismatch = _gf_false;
+    struct afr_reply *replies = NULL;
+    int ret = 0;
+    unsigned char *par_readables = NULL;
+    unsigned char *success = NULL;
+    int32_t op_errno = 0;
+    uuid_t gfid = {0};
+
+    local = frame->local;
+    replies = local->replies;
+    priv = this->private;
+    par_readables = alloca0(priv->child_count);
+    success = alloca0(priv->child_count);
+
+    ret = afr_inode_read_subvol_get(local->loc.parent, this, par_readables,
+                                    NULL, NULL);
+    if (ret < 0 || AFR_COUNT(par_readables, priv->child_count) == 0) {
+        /* In this case set par_readables to all 1 so that name_heal
+         * need checks at the end of this function will flag missing
+         * entry when name state mismatches*/
+        memset(par_readables, 1, priv->child_count);
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid)
+            continue;
+
+        if (replies[i].op_ret == 0) {
+            if (gf_uuid_is_null(gfid)) {
+                gf_uuid_copy(gfid, replies[i].poststat.ia_gfid);
+            }
+            success[i] = 1;
         } else {
-                LOCK (&priv->read_child_lock);
-                {
-                        if (priv->hash_mode) {
-                                local->read_child_index = -1;
-                        }
-                        else {
-                                local->read_child_index =
-                                        (++priv->read_child_rr) %
-                                        (priv->child_count);
-                        }
-                }
-                UNLOCK (&priv->read_child_lock);
-                local->cont.lookup.fresh_lookup = _gf_true;
+            if ((replies[i].op_errno != ENOTCONN) &&
+                (replies[i].op_errno != ENOENT) &&
+                (replies[i].op_errno != ESTALE)) {
+                op_errno = replies[i].op_errno;
+            }
         }
 
-        local->child_up = memdup (priv->child_up,
-                                  sizeof (*local->child_up) * priv->child_count);
-        if (NULL == local->child_up) {
-                op_errno = ENOMEM;
-                goto out;
+        /*gfid is missing, needs heal*/
+        if ((replies[i].op_ret == -1) && (replies[i].op_errno == ENODATA)) {
+            goto name_heal;
         }
 
-        ret = afr_lookup_cont_init (local, priv->child_count);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
+        if (first == -1) {
+            first = i;
+            continue;
         }
 
-        local->call_count = afr_up_children_count (local->child_up,
-                                                   priv->child_count);
-        call_count = local->call_count;
-        if (local->call_count == 0) {
-                ret      = -1;
-                op_errno = ENOTCONN;
-                goto out;
+        if (replies[i].op_ret != replies[first].op_ret) {
+            name_state_mismatch = _gf_true;
         }
 
-        /* By default assume ENOTCONN. On success it will be set to 0. */
-        local->op_errno = ENOTCONN;
-
-        ret = afr_lookup_xattr_req_prepare (local, this, xattr_req, &local->loc,
-                                            &gfid_req);
-        if (ret) {
-                local->op_errno = -ret;
-                goto out;
-        }
-        afr_lookup_save_gfid (local->cont.lookup.gfid_req, gfid_req,
-                              &local->loc);
-        local->fop = GF_FOP_LOOKUP;
-        if (priv->choose_local && !priv->did_discovery) {
-                if (gfid_req && __is_root_gfid(gfid_req)) {
-                        local->do_discovery = _gf_true;
-                        priv->did_discovery = _gf_true;
-                }
+        if (replies[i].op_ret == 0) {
+            /* Rename after this lookup may succeed if we don't do
+             * a name-heal and the destination may not have pending xattrs
+             * to indicate which name is good and which is bad so always do
+             * this heal*/
+            if (gf_uuid_compare(replies[i].poststat.ia_gfid, gfid)) {
+                goto name_heal;
+            }
         }
+    }
+
+    if (name_state_mismatch) {
+        if (!priv->quorum_count)
+            goto name_heal;
+        if (!afr_has_quorum(success, this, NULL))
+            goto name_heal;
+        if (op_errno)
+            goto name_heal;
         for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND_COOKIE (frame, afr_lookup_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->lookup,
-                                           &local->loc, local->xattr_req);
-                        if (!--call_count)
-                                break;
-                }
+            if (!replies[i].valid)
+                continue;
+            if (par_readables[i] && replies[i].op_ret < 0 &&
+                replies[i].op_errno != ENOTCONN) {
+                goto name_heal;
+            }
         }
+    }
 
-        ret = 0;
-out:
-        if (ret)
-                AFR_STACK_UNWIND (lookup, frame, -1, op_errno,
-                                  NULL, NULL, NULL, NULL);
-
-        return 0;
-}
+    goto metadata_heal;
 
+name_heal:
+    heal = afr_frame_create(this, NULL);
+    if (!heal)
+        goto metadata_heal;
 
-/* {{{ open */
+    ret = synctask_new(this->ctx->env, afr_lookup_selfheal_wrap,
+                       afr_refresh_selfheal_done, heal, frame);
+    if (ret) {
+        AFR_STACK_DESTROY(heal);
+        goto metadata_heal;
+    }
+    return ret;
 
-int
-__afr_fd_ctx_set (xlator_t *this, fd_t *fd)
-{
-        afr_private_t * priv   = NULL;
-        int             ret    = -1;
-        uint64_t        ctx    = 0;
-        afr_fd_ctx_t *  fd_ctx = NULL;
+metadata_heal:
+    ret = afr_lookup_metadata_heal_check(frame, this);
 
-        VALIDATE_OR_GOTO (this->private, out);
-        VALIDATE_OR_GOTO (fd, out);
+    return ret;
+}
 
-        priv = this->private;
+int
+afr_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, inode_t *inode, struct iatt *buf, dict_t *xdata,
+               struct iatt *postparent)
+{
+    afr_local_t *local = NULL;
+    int call_count = -1;
+    int child_index = -1;
+    GF_UNUSED int ret = 0;
+    int8_t need_heal = 1;
+
+    child_index = (long)cookie;
+
+    local = frame->local;
+
+    local->replies[child_index].valid = 1;
+    local->replies[child_index].op_ret = op_ret;
+    local->replies[child_index].op_errno = op_errno;
+    /*
+     * On revalidate lookup if the gfid-changed, afr should unwind the fop
+     * with ESTALE so that a fresh lookup will be sent by the top xlator.
+     * So remember it.
+     */
+    if (xdata && dict_get_sizen(xdata, "gfid-changed"))
+        local->cont.lookup.needs_fresh_lookup = _gf_true;
+
+    if (xdata) {
+        ret = dict_get_int8(xdata, "link-count", &need_heal);
+        local->replies[child_index].need_heal = need_heal;
+    } else {
+        local->replies[child_index].need_heal = need_heal;
+    }
+    if (op_ret != -1) {
+        local->replies[child_index].poststat = *buf;
+        local->replies[child_index].postparent = *postparent;
+        if (xdata)
+            local->replies[child_index].xdata = dict_ref(xdata);
+    }
+
+    call_count = afr_frame_return(frame);
+    if (call_count == 0) {
+        afr_set_need_heal(this, local);
+        afr_lookup_entry_heal(frame, this);
+    }
+
+    return 0;
+}
 
-        ret = __fd_ctx_get (fd, this, &ctx);
+static void
+afr_discover_unwind(call_frame_t *frame, xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int read_subvol = -1;
+    int ret = 0;
+    unsigned char *data_readable = NULL;
+    unsigned char *success_replies = NULL;
 
-        if (ret == 0)
-                goto out;
+    priv = this->private;
+    local = frame->local;
+    data_readable = alloca0(priv->child_count);
+    success_replies = alloca0(priv->child_count);
 
-        fd_ctx = GF_CALLOC (1, sizeof (afr_fd_ctx_t),
-                            gf_afr_mt_afr_fd_ctx_t);
-        if (!fd_ctx) {
-                ret = -ENOMEM;
-                goto out;
-        }
+    afr_fill_success_replies(local, priv, success_replies);
+    if (AFR_COUNT(success_replies, priv->child_count) > 0)
+        local->op_ret = 0;
 
-        fd_ctx->pre_op_done = GF_CALLOC (sizeof (*fd_ctx->pre_op_done),
-                                         priv->child_count,
-                                         gf_afr_mt_char);
-        if (!fd_ctx->pre_op_done) {
-                ret = -ENOMEM;
-                goto out;
-        }
+    if (local->op_ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = afr_final_errno(frame->local, this->private);
+        goto error;
+    }
 
-        fd_ctx->pre_op_piggyback = GF_CALLOC (sizeof (*fd_ctx->pre_op_piggyback),
-                                              priv->child_count,
-                                              gf_afr_mt_char);
-        if (!fd_ctx->pre_op_piggyback) {
-                ret = -ENOMEM;
-                goto out;
-        }
+    if (!afr_has_quorum(success_replies, this, frame))
+        goto unwind;
 
-        fd_ctx->opened_on = GF_CALLOC (sizeof (*fd_ctx->opened_on),
-                                       priv->child_count,
-                                       gf_afr_mt_int32_t);
-        if (!fd_ctx->opened_on) {
-                ret = -ENOMEM;
-                goto out;
-        }
+    ret = afr_replies_interpret(frame, this, local->inode, NULL);
+    if (ret) {
+        afr_inode_need_refresh_set(local->inode, this);
+    }
 
-        fd_ctx->lock_piggyback = GF_CALLOC (sizeof (*fd_ctx->lock_piggyback),
-                                            priv->child_count,
-                                            gf_afr_mt_char);
-        if (!fd_ctx->lock_piggyback) {
-                ret = -ENOMEM;
-                goto out;
-        }
+    read_subvol = afr_read_subvol_decide(local->inode, this, NULL,
+                                         data_readable);
 
-        fd_ctx->lock_acquired = GF_CALLOC (sizeof (*fd_ctx->lock_acquired),
-                                           priv->child_count,
-                                           gf_afr_mt_char);
-        if (!fd_ctx->lock_acquired) {
-                ret = -ENOMEM;
-                goto out;
-        }
+unwind:
+    afr_attempt_readsubvol_set(frame, this, success_replies, data_readable,
+                               &read_subvol);
+    if (read_subvol == -1)
+        goto error;
 
-        fd_ctx->up_count   = priv->up_count;
-        fd_ctx->down_count = priv->down_count;
+    if (AFR_IS_ARBITER_BRICK(priv, read_subvol) && local->op_ret == 0) {
+        local->op_ret = -1;
+        local->op_errno = ENOTCONN;
+        gf_msg_debug(this->name, 0,
+                     "Arbiter cannot be a read subvol "
+                     "for %s",
+                     local->loc.path);
+    }
 
-        fd_ctx->locked_on = GF_CALLOC (sizeof (*fd_ctx->locked_on),
-                                       priv->child_count,
-                                       gf_afr_mt_char);
-        if (!fd_ctx->locked_on) {
-                ret = -ENOMEM;
-                goto out;
-        }
+    AFR_STACK_UNWIND(lookup, frame, local->op_ret, local->op_errno,
+                     local->inode, &local->replies[read_subvol].poststat,
+                     local->replies[read_subvol].xdata,
+                     &local->replies[read_subvol].postparent);
+    return;
 
-	pthread_mutex_init (&fd_ctx->delay_lock, NULL);
-        INIT_LIST_HEAD (&fd_ctx->entries);
-        fd_ctx->call_child = -1;
+error:
+    AFR_STACK_UNWIND(lookup, frame, local->op_ret, local->op_errno, NULL, NULL,
+                     NULL, NULL);
+}
 
-        INIT_LIST_HEAD (&fd_ctx->eager_locked);
+static int
+afr_ta_id_file_check(void *opaque)
+{
+    afr_private_t *priv = NULL;
+    xlator_t *this = NULL;
+    loc_t loc = {
+        0,
+    };
+    struct iatt stbuf = {
+        0,
+    };
+    dict_t *dict = NULL;
+    uuid_t gfid = {
+        0,
+    };
+    fd_t *fd = NULL;
+    int ret = 0;
+
+    this = opaque;
+    priv = this->private;
+
+    ret = afr_fill_ta_loc(this, &loc, _gf_false);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed to populate thin-arbiter loc for: %s.", loc.name);
+        goto out;
+    }
+
+    ret = syncop_lookup(priv->children[THIN_ARBITER_BRICK_INDEX], &loc, &stbuf,
+                        0, 0, 0);
+    if (ret == 0) {
+        goto out;
+    } else if (ret == -ENOENT) {
+        fd = fd_create(loc.inode, getpid());
+        if (!fd)
+            goto out;
+        dict = dict_new();
+        if (!dict)
+            goto out;
+        gf_uuid_generate(gfid);
+        ret = dict_set_gfuuid(dict, "gfid-req", gfid, true);
+        ret = syncop_create(priv->children[THIN_ARBITER_BRICK_INDEX], &loc,
+                            O_RDWR, 0664, fd, &stbuf, dict, NULL);
+    }
 
-        ret = __fd_ctx_set (fd, this, (uint64_t)(long) fd_ctx);
-        if (ret)
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "failed to set fd ctx (%p)", fd);
 out:
-        return ret;
+    if (ret == 0) {
+        gf_uuid_copy(priv->ta_gfid, stbuf.ia_gfid);
+    } else {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed to lookup/create thin-arbiter id file.");
+    }
+    if (dict)
+        dict_unref(dict);
+    if (fd)
+        fd_unref(fd);
+    loc_wipe(&loc);
+
+    return 0;
 }
 
+static int
+afr_ta_id_file_check_cbk(int ret, call_frame_t *ta_frame, void *opaque)
+{
+    return 0;
+}
 
-int
-afr_fd_ctx_set (xlator_t *this, fd_t *fd)
+static void
+afr_discover_done(call_frame_t *frame, xlator_t *this)
 {
-        int ret = -1;
+    int ret = 0;
+    afr_private_t *priv = NULL;
 
-        LOCK (&fd->lock);
-        {
-                ret = __afr_fd_ctx_set (this, fd);
-        }
-        UNLOCK (&fd->lock);
+    priv = this->private;
+    if (!priv->thin_arbiter_count)
+        goto unwind;
+    if (!gf_uuid_is_null(priv->ta_gfid))
+        goto unwind;
 
-        return ret;
+    ret = synctask_new(this->ctx->env, afr_ta_id_file_check,
+                       afr_ta_id_file_check_cbk, NULL, this);
+    if (ret)
+        goto unwind;
+unwind:
+    afr_discover_unwind(frame, this);
 }
 
-/* {{{ flush */
-
 int
-afr_flush_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
-              int32_t op_ret, int32_t op_errno, dict_t *xdata)
+afr_discover_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, inode_t *inode, struct iatt *buf, dict_t *xdata,
+                 struct iatt *postparent)
 {
-        afr_local_t *   local = NULL;
-        int call_count  = -1;
-
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-                        }
-                        local->success_count++;
-                }
+    afr_local_t *local = NULL;
+    int call_count = -1;
+    int child_index = -1;
+    GF_UNUSED int ret = 0;
+    int8_t need_heal = 1;
 
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+    child_index = (long)cookie;
 
-	call_count = afr_frame_return (frame);
+    local = frame->local;
 
-	if (call_count == 0)
-		AFR_STACK_UNWIND(flush, frame, local->op_ret,
-				 local->op_errno, NULL);
+    local->replies[child_index].valid = 1;
+    local->replies[child_index].op_ret = op_ret;
+    local->replies[child_index].op_errno = op_errno;
+    if (op_ret != -1) {
+        local->replies[child_index].poststat = *buf;
+        local->replies[child_index].postparent = *postparent;
+        if (xdata)
+            local->replies[child_index].xdata = dict_ref(xdata);
+    }
 
-        return 0;
-}
+    if (local->do_discovery && (op_ret == 0))
+        afr_attempt_local_discovery(this, child_index);
 
-static int
-afr_flush_wrapper (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
-{
-        int           i      = 0;
-        afr_local_t   *local = NULL;
-        afr_private_t *priv  = NULL;
-        int call_count       = -1;
+    if (xdata) {
+        ret = dict_get_int8(xdata, "link-count", &need_heal);
+        local->replies[child_index].need_heal = need_heal;
+    } else {
+        local->replies[child_index].need_heal = need_heal;
+    }
 
-        priv = this->private;
-        local = frame->local;
-        call_count = local->call_count;
+    call_count = afr_frame_return(frame);
+    if (call_count == 0) {
+        afr_set_need_heal(this, local);
+        afr_lookup_metadata_heal_check(frame, this);
+    }
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND_COOKIE (frame, afr_flush_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->flush,
-                                           local->fd, NULL);
-                        if (!--call_count)
-                                break;
+    return 0;
+}
 
-                }
+int
+afr_discover_do(call_frame_t *frame, xlator_t *this, int err)
+{
+    int ret = 0;
+    int i = 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int call_count = 0;
+
+    local = frame->local;
+    priv = this->private;
+
+    if (err) {
+        local->op_errno = err;
+        goto out;
+    }
+
+    call_count = local->call_count = AFR_COUNT(local->child_up,
+                                               priv->child_count);
+
+    ret = afr_lookup_xattr_req_prepare(local, this, local->xattr_req,
+                                       &local->loc);
+    if (ret) {
+        local->op_errno = -ret;
+        goto out;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->child_up[i]) {
+            STACK_WIND_COOKIE(
+                frame, afr_discover_cbk, (void *)(long)i, priv->children[i],
+                priv->children[i]->fops->lookup, &local->loc, local->xattr_req);
+            if (!--call_count)
+                break;
         }
+    }
 
-        return 0;
+    return 0;
+out:
+    AFR_STACK_UNWIND(lookup, frame, -1, local->op_errno, 0, 0, 0, 0);
+    return 0;
 }
 
 int
-afr_flush (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+afr_discover(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req)
 {
-        afr_private_t *priv  = NULL;
-        afr_local_t   *local = NULL;
-        call_stub_t   *stub = NULL;
-        int            ret        = -1;
-        int            op_errno   = 0;
+    int op_errno = ENOMEM;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int event = 0;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    priv = this->private;
 
-        priv = this->private;
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-	AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-	local = frame->local;
+    if (!local->call_count) {
+        op_errno = ENOTCONN;
+        goto out;
+    }
 
-	ret = afr_local_init(local, priv, &op_errno);
-	if (ret < 0)
-		goto out;
+    if (__is_root_gfid(loc->inode->gfid)) {
+        if (!priv->root_inode)
+            priv->root_inode = inode_ref(loc->inode);
 
-	local->fd = fd_ref(fd);
-        stub = fop_flush_stub (frame, afr_flush_wrapper, fd, xdata);
-        if (!stub) {
-                ret = -1;
-                op_errno = ENOMEM;
-                goto out;
+        if (priv->choose_local && !priv->did_discovery) {
+            /* Logic to detect which subvolumes of AFR are
+               local, in order to prefer them for reads
+            */
+            local->do_discovery = _gf_true;
+            priv->did_discovery = _gf_true;
         }
+    }
 
-        afr_delayed_changelog_wake_resume (this, fd, stub);
-	ret = 0;
+    local->op = GF_FOP_LOOKUP;
 
-out:
-	if (ret < 0)
-		AFR_STACK_UNWIND(flush, frame, -1, op_errno, NULL);
+    loc_copy(&local->loc, loc);
+
+    local->inode = inode_ref(loc->inode);
 
+    if (xattr_req) {
+        /* If xattr_req was null, afr_lookup_xattr_req_prepare() will
+           allocate one for us */
+        local->xattr_req = dict_copy_with_ref(xattr_req, NULL);
+        if (!local->xattr_req) {
+            op_errno = ENOMEM;
+            goto out;
+        }
+    }
+
+    if (gf_uuid_is_null(loc->inode->gfid)) {
+        afr_discover_do(frame, this, 0);
         return 0;
+    }
+
+    afr_read_subvol_get(loc->inode, this, NULL, NULL, &event,
+                        AFR_DATA_TRANSACTION, NULL);
+
+    afr_discover_do(frame, this, 0);
+
+    return 0;
+out:
+    AFR_STACK_UNWIND(lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+    return 0;
 }
 
-/* }}} */
+int
+afr_lookup_do(call_frame_t *frame, xlator_t *this, int err)
+{
+    int ret = 0;
+    int i = 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int call_count = 0;
+
+    local = frame->local;
+    priv = this->private;
+
+    if (err < 0) {
+        local->op_errno = err;
+        goto out;
+    }
+
+    call_count = local->call_count = AFR_COUNT(local->child_up,
+                                               priv->child_count);
+
+    ret = afr_lookup_xattr_req_prepare(local, this, local->xattr_req,
+                                       &local->loc);
+    if (ret) {
+        local->op_errno = -ret;
+        goto out;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->child_up[i]) {
+            STACK_WIND_COOKIE(
+                frame, afr_lookup_cbk, (void *)(long)i, priv->children[i],
+                priv->children[i]->fops->lookup, &local->loc, local->xattr_req);
+            if (!--call_count)
+                break;
+        }
+    }
+    return 0;
+out:
+    AFR_STACK_UNWIND(lookup, frame, -1, local->op_errno, 0, 0, 0, 0);
+    return 0;
+}
 
+/*
+ * afr_lookup()
+ *
+ * The goal here is to figure out what the element getting looked up is.
+ * i.e what is the GFID, inode type and a conservative estimate of the
+ * inode attributes are.
+ *
+ * As we lookup, operations may be underway on the entry name and the
+ * inode. In lookup() we are primarily concerned only with the entry
+ * operations. If the entry is getting unlinked or renamed, we detect
+ * what operation is underway by querying for on-going transactions and
+ * pending self-healing on the entry through xdata.
+ *
+ * If the entry is a file/dir, it may need self-heal and/or in a
+ * split-brain condition. Lookup is not the place to worry about these
+ * conditions. Outcast marking will naturally handle them in the read
+ * paths.
+ *
+ * Here is a brief goal of what we are trying to achieve:
+ *
+ * - LOOKUP on all subvolumes concurrently, querying on-going transaction
+ *   and pending self-heal info from the servers.
+ *
+ * - If all servers reply the same inode type and GFID, the overall call
+ *   MUST be a success.
+ *
+ * - If inode types or GFIDs mismatch, and there IS either an on-going
+ *   transaction or pending self-heal, inspect what the nature of the
+ *   transaction or pending heal is, and select the appropriate subvolume's
+ *   reply as the winner.
+ *
+ * - If inode types or GFIDs mismatch, and there are no on-going transactions
+ *   or pending self-heal on the entry name on any of the servers, fail the
+ *   lookup with EIO. Something has gone wrong beyond reasonable action.
+ */
 
 int
-afr_cleanup_fd_ctx (xlator_t *this, fd_t *fd)
+afr_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req)
 {
-        uint64_t        ctx = 0;
-        afr_fd_ctx_t    *fd_ctx = NULL;
-        int             ret = 0;
-
-        ret = fd_ctx_get (fd, this, &ctx);
-        if (ret < 0)
-                goto out;
+    afr_local_t *local = NULL;
+    int32_t op_errno = 0;
+    int event = 0;
+    int ret = 0;
 
-        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+    if (loc_is_nameless(loc)) {
+        if (xattr_req)
+            dict_del_sizen(xattr_req, "gfid-req");
+        afr_discover(frame, this, loc, xattr_req);
+        return 0;
+    }
 
-        if (fd_ctx) {
-                GF_FREE (fd_ctx->pre_op_done);
+    if (afr_is_private_directory(this->private, loc->parent->gfid, loc->name,
+                                 frame->root->pid)) {
+        op_errno = EPERM;
+        goto out;
+    }
 
-                GF_FREE (fd_ctx->opened_on);
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-                GF_FREE (fd_ctx->locked_on);
+    if (!local->call_count) {
+        op_errno = ENOTCONN;
+        goto out;
+    }
 
-                GF_FREE (fd_ctx->pre_op_piggyback);
-                GF_FREE (fd_ctx->lock_piggyback);
+    local->op = GF_FOP_LOOKUP;
 
-                GF_FREE (fd_ctx->lock_acquired);
+    loc_copy(&local->loc, loc);
 
-		pthread_mutex_destroy (&fd_ctx->delay_lock);
+    local->inode = inode_ref(loc->inode);
 
-                GF_FREE (fd_ctx);
+    if (xattr_req) {
+        /* If xattr_req was null, afr_lookup_xattr_req_prepare() will
+           allocate one for us */
+        local->xattr_req = dict_copy_with_ref(xattr_req, NULL);
+        if (!local->xattr_req) {
+            op_errno = ENOMEM;
+            goto out;
+        }
+        ret = dict_get_gfuuid(local->xattr_req, "gfid-req",
+                              &local->cont.lookup.gfid_req);
+        if (ret == 0) {
+            dict_del_sizen(local->xattr_req, "gfid-req");
         }
+    }
 
+    afr_read_subvol_get(loc->parent, this, NULL, NULL, &event,
+                        AFR_DATA_TRANSACTION, NULL);
+
+    afr_lookup_do(frame, this, 0);
+
+    return 0;
 out:
-        return 0;
+    AFR_STACK_UNWIND(lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+
+    return 0;
 }
 
+void
+_afr_cleanup_fd_ctx(xlator_t *this, afr_fd_ctx_t *fd_ctx)
+{
+    afr_private_t *priv = this->private;
+
+    if (fd_ctx->lk_heal_info) {
+        LOCK(&priv->lock);
+        {
+            list_del(&fd_ctx->lk_heal_info->pos);
+        }
+        afr_lk_heal_info_cleanup(fd_ctx->lk_heal_info);
+        fd_ctx->lk_heal_info = NULL;
+    }
+    GF_FREE(fd_ctx->opened_on);
+    GF_FREE(fd_ctx);
+    return;
+}
 
 int
-afr_release (xlator_t *this, fd_t *fd)
+afr_cleanup_fd_ctx(xlator_t *this, fd_t *fd)
 {
-        afr_locked_fd_t *locked_fd = NULL;
-        afr_locked_fd_t *tmp       = NULL;
-        afr_private_t   *priv      = NULL;
+    uint64_t ctx = 0;
+    afr_fd_ctx_t *fd_ctx = NULL;
+    int ret = 0;
 
-        priv = this->private;
+    ret = fd_ctx_get(fd, this, &ctx);
+    if (ret < 0)
+        goto out;
 
-        afr_cleanup_fd_ctx (this, fd);
+    fd_ctx = (afr_fd_ctx_t *)(long)ctx;
 
-        list_for_each_entry_safe (locked_fd, tmp, &priv->saved_fds,
-                                  list) {
+    if (fd_ctx) {
+        _afr_cleanup_fd_ctx(this, fd_ctx);
+    }
 
-                if (locked_fd->fd == fd) {
-                        list_del_init (&locked_fd->list);
-                        GF_FREE (locked_fd);
-                }
+out:
+    return 0;
+}
 
-        }
+int
+afr_release(xlator_t *this, fd_t *fd)
+{
+    afr_cleanup_fd_ctx(this, fd);
 
-        return 0;
+    return 0;
 }
 
+afr_fd_ctx_t *
+__afr_fd_ctx_get(fd_t *fd, xlator_t *this)
+{
+    uint64_t ctx = 0;
+    int ret = 0;
+    afr_fd_ctx_t *fd_ctx = NULL;
 
-/* {{{ fsync */
+    ret = __fd_ctx_get(fd, this, &ctx);
 
-int
-afr_fsync_unwind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                      struct iatt *postbuf, dict_t *xdata)
+    if (ret < 0) {
+        ret = __afr_fd_ctx_set(this, fd);
+        if (ret < 0)
+            goto out;
+
+        ret = __fd_ctx_get(fd, this, &ctx);
+        if (ret < 0)
+            goto out;
+    }
+
+    fd_ctx = (afr_fd_ctx_t *)(long)ctx;
+out:
+    return fd_ctx;
+}
+
+afr_fd_ctx_t *
+afr_fd_ctx_get(fd_t *fd, xlator_t *this)
 {
-        AFR_STACK_UNWIND (fsync, frame, op_ret, op_errno, prebuf, postbuf,
-                          xdata);
-        return 0;
+    afr_fd_ctx_t *fd_ctx = NULL;
+
+    LOCK(&fd->lock);
+    {
+        fd_ctx = __afr_fd_ctx_get(fd, this);
+    }
+    UNLOCK(&fd->lock);
+
+    return fd_ctx;
 }
 
 int
-afr_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-               int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-               struct iatt *postbuf, dict_t *xdata)
+__afr_fd_ctx_set(xlator_t *this, fd_t *fd)
 {
-        afr_local_t *local = NULL;
-        int call_count = -1;
-        int child_index = (long) cookie;
-        int read_child  = 0;
-	call_stub_t *stub = NULL;
+    afr_private_t *priv = NULL;
+    int ret = -1;
+    uint64_t ctx = 0;
+    afr_fd_ctx_t *fd_ctx = NULL;
+    int i = 0;
 
-        local = frame->local;
+    VALIDATE_OR_GOTO(this->private, out);
+    VALIDATE_OR_GOTO(fd, out);
 
-        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
+    priv = this->private;
 
-        LOCK (&frame->lock);
-        {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
+    ret = __fd_ctx_get(fd, this, &ctx);
 
-                if (op_ret == 0) {
-                        local->op_ret = 0;
+    if (ret == 0)
+        goto out;
 
-                        if (local->success_count == 0) {
-                                local->cont.inode_wfop.prebuf  = *prebuf;
-                                local->cont.inode_wfop.postbuf = *postbuf;
-                        }
+    fd_ctx = GF_CALLOC(1, sizeof(afr_fd_ctx_t), gf_afr_mt_afr_fd_ctx_t);
+    if (!fd_ctx) {
+        ret = -ENOMEM;
+        goto out;
+    }
 
-                        if (child_index == read_child) {
-                                local->cont.inode_wfop.prebuf  = *prebuf;
-                                local->cont.inode_wfop.postbuf = *postbuf;
-                        }
+    fd_ctx->opened_on = GF_CALLOC(sizeof(*fd_ctx->opened_on), priv->child_count,
+                                  gf_afr_mt_int32_t);
+    if (!fd_ctx->opened_on) {
+        ret = -ENOMEM;
+        goto out;
+    }
 
-                        local->success_count++;
-                }
+    for (i = 0; i < priv->child_count; i++) {
+        if (fd_is_anonymous(fd))
+            fd_ctx->opened_on[i] = AFR_FD_OPENED;
+        else
+            fd_ctx->opened_on[i] = AFR_FD_NOT_OPENED;
+    }
 
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-		/* Make a stub out of the frame, and register it
-		   with the waking up post-op. When the call-stub resumes,
-		   we are guaranteed that there was no post-op pending
-		   (i.e changelogs were unset in the server). This is an
-		   essential "guarantee", that fsync() returns only after
-		   completely finishing EVERYTHING, including the delayed
-		   post-op. This guarantee is expected by FUSE graph switching
-		   for example.
-		*/
-		stub = fop_fsync_cbk_stub (frame, afr_fsync_unwind_cbk,
-                                           local->op_ret, local->op_errno,
-                                           &local->cont.inode_wfop.prebuf,
-                                           &local->cont.inode_wfop.postbuf,
-                                           xdata);
-		if (!stub) {
-			AFR_STACK_UNWIND (fsync, frame, -1, ENOMEM, 0, 0, 0);
-			return 0;
-		}
-
-		/* If no new unstable writes happened between the
-		   time we cleared the unstable write witness flag in afr_fsync
-		   and now, calling afr_delayed_changelog_wake_up() should
-		   wake up and skip over the fsync phase and go straight to
-		   afr_changelog_post_op_now()
-		*/
-		afr_delayed_changelog_wake_resume (this, local->fd, stub);
-        }
+    fd_ctx->readdir_subvol = -1;
+    fd_ctx->lk_heal_info = NULL;
 
-        return 0;
+    ret = __fd_ctx_set(fd, this, (uint64_t)(long)fd_ctx);
+    if (ret)
+        gf_msg_debug(this->name, 0, "failed to set fd ctx (%p)", fd);
+out:
+    if (ret && fd_ctx)
+        _afr_cleanup_fd_ctx(this, fd_ctx);
+    return ret;
 }
 
+/* {{{ flush */
 
 int
-afr_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd,
-           int32_t datasync, dict_t *xdata)
+afr_flush_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, dict_t *xdata)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t *local = NULL;
-        int ret = -1;
-        int i = 0;
-        int32_t call_count = 0;
-        int32_t op_errno = 0;
+    afr_local_t *local = NULL;
+    int call_count = -1;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    local = frame->local;
 
-        priv = this->private;
+    LOCK(&frame->lock);
+    {
+        if (op_ret != -1) {
+            local->op_ret = op_ret;
+            if (!local->xdata_rsp && xdata)
+                local->xdata_rsp = dict_ref(xdata);
+        } else {
+            local->op_errno = op_errno;
+        }
+        call_count = --local->call_count;
+    }
+    UNLOCK(&frame->lock);
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    if (call_count == 0)
+        AFR_STACK_UNWIND(flush, frame, local->op_ret, local->op_errno,
+                         local->xdata_rsp);
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    return 0;
+}
 
-        call_count = local->call_count;
+static int
+afr_flush_wrapper(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    int i = 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int call_count = -1;
+
+    priv = this->private;
+    local = frame->local;
+    call_count = local->call_count;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->child_up[i]) {
+            STACK_WIND_COOKIE(frame, afr_flush_cbk, (void *)(long)i,
+                              priv->children[i], priv->children[i]->fops->flush,
+                              local->fd, xdata);
+            if (!--call_count)
+                break;
+        }
+    }
 
-        local->fd             = fd_ref (fd);
+    return 0;
+}
 
-	if (afr_fd_has_witnessed_unstable_write (this, fd)) {
-		/* don't care. we only wanted to CLEAR the bit */
-	}
+afr_local_t *
+afr_wakeup_same_fd_delayed_op(xlator_t *this, afr_lock_t *lock, fd_t *fd)
+{
+    afr_local_t *local = NULL;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND_COOKIE (frame, afr_fsync_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->fsync,
-                                           fd, datasync, xdata);
-                        if (!--call_count)
-                                break;
-                }
+    if (lock->delay_timer) {
+        local = list_entry(lock->post_op.next, afr_local_t,
+                           transaction.owner_list);
+        if (fd == local->fd) {
+            if (gf_timer_call_cancel(this->ctx, lock->delay_timer)) {
+                local = NULL;
+            } else {
+                lock->delay_timer = NULL;
+            }
+        } else {
+            local = NULL;
         }
+    }
 
-        ret = 0;
-out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (fsync, frame, -1, op_errno, NULL, NULL, NULL);
-        return 0;
+    return local;
 }
 
-/* }}} */
-
-/* {{{ fsync */
+void
+afr_delayed_changelog_wake_resume(xlator_t *this, inode_t *inode,
+                                  call_stub_t *stub)
+{
+    afr_inode_ctx_t *ctx = NULL;
+    afr_lock_t *lock = NULL;
+    afr_local_t *metadata_local = NULL;
+    afr_local_t *data_local = NULL;
+    LOCK(&inode->lock);
+    {
+        (void)__afr_inode_ctx_get(this, inode, &ctx);
+        lock = &ctx->lock[AFR_DATA_TRANSACTION];
+        data_local = afr_wakeup_same_fd_delayed_op(this, lock, stub->args.fd);
+        lock = &ctx->lock[AFR_METADATA_TRANSACTION];
+        metadata_local = afr_wakeup_same_fd_delayed_op(this, lock,
+                                                       stub->args.fd);
+    }
+    UNLOCK(&inode->lock);
+
+    if (data_local) {
+        data_local->transaction.resume_stub = stub;
+    } else if (metadata_local) {
+        metadata_local->transaction.resume_stub = stub;
+    } else {
+        call_resume(stub);
+    }
+    if (data_local) {
+        afr_delayed_changelog_wake_up_cbk(data_local);
+    }
+    if (metadata_local) {
+        afr_delayed_changelog_wake_up_cbk(metadata_local);
+    }
+}
 
-int32_t
-afr_fsyncdir_cbk (call_frame_t *frame, void *cookie,
-                  xlator_t *this, int32_t op_ret, int32_t op_errno,
-                  dict_t *xdata)
+int
+afr_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
-        afr_local_t *local = NULL;
-        int call_count = -1;
+    afr_local_t *local = NULL;
+    call_stub_t *stub = NULL;
+    int op_errno = ENOMEM;
 
-        local = frame->local;
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == 0)
-                        local->op_ret = 0;
+    local->op = GF_FOP_FLUSH;
+    if (!afr_is_consistent_io_possible(local, this->private, &op_errno))
+        goto out;
 
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+    local->fd = fd_ref(fd);
 
-        call_count = afr_frame_return (frame);
+    stub = fop_flush_stub(frame, afr_flush_wrapper, fd, xdata);
+    if (!stub)
+        goto out;
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (fsyncdir, frame, local->op_ret,
-                                  local->op_errno, xdata);
+    afr_delayed_changelog_wake_resume(this, fd->inode, stub);
 
-        return 0;
+    return 0;
+out:
+    AFR_STACK_UNWIND(flush, frame, -1, op_errno, NULL);
+    return 0;
 }
 
-
-int32_t
-afr_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd,
-              int32_t datasync, dict_t *xdata)
+int
+afr_fsyncdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t *local = NULL;
-        int ret = -1;
-        int i = 0;
-        int32_t call_count = 0;
-        int32_t op_errno = 0;
+    afr_local_t *local = NULL;
+    int call_count = -1;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    local = frame->local;
 
-        priv = this->private;
+    LOCK(&frame->lock);
+    {
+        if (op_ret == 0) {
+            local->op_ret = 0;
+            if (!local->xdata_rsp && xdata)
+                local->xdata_rsp = dict_ref(xdata);
+        } else {
+            local->op_errno = op_errno;
+        }
+        call_count = --local->call_count;
+    }
+    UNLOCK(&frame->lock);
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    if (call_count == 0)
+        AFR_STACK_UNWIND(fsyncdir, frame, local->op_ret, local->op_errno,
+                         local->xdata_rsp);
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    return 0;
+}
 
-        call_count = local->call_count;
+int
+afr_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+             dict_t *xdata)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int i = 0;
+    int32_t call_count = 0;
+    int32_t op_errno = ENOMEM;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND (frame, afr_fsyncdir_cbk,
-                                    priv->children[i],
-                                    priv->children[i]->fops->fsyncdir,
-                                    fd, datasync, xdata);
-                        if (!--call_count)
-                                break;
-                }
+    priv = this->private;
+
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
+
+    local->op = GF_FOP_FSYNCDIR;
+    if (!afr_is_consistent_io_possible(local, priv, &op_errno))
+        goto out;
+
+    call_count = local->call_count;
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->child_up[i]) {
+            STACK_WIND(frame, afr_fsyncdir_cbk, priv->children[i],
+                       priv->children[i]->fops->fsyncdir, fd, datasync, xdata);
+            if (!--call_count)
+                break;
         }
+    }
 
-        ret = 0;
+    return 0;
 out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (fsyncdir, frame, -1, op_errno, NULL);
-        return 0;
+    AFR_STACK_UNWIND(fsyncdir, frame, -1, op_errno, NULL);
+
+    return 0;
 }
 
 /* }}} */
 
-/* {{{ xattrop */
+static int
+afr_serialized_lock_wind(call_frame_t *frame, xlator_t *this);
 
-int32_t
-afr_xattrop_cbk (call_frame_t *frame, void *cookie,
-                 xlator_t *this, int32_t op_ret, int32_t op_errno,
-                 dict_t *xattr, dict_t *xdata)
+static gf_boolean_t
+afr_is_conflicting_lock_present(int32_t op_ret, int32_t op_errno)
 {
-        afr_local_t *local = NULL;
-        int call_count = -1;
-
-        local = frame->local;
+    if (op_ret == -1 && op_errno == EAGAIN)
+        return _gf_true;
+    return _gf_false;
+}
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == 0) {
-                        if (!local->cont.xattrop.xattr)
-                                local->cont.xattrop.xattr = dict_ref (xattr);
-                        local->op_ret = 0;
-                }
+static void
+afr_fop_lock_unwind(call_frame_t *frame, glusterfs_fop_t op, int32_t op_ret,
+                    int32_t op_errno, dict_t *xdata)
+{
+    switch (op) {
+        case GF_FOP_INODELK:
+            AFR_STACK_UNWIND(inodelk, frame, op_ret, op_errno, xdata);
+            break;
+        case GF_FOP_FINODELK:
+            AFR_STACK_UNWIND(finodelk, frame, op_ret, op_errno, xdata);
+            break;
+        case GF_FOP_ENTRYLK:
+            AFR_STACK_UNWIND(entrylk, frame, op_ret, op_errno, xdata);
+            break;
+        case GF_FOP_FENTRYLK:
+            AFR_STACK_UNWIND(fentrylk, frame, op_ret, op_errno, xdata);
+            break;
+        default:
+            break;
+    }
+}
 
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+static void
+afr_fop_lock_wind(call_frame_t *frame, xlator_t *this, int child_index,
+                  int32_t (*lock_cbk)(call_frame_t *, void *, xlator_t *,
+                                      int32_t, int32_t, dict_t *))
+{
+    afr_local_t *local = frame->local;
+    afr_private_t *priv = this->private;
+    int i = child_index;
+
+    switch (local->op) {
+        case GF_FOP_INODELK:
+            STACK_WIND_COOKIE(
+                frame, lock_cbk, (void *)(long)i, priv->children[i],
+                priv->children[i]->fops->inodelk,
+                (const char *)local->cont.inodelk.volume, &local->loc,
+                local->cont.inodelk.cmd, &local->cont.inodelk.flock,
+                local->cont.inodelk.xdata);
+            break;
+        case GF_FOP_FINODELK:
+            STACK_WIND_COOKIE(
+                frame, lock_cbk, (void *)(long)i, priv->children[i],
+                priv->children[i]->fops->finodelk,
+                (const char *)local->cont.inodelk.volume, local->fd,
+                local->cont.inodelk.cmd, &local->cont.inodelk.flock,
+                local->cont.inodelk.xdata);
+            break;
+        case GF_FOP_ENTRYLK:
+            STACK_WIND_COOKIE(
+                frame, lock_cbk, (void *)(long)i, priv->children[i],
+                priv->children[i]->fops->entrylk, local->cont.entrylk.volume,
+                &local->loc, local->cont.entrylk.basename,
+                local->cont.entrylk.cmd, local->cont.entrylk.type,
+                local->cont.entrylk.xdata);
+            break;
+        case GF_FOP_FENTRYLK:
+            STACK_WIND_COOKIE(
+                frame, lock_cbk, (void *)(long)i, priv->children[i],
+                priv->children[i]->fops->fentrylk, local->cont.entrylk.volume,
+                local->fd, local->cont.entrylk.basename,
+                local->cont.entrylk.cmd, local->cont.entrylk.type,
+                local->cont.entrylk.xdata);
+            break;
+        default:
+            break;
+    }
+}
 
-        call_count = afr_frame_return (frame);
+void
+afr_fop_lock_proceed(call_frame_t *frame)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (xattrop, frame, local->op_ret, local->op_errno,
-                local->cont.xattrop.xattr, xdata);
+    local = frame->local;
+    priv = frame->this->private;
 
-        return 0;
+    if (local->fop_lock_state != AFR_FOP_LOCK_PARALLEL) {
+        afr_fop_lock_unwind(frame, local->op, local->op_ret, local->op_errno,
+                            local->xdata_rsp);
+        return;
+    }
+    /* At least one child is up */
+    /*
+     * Non-blocking locks also need to be serialized.  Otherwise there is
+     * a chance that both the mounts which issued same non-blocking inodelk
+     * may endup not acquiring the lock on any-brick.
+     * Ex: Mount1 and Mount2
+     * request for full length lock on file f1.  Mount1 afr may acquire the
+     * partial lock on brick-1 and may not acquire the lock on brick-2
+     * because Mount2 already got the lock on brick-2, vice versa.  Since
+     * both the mounts only got partial locks, afr treats them as failure in
+     * gaining the locks and unwinds with EAGAIN errno.
+     */
+    local->op_ret = -1;
+    local->op_errno = EUCLEAN;
+    local->fop_lock_state = AFR_FOP_LOCK_SERIAL;
+    afr_local_replies_wipe(local, priv);
+    if (local->xdata_rsp)
+        dict_unref(local->xdata_rsp);
+    local->xdata_rsp = NULL;
+    switch (local->op) {
+        case GF_FOP_INODELK:
+        case GF_FOP_FINODELK:
+            local->cont.inodelk.cmd = local->cont.inodelk.in_cmd;
+            local->cont.inodelk.flock = local->cont.inodelk.in_flock;
+            if (local->cont.inodelk.xdata)
+                dict_unref(local->cont.inodelk.xdata);
+            local->cont.inodelk.xdata = NULL;
+            if (local->xdata_req)
+                local->cont.inodelk.xdata = dict_ref(local->xdata_req);
+            break;
+        case GF_FOP_ENTRYLK:
+        case GF_FOP_FENTRYLK:
+            local->cont.entrylk.cmd = local->cont.entrylk.in_cmd;
+            if (local->cont.entrylk.xdata)
+                dict_unref(local->cont.entrylk.xdata);
+            local->cont.entrylk.xdata = NULL;
+            if (local->xdata_req)
+                local->cont.entrylk.xdata = dict_ref(local->xdata_req);
+            break;
+        default:
+            break;
+    }
+    afr_serialized_lock_wind(frame, frame->this);
 }
 
+static int32_t
+afr_unlock_partial_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int32_t op_ret, int32_t op_errno, dict_t *xdata)
 
-int32_t
-afr_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc,
-             gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t *local  = NULL;
-        int ret = -1;
-        int i = 0;
-        int32_t call_count = 0;
-        int32_t op_errno = 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int call_count = -1;
+    int child_index = (long)cookie;
+    uuid_t gfid = {0};
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    local = frame->local;
+    priv = this->private;
 
-        priv = this->private;
+    if (op_ret < 0 && op_errno != ENOTCONN) {
+        if (local->fd)
+            gf_uuid_copy(gfid, local->fd->inode->gfid);
+        else
+            loc_gfid(&local->loc, gfid);
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, AFR_MSG_UNLOCK_FAIL,
+               "%s: Failed to unlock %s on %s "
+               "with lk_owner: %s",
+               uuid_utoa(gfid), gf_fop_list[local->op],
+               priv->children[child_index]->name,
+               lkowner_utoa(&frame->root->lk_owner));
+    }
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    call_count = afr_frame_return(frame);
+    if (call_count == 0)
+        afr_fop_lock_proceed(frame);
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    return 0;
+}
 
-        call_count = local->call_count;
+static int32_t
+afr_unlock_locks_and_proceed(call_frame_t *frame, xlator_t *this,
+                             int call_count)
+{
+    int i = 0;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+
+    if (call_count == 0) {
+        afr_fop_lock_proceed(frame);
+        goto out;
+    }
+
+    local = frame->local;
+    priv = this->private;
+    local->call_count = call_count;
+    switch (local->op) {
+        case GF_FOP_INODELK:
+        case GF_FOP_FINODELK:
+            local->cont.inodelk.flock.l_type = F_UNLCK;
+            local->cont.inodelk.cmd = F_SETLK;
+            if (local->cont.inodelk.xdata)
+                dict_unref(local->cont.inodelk.xdata);
+            local->cont.inodelk.xdata = NULL;
+            break;
+        case GF_FOP_ENTRYLK:
+        case GF_FOP_FENTRYLK:
+            local->cont.entrylk.cmd = ENTRYLK_UNLOCK;
+            if (local->cont.entrylk.xdata)
+                dict_unref(local->cont.entrylk.xdata);
+            local->cont.entrylk.xdata = NULL;
+            break;
+        default:
+            break;
+    }
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND (frame, afr_xattrop_cbk,
-                                    priv->children[i],
-                                    priv->children[i]->fops->xattrop,
-                                    loc, optype, xattr, xdata);
-                        if (!--call_count)
-                                break;
-                }
-        }
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
 
-        ret = 0;
-out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (xattrop, frame, -1, op_errno, NULL, NULL);
-        return 0;
-}
+        if (local->replies[i].op_ret == -1)
+            continue;
 
-/* }}} */
+        afr_fop_lock_wind(frame, this, i, afr_unlock_partial_lock_cbk);
 
-/* {{{ fxattrop */
+        if (!--call_count)
+            break;
+    }
+
+out:
+    return 0;
+}
 
 int32_t
-afr_fxattrop_cbk (call_frame_t *frame, void *cookie,
-                  xlator_t *this, int32_t op_ret, int32_t op_errno,
-                  dict_t *xattr, dict_t *xdata)
+afr_fop_lock_done(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *local = NULL;
+    int i = 0;
+    int lock_count = 0;
+    unsigned char *success = NULL;
 
-        int call_count = -1;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        local = frame->local;
+    local = frame->local;
+    priv = this->private;
+    success = alloca0(priv->child_count);
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == 0) {
-                        if (!local->cont.fxattrop.xattr)
-                                local->cont.fxattrop.xattr = dict_ref (xattr);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
 
-                        local->op_ret = 0;
-                }
+        if (local->replies[i].op_ret == 0) {
+            lock_count++;
+            success[i] = 1;
+        }
+
+        if (local->op_ret == -1 && local->op_errno == EAGAIN)
+            continue;
 
-                local->op_errno = op_errno;
+        if ((local->replies[i].op_ret == -1) &&
+            (local->replies[i].op_errno == EAGAIN)) {
+            local->op_ret = -1;
+            local->op_errno = EAGAIN;
+            continue;
         }
-        UNLOCK (&frame->lock);
 
-        call_count = afr_frame_return (frame);
+        if (local->replies[i].op_ret == 0)
+            local->op_ret = 0;
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (fxattrop, frame, local->op_ret, local->op_errno,
-                                  local->cont.fxattrop.xattr, xdata);
+        local->op_errno = local->replies[i].op_errno;
+    }
 
-        return 0;
+    if (afr_fop_lock_is_unlock(frame))
+        goto unwind;
+
+    if (afr_is_conflicting_lock_present(local->op_ret, local->op_errno)) {
+        afr_unlock_locks_and_proceed(frame, this, lock_count);
+    } else if (priv->quorum_count && !afr_has_quorum(success, this, NULL)) {
+        local->fop_lock_state = AFR_FOP_LOCK_QUORUM_FAILED;
+        local->op_ret = -1;
+        local->op_errno = afr_final_errno(local, priv);
+        if (local->op_errno == 0)
+            local->op_errno = afr_quorum_errno(priv);
+        afr_unlock_locks_and_proceed(frame, this, lock_count);
+    } else {
+        goto unwind;
+    }
+
+    return 0;
+unwind:
+    afr_fop_lock_unwind(frame, local->op, local->op_ret, local->op_errno,
+                        local->xdata_rsp);
+    return 0;
 }
 
+static int
+afr_common_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int child_index = (long)cookie;
+
+    local = frame->local;
+
+    local->replies[child_index].valid = 1;
+    local->replies[child_index].op_ret = op_ret;
+    local->replies[child_index].op_errno = op_errno;
+    if (op_ret == 0 && xdata) {
+        local->replies[child_index].xdata = dict_ref(xdata);
+        LOCK(&frame->lock);
+        {
+            if (!local->xdata_rsp)
+                local->xdata_rsp = dict_ref(xdata);
+        }
+        UNLOCK(&frame->lock);
+    }
+    return 0;
+}
+
+static int32_t
+afr_serialized_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
 
-int32_t
-afr_fxattrop (call_frame_t *frame, xlator_t *this, fd_t *fd,
-              gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t *local  = NULL;
-        int ret = -1;
-        int i = 0;
-        int32_t call_count = 0;
-        int32_t op_errno = 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int child_index = (long)cookie;
+    int next_child = 0;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    local = frame->local;
+    priv = this->private;
 
-        priv = this->private;
+    afr_common_lock_cbk(frame, cookie, this, op_ret, op_errno, xdata);
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    for (next_child = child_index + 1; next_child < priv->child_count;
+         next_child++) {
+        if (local->child_up[next_child])
+            break;
+    }
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    if (afr_is_conflicting_lock_present(op_ret, op_errno) ||
+        (next_child == priv->child_count)) {
+        afr_fop_lock_done(frame, this);
+    } else {
+        afr_fop_lock_wind(frame, this, next_child, afr_serialized_lock_cbk);
+    }
 
-        call_count = local->call_count;
+    return 0;
+}
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND (frame, afr_fxattrop_cbk,
-                                    priv->children[i],
-                                    priv->children[i]->fops->fxattrop,
-                                    fd, optype, xattr, xdata);
-                        if (!--call_count)
-                                break;
-                }
+static int
+afr_serialized_lock_wind(call_frame_t *frame, xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int i = 0;
+
+    priv = this->private;
+    local = frame->local;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->child_up[i]) {
+            afr_fop_lock_wind(frame, this, i, afr_serialized_lock_cbk);
+            break;
         }
+    }
+    return 0;
+}
 
-        ret = 0;
-out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (fxattrop, frame, -1, op_errno, NULL, NULL);
-        return 0;
+static int32_t
+afr_parallel_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
+
+{
+    int call_count = 0;
+
+    afr_common_lock_cbk(frame, cookie, this, op_ret, op_errno, xdata);
+
+    call_count = afr_frame_return(frame);
+    if (call_count == 0)
+        afr_fop_lock_done(frame, this);
+
+    return 0;
 }
 
-/* }}} */
+static int
+afr_parallel_lock_wind(call_frame_t *frame, xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int call_count = 0;
+    int i = 0;
 
+    priv = this->private;
+    local = frame->local;
+    call_count = local->call_count;
 
-int32_t
-afr_inodelk_cbk (call_frame_t *frame, void *cookie,
-                 xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata)
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->child_up[i])
+            continue;
+        afr_fop_lock_wind(frame, this, i, afr_parallel_lock_cbk);
+        if (!--call_count)
+            break;
+    }
+    return 0;
+}
 
+static int
+afr_fop_handle_lock(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *local = NULL;
-        int call_count = -1;
+    afr_local_t *local = frame->local;
+    int op_errno = 0;
 
-        local = frame->local;
+    if (!afr_fop_lock_is_unlock(frame)) {
+        if (!afr_is_consistent_io_possible(local, this->private, &op_errno))
+            goto out;
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == 0)
-                        local->op_ret = 0;
+        switch (local->op) {
+            case GF_FOP_INODELK:
+            case GF_FOP_FINODELK:
+                local->cont.inodelk.cmd = F_SETLK;
+                break;
+            case GF_FOP_ENTRYLK:
+            case GF_FOP_FENTRYLK:
+                local->cont.entrylk.cmd = ENTRYLK_LOCK_NB;
+                break;
+            default:
+                break;
+        }
+    }
 
-                local->op_errno = op_errno;
+    if (local->xdata_req) {
+        switch (local->op) {
+            case GF_FOP_INODELK:
+            case GF_FOP_FINODELK:
+                local->cont.inodelk.xdata = dict_ref(local->xdata_req);
+                break;
+            case GF_FOP_ENTRYLK:
+            case GF_FOP_FENTRYLK:
+                local->cont.entrylk.xdata = dict_ref(local->xdata_req);
+                break;
+            default:
+                break;
         }
-        UNLOCK (&frame->lock);
+    }
 
-        call_count = afr_frame_return (frame);
+    local->fop_lock_state = AFR_FOP_LOCK_PARALLEL;
+    afr_parallel_lock_wind(frame, this);
+out:
+    return -op_errno;
+}
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (inodelk, frame, local->op_ret,
-                                  local->op_errno, xdata);
+static int32_t
+afr_handle_inodelk(call_frame_t *frame, xlator_t *this, glusterfs_fop_t fop,
+                   const char *volume, loc_t *loc, fd_t *fd, int32_t cmd,
+                   struct gf_flock *flock, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int32_t op_errno = ENOMEM;
+
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
+
+    local->op = fop;
+    if (loc)
+        loc_copy(&local->loc, loc);
+    if (fd && (flock->l_type != F_UNLCK)) {
+        AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+        local->fd = fd_ref(fd);
+    }
+
+    local->cont.inodelk.volume = gf_strdup(volume);
+    if (!local->cont.inodelk.volume) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    local->cont.inodelk.in_cmd = cmd;
+    local->cont.inodelk.cmd = cmd;
+    local->cont.inodelk.in_flock = *flock;
+    local->cont.inodelk.flock = *flock;
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
+
+    op_errno = -afr_fop_handle_lock(frame, frame->this);
+    if (op_errno)
+        goto out;
+    return 0;
+out:
+    afr_fop_lock_unwind(frame, fop, -1, op_errno, NULL);
 
-        return 0;
+    return 0;
 }
 
+int32_t
+afr_inodelk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+            int32_t cmd, struct gf_flock *flock, dict_t *xdata)
+{
+    afr_handle_inodelk(frame, this, GF_FOP_INODELK, volume, loc, NULL, cmd,
+                       flock, xdata);
+    return 0;
+}
 
 int32_t
-afr_inodelk (call_frame_t *frame, xlator_t *this,
-             const char *volume, loc_t *loc, int32_t cmd,
-             struct gf_flock *flock, dict_t *xdata)
+afr_finodelk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+             int32_t cmd, struct gf_flock *flock, dict_t *xdata)
+{
+    afr_handle_inodelk(frame, this, GF_FOP_FINODELK, volume, NULL, fd, cmd,
+                       flock, xdata);
+    return 0;
+}
+
+static int
+afr_handle_entrylk(call_frame_t *frame, xlator_t *this, glusterfs_fop_t fop,
+                   const char *volume, loc_t *loc, fd_t *fd,
+                   const char *basename, entrylk_cmd cmd, entrylk_type type,
+                   dict_t *xdata)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t *local  = NULL;
-        int ret = -1;
-        int i = 0;
-        int32_t call_count = 0;
-        int32_t op_errno = 0;
+    afr_local_t *local = NULL;
+    int32_t op_errno = ENOMEM;
+
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
+
+    local->op = fop;
+    if (loc)
+        loc_copy(&local->loc, loc);
+    if (fd && (cmd != ENTRYLK_UNLOCK)) {
+        AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+        local->fd = fd_ref(fd);
+    }
+    local->cont.entrylk.cmd = cmd;
+    local->cont.entrylk.in_cmd = cmd;
+    local->cont.entrylk.type = type;
+    local->cont.entrylk.volume = gf_strdup(volume);
+    local->cont.entrylk.basename = gf_strdup(basename);
+    if (!local->cont.entrylk.volume || !local->cont.entrylk.basename) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
+    op_errno = -afr_fop_handle_lock(frame, frame->this);
+    if (op_errno)
+        goto out;
+
+    return 0;
+out:
+    afr_fop_lock_unwind(frame, fop, -1, op_errno, NULL);
+    return 0;
+}
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+int
+afr_entrylk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+            const char *basename, entrylk_cmd cmd, entrylk_type type,
+            dict_t *xdata)
+{
+    afr_handle_entrylk(frame, this, GF_FOP_ENTRYLK, volume, loc, NULL, basename,
+                       cmd, type, xdata);
+    return 0;
+}
 
-        priv = this->private;
+int
+afr_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+             const char *basename, entrylk_cmd cmd, entrylk_type type,
+             dict_t *xdata)
+{
+    afr_handle_entrylk(frame, this, GF_FOP_FENTRYLK, volume, NULL, fd, basename,
+                       cmd, type, xdata);
+    return 0;
+}
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+int
+afr_statfs_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, struct statvfs *statvfs, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int call_count = 0;
+    struct statvfs *buf = NULL;
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    local = frame->local;
 
-        call_count = local->call_count;
+    LOCK(&frame->lock);
+    {
+        if (op_ret != 0) {
+            local->op_errno = op_errno;
+            goto unlock;
+        }
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND (frame, afr_inodelk_cbk,
-                                    priv->children[i],
-                                    priv->children[i]->fops->inodelk,
-                                    volume, loc, cmd, flock, xdata);
-
-                        if (!--call_count)
-                                break;
+        local->op_ret = op_ret;
+
+        buf = &local->cont.statfs.buf;
+        if (local->cont.statfs.buf_set) {
+            if (statvfs->f_bavail < buf->f_bavail) {
+                *buf = *statvfs;
+                if (xdata) {
+                    if (local->xdata_rsp)
+                        dict_unref(local->xdata_rsp);
+                    local->xdata_rsp = dict_ref(xdata);
                 }
+            }
+        } else {
+            *buf = *statvfs;
+            local->cont.statfs.buf_set = 1;
+            if (xdata)
+                local->xdata_rsp = dict_ref(xdata);
         }
+    }
+unlock:
+    call_count = --local->call_count;
+    UNLOCK(&frame->lock);
 
-        ret = 0;
-out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (inodelk, frame, -1, op_errno, NULL);
-        return 0;
+    if (call_count == 0)
+        AFR_STACK_UNWIND(statfs, frame, local->op_ret, local->op_errno,
+                         &local->cont.statfs.buf, local->xdata_rsp);
+
+    return 0;
 }
 
+int
+afr_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int call_count = 0;
+    int32_t op_errno = ENOMEM;
+
+    priv = this->private;
+
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
+
+    local->op = GF_FOP_STATFS;
+    if (!afr_is_consistent_io_possible(local, priv, &op_errno))
+        goto out;
+
+    if (priv->arbiter_count == 1 && local->child_up[ARBITER_BRICK_INDEX])
+        local->call_count--;
+    call_count = local->call_count;
+    if (!call_count) {
+        op_errno = ENOTCONN;
+        goto out;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->child_up[i]) {
+            if (AFR_IS_ARBITER_BRICK(priv, i))
+                continue;
+            STACK_WIND(frame, afr_statfs_cbk, priv->children[i],
+                       priv->children[i]->fops->statfs, loc, xdata);
+            if (!--call_count)
+                break;
+        }
+    }
+
+    return 0;
+out:
+    AFR_STACK_UNWIND(statfs, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
+}
 
 int32_t
-afr_finodelk_cbk (call_frame_t *frame, void *cookie,
-                  xlator_t *this, int32_t op_ret, int32_t op_errno,
+afr_lk_unlock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct gf_flock *lock,
                   dict_t *xdata)
-
 {
-        afr_local_t *local = NULL;
-        int call_count = -1;
-
-        local = frame->local;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = this->private;
+    int call_count = -1;
+    int child_index = (long)cookie;
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == 0)
-                        local->op_ret = 0;
-
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+    local = frame->local;
 
-        call_count = afr_frame_return (frame);
+    if (op_ret < 0 && op_errno != ENOTCONN && op_errno != EBADFD) {
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, AFR_MSG_UNLOCK_FAIL,
+               "gfid=%s: unlock failed on subvolume %s "
+               "with lock owner %s",
+               uuid_utoa(local->fd->inode->gfid),
+               priv->children[child_index]->name,
+               lkowner_utoa(&frame->root->lk_owner));
+    }
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (finodelk, frame, local->op_ret,
-                                  local->op_errno, xdata);
+    call_count = afr_frame_return(frame);
+    if (call_count == 0) {
+        AFR_STACK_UNWIND(lk, frame, local->op_ret, local->op_errno, NULL,
+                         local->xdata_rsp);
+    }
 
-        return 0;
+    return 0;
 }
 
-
 int32_t
-afr_finodelk (call_frame_t *frame, xlator_t *this,
-              const char *volume, fd_t *fd, int32_t cmd, struct gf_flock *flock,
-              dict_t *xdata)
+afr_lk_unlock(call_frame_t *frame, xlator_t *this)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t *local  = NULL;
-        int ret = -1;
-        int i = 0;
-        int32_t call_count = 0;
-        int32_t op_errno = 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int call_count = 0;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    local = frame->local;
+    priv = this->private;
 
-        priv = this->private;
+    call_count = afr_locked_nodes_count(local->cont.lk.locked_nodes,
+                                        priv->child_count);
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    if (call_count == 0) {
+        AFR_STACK_UNWIND(lk, frame, local->op_ret, local->op_errno, NULL,
+                         local->xdata_rsp);
+        return 0;
+    }
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    local->call_count = call_count;
 
-        call_count = local->call_count;
+    local->cont.lk.user_flock.l_type = F_UNLCK;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND (frame, afr_finodelk_cbk,
-                                    priv->children[i],
-                                    priv->children[i]->fops->finodelk,
-                                    volume, fd, cmd, flock, xdata);
-
-                        if (!--call_count)
-                                break;
-                }
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->cont.lk.locked_nodes[i]) {
+            STACK_WIND_COOKIE(frame, afr_lk_unlock_cbk, (void *)(long)i,
+                              priv->children[i], priv->children[i]->fops->lk,
+                              local->fd, F_SETLK, &local->cont.lk.user_flock,
+                              NULL);
+
+            if (!--call_count)
+                break;
         }
+    }
 
-        ret = 0;
-out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (finodelk, frame, -1, op_errno, NULL);
-        return 0;
+    return 0;
 }
 
-
 int32_t
-afr_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
+afr_lk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+           int32_t op_errno, struct gf_flock *lock, dict_t *xdata)
 {
-        afr_local_t *local = NULL;
-        int call_count = -1;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int child_index = -1;
 
-        local = frame->local;
+    local = frame->local;
+    priv = this->private;
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == 0)
-                        local->op_ret = 0;
+    child_index = (long)cookie;
 
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+    afr_common_lock_cbk(frame, cookie, this, op_ret, op_errno, xdata);
+    if (op_ret < 0 && op_errno == EAGAIN) {
+        local->op_ret = -1;
+        local->op_errno = EAGAIN;
 
-        call_count = afr_frame_return (frame);
+        afr_lk_unlock(frame, this);
+        return 0;
+    }
+
+    if (op_ret == 0) {
+        local->op_ret = 0;
+        local->op_errno = 0;
+        local->cont.lk.locked_nodes[child_index] = 1;
+        local->cont.lk.ret_flock = *lock;
+    }
+
+    child_index++;
+
+    if (child_index < priv->child_count) {
+        STACK_WIND_COOKIE(frame, afr_lk_cbk, (void *)(long)child_index,
+                          priv->children[child_index],
+                          priv->children[child_index]->fops->lk, local->fd,
+                          local->cont.lk.cmd, &local->cont.lk.user_flock,
+                          local->xdata_req);
+    } else if (priv->quorum_count &&
+               !afr_has_quorum(local->cont.lk.locked_nodes, this, NULL)) {
+        local->op_ret = -1;
+        local->op_errno = afr_final_errno(local, priv);
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (entrylk, frame, local->op_ret,
-                                  local->op_errno, xdata);
+        afr_lk_unlock(frame, this);
+    } else {
+        if (local->op_ret < 0)
+            local->op_errno = afr_final_errno(local, priv);
 
-        return 0;
-}
+        AFR_STACK_UNWIND(lk, frame, local->op_ret, local->op_errno,
+                         &local->cont.lk.ret_flock, local->xdata_rsp);
+    }
 
+    return 0;
+}
 
-int32_t
-afr_entrylk (call_frame_t *frame, xlator_t *this,
-             const char *volume, loc_t *loc,
-             const char *basename, entrylk_cmd cmd, entrylk_type type,
-             dict_t *xdata)
+int
+afr_lk_transaction_cbk(int ret, call_frame_t *frame, void *opaque)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t *local  = NULL;
-        int ret = -1;
-        int i = 0;
-        int32_t call_count = 0;
-        int32_t op_errno = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    return 0;
+}
 
-        priv = this->private;
+int
+afr_lk_txn_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct gf_flock *lock,
+                    dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int child_index = -1;
+
+    local = frame->local;
+    child_index = (long)cookie;
+    afr_common_lock_cbk(frame, cookie, this, op_ret, op_errno, xdata);
+    if (op_ret == 0) {
+        local->op_ret = 0;
+        local->op_errno = 0;
+        local->cont.lk.locked_nodes[child_index] = 1;
+        local->cont.lk.ret_flock = *lock;
+    }
+    syncbarrier_wake(&local->barrier);
+    return 0;
+}
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+int
+afr_lk_txn_unlock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct gf_flock *lock,
+                      dict_t *xdata)
+{
+    afr_local_t *local = frame->local;
+    afr_private_t *priv = this->private;
+    int child_index = (long)cookie;
+
+    if (op_ret < 0 && op_errno != ENOTCONN && op_errno != EBADFD) {
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, AFR_MSG_UNLOCK_FAIL,
+               "gfid=%s: unlock failed on subvolume %s "
+               "with lock owner %s",
+               uuid_utoa(local->fd->inode->gfid),
+               priv->children[child_index]->name,
+               lkowner_utoa(&frame->root->lk_owner));
+    }
+    return 0;
+}
+int
+afr_lk_transaction(void *opaque)
+{
+    call_frame_t *frame = NULL;
+    xlator_t *this = NULL;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    char *wind_on = NULL;
+    int op_errno = 0;
+    int i = 0;
+    int ret = 0;
+
+    frame = (call_frame_t *)opaque;
+    local = frame->local;
+    this = frame->this;
+    priv = this->private;
+    wind_on = alloca0(priv->child_count);
+
+    if (priv->arbiter_count || priv->child_count != 3) {
+        op_errno = ENOTSUP;
+        gf_msg(frame->this->name, GF_LOG_ERROR, op_errno, AFR_MSG_LK_HEAL_DOM,
+               "%s: Lock healing supported only for replica 3 volumes.",
+               uuid_utoa(local->fd->inode->gfid));
+        goto err;
+    }
+
+    op_errno = -afr_dom_lock_acquire(frame);  // Released during
+                                              // AFR_STACK_UNWIND
+    if (op_errno != 0) {
+        goto err;
+    }
+    if (priv->quorum_count &&
+        !afr_has_quorum(local->cont.lk.dom_locked_nodes, this, NULL)) {
+        op_errno = afr_final_errno(local, priv);
+        goto err;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (priv->child_up[i] && local->cont.lk.dom_locked_nodes[i])
+            wind_on[i] = 1;
+    }
+    AFR_ONLIST(wind_on, frame, afr_lk_txn_wind_cbk, lk, local->fd,
+               local->cont.lk.cmd, &local->cont.lk.user_flock,
+               local->xdata_req);
+
+    if (priv->quorum_count &&
+        !afr_has_quorum(local->cont.lk.locked_nodes, this, NULL)) {
+        local->op_ret = -1;
+        local->op_errno = afr_final_errno(local, priv);
+        goto unlock;
+    } else {
+        if (local->cont.lk.user_flock.l_type == F_UNLCK)
+            ret = afr_remove_lock_from_saved_locks(local, this);
+        else
+            ret = afr_add_lock_to_saved_locks(frame, this);
+        if (ret) {
+            local->op_ret = -1;
+            local->op_errno = -ret;
+            goto unlock;
+        }
+        AFR_STACK_UNWIND(lk, frame, local->op_ret, local->op_errno,
+                         &local->cont.lk.ret_flock, local->xdata_rsp);
+    }
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    return 0;
 
-        call_count = local->call_count;
+unlock:
+    local->cont.lk.user_flock.l_type = F_UNLCK;
+    AFR_ONLIST(local->cont.lk.locked_nodes, frame, afr_lk_txn_unlock_cbk, lk,
+               local->fd, F_SETLK, &local->cont.lk.user_flock, NULL);
+err:
+    AFR_STACK_UNWIND(lk, frame, -1, op_errno, NULL, NULL);
+    return -1;
+}
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND (frame, afr_entrylk_cbk,
-                                    priv->children[i],
-                                    priv->children[i]->fops->entrylk,
-                                    volume, loc, basename, cmd, type, xdata);
-
-                        if (!--call_count)
-                                break;
-                }
+int
+afr_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
+       struct gf_flock *flock, dict_t *xdata)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int ret = 0;
+    int i = 0;
+    int32_t op_errno = ENOMEM;
+
+    priv = this->private;
+
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
+
+    local->op = GF_FOP_LK;
+    if (!afr_lk_is_unlock(cmd, flock)) {
+        AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+        if (!afr_is_consistent_io_possible(local, priv, &op_errno))
+            goto out;
+    }
+
+    local->cont.lk.locked_nodes = GF_CALLOC(
+        priv->child_count, sizeof(*local->cont.lk.locked_nodes),
+        gf_afr_mt_char);
+
+    if (!local->cont.lk.locked_nodes) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    local->fd = fd_ref(fd);
+    local->cont.lk.cmd = cmd;
+    local->cont.lk.user_flock = *flock;
+    local->cont.lk.ret_flock = *flock;
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
+
+    if (afr_is_lock_mode_mandatory(xdata)) {
+        ret = synctask_new(this->ctx->env, afr_lk_transaction,
+                           afr_lk_transaction_cbk, frame, frame);
+        if (ret) {
+            op_errno = ENOMEM;
+            goto out;
         }
-
-        ret = 0;
-out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (entrylk, frame, -1, op_errno, NULL);
         return 0;
-}
+    }
 
+    STACK_WIND_COOKIE(frame, afr_lk_cbk, (void *)(long)0, priv->children[i],
+                      priv->children[i]->fops->lk, fd, cmd, flock,
+                      local->xdata_req);
 
+    return 0;
+out:
+    AFR_STACK_UNWIND(lk, frame, -1, op_errno, NULL, NULL);
 
-int32_t
-afr_fentrylk_cbk (call_frame_t *frame, void *cookie,
-                  xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata)
+    return 0;
+}
 
+int32_t
+afr_lease_unlock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct gf_lease *lease,
+                     dict_t *xdata)
 {
-        afr_local_t *local = NULL;
-        int call_count = -1;
-
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == 0)
-                        local->op_ret = 0;
-
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+    afr_local_t *local = NULL;
+    int call_count = -1;
 
-        call_count = afr_frame_return (frame);
+    local = frame->local;
+    call_count = afr_frame_return(frame);
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (fentrylk, frame, local->op_ret,
-                                  local->op_errno, xdata);
+    if (call_count == 0)
+        AFR_STACK_UNWIND(lease, frame, local->op_ret, local->op_errno, lease,
+                         xdata);
 
-        return 0;
+    return 0;
 }
 
-
 int32_t
-afr_fentrylk (call_frame_t *frame, xlator_t *this,
-              const char *volume, fd_t *fd,
-              const char *basename, entrylk_cmd cmd,
-              entrylk_type type, dict_t *xdata)
+afr_lease_unlock(call_frame_t *frame, xlator_t *this)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t *local  = NULL;
-        int ret = -1;
-        int i = 0;
-        int32_t call_count = 0;
-        int32_t op_errno = 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int call_count = 0;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    local = frame->local;
+    priv = this->private;
 
-        priv = this->private;
+    call_count = afr_locked_nodes_count(local->cont.lease.locked_nodes,
+                                        priv->child_count);
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    if (call_count == 0) {
+        AFR_STACK_UNWIND(lease, frame, local->op_ret, local->op_errno,
+                         &local->cont.lease.ret_lease, NULL);
+        return 0;
+    }
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    local->call_count = call_count;
 
-        call_count = local->call_count;
+    local->cont.lease.user_lease.cmd = GF_UNLK_LEASE;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND (frame, afr_fentrylk_cbk,
-                                    priv->children[i],
-                                    priv->children[i]->fops->fentrylk,
-                                    volume, fd, basename, cmd, type, xdata);
-
-                        if (!--call_count)
-                                break;
-                }
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->cont.lease.locked_nodes[i]) {
+            STACK_WIND(frame, afr_lease_unlock_cbk, priv->children[i],
+                       priv->children[i]->fops->lease, &local->loc,
+                       &local->cont.lease.user_lease, NULL);
+
+            if (!--call_count)
+                break;
         }
+    }
 
-        ret = 0;
-out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (fentrylk, frame, -1, op_errno, NULL);
-        return 0;
+    return 0;
 }
 
 int32_t
-afr_statfs_cbk (call_frame_t *frame, void *cookie,
-                xlator_t *this, int32_t op_ret, int32_t op_errno,
-                struct statvfs *statvfs, dict_t *xdata)
+afr_lease_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct gf_lease *lease, dict_t *xdata)
 {
-        afr_local_t *local = NULL;
-        int call_count = 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int child_index = -1;
 
-        LOCK (&frame->lock);
-        {
-                local = frame->local;
-
-                if (op_ret == 0) {
-                        local->op_ret   = op_ret;
-
-                        if (local->cont.statfs.buf_set) {
-                                if (statvfs->f_bavail < local->cont.statfs.buf.f_bavail)
-                                        local->cont.statfs.buf = *statvfs;
-                        } else {
-                                local->cont.statfs.buf = *statvfs;
-                                local->cont.statfs.buf_set = 1;
-                        }
-                }
+    local = frame->local;
+    priv = this->private;
 
-                if (op_ret == -1)
-                        local->op_errno = op_errno;
+    child_index = (long)cookie;
 
-        }
-        UNLOCK (&frame->lock);
+    afr_common_lock_cbk(frame, cookie, this, op_ret, op_errno, xdata);
+    if (op_ret < 0 && op_errno == EAGAIN) {
+        local->op_ret = -1;
+        local->op_errno = EAGAIN;
 
-        call_count = afr_frame_return (frame);
+        afr_lease_unlock(frame, this);
+        return 0;
+    }
+
+    if (op_ret == 0) {
+        local->op_ret = 0;
+        local->op_errno = 0;
+        local->cont.lease.locked_nodes[child_index] = 1;
+        local->cont.lease.ret_lease = *lease;
+    }
+
+    child_index++;
+    if (child_index < priv->child_count) {
+        STACK_WIND_COOKIE(frame, afr_lease_cbk, (void *)(long)child_index,
+                          priv->children[child_index],
+                          priv->children[child_index]->fops->lease, &local->loc,
+                          &local->cont.lease.user_lease, xdata);
+    } else if (priv->quorum_count &&
+               !afr_has_quorum(local->cont.lease.locked_nodes, this, NULL)) {
+        local->op_ret = -1;
+        local->op_errno = afr_final_errno(local, priv);
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (statfs, frame, local->op_ret, local->op_errno,
-                                  &local->cont.statfs.buf, xdata);
+        afr_lease_unlock(frame, this);
+    } else {
+        if (local->op_ret < 0)
+            local->op_errno = afr_final_errno(local, priv);
+        AFR_STACK_UNWIND(lease, frame, local->op_ret, local->op_errno,
+                         &local->cont.lease.ret_lease, NULL);
+    }
 
-        return 0;
+    return 0;
 }
 
-
-int32_t
-afr_statfs (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, dict_t *xdata)
+int
+afr_lease(call_frame_t *frame, xlator_t *this, loc_t *loc,
+          struct gf_lease *lease, dict_t *xdata)
 {
-        afr_private_t *  priv        = NULL;
-        int              child_count = 0;
-        afr_local_t   *  local       = NULL;
-        int              i           = 0;
-        int              ret = -1;
-        int              call_count = 0;
-        int32_t          op_errno    = 0;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int32_t op_errno = ENOMEM;
 
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
-        VALIDATE_OR_GOTO (loc, out);
+    priv = this->private;
 
-        priv = this->private;
-        child_count = priv->child_count;
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    local->op = GF_FOP_LEASE;
+    local->cont.lease.locked_nodes = GF_CALLOC(
+        priv->child_count, sizeof(*local->cont.lease.locked_nodes),
+        gf_afr_mt_char);
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    if (!local->cont.lease.locked_nodes) {
+        op_errno = ENOMEM;
+        goto out;
+    }
 
-        call_count = local->call_count;
+    loc_copy(&local->loc, loc);
+    local->cont.lease.user_lease = *lease;
+    local->cont.lease.ret_lease = *lease;
 
-        for (i = 0; i < child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND (frame, afr_statfs_cbk,
-                                    priv->children[i],
-                                    priv->children[i]->fops->statfs,
-                                    loc, xdata);
-                        if (!--call_count)
-                                break;
-                }
-        }
+    STACK_WIND_COOKIE(frame, afr_lease_cbk, (void *)(long)0, priv->children[0],
+                      priv->children[0]->fops->lease, loc, lease, xdata);
 
-        ret = 0;
+    return 0;
 out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (statfs, frame, -1, op_errno, NULL, NULL);
-        return 0;
-}
-
+    AFR_STACK_UNWIND(lease, frame, -1, op_errno, NULL, NULL);
 
-int32_t
-afr_lk_unlock_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct gf_flock *lock,
-                   dict_t *xdata)
-{
-        afr_local_t * local = NULL;
-        int call_count = -1;
+    return 0;
+}
 
-        local = frame->local;
-        call_count = afr_frame_return (frame);
+int
+afr_ipc_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int child_index = (long)cookie;
+    int call_count = 0;
+    gf_boolean_t failed = _gf_false;
+    gf_boolean_t succeeded = _gf_false;
+    int i = 0;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    local->replies[child_index].valid = 1;
+    local->replies[child_index].op_ret = op_ret;
+    local->replies[child_index].op_errno = op_errno;
+    if (xdata)
+        local->replies[child_index].xdata = dict_ref(xdata);
+
+    call_count = afr_frame_return(frame);
+    if (call_count)
+        goto out;
+    /* If any of the subvolumes failed with other than ENOTCONN
+     * return error else return success unless all the subvolumes
+     * failed.
+     * TODO: In case of failure, we need to unregister the xattrs
+     * from the other subvolumes where it succeeded (once upcall
+     * fixes the Bz-1371622)*/
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
+        if (local->replies[i].op_ret < 0 &&
+            local->replies[i].op_errno != ENOTCONN) {
+            local->op_ret = local->replies[i].op_ret;
+            local->op_errno = local->replies[i].op_errno;
+            if (local->xdata_rsp)
+                dict_unref(local->xdata_rsp);
+            local->xdata_rsp = NULL;
+            if (local->replies[i].xdata) {
+                local->xdata_rsp = dict_ref(local->replies[i].xdata);
+            }
+            failed = _gf_true;
+            break;
+        }
+        if (local->replies[i].op_ret == 0) {
+            succeeded = _gf_true;
+            local->op_ret = 0;
+            local->op_errno = 0;
+            if (!local->xdata_rsp && local->replies[i].xdata) {
+                local->xdata_rsp = dict_ref(local->replies[i].xdata);
+            }
+        }
+    }
+
+    if (!succeeded && !failed) {
+        local->op_ret = -1;
+        local->op_errno = ENOTCONN;
+    }
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (lk, frame, local->op_ret, local->op_errno,
-                                  lock, xdata);
+    AFR_STACK_UNWIND(ipc, frame, local->op_ret, local->op_errno,
+                     local->xdata_rsp);
 
-        return 0;
+out:
+    return 0;
 }
 
-
-int32_t
-afr_lk_unlock (call_frame_t *frame, xlator_t *this)
+int
+afr_ipc(call_frame_t *frame, xlator_t *this, int32_t op, dict_t *xdata)
 {
-        afr_local_t   * local = NULL;
-        afr_private_t * priv  = NULL;
-        int i = 0;
-        int call_count = 0;
+    afr_local_t *local = NULL;
+    int32_t op_errno = -1;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int call_cnt = -1;
 
-        local = frame->local;
-        priv  = this->private;
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
 
-        call_count = afr_locked_nodes_count (local->cont.lk.locked_nodes,
-                                             priv->child_count);
+    if (op != GF_IPC_TARGET_UPCALL)
+        goto wind_default;
 
-        if (call_count == 0) {
-                AFR_STACK_UNWIND (lk, frame, local->op_ret, local->op_errno,
-                                  &local->cont.lk.ret_flock, NULL);
-                return 0;
-        }
+    VALIDATE_OR_GOTO(this->private, err);
+    priv = this->private;
 
-        local->call_count = call_count;
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto err;
 
-        local->cont.lk.user_flock.l_type = F_UNLCK;
+    call_cnt = local->call_count;
 
+    if (xdata) {
         for (i = 0; i < priv->child_count; i++) {
-                if (local->cont.lk.locked_nodes[i]) {
-                        STACK_WIND (frame, afr_lk_unlock_cbk,
-                                    priv->children[i],
-                                    priv->children[i]->fops->lk,
-                                    local->fd, F_SETLK,
-                                    &local->cont.lk.user_flock, NULL);
-
-                        if (!--call_count)
-                                break;
-                }
+            if (dict_set_int8(xdata, priv->pending_key[i], 0) < 0)
+                goto err;
         }
+    }
 
-        return 0;
-}
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->child_up[i])
+            continue;
 
+        STACK_WIND_COOKIE(frame, afr_ipc_cbk, (void *)(long)i,
+                          priv->children[i], priv->children[i]->fops->ipc, op,
+                          xdata);
+        if (!--call_cnt)
+            break;
+    }
+    return 0;
 
-int32_t
-afr_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-            int32_t op_ret, int32_t op_errno, struct gf_flock *lock, dict_t *xdata)
+err:
+    if (op_errno == -1)
+        op_errno = errno;
+    AFR_STACK_UNWIND(ipc, frame, -1, op_errno, NULL);
+
+    return 0;
+
+wind_default:
+    STACK_WIND(frame, default_ipc_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ipc, op, xdata);
+    return 0;
+}
+
+int
+afr_forget(xlator_t *this, inode_t *inode)
 {
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int child_index = -1;
-/*        int            ret  = 0; */
+    uint64_t ctx_int = 0;
+    afr_inode_ctx_t *ctx = NULL;
 
+    afr_spb_choice_timeout_cancel(this, inode);
+    inode_ctx_del(inode, this, &ctx_int);
+    if (!ctx_int)
+        return 0;
 
-        local = frame->local;
-        priv  = this->private;
+    ctx = (afr_inode_ctx_t *)(uintptr_t)ctx_int;
+    afr_inode_ctx_destroy(ctx);
+    return 0;
+}
 
-        child_index = (long) cookie;
+int
+afr_priv_dump(xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    char key_prefix[GF_DUMP_MAX_BUF_LEN];
+    char key[GF_DUMP_MAX_BUF_LEN];
+    int i = 0;
+
+    GF_ASSERT(this);
+    priv = this->private;
+
+    GF_ASSERT(priv);
+    snprintf(key_prefix, GF_DUMP_MAX_BUF_LEN, "%s.%s", this->type, this->name);
+    gf_proc_dump_add_section("%s", key_prefix);
+    gf_proc_dump_write("child_count", "%u", priv->child_count);
+    for (i = 0; i < priv->child_count; i++) {
+        sprintf(key, "child_up[%d]", i);
+        gf_proc_dump_write(key, "%d", priv->child_up[i]);
+        sprintf(key, "pending_key[%d]", i);
+        gf_proc_dump_write(key, "%s", priv->pending_key[i]);
+        sprintf(key, "pending_reads[%d]", i);
+        gf_proc_dump_write(key, "%" PRId64,
+                           GF_ATOMIC_GET(priv->pending_reads[i]));
+        sprintf(key, "child_latency[%d]", i);
+        gf_proc_dump_write(key, "%" PRId64, priv->child_latency[i]);
+        sprintf(key, "halo_child_up[%d]", i);
+        gf_proc_dump_write(key, "%d", priv->halo_child_up[i]);
+    }
+    gf_proc_dump_write("data_self_heal", "%d", priv->data_self_heal);
+    gf_proc_dump_write("metadata_self_heal", "%d", priv->metadata_self_heal);
+    gf_proc_dump_write("entry_self_heal", "%d", priv->entry_self_heal);
+    gf_proc_dump_write("read_child", "%d", priv->read_child);
+    gf_proc_dump_write("wait_count", "%u", priv->wait_count);
+    gf_proc_dump_write("heal-wait-queue-length", "%d", priv->heal_wait_qlen);
+    gf_proc_dump_write("heal-waiters", "%d", priv->heal_waiters);
+    gf_proc_dump_write("background-self-heal-count", "%d",
+                       priv->background_self_heal_count);
+    gf_proc_dump_write("healers", "%d", priv->healers);
+    gf_proc_dump_write("read-hash-mode", "%d", priv->hash_mode);
+    gf_proc_dump_write("use-anonymous-inode", "%d", priv->use_anon_inode);
+    if (priv->quorum_count == AFR_QUORUM_AUTO) {
+        gf_proc_dump_write("quorum-type", "auto");
+    } else if (priv->quorum_count == 0) {
+        gf_proc_dump_write("quorum-type", "none");
+    } else {
+        gf_proc_dump_write("quorum-type", "fixed");
+        gf_proc_dump_write("quorum-count", "%d", priv->quorum_count);
+    }
+    gf_proc_dump_write("up", "%u", afr_has_quorum(priv->child_up, this, NULL));
+    if (priv->thin_arbiter_count) {
+        gf_proc_dump_write("ta_child_up", "%d", priv->ta_child_up);
+        gf_proc_dump_write("ta_bad_child_index", "%d",
+                           priv->ta_bad_child_index);
+        gf_proc_dump_write("ta_notify_dom_lock_offset", "%" PRId64,
+                           priv->ta_notify_dom_lock_offset);
+    }
+
+    return 0;
+}
 
-        if (!child_went_down (op_ret, op_errno) && (op_ret == -1)) {
-                local->op_ret   = -1;
-                local->op_errno = op_errno;
+/**
+ * find_child_index - find the child's index in the array of subvolumes
+ * @this: AFR
+ * @child: child
+ */
 
-                afr_lk_unlock (frame, this);
-                return 0;
-        }
+static int
+afr_find_child_index(xlator_t *this, xlator_t *child)
+{
+    afr_private_t *priv = NULL;
+    int child_count = -1;
+    int i = -1;
 
-        if (op_ret == 0) {
-                local->op_ret        = 0;
-                local->op_errno      = 0;
-                local->cont.lk.locked_nodes[child_index] = 1;
-                local->cont.lk.ret_flock = *lock;
-        }
+    priv = this->private;
+    child_count = priv->child_count;
+    if (priv->thin_arbiter_count) {
+        child_count++;
+    }
 
-        child_index++;
+    for (i = 0; i < child_count; i++) {
+        if ((xlator_t *)child == priv->children[i])
+            break;
+    }
 
-        if (child_index < priv->child_count) {
-                STACK_WIND_COOKIE (frame, afr_lk_cbk, (void *) (long) child_index,
-                                   priv->children[child_index],
-                                   priv->children[child_index]->fops->lk,
-                                   local->fd, local->cont.lk.cmd,
-                                   &local->cont.lk.user_flock, xdata);
-        } else if (local->op_ret == -1) {
-                /* all nodes have gone down */
+    return i;
+}
 
-                AFR_STACK_UNWIND (lk, frame, -1, ENOTCONN,
-                                  &local->cont.lk.ret_flock, NULL);
-        } else {
-                /* locking has succeeded on all nodes that are up */
+int
+__afr_get_up_children_count(afr_private_t *priv)
+{
+    int up_children = 0;
+    int i = 0;
 
-                /* temporarily
-                   ret = afr_mark_locked_nodes (this, local->fd,
-                   local->cont.lk.locked_nodes);
-                   if (ret)
-                   gf_log (this->name, GF_LOG_DEBUG,
-                   "Could not save locked nodes info in fdctx");
+    for (i = 0; i < priv->child_count; i++)
+        if (priv->child_up[i] == 1)
+            up_children++;
 
-                   ret = afr_save_locked_fd (this, local->fd);
-                   if (ret)
-                   gf_log (this->name, GF_LOG_DEBUG,
-                   "Could not save locked fd");
+    return up_children;
+}
 
-                */
-                AFR_STACK_UNWIND (lk, frame, local->op_ret, local->op_errno,
-                                  &local->cont.lk.ret_flock, NULL);
-        }
+static int
+__get_heard_from_all_status(xlator_t *this)
+{
+    afr_private_t *priv = this->private;
+    int i;
 
+    for (i = 0; i < priv->child_count; i++) {
+        if (!priv->last_event[i]) {
+            return 0;
+        }
+    }
+    if (priv->thin_arbiter_count && !priv->ta_child_up) {
         return 0;
+    }
+    return 1;
 }
 
-
-int
-afr_lk (call_frame_t *frame, xlator_t *this,
-        fd_t *fd, int32_t cmd, struct gf_flock *flock, dict_t *xdata)
+glusterfs_event_t
+__afr_transform_event_from_state(xlator_t *this)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t *local = NULL;
-        int i = 0;
-        int32_t op_errno = 0;
-        int     ret      = -1;
+    int i = 0;
+    int up_children = 0;
+    afr_private_t *priv = this->private;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    if (__get_heard_from_all_status(this))
+        /* have_heard_from_all. Let afr_notify() do the propagation. */
+        return GF_EVENT_MAXVAL;
 
-        priv = this->private;
+    up_children = __afr_get_up_children_count(priv);
+    /* Treat the children with pending notification, as having sent a
+     * GF_EVENT_CHILD_DOWN. i.e. set the event as GF_EVENT_SOME_DESCENDENT_DOWN,
+     * as done in afr_notify() */
+    for (i = 0; i < priv->child_count; i++) {
+        if (priv->last_event[i])
+            continue;
+        priv->last_event[i] = GF_EVENT_SOME_DESCENDENT_DOWN;
+        priv->child_up[i] = 0;
+    }
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    if (up_children)
+        /* We received at least one child up */
+        return GF_EVENT_CHILD_UP;
+    else
+        return GF_EVENT_CHILD_DOWN;
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    return GF_EVENT_MAXVAL;
+}
+
+static void
+afr_notify_cbk(void *data)
+{
+    xlator_t *this = data;
+    afr_private_t *priv = this->private;
+    glusterfs_event_t event = GF_EVENT_MAXVAL;
+    gf_boolean_t propagate = _gf_false;
+
+    LOCK(&priv->lock);
+    {
+        if (!priv->timer) {
+            /*
+             * Either child_up/child_down is already sent to parent.
+             * This is a spurious wake up.
+             */
+            goto unlock;
+        }
+        priv->timer = NULL;
+        event = __afr_transform_event_from_state(this);
+        if (event != GF_EVENT_MAXVAL)
+            propagate = _gf_true;
+    }
+unlock:
+    UNLOCK(&priv->lock);
+    if (propagate)
+        default_notify(this, event, NULL);
+}
 
-        local->cont.lk.locked_nodes = GF_CALLOC (priv->child_count,
-                                                 sizeof (*local->cont.lk.locked_nodes),
-                                                 gf_afr_mt_char);
+static void
+__afr_launch_notify_timer(xlator_t *this, afr_private_t *priv)
+{
+    struct timespec delay = {
+        0,
+    };
 
-        if (!local->cont.lk.locked_nodes) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+    gf_msg_debug(this->name, 0, "Initiating child-down timer");
+    delay.tv_sec = 10;
+    delay.tv_nsec = 0;
+    priv->timer = gf_timer_call_after(this->ctx, delay, afr_notify_cbk, this);
+    if (priv->timer == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_TIMER_CREATE_FAIL,
+               "Cannot create timer for delayed initialization");
+    }
+}
 
-        local->fd            = fd_ref (fd);
-        local->cont.lk.cmd   = cmd;
-        local->cont.lk.user_flock = *flock;
-        local->cont.lk.ret_flock = *flock;
+static int
+find_best_down_child(xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    int i = -1;
+    int32_t best_child = -1;
+    int64_t best_latency = INT64_MAX;
 
-        STACK_WIND_COOKIE (frame, afr_lk_cbk, (void *) (long) 0,
-                           priv->children[i],
-                           priv->children[i]->fops->lk,
-                           fd, cmd, flock, xdata);
+    priv = this->private;
 
-        ret = 0;
-out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (lk, frame, -1, op_errno, NULL, NULL);
-        return 0;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!priv->child_up[i] && priv->child_latency[i] >= 0 &&
+            priv->child_latency[i] < best_latency) {
+            best_child = i;
+            best_latency = priv->child_latency[i];
+        }
+    }
+    if (best_child >= 0) {
+        gf_msg_debug(this->name, 0,
+                     "Found best down child (%d) @ %" PRId64 " ms latency",
+                     best_child, best_latency);
+    }
+    return best_child;
 }
 
 int
-afr_forget (xlator_t *this, inode_t *inode)
+find_worst_up_child(xlator_t *this)
 {
-        uint64_t        ctx_addr = 0;
-        afr_inode_ctx_t *ctx     = NULL;
+    afr_private_t *priv = NULL;
+    int i = -1;
+    int32_t worst_child = -1;
+    int64_t worst_latency = INT64_MIN;
 
-        inode_ctx_get (inode, this, &ctx_addr);
+    priv = this->private;
 
-        if (!ctx_addr)
-                goto out;
+    for (i = 0; i < priv->child_count; i++) {
+        if (priv->child_up[i] && priv->child_latency[i] >= 0 &&
+            priv->child_latency[i] > worst_latency) {
+            worst_child = i;
+            worst_latency = priv->child_latency[i];
+        }
+    }
+    if (worst_child >= 0) {
+        gf_msg_debug(this->name, 0,
+                     "Found worst up child (%d) @ %" PRId64 " ms latency",
+                     worst_child, worst_latency);
+    }
+    return worst_child;
+}
 
-        ctx = (afr_inode_ctx_t *)(long)ctx_addr;
-        GF_FREE (ctx->fresh_children);
-        GF_FREE (ctx);
-out:
-        return 0;
+void
+__afr_handle_ping_event(xlator_t *this, xlator_t *child_xlator, const int idx,
+                        int64_t halo_max_latency_msec, int32_t *event,
+                        int64_t child_latency_msec)
+{
+    afr_private_t *priv = NULL;
+    int up_children = 0;
+
+    priv = this->private;
+
+    priv->child_latency[idx] = child_latency_msec;
+    gf_msg_debug(child_xlator->name, 0, "Client ping @ %" PRId64 " ms",
+                 child_latency_msec);
+    if (priv->shd.iamshd)
+        return;
+
+    up_children = __afr_get_up_children_count(priv);
+
+    if (child_latency_msec > halo_max_latency_msec &&
+        priv->child_up[idx] == 1 && up_children > priv->halo_min_replicas) {
+        if ((up_children - 1) < priv->halo_min_replicas) {
+            gf_log(child_xlator->name, GF_LOG_INFO,
+                   "Overriding halo threshold, "
+                   "min replicas: %d",
+                   priv->halo_min_replicas);
+        } else {
+            gf_log(child_xlator->name, GF_LOG_INFO,
+                   "Child latency (%" PRId64
+                   " ms) "
+                   "exceeds halo threshold (%" PRId64
+                   "), "
+                   "marking child down.",
+                   child_latency_msec, halo_max_latency_msec);
+            if (priv->halo_child_up[idx]) {
+                *event = GF_EVENT_CHILD_DOWN;
+            }
+        }
+    } else if (child_latency_msec < halo_max_latency_msec &&
+               priv->child_up[idx] == 0) {
+        if (up_children < priv->halo_max_replicas) {
+            gf_log(child_xlator->name, GF_LOG_INFO,
+                   "Child latency (%" PRId64
+                   " ms) "
+                   "below halo threshold (%" PRId64
+                   "), "
+                   "marking child up.",
+                   child_latency_msec, halo_max_latency_msec);
+            if (priv->halo_child_up[idx]) {
+                *event = GF_EVENT_CHILD_UP;
+            }
+        } else {
+            gf_log(child_xlator->name, GF_LOG_INFO,
+                   "Not marking child %d up, "
+                   "max replicas (%d) reached.",
+                   idx, priv->halo_max_replicas);
+        }
+    }
 }
 
-int
-afr_priv_dump (xlator_t *this)
+static int64_t
+afr_get_halo_latency(xlator_t *this)
 {
-        afr_private_t *priv = NULL;
-        char  key_prefix[GF_DUMP_MAX_BUF_LEN];
-        char  key[GF_DUMP_MAX_BUF_LEN];
-        int   i = 0;
+    afr_private_t *priv = NULL;
+    int64_t halo_max_latency_msec = 0;
 
+    priv = this->private;
 
-        GF_ASSERT (this);
-        priv = this->private;
+    if (priv->shd.iamshd) {
+        halo_max_latency_msec = priv->shd.halo_max_latency_msec;
+    } else if (priv->nfsd.iamnfsd) {
+        halo_max_latency_msec = priv->nfsd.halo_max_latency_msec;
+    } else {
+        halo_max_latency_msec = priv->halo_max_latency_msec;
+    }
+    gf_msg_debug(this->name, 0, "Using halo latency %" PRId64,
+                 halo_max_latency_msec);
+    return halo_max_latency_msec;
+}
 
-        GF_ASSERT (priv);
-        snprintf(key_prefix, GF_DUMP_MAX_BUF_LEN, "%s.%s", this->type, this->name);
-        gf_proc_dump_add_section(key_prefix);
-        gf_proc_dump_write("child_count", "%u", priv->child_count);
-        gf_proc_dump_write("read_child_rr", "%u", priv->read_child_rr);
-        for (i = 0; i < priv->child_count; i++) {
-                sprintf (key, "child_up[%d]", i);
-                gf_proc_dump_write(key, "%d", priv->child_up[i]);
-                sprintf (key, "pending_key[%d]", i);
-                gf_proc_dump_write(key, "%s", priv->pending_key[i]);
-        }
-        gf_proc_dump_write("data_self_heal", "%s", priv->data_self_heal);
-        gf_proc_dump_write("metadata_self_heal", "%d", priv->metadata_self_heal);
-        gf_proc_dump_write("entry_self_heal", "%d", priv->entry_self_heal);
-        gf_proc_dump_write("data_change_log", "%d", priv->data_change_log);
-        gf_proc_dump_write("metadata_change_log", "%d", priv->metadata_change_log);
-        gf_proc_dump_write("entry-change_log", "%d", priv->entry_change_log);
-        gf_proc_dump_write("read_child", "%d", priv->read_child);
-        gf_proc_dump_write("favorite_child", "%d", priv->favorite_child);
-        gf_proc_dump_write("wait_count", "%u", priv->wait_count);
+void
+__afr_handle_child_up_event(xlator_t *this, xlator_t *child_xlator,
+                            const int idx, int64_t child_latency_msec,
+                            int32_t *event, int32_t *call_psh,
+                            int32_t *up_child)
+{
+    afr_private_t *priv = NULL;
+    int up_children = 0;
+    int worst_up_child = -1;
+    int64_t halo_max_latency_msec = afr_get_halo_latency(this);
+
+    priv = this->private;
+
+    /*
+     * This only really counts if the child was never up
+     * (value = -1) or had been down (value = 0).  See
+     * comment at GF_EVENT_CHILD_DOWN for a more detailed
+     * explanation.
+     */
+    if (priv->child_up[idx] != 1) {
+        priv->event_generation++;
+    }
+    priv->child_up[idx] = 1;
+
+    *call_psh = 1;
+    *up_child = idx;
+    up_children = __afr_get_up_children_count(priv);
+    /*
+     * If this is an _actual_ CHILD_UP event, we
+     * want to set the child_latency to MAX to indicate
+     * the child needs ping data to be available before doing child-up
+     */
+    if (!priv->halo_enabled)
+        goto out;
+
+    if (child_latency_msec < 0) {
+        /*set to INT64_MAX-1 so that it is found for best_down_child*/
+        priv->halo_child_up[idx] = 1;
+        if (priv->child_latency[idx] < 0) {
+            priv->child_latency[idx] = AFR_HALO_MAX_LATENCY;
+        }
+    }
+
+    /*
+     * Handle the edge case where we exceed
+     * halo_min_replicas and we've got a child which is
+     * marked up as it was helping to satisfy the
+     * halo_min_replicas even though it's latency exceeds
+     * halo_max_latency_msec.
+     */
+    if (up_children > priv->halo_min_replicas) {
+        worst_up_child = find_worst_up_child(this);
+        if (worst_up_child >= 0 &&
+            priv->child_latency[worst_up_child] > halo_max_latency_msec) {
+            gf_msg_debug(this->name, 0,
+                         "Marking child %d down, "
+                         "doesn't meet halo threshold (%" PRId64
+                         "), and > "
+                         "halo_min_replicas (%d)",
+                         worst_up_child, halo_max_latency_msec,
+                         priv->halo_min_replicas);
+            priv->child_up[worst_up_child] = 0;
+            up_children--;
+        }
+    }
+
+    if (up_children > priv->halo_max_replicas && !priv->shd.iamshd) {
+        worst_up_child = find_worst_up_child(this);
+        if (worst_up_child < 0) {
+            worst_up_child = idx;
+        }
+        priv->child_up[worst_up_child] = 0;
+        up_children--;
+        gf_msg_debug(this->name, 0,
+                     "Marking child %d down, "
+                     "up_children (%d) > halo_max_replicas (%d)",
+                     worst_up_child, up_children, priv->halo_max_replicas);
+    }
+out:
+    if (up_children == 1) {
+        gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_SUBVOL_UP,
+               "Subvolume '%s' came back up; "
+               "going online.",
+               child_xlator->name);
+        gf_event(EVENT_AFR_SUBVOL_UP, "client-pid=%d; subvol=%s",
+                 this->ctx->cmd_args.client_pid, this->name);
+    } else {
+        *event = GF_EVENT_SOME_DESCENDENT_UP;
+    }
 
-        return 0;
+    priv->last_event[idx] = *event;
+}
+
+void
+__afr_handle_child_down_event(xlator_t *this, xlator_t *child_xlator, int idx,
+                              int64_t child_latency_msec, int32_t *event,
+                              int32_t *call_psh, int32_t *up_child)
+{
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int up_children = 0;
+    int down_children = 0;
+    int best_down_child = -1;
+
+    priv = this->private;
+
+    /*
+     * If a brick is down when we start, we'll get a
+     * CHILD_DOWN to indicate its initial state.  There
+     * was never a CHILD_UP in this case, so if we
+     * increment "down_count" the difference between than
+     * and "up_count" will no longer be the number of
+     * children that are currently up.  This has serious
+     * implications e.g. for quorum enforcement, so we
+     * don't increment these values unless the event
+     * represents an actual state transition between "up"
+     * (value = 1) and anything else.
+     */
+    if (priv->child_up[idx] == 1) {
+        priv->event_generation++;
+    }
+
+    /*
+     * If this is an _actual_ CHILD_DOWN event, we
+     * want to set the child_latency to < 0 to indicate
+     * the child is really disconnected.
+     */
+    if (child_latency_msec < 0) {
+        priv->child_latency[idx] = child_latency_msec;
+        priv->halo_child_up[idx] = 0;
+    }
+    priv->child_up[idx] = 0;
+
+    up_children = __afr_get_up_children_count(priv);
+    /*
+     * Handle the edge case where we need to find the
+     * next best child (to mark up) as marking this child
+     * down would cause us to fall below halo_min_replicas.
+     * We will also force the SHD to heal this child _now_
+     * as we want it to be up to date if we are going to
+     * begin using it synchronously.
+     */
+    if (priv->halo_enabled && up_children < priv->halo_min_replicas) {
+        best_down_child = find_best_down_child(this);
+        if (best_down_child >= 0) {
+            gf_msg_debug(this->name, 0,
+                         "Swapping out child %d for "
+                         "child %d to satisfy halo_min_replicas (%d).",
+                         idx, best_down_child, priv->halo_min_replicas);
+            priv->child_up[best_down_child] = 1;
+            *call_psh = 1;
+            *up_child = best_down_child;
+        }
+    }
+    for (i = 0; i < priv->child_count; i++)
+        if (priv->child_up[i] == 0)
+            down_children++;
+    if (down_children == priv->child_count) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SUBVOLS_DOWN,
+               "All subvolumes are down. Going "
+               "offline until at least one of them "
+               "comes back up.");
+        gf_event(EVENT_AFR_SUBVOLS_DOWN, "client-pid=%d; subvol=%s",
+                 this->ctx->cmd_args.client_pid, this->name);
+    } else {
+        *event = GF_EVENT_SOME_DESCENDENT_DOWN;
+    }
+    priv->last_event[idx] = *event;
 }
 
+void
+afr_ta_lock_release_synctask(xlator_t *this)
+{
+    call_frame_t *ta_frame = NULL;
+    int ret = 0;
 
-/**
- * find_child_index - find the child's index in the array of subvolumes
- * @this: AFR
- * @child: child
- */
+    ta_frame = afr_ta_frame_create(this);
+    if (!ta_frame) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_THIN_ARB,
+               "Failed to create ta_frame");
+        return;
+    }
 
-static int
-find_child_index (xlator_t *this, xlator_t *child)
+    ret = synctask_new(this->ctx->env, afr_release_notify_lock_for_ta,
+                       afr_ta_lock_release_done, ta_frame, this);
+    if (ret) {
+        STACK_DESTROY(ta_frame->root);
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_THIN_ARB,
+               "Failed to release "
+               "AFR_TA_DOM_NOTIFY lock.");
+    }
+}
+
+static void
+afr_handle_inodelk_contention(xlator_t *this, struct gf_upcall *upcall)
 {
-        afr_private_t *priv = NULL;
-        int i = -1;
+    struct gf_upcall_inodelk_contention *lc = NULL;
+    unsigned int inmem_count = 0;
+    unsigned int onwire_count = 0;
+    afr_private_t *priv = this->private;
 
-        priv = this->private;
+    lc = upcall->data;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if ((xlator_t *) child == priv->children[i])
-                        break;
-        }
+    if (strcmp(lc->domain, AFR_TA_DOM_NOTIFY) != 0)
+        return;
 
-        return i;
+    if (priv->shd.iamshd) {
+        /* shd should ignore AFR_TA_DOM_NOTIFY release requests. */
+        return;
+    }
+    LOCK(&priv->lock);
+    {
+        if (priv->release_ta_notify_dom_lock == _gf_true) {
+            /* Ignore multiple release requests from shds.*/
+            UNLOCK(&priv->lock);
+            return;
+        }
+        priv->release_ta_notify_dom_lock = _gf_true;
+        inmem_count = priv->ta_in_mem_txn_count;
+        onwire_count = priv->ta_on_wire_txn_count;
+    }
+    UNLOCK(&priv->lock);
+    if (inmem_count || onwire_count)
+        /* lock release will happen in txn code path after
+         * in-memory or on-wire txns are over.*/
+        return;
+
+    afr_ta_lock_release_synctask(this);
+}
+
+static void
+afr_handle_upcall_event(xlator_t *this, struct gf_upcall *upcall)
+{
+    struct gf_upcall_cache_invalidation *up_ci = NULL;
+    afr_private_t *priv = this->private;
+    inode_t *inode = NULL;
+    inode_table_t *itable = NULL;
+    int i = 0;
+
+    switch (upcall->event_type) {
+        case GF_UPCALL_INODELK_CONTENTION:
+            afr_handle_inodelk_contention(this, upcall);
+            break;
+        case GF_UPCALL_CACHE_INVALIDATION:
+            up_ci = (struct gf_upcall_cache_invalidation *)upcall->data;
+
+            /* Since md-cache will be aggressively filtering
+             * lookups, the stale read issue will be more
+             * pronounced. Hence when a pending xattr is set notify
+             * all the md-cache clients to invalidate the existing
+             * stat cache and send the lookup next time */
+            if (!up_ci->dict)
+                break;
+            for (i = 0; i < priv->child_count; i++) {
+                if (!dict_get(up_ci->dict, priv->pending_key[i]))
+                    continue;
+                up_ci->flags |= UP_INVAL_ATTR;
+                itable = ((xlator_t *)this->graph->top)->itable;
+                /*Internal processes may not have itable for
+                 *top xlator*/
+                if (itable)
+                    inode = inode_find(itable, upcall->gfid);
+                if (inode)
+                    afr_inode_need_refresh_set(inode, this);
+                break;
+            }
+            break;
+        default:
+            break;
+    }
 }
 
 int32_t
-afr_notify (xlator_t *this, int32_t event,
-            void *data, void *data2)
-{
-        afr_private_t   *priv               = NULL;
-        int             i                   = -1;
-        int             up_children         = 0;
-        int             down_children       = 0;
-        int             propagate           = 0;
-        int             had_heard_from_all  = 0;
-        int             have_heard_from_all = 0;
-        int             idx                 = -1;
-        int             ret                 = -1;
-        int             call_psh            = 0;
-        int             up_child            = AFR_ALL_CHILDREN;
-        dict_t          *input              = NULL;
-        dict_t          *output             = NULL;
-
-        priv = this->private;
-
-        if (!priv)
-                return 0;
-
-        /*
-         * We need to reset this in case children come up in "staggered"
-         * fashion, so that we discover a late-arriving local subvolume.  Note
-         * that we could end up issuing N lookups to the first subvolume, and
-         * O(N^2) overall, but N is small for AFR so it shouldn't be an issue.
-         */
-        priv->did_discovery = _gf_false;
+afr_notify(xlator_t *this, int32_t event, void *data, void *data2)
+{
+    afr_private_t *priv = NULL;
+    xlator_t *child_xlator = NULL;
+    int i = -1;
+    int propagate = 0;
+    int had_heard_from_all = 0;
+    int have_heard_from_all = 0;
+    int idx = -1;
+    int ret = -1;
+    int call_psh = 0;
+    int up_child = -1;
+    dict_t *input = NULL;
+    dict_t *output = NULL;
+    gf_boolean_t had_quorum = _gf_false;
+    gf_boolean_t has_quorum = _gf_false;
+    int64_t halo_max_latency_msec = 0;
+    int64_t child_latency_msec = -1;
+
+    child_xlator = (xlator_t *)data;
+
+    priv = this->private;
+
+    if (!priv)
+        return 0;
 
-        had_heard_from_all = 1;
-        for (i = 0; i < priv->child_count; i++) {
-                if (!priv->last_event[i]) {
-                        had_heard_from_all = 0;
-                }
+    /*
+     * We need to reset this in case children come up in "staggered"
+     * fashion, so that we discover a late-arriving local subvolume.  Note
+     * that we could end up issuing N lookups to the first subvolume, and
+     * O(N^2) overall, but N is small for AFR so it shouldn't be an issue.
+     */
+    priv->did_discovery = _gf_false;
+
+    /* parent xlators don't need to know about every child_up, child_down
+     * because of afr ha. If all subvolumes go down, child_down has
+     * to be triggered. In that state when 1 subvolume comes up child_up
+     * needs to be triggered. dht optimizes revalidate lookup by sending
+     * it only to one of its subvolumes. When child up/down happens
+     * for afr's subvolumes dht should be notified by child_modified. The
+     * subsequent revalidate lookup happens on all the dht's subvolumes
+     * which triggers afr self-heals if any.
+     */
+    idx = afr_find_child_index(this, child_xlator);
+    if (idx < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_INVALID_CHILD_UP,
+               "Received child_up from invalid subvolume");
+        goto out;
+    }
+
+    had_quorum = priv->quorum_count &&
+                 afr_has_quorum(priv->child_up, this, NULL);
+    if (event == GF_EVENT_CHILD_PING) {
+        child_latency_msec = (int64_t)(uintptr_t)data2;
+        if (priv->halo_enabled) {
+            halo_max_latency_msec = afr_get_halo_latency(this);
+
+            /* Calculates the child latency and sets event
+             */
+            LOCK(&priv->lock);
+            {
+                __afr_handle_ping_event(this, child_xlator, idx,
+                                        halo_max_latency_msec, &event,
+                                        child_latency_msec);
+            }
+            UNLOCK(&priv->lock);
+        } else {
+            LOCK(&priv->lock);
+            {
+                priv->child_latency[idx] = child_latency_msec;
+            }
+            UNLOCK(&priv->lock);
         }
+    }
 
-        /* parent xlators dont need to know about every child_up, child_down
-         * because of afr ha. If all subvolumes go down, child_down has
-         * to be triggered. In that state when 1 subvolume comes up child_up
-         * needs to be triggered. dht optimizes revalidate lookup by sending
-         * it only to one of its subvolumes. When child up/down happens
-         * for afr's subvolumes dht should be notified by child_modified. The
-         * subsequent revalidate lookup happens on all the dht's subvolumes
-         * which triggers afr self-heals if any.
+    if (event == GF_EVENT_CHILD_PING) {
+        /* This is the only xlator that handles PING, no reason to
+         * propagate.
          */
-        idx = find_child_index (this, data);
-        if (idx < 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Received child_up "
-                        "from invalid subvolume");
-                goto out;
+        goto out;
+    }
+
+    if (event == GF_EVENT_TRANSLATOR_OP) {
+        LOCK(&priv->lock);
+        {
+            had_heard_from_all = __get_heard_from_all_status(this);
+        }
+        UNLOCK(&priv->lock);
+
+        if (!had_heard_from_all) {
+            ret = -1;
+        } else {
+            input = data;
+            output = data2;
+            ret = afr_xl_op(this, input, output);
         }
+        goto out;
+    }
 
+    if (event == GF_EVENT_UPCALL) {
+        afr_handle_upcall_event(this, data);
+    }
+
+    LOCK(&priv->lock);
+    {
+        had_heard_from_all = __get_heard_from_all_status(this);
         switch (event) {
-        case GF_EVENT_CHILD_UP:
-                LOCK (&priv->lock);
-                {
-                        /*
-                         * This only really counts if the child was never up
-                         * (value = -1) or had been down (value = 0).  See
-                         * comment at GF_EVENT_CHILD_DOWN for a more detailed
-                         * explanation.
-                         */
-                        if (priv->child_up[idx] != 1) {
-                                priv->up_count++;
-                        }
-                        priv->child_up[idx] = 1;
-
-                        call_psh = 1;
-                        up_child = idx;
-                        for (i = 0; i < priv->child_count; i++)
-                                if (priv->child_up[i] == 1)
-                                        up_children++;
-                        if (up_children == 1) {
-                                gf_log (this->name, GF_LOG_INFO,
-                                        "Subvolume '%s' came back up; "
-                                        "going online.", ((xlator_t *)data)->name);
-                        } else {
-                                event = GF_EVENT_CHILD_MODIFIED;
-                        }
-
-                        priv->last_event[idx] = event;
+            case GF_EVENT_PARENT_UP:
+                __afr_launch_notify_timer(this, priv);
+                propagate = 1;
+                break;
+            case GF_EVENT_CHILD_UP:
+                if (priv->thin_arbiter_count &&
+                    (idx == AFR_CHILD_THIN_ARBITER)) {
+                    priv->ta_child_up = 1;
+                    priv->ta_event_gen++;
+                    break;
                 }
-                UNLOCK (&priv->lock);
-
+                __afr_handle_child_up_event(this, child_xlator, idx,
+                                            child_latency_msec, &event,
+                                            &call_psh, &up_child);
+                __afr_lock_heal_synctask(this, priv, idx);
                 break;
 
-        case GF_EVENT_CHILD_DOWN:
-                LOCK (&priv->lock);
-                {
-                        /*
-                         * If a brick is down when we start, we'll get a
-                         * CHILD_DOWN to indicate its initial state.  There
-                         * was never a CHILD_UP in this case, so if we
-                         * increment "down_count" the difference between than
-                         * and "up_count" will no longer be the number of
-                         * children that are currently up.  This has serious
-                         * implications e.g. for quorum enforcement, so we
-                         * don't increment these values unless the event
-                         * represents an actual state transition between "up"
-                         * (value = 1) and anything else.
-                         */
-                        if (priv->child_up[idx] == 1) {
-                                priv->down_count++;
-                        }
-                        priv->child_up[idx] = 0;
-
-                        for (i = 0; i < priv->child_count; i++)
-                                if (priv->child_up[i] == 0)
-                                        down_children++;
-                        if (down_children == priv->child_count) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "All subvolumes are down. Going offline "
-                                        "until atleast one of them comes back up.");
-                        } else {
-                                event = GF_EVENT_CHILD_MODIFIED;
-                        }
-
-                        priv->last_event[idx] = event;
+            case GF_EVENT_CHILD_DOWN:
+                if (priv->thin_arbiter_count &&
+                    (idx == AFR_CHILD_THIN_ARBITER)) {
+                    priv->ta_child_up = 0;
+                    priv->ta_event_gen++;
+                    afr_ta_locked_priv_invalidate(priv);
+                    break;
                 }
-                UNLOCK (&priv->lock);
-
+                __afr_handle_child_down_event(this, child_xlator, idx,
+                                              child_latency_msec, &event,
+                                              &call_psh, &up_child);
+                __afr_mark_pending_lk_heal(this, priv, idx);
                 break;
 
-        case GF_EVENT_CHILD_CONNECTING:
-                LOCK (&priv->lock);
-                {
-                        priv->last_event[idx] = event;
-                }
-                UNLOCK (&priv->lock);
+            case GF_EVENT_CHILD_CONNECTING:
+                priv->last_event[idx] = event;
 
                 break;
 
-        case GF_EVENT_TRANSLATOR_OP:
-                input = data;
-                output = data2;
-                ret = afr_xl_op (this, input, output);
-                goto out;
+            case GF_EVENT_SOME_DESCENDENT_DOWN:
+                priv->last_event[idx] = event;
                 break;
-
-        default:
+            default:
                 propagate = 1;
                 break;
         }
-
-        /* have all subvolumes reported status once by now? */
-        have_heard_from_all = 1;
-        for (i = 0; i < priv->child_count; i++) {
-                if (!priv->last_event[i])
-                        have_heard_from_all = 0;
-        }
-
-        /* if all subvols have reported status, no need to hide anything
-           or wait for anything else. Just propagate blindly */
-        if (have_heard_from_all)
-                propagate = 1;
-
+        have_heard_from_all = __get_heard_from_all_status(this);
         if (!had_heard_from_all && have_heard_from_all) {
-                /* This is the first event which completes aggregation
-                   of events from all subvolumes. If at least one subvol
-                   had come up, propagate CHILD_UP, but only this time
-                */
-                event = GF_EVENT_CHILD_DOWN;
-
-                LOCK (&priv->lock);
-                {
-                        up_children = afr_up_children_count (priv->child_up,
-                                                             priv->child_count);
-                        for (i = 0; i < priv->child_count; i++) {
-                                if (priv->last_event[i] == GF_EVENT_CHILD_UP) {
-                                        event = GF_EVENT_CHILD_UP;
-                                        break;
-                                }
-
-                                if (priv->last_event[i] ==
-                                                GF_EVENT_CHILD_CONNECTING) {
-                                        event = GF_EVENT_CHILD_CONNECTING;
-                                        /* continue to check other events for CHILD_UP */
-                                }
-                        }
+            if (priv->timer) {
+                gf_timer_call_cancel(this->ctx, priv->timer);
+                priv->timer = NULL;
+            }
+            /* This is the first event which completes aggregation
+               of events from all subvolumes. If at least one subvol
+               had come up, propagate CHILD_UP, but only this time
+            */
+            event = GF_EVENT_CHILD_DOWN;
+            for (i = 0; i < priv->child_count; i++) {
+                if (priv->last_event[i] == GF_EVENT_CHILD_UP) {
+                    event = GF_EVENT_CHILD_UP;
+                    break;
                 }
-                UNLOCK (&priv->lock);
-        }
 
-        ret = 0;
-        if (propagate)
-                ret = default_notify (this, event, data);
-        if (call_psh && priv->shd.iamshd)
-                afr_proactive_self_heal ((void*) (long) up_child);
+                if (priv->last_event[i] == GF_EVENT_CHILD_CONNECTING) {
+                    event = GF_EVENT_CHILD_CONNECTING;
+                    /* continue to check other events for CHILD_UP */
+                }
+            }
+        }
+    }
+    UNLOCK(&priv->lock);
+
+    if (priv->quorum_count) {
+        has_quorum = afr_has_quorum(priv->child_up, this, NULL);
+        if (!had_quorum && has_quorum) {
+            gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_QUORUM_MET,
+                   "Client-quorum is met");
+            gf_event(EVENT_AFR_QUORUM_MET, "client-pid=%d; subvol=%s",
+                     this->ctx->cmd_args.client_pid, this->name);
+        }
+        if (had_quorum && !has_quorum) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_QUORUM_FAIL,
+                   "Client-quorum is not met");
+            gf_event(EVENT_AFR_QUORUM_FAIL, "client-pid=%d; subvol=%s",
+                     this->ctx->cmd_args.client_pid, this->name);
+        }
+    }
+
+    /* if all subvols have reported status, no need to hide anything
+       or wait for anything else. Just propagate blindly */
+    if (have_heard_from_all)
+        propagate = 1;
+
+    ret = 0;
+    if (propagate)
+        ret = default_notify(this, event, data);
+
+    if ((!had_heard_from_all) || call_psh) {
+        /* Launch self-heal on all local subvolumes if:
+         * a) We have_heard_from_all for the first time
+         * b) Already heard from everyone, but we now got a child-up
+         *    event.
+         */
+        if (have_heard_from_all) {
+            afr_selfheal_childup(this, priv);
+        }
+    }
+out:
+    return ret;
+}
 
+int
+afr_local_init(afr_local_t *local, afr_private_t *priv, int32_t *op_errno)
+{
+    int __ret = -1;
+    local->op_ret = -1;
+    local->op_errno = EUCLEAN;
+
+    __ret = syncbarrier_init(&local->barrier);
+    if (__ret) {
+        if (op_errno)
+            *op_errno = __ret;
+        goto out;
+    }
+
+    local->child_up = GF_MALLOC(priv->child_count * sizeof(*local->child_up),
+                                gf_afr_mt_char);
+    if (!local->child_up) {
+        if (op_errno)
+            *op_errno = ENOMEM;
+        goto out;
+    }
+
+    memcpy(local->child_up, priv->child_up,
+           sizeof(*local->child_up) * priv->child_count);
+    local->call_count = AFR_COUNT(local->child_up, priv->child_count);
+    if (local->call_count == 0) {
+        gf_msg(THIS->name, GF_LOG_INFO, 0, AFR_MSG_SUBVOLS_DOWN,
+               "no subvolumes up");
+        if (op_errno)
+            *op_errno = ENOTCONN;
+        goto out;
+    }
+
+    local->event_generation = priv->event_generation;
+
+    local->read_attempted = GF_CALLOC(priv->child_count, sizeof(char),
+                                      gf_afr_mt_char);
+    if (!local->read_attempted) {
+        if (op_errno)
+            *op_errno = ENOMEM;
+        goto out;
+    }
+
+    local->readable = GF_CALLOC(priv->child_count, sizeof(char),
+                                gf_afr_mt_char);
+    if (!local->readable) {
+        if (op_errno)
+            *op_errno = ENOMEM;
+        goto out;
+    }
+
+    local->readable2 = GF_CALLOC(priv->child_count, sizeof(char),
+                                 gf_afr_mt_char);
+    if (!local->readable2) {
+        if (op_errno)
+            *op_errno = ENOMEM;
+        goto out;
+    }
+
+    local->read_subvol = -1;
+
+    local->replies = GF_CALLOC(priv->child_count, sizeof(*local->replies),
+                               gf_afr_mt_reply_t);
+    if (!local->replies) {
+        if (op_errno)
+            *op_errno = ENOMEM;
+        goto out;
+    }
+
+    local->need_full_crawl = _gf_false;
+    if (priv->thin_arbiter_count) {
+        local->ta_child_up = priv->ta_child_up;
+        local->ta_failed_subvol = AFR_CHILD_UNKNOWN;
+        local->read_txn_query_child = AFR_CHILD_UNKNOWN;
+        local->ta_event_gen = priv->ta_event_gen;
+        local->fop_state = TA_SUCCESS;
+    }
+    local->is_new_entry = _gf_false;
+
+    INIT_LIST_HEAD(&local->healer);
+    return 0;
 out:
-        return ret;
+    return -1;
 }
 
 int
-afr_first_up_child (unsigned char *child_up, size_t child_count)
+afr_internal_lock_init(afr_internal_lock_t *lk, size_t child_count)
 {
-        int         ret      = -1;
-        int         i        = 0;
+    int ret = -ENOMEM;
 
-        GF_ASSERT (child_up);
+    lk->lower_locked_nodes = GF_CALLOC(sizeof(*lk->lower_locked_nodes),
+                                       child_count, gf_afr_mt_char);
+    if (NULL == lk->lower_locked_nodes)
+        goto out;
 
-        for (i = 0; i < child_count; i++) {
-                if (child_up[i]) {
-                        ret = i;
-                        break;
-                }
-        }
+    lk->lock_op_ret = -1;
+    lk->lock_op_errno = EUCLEAN;
 
-        return ret;
+    ret = 0;
+out:
+    return ret;
 }
 
-int
-afr_local_init (afr_local_t *local, afr_private_t *priv, int32_t *op_errno)
+void
+afr_matrix_cleanup(int32_t **matrix, unsigned int m)
 {
-        int     ret = -1;
+    int i = 0;
 
-        local->op_ret = -1;
-        local->op_errno = EUCLEAN;
-
-        local->child_up = GF_CALLOC (priv->child_count,
-                                     sizeof (*local->child_up),
-                                     gf_afr_mt_char);
-        if (!local->child_up) {
-                if (op_errno)
-                        *op_errno = ENOMEM;
-                goto out;
-        }
-
-        memcpy (local->child_up, priv->child_up,
-                sizeof (*local->child_up) * priv->child_count);
-        local->call_count = afr_up_children_count (local->child_up,
-                                                   priv->child_count);
-        if (local->call_count == 0) {
-                gf_log (THIS->name, GF_LOG_INFO, "no subvolumes up");
-                if (op_errno)
-                        *op_errno = ENOTCONN;
-                goto out;
-        }
+    if (!matrix)
+        goto out;
+    for (i = 0; i < m; i++) {
+        GF_FREE(matrix[i]);
+    }
 
-        local->child_errno = GF_CALLOC (priv->child_count,
-                                        sizeof (*local->child_errno),
-                                        gf_afr_mt_int32_t);
-        if (!local->child_errno) {
-                if (op_errno)
-                        *op_errno = ENOMEM;
-                goto out;
-        }
+    GF_FREE(matrix);
+out:
+    return;
+}
 
-        local->transaction.postop_piggybacked = GF_CALLOC (priv->child_count,
-							   sizeof (int),
-							   gf_afr_mt_int32_t);
-        if (!local->transaction.postop_piggybacked) {
-                if (op_errno)
-                        *op_errno = ENOMEM;
-                goto out;
-        }
+int32_t **
+afr_matrix_create(unsigned int m, unsigned int n)
+{
+    int32_t **matrix = NULL;
+    int i = 0;
 
-	local->append_write = _gf_false;
+    matrix = GF_CALLOC(sizeof(*matrix), m, gf_afr_mt_int32_t);
+    if (!matrix)
+        goto out;
 
-        ret = 0;
+    for (i = 0; i < m; i++) {
+        matrix[i] = GF_CALLOC(sizeof(*matrix[i]), n, gf_afr_mt_int32_t);
+        if (!matrix[i])
+            goto out;
+    }
+    return matrix;
 out:
-        return ret;
+    afr_matrix_cleanup(matrix, m);
+    return NULL;
 }
 
 int
-afr_internal_lock_init (afr_internal_lock_t *lk, size_t child_count,
-                        transaction_lk_type_t lk_type)
+afr_transaction_local_init(afr_local_t *local, xlator_t *this)
+{
+    int ret = -ENOMEM;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+    INIT_LIST_HEAD(&local->transaction.wait_list);
+    INIT_LIST_HEAD(&local->transaction.owner_list);
+    INIT_LIST_HEAD(&local->ta_waitq);
+    INIT_LIST_HEAD(&local->ta_onwireq);
+    ret = afr_internal_lock_init(&local->internal_lock, priv->child_count);
+    if (ret < 0)
+        goto out;
+
+    ret = -ENOMEM;
+    local->pre_op_compat = priv->pre_op_compat;
+
+    local->transaction.pre_op = GF_CALLOC(sizeof(*local->transaction.pre_op),
+                                          priv->child_count, gf_afr_mt_char);
+    if (!local->transaction.pre_op)
+        goto out;
+
+    local->transaction.changelog_xdata = GF_CALLOC(
+        sizeof(*local->transaction.changelog_xdata), priv->child_count,
+        gf_afr_mt_dict_t);
+    if (!local->transaction.changelog_xdata)
+        goto out;
+
+    if (priv->arbiter_count == 1) {
+        local->transaction.pre_op_sources = GF_CALLOC(
+            sizeof(*local->transaction.pre_op_sources), priv->child_count,
+            gf_afr_mt_char);
+        if (!local->transaction.pre_op_sources)
+            goto out;
+    }
+
+    local->transaction.failed_subvols = GF_CALLOC(
+        sizeof(*local->transaction.failed_subvols), priv->child_count,
+        gf_afr_mt_char);
+    if (!local->transaction.failed_subvols)
+        goto out;
+
+    local->pending = afr_matrix_create(priv->child_count, AFR_NUM_CHANGE_LOGS);
+    if (!local->pending)
+        goto out;
+
+    ret = 0;
+out:
+    return ret;
+}
+
+void
+afr_set_low_priority(call_frame_t *frame)
 {
-        int             ret = -ENOMEM;
+    frame->root->pid = LOW_PRIO_PROC_PID;
+}
 
-        lk->locked_nodes = GF_CALLOC (sizeof (*lk->locked_nodes),
-                                      child_count, gf_afr_mt_char);
-        if (NULL == lk->locked_nodes)
-                goto out;
+void
+afr_priv_destroy(afr_private_t *priv)
+{
+    int i = 0;
+    int child_count = -1;
 
-        lk->lower_locked_nodes = GF_CALLOC (sizeof (*lk->lower_locked_nodes),
-                                            child_count, gf_afr_mt_char);
-        if (NULL == lk->lower_locked_nodes)
-                goto out;
+    if (!priv)
+        goto out;
 
-        lk->lock_op_ret   = -1;
-        lk->lock_op_errno = EUCLEAN;
-        lk->transaction_lk_type = lk_type;
+    GF_FREE(priv->sh_domain);
+    GF_FREE(priv->last_event);
 
-        ret = 0;
+    child_count = priv->child_count;
+    if (priv->thin_arbiter_count) {
+        child_count++;
+    }
+    if (priv->pending_key) {
+        for (i = 0; i < child_count; i++)
+            GF_FREE(priv->pending_key[i]);
+    }
+
+    GF_FREE(priv->pending_reads);
+    GF_FREE(priv->local);
+    GF_FREE(priv->pending_key);
+    GF_FREE(priv->children);
+    GF_FREE(priv->anon_inode);
+    GF_FREE(priv->child_up);
+    GF_FREE(priv->halo_child_up);
+    GF_FREE(priv->child_latency);
+    LOCK_DESTROY(&priv->lock);
+
+    GF_FREE(priv);
 out:
-        return ret;
+    return;
 }
 
-void
-afr_matrix_cleanup (int32_t **matrix, unsigned int m)
+int **
+afr_mark_pending_changelog(afr_private_t *priv, unsigned char *pending,
+                           dict_t *xattr, ia_type_t iat)
 {
-        int             i         = 0;
+    int i = 0;
+    int **changelog = NULL;
+    int idx = -1;
+    int m_idx = 0;
+    int d_idx = 0;
+    int ret = 0;
 
-        if (!matrix)
-                goto out;
-        for (i = 0; i < m; i++) {
-                GF_FREE (matrix[i]);
-        }
+    m_idx = afr_index_for_transaction_type(AFR_METADATA_TRANSACTION);
+    d_idx = afr_index_for_transaction_type(AFR_DATA_TRANSACTION);
+
+    idx = afr_index_from_ia_type(iat);
 
-        GF_FREE (matrix);
+    changelog = afr_matrix_create(priv->child_count, AFR_NUM_CHANGE_LOGS);
+    if (!changelog)
+        goto out;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!pending[i])
+            continue;
+
+        changelog[i][m_idx] = hton32(1);
+        if (idx != -1)
+            changelog[i][idx] = hton32(1);
+        /* If the newentry marking is on a newly created directory,
+         * then mark it with the full-heal indicator.
+         */
+        if ((IA_ISDIR(iat)) && (priv->esh_granular))
+            changelog[i][d_idx] = hton32(1);
+    }
+    ret = afr_set_pending_dict(priv, xattr, changelog);
+    if (ret < 0) {
+        afr_matrix_cleanup(changelog, priv->child_count);
+        return NULL;
+    }
 out:
-        return;
+    return changelog;
 }
 
-int32_t**
-afr_matrix_create (unsigned int m, unsigned int n)
+static dict_t *
+afr_set_heal_info(char *status)
 {
-        int32_t         **matrix = NULL;
-        int             i       = 0;
+    dict_t *dict = NULL;
+    int ret = -1;
 
-        matrix = GF_CALLOC (sizeof (*matrix), m, gf_afr_mt_int32_t);
-        if (!matrix)
-                goto out;
+    dict = dict_new();
+    if (!dict) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    ret = dict_set_dynstr_sizen(dict, "heal-info", status);
+    if (ret)
+        gf_msg("", GF_LOG_WARNING, -ret, AFR_MSG_DICT_SET_FAILED,
+               "Failed to set heal-info key to "
+               "%s",
+               status);
+out:
+    /* Any error other than EINVAL, dict_set_dynstr frees status */
+    if (ret == -ENOMEM || ret == -EINVAL) {
+        GF_FREE(status);
+    }
+
+    if (ret && dict) {
+        dict_unref(dict);
+        dict = NULL;
+    }
+    return dict;
+}
+
+static gf_boolean_t
+afr_is_dirty_count_non_unary_for_txn(xlator_t *this, struct afr_reply *replies,
+                                     afr_transaction_type type)
+{
+    afr_private_t *priv = this->private;
+    int *dirty = alloca0(priv->child_count * sizeof(int));
+    int i = 0;
+
+    afr_selfheal_extract_xattr(this, replies, type, dirty, NULL);
+    for (i = 0; i < priv->child_count; i++) {
+        if (dirty[i] > 1)
+            return _gf_true;
+    }
+
+    return _gf_false;
+}
+
+static gf_boolean_t
+afr_is_dirty_count_non_unary(xlator_t *this, struct afr_reply *replies,
+                             ia_type_t ia_type)
+{
+    gf_boolean_t data_chk = _gf_false;
+    gf_boolean_t mdata_chk = _gf_false;
+    gf_boolean_t entry_chk = _gf_false;
+
+    switch (ia_type) {
+        case IA_IFDIR:
+            mdata_chk = _gf_true;
+            entry_chk = _gf_true;
+            break;
+        case IA_IFREG:
+            mdata_chk = _gf_true;
+            data_chk = _gf_true;
+            break;
+        default:
+            /*IA_IFBLK, IA_IFCHR, IA_IFLNK, IA_IFIFO, IA_IFSOCK*/
+            mdata_chk = _gf_true;
+            break;
+    }
+
+    if (data_chk && afr_is_dirty_count_non_unary_for_txn(
+                        this, replies, AFR_DATA_TRANSACTION)) {
+        return _gf_true;
+    } else if (mdata_chk && afr_is_dirty_count_non_unary_for_txn(
+                                this, replies, AFR_METADATA_TRANSACTION)) {
+        return _gf_true;
+    } else if (entry_chk && afr_is_dirty_count_non_unary_for_txn(
+                                this, replies, AFR_ENTRY_TRANSACTION)) {
+        return _gf_true;
+    }
+
+    return _gf_false;
+}
 
-        for (i = 0; i < m; i++) {
-                matrix[i] = GF_CALLOC (sizeof (*matrix[i]), n,
-                                       gf_afr_mt_int32_t);
-                if (!matrix[i])
-                        goto out;
+static int
+afr_update_heal_status(xlator_t *this, struct afr_reply *replies,
+                       ia_type_t ia_type, gf_boolean_t *esh, gf_boolean_t *dsh,
+                       gf_boolean_t *msh, unsigned char pending)
+{
+    int ret = -1;
+    GF_UNUSED int ret1 = 0;
+    int i = 0;
+    int io_domain_lk_count = 0;
+    int shd_domain_lk_count = 0;
+    afr_private_t *priv = NULL;
+    char *key1 = NULL;
+    char *key2 = NULL;
+
+    priv = this->private;
+    key1 = alloca0(strlen(GLUSTERFS_INODELK_DOM_PREFIX) + 2 +
+                   strlen(this->name));
+    key2 = alloca0(strlen(GLUSTERFS_INODELK_DOM_PREFIX) + 2 +
+                   strlen(priv->sh_domain));
+    sprintf(key1, "%s:%s", GLUSTERFS_INODELK_DOM_PREFIX, this->name);
+    sprintf(key2, "%s:%s", GLUSTERFS_INODELK_DOM_PREFIX, priv->sh_domain);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if ((replies[i].valid != 1) || (replies[i].op_ret != 0))
+            continue;
+        if (!io_domain_lk_count) {
+            ret1 = dict_get_int32(replies[i].xdata, key1, &io_domain_lk_count);
+        }
+        if (!shd_domain_lk_count) {
+            ret1 = dict_get_int32(replies[i].xdata, key2, &shd_domain_lk_count);
+        }
+    }
+
+    if (!pending) {
+        if ((afr_is_dirty_count_non_unary(this, replies, ia_type)) ||
+            (!io_domain_lk_count)) {
+            /* Needs heal. */
+            ret = 0;
+        } else {
+            /* No heal needed. */
+            *dsh = *esh = *msh = 0;
         }
-        return matrix;
-out:
-        afr_matrix_cleanup (matrix, m);
-        return NULL;
+    } else {
+        if (shd_domain_lk_count) {
+            ret = -EAGAIN; /*For 'possibly-healing'. */
+        } else {
+            ret = 0; /*needs heal. Just set a non -ve value so that it is
+                       assumed as the source index.*/
+        }
+    }
+    return ret;
 }
 
+/*return EIO, EAGAIN or pending*/
 int
-afr_inodelk_init (afr_inodelk_t *lk, char *dom, size_t child_count)
-{
-        int             ret = -ENOMEM;
+afr_lockless_inspect(call_frame_t *frame, xlator_t *this, uuid_t gfid,
+                     inode_t **inode, gf_boolean_t *entry_selfheal,
+                     gf_boolean_t *data_selfheal,
+                     gf_boolean_t *metadata_selfheal, unsigned char *pending)
+{
+    int ret = -1;
+    int i = 0;
+    afr_private_t *priv = NULL;
+    struct afr_reply *replies = NULL;
+    gf_boolean_t dsh = _gf_false;
+    gf_boolean_t msh = _gf_false;
+    gf_boolean_t esh = _gf_false;
+    unsigned char *sources = NULL;
+    unsigned char *sinks = NULL;
+    unsigned char *valid_on = NULL;
+    uint64_t *witness = NULL;
+
+    priv = this->private;
+    replies = alloca0(sizeof(*replies) * priv->child_count);
+    sources = alloca0(sizeof(*sources) * priv->child_count);
+    sinks = alloca0(sizeof(*sinks) * priv->child_count);
+    witness = alloca0(sizeof(*witness) * priv->child_count);
+    valid_on = alloca0(sizeof(*valid_on) * priv->child_count);
+
+    ret = afr_selfheal_unlocked_inspect(frame, this, gfid, inode, &dsh, &msh,
+                                        &esh, replies);
+    if (ret)
+        goto out;
+    for (i = 0; i < priv->child_count; i++) {
+        if (replies[i].valid && replies[i].op_ret == 0) {
+            valid_on[i] = 1;
+        }
+    }
+    if (msh) {
+        ret = afr_selfheal_find_direction(frame, this, replies,
+                                          AFR_METADATA_TRANSACTION, valid_on,
+                                          sources, sinks, witness, pending);
+        if (*pending & PFLAG_SBRAIN)
+            ret = -EIO;
+        if (ret)
+            goto out;
+    }
+    if (dsh) {
+        ret = afr_selfheal_find_direction(frame, this, replies,
+                                          AFR_DATA_TRANSACTION, valid_on,
+                                          sources, sinks, witness, pending);
+        if (*pending & PFLAG_SBRAIN)
+            ret = -EIO;
+        if (ret)
+            goto out;
+    }
+    if (esh) {
+        ret = afr_selfheal_find_direction(frame, this, replies,
+                                          AFR_ENTRY_TRANSACTION, valid_on,
+                                          sources, sinks, witness, pending);
+        if (*pending & PFLAG_SBRAIN)
+            ret = -EIO;
+        if (ret)
+            goto out;
+    }
 
-        lk->domain = dom;
-        lk->locked_nodes = GF_CALLOC (sizeof (*lk->locked_nodes),
-                                      child_count, gf_afr_mt_char);
-        if (NULL == lk->locked_nodes)
+    ret = afr_update_heal_status(this, replies, (*inode)->ia_type, &esh, &dsh,
+                                 &msh, *pending);
+out:
+    *data_selfheal = dsh;
+    *entry_selfheal = esh;
+    *metadata_selfheal = msh;
+    if (replies)
+        afr_replies_wipe(replies, priv->child_count);
+    return ret;
+}
+
+int
+afr_get_heal_info(call_frame_t *frame, xlator_t *this, loc_t *loc)
+{
+    gf_boolean_t data_selfheal = _gf_false;
+    gf_boolean_t metadata_selfheal = _gf_false;
+    gf_boolean_t entry_selfheal = _gf_false;
+    unsigned char pending = 0;
+    dict_t *dict = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
+    inode_t *inode = NULL;
+    char *substr = NULL;
+    char *status = NULL;
+    call_frame_t *heal_frame = NULL;
+    afr_local_t *heal_local = NULL;
+
+    /*Use frame with lk-owner set*/
+    heal_frame = afr_frame_create(frame->this, &op_errno);
+    if (!heal_frame) {
+        ret = -1;
+        goto out;
+    }
+    heal_local = heal_frame->local;
+    heal_frame->local = frame->local;
+
+    ret = afr_lockless_inspect(heal_frame, this, loc->gfid, &inode,
+                               &entry_selfheal, &data_selfheal,
+                               &metadata_selfheal, &pending);
+
+    if (ret == -ENOMEM) {
+        ret = -1;
+        goto out;
+    }
+
+    if (pending & PFLAG_PENDING) {
+        gf_asprintf(&substr, "-pending");
+        if (!substr)
+            goto out;
+    }
+
+    if (ret == -EIO) {
+        ret = gf_asprintf(&status, "split-brain%s", substr ? substr : "");
+        if (ret < 0) {
+            goto out;
+        }
+        dict = afr_set_heal_info(status);
+        if (!dict) {
+            ret = -1;
+            goto out;
+        }
+    } else if (ret == -EAGAIN) {
+        ret = gf_asprintf(&status, "possibly-healing%s", substr ? substr : "");
+        if (ret < 0) {
+            goto out;
+        }
+        dict = afr_set_heal_info(status);
+        if (!dict) {
+            ret = -1;
+            goto out;
+        }
+    } else if (ret >= 0) {
+        /* value of ret = source index
+         * so ret >= 0 and at least one of the 3 booleans set to
+         * true means a source is identified; heal is required.
+         */
+        if (!data_selfheal && !entry_selfheal && !metadata_selfheal) {
+            status = gf_strdup("no-heal");
+            if (!status) {
+                ret = -1;
                 goto out;
-        ret = 0;
+            }
+            dict = afr_set_heal_info(status);
+            if (!dict) {
+                ret = -1;
+                goto out;
+            }
+        } else {
+            ret = gf_asprintf(&status, "heal%s", substr ? substr : "");
+            if (ret < 0) {
+                goto out;
+            }
+            dict = afr_set_heal_info(status);
+            if (!dict) {
+                ret = -1;
+                goto out;
+            }
+        }
+    } else if (ret < 0) {
+        /* Apart from above checked -ve ret values, there are
+         * other possible ret values like ENOTCONN
+         * (returned when number of valid replies received are
+         * less than 2)
+         * in which case heal is required when one of the
+         * selfheal booleans is set.
+         */
+        if (data_selfheal || entry_selfheal || metadata_selfheal) {
+            ret = gf_asprintf(&status, "heal%s", substr ? substr : "");
+            if (ret < 0) {
+                goto out;
+            }
+            dict = afr_set_heal_info(status);
+            if (!dict) {
+                ret = -1;
+                goto out;
+            }
+        }
+    }
+
+    ret = 0;
+    op_errno = 0;
+
 out:
+    if (heal_frame) {
+        heal_frame->local = heal_local;
+        AFR_STACK_DESTROY(heal_frame);
+    }
+    AFR_STACK_UNWIND(getxattr, frame, ret, op_errno, dict, NULL);
+    if (dict)
+        dict_unref(dict);
+    if (inode)
+        inode_unref(inode);
+    GF_FREE(substr);
+    return ret;
+}
+
+int
+_afr_is_split_brain(call_frame_t *frame, xlator_t *this,
+                    struct afr_reply *replies, afr_transaction_type type,
+                    gf_boolean_t *spb)
+{
+    afr_private_t *priv = NULL;
+    uint64_t *witness = NULL;
+    unsigned char *sources = NULL;
+    unsigned char *sinks = NULL;
+    int sources_count = 0;
+    int ret = 0;
+
+    priv = this->private;
+
+    sources = alloca0(priv->child_count);
+    sinks = alloca0(priv->child_count);
+    witness = alloca0(priv->child_count * sizeof(*witness));
+
+    ret = afr_selfheal_find_direction(frame, this, replies, type,
+                                      priv->child_up, sources, sinks, witness,
+                                      NULL);
+    if (ret)
         return ret;
+
+    sources_count = AFR_COUNT(sources, priv->child_count);
+    if (!sources_count)
+        *spb = _gf_true;
+
+    return ret;
 }
 
 int
-afr_transaction_local_init (afr_local_t *local, xlator_t *this)
+afr_is_split_brain(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                   uuid_t gfid, gf_boolean_t *d_spb, gf_boolean_t *m_spb)
 {
-        int            child_up_count = 0;
-        int            ret = -ENOMEM;
-        afr_private_t *priv = NULL;
+    int ret = -1;
+    afr_private_t *priv = NULL;
+    struct afr_reply *replies = NULL;
 
-        priv = this->private;
-        ret = afr_internal_lock_init (&local->internal_lock, priv->child_count,
-                                      AFR_TRANSACTION_LK);
-        if (ret < 0)
-                goto out;
+    priv = this->private;
 
-        if ((local->transaction.type == AFR_DATA_TRANSACTION) ||
-            (local->transaction.type == AFR_METADATA_TRANSACTION)) {
-                ret = afr_inodelk_init (&local->internal_lock.inodelk[0],
-                                        this->name, priv->child_count);
-                if (ret < 0)
-                        goto out;
-        }
-
-        ret = -ENOMEM;
-        child_up_count = afr_up_children_count (local->child_up,
-                                                priv->child_count);
-        if (priv->optimistic_change_log && child_up_count == priv->child_count)
-                local->optimistic_change_log = 1;
+    replies = alloca0(sizeof(*replies) * priv->child_count);
 
-        local->first_up_child = afr_first_up_child (local->child_up,
-                                                    priv->child_count);
+    ret = afr_selfheal_unlocked_discover(frame, inode, gfid, replies);
+    if (ret)
+        goto out;
 
-        local->transaction.eager_lock =
-                GF_CALLOC (sizeof (*local->transaction.eager_lock),
-                           priv->child_count,
-                           gf_afr_mt_int32_t);
+    if (!afr_can_decide_split_brain_source_sinks(replies, priv->child_count)) {
+        ret = -EAGAIN;
+        goto out;
+    }
 
-        if (!local->transaction.eager_lock)
-                goto out;
+    ret = _afr_is_split_brain(frame, this, replies, AFR_DATA_TRANSACTION,
+                              d_spb);
+    if (ret)
+        goto out;
 
-        local->fresh_children = afr_children_create (priv->child_count);
-        if (!local->fresh_children)
-                goto out;
+    ret = _afr_is_split_brain(frame, this, replies, AFR_METADATA_TRANSACTION,
+                              m_spb);
+out:
+    if (replies) {
+        afr_replies_wipe(replies, priv->child_count);
+        replies = NULL;
+    }
+    return ret;
+}
 
-        local->transaction.pre_op = GF_CALLOC (sizeof (*local->transaction.pre_op),
-                                               priv->child_count,
-                                               gf_afr_mt_char);
-        if (!local->transaction.pre_op)
-                goto out;
+int
+afr_get_split_brain_status_cbk(int ret, call_frame_t *frame, void *opaque)
+{
+    GF_FREE(opaque);
+    return 0;
+}
 
-        local->pending = afr_matrix_create (priv->child_count,
-                                            AFR_NUM_CHANGE_LOGS);
-        if (!local->pending)
-                goto out;
+int
+afr_get_split_brain_status(void *opaque)
+{
+    gf_boolean_t d_spb = _gf_false;
+    gf_boolean_t m_spb = _gf_false;
+    int ret = -1;
+    int op_errno = 0;
+    int i = 0;
+    char *choices = NULL;
+    char *status = NULL;
+    dict_t *dict = NULL;
+    inode_t *inode = NULL;
+    afr_private_t *priv = NULL;
+    xlator_t **children = NULL;
+    call_frame_t *frame = NULL;
+    xlator_t *this = NULL;
+    loc_t *loc = NULL;
+    afr_spb_status_t *data = NULL;
+
+    data = opaque;
+    frame = data->frame;
+    this = frame->this;
+    loc = data->loc;
+    priv = this->private;
+    children = priv->children;
+
+    inode = afr_inode_find(this, loc->gfid);
+    if (!inode)
+        goto out;
+
+    dict = dict_new();
+    if (!dict) {
+        op_errno = ENOMEM;
+        ret = -1;
+        goto out;
+    }
+
+    /* Calculation for string length :
+     * (child_count X length of child-name) + SLEN("    Choices :")
+     * child-name consists of :
+     * a) 251 = max characters for volname according to GD_VOLUME_NAME_MAX
+     * b) strlen("-client-00,") assuming 16 replicas
+     */
+    choices = alloca0(priv->child_count * (256 + SLEN("-client-00,")) +
+                      SLEN("    Choices:"));
+
+    ret = afr_is_split_brain(frame, this, inode, loc->gfid, &d_spb, &m_spb);
+    if (ret) {
+        op_errno = -ret;
+        if (ret == -EAGAIN) {
+            ret = dict_set_sizen_str_sizen(dict, GF_AFR_SBRAIN_STATUS,
+                                           SBRAIN_HEAL_NO_GO_MSG);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_WARNING, -ret,
+                       AFR_MSG_DICT_SET_FAILED,
+                       "Failed to set GF_AFR_SBRAIN_STATUS in dict");
+            }
+        }
+        ret = -1;
+        goto out;
+    }
+
+    if (d_spb || m_spb) {
+        sprintf(choices, "    Choices:");
+        for (i = 0; i < priv->child_count; i++) {
+            strcat(choices, children[i]->name);
+            strcat(choices, ",");
+        }
+        choices[strlen(choices) - 1] = '\0';
 
-        local->transaction.txn_changelog = afr_matrix_create (priv->child_count,
-                                                           AFR_NUM_CHANGE_LOGS);
-        if (!local->transaction.txn_changelog)
-                goto out;
+        ret = gf_asprintf(&status,
+                          "data-split-brain:%s    "
+                          "metadata-split-brain:%s%s",
+                          (d_spb) ? "yes" : "no", (m_spb) ? "yes" : "no",
+                          choices);
 
-	INIT_LIST_HEAD (&local->transaction.eager_locked);
+        if (-1 == ret) {
+            op_errno = ENOMEM;
+            goto out;
+        }
+        ret = dict_set_dynstr_sizen(dict, GF_AFR_SBRAIN_STATUS, status);
+        if (ret) {
+            op_errno = -ret;
+            ret = -1;
+            goto out;
+        }
+    } else {
+        ret = dict_set_sizen_str_sizen(dict, GF_AFR_SBRAIN_STATUS,
+                                       SFILE_NOT_UNDER_DATA);
+        if (ret) {
+            op_errno = -ret;
+            ret = -1;
+            goto out;
+        }
+    }
 
-        ret = 0;
+    ret = 0;
 out:
-        return ret;
+    AFR_STACK_UNWIND(getxattr, frame, ret, op_errno, dict, NULL);
+    if (dict)
+        dict_unref(dict);
+    if (inode)
+        inode_unref(inode);
+    return ret;
 }
 
-void
-afr_reset_children (int32_t *fresh_children, int32_t child_count)
-{
-        unsigned int i = 0;
-        for (i = 0; i < child_count; i++)
-                fresh_children[i] = -1;
+int32_t
+afr_heal_splitbrain_file(call_frame_t *frame, xlator_t *this, loc_t *loc)
+{
+    int ret = 0;
+    int op_errno = 0;
+    dict_t *dict = NULL;
+    afr_local_t *local = NULL;
+    afr_local_t *heal_local = NULL;
+    call_frame_t *heal_frame = NULL;
+
+    local = frame->local;
+    dict = dict_new();
+    if (!dict) {
+        op_errno = ENOMEM;
+        ret = -1;
+        goto out;
+    }
+
+    heal_frame = afr_frame_create(this, &op_errno);
+    if (!heal_frame) {
+        ret = -1;
+        goto out;
+    }
+    heal_local = heal_frame->local;
+    heal_frame->local = frame->local;
+    /*Initiate heal with heal_frame with lk-owner set so that inodelk/entrylk
+     * work correctly*/
+    ret = afr_selfheal_do(heal_frame, this, loc->gfid);
+
+    if (ret == 1 || ret == 2) {
+        ret = dict_set_sizen_str_sizen(dict, "sh-fail-msg",
+                                       SFILE_NOT_IN_SPLIT_BRAIN);
+        if (ret)
+            gf_msg(this->name, GF_LOG_WARNING, -ret, AFR_MSG_DICT_SET_FAILED,
+                   "Failed to set sh-fail-msg in dict");
+        ret = 0;
+        goto out;
+    } else {
+        if (local->xdata_rsp) {
+            /* 'sh-fail-msg' has been set in the dict during self-heal.*/
+            dict_copy(local->xdata_rsp, dict);
+            ret = 0;
+        } else if (ret < 0) {
+            op_errno = -ret;
+            ret = -1;
+        }
+    }
+
+out:
+    if (heal_frame) {
+        heal_frame->local = heal_local;
+        AFR_STACK_DESTROY(heal_frame);
+    }
+    if (local->op == GF_FOP_GETXATTR)
+        AFR_STACK_UNWIND(getxattr, frame, ret, op_errno, dict, NULL);
+    else if (local->op == GF_FOP_SETXATTR)
+        AFR_STACK_UNWIND(setxattr, frame, ret, op_errno, NULL);
+    if (dict)
+        dict_unref(dict);
+    return ret;
 }
 
-int32_t*
-afr_children_create (int32_t child_count)
+int
+afr_get_child_index_from_name(xlator_t *this, char *name)
 {
-        int32_t           *children = NULL;
-        int               i               = 0;
-
-        GF_ASSERT (child_count > 0);
+    afr_private_t *priv = this->private;
+    int index = -1;
 
-        children = GF_CALLOC (child_count, sizeof (*children),
-                              gf_afr_mt_int32_t);
-        if (NULL == children)
-                goto out;
-        for (i = 0; i < child_count; i++)
-                children[i] = -1;
+    for (index = 0; index < priv->child_count; index++) {
+        if (!strcmp(priv->children[index]->name, name))
+            goto out;
+    }
+    index = -1;
 out:
-        return children;
+    return index;
 }
 
 void
-afr_children_add_child (int32_t *children, int32_t child,
-                        int32_t child_count)
-{
-        gf_boolean_t child_found = _gf_false;
-        int          i               = 0;
-
-        for (i = 0; i < child_count; i++) {
-                if (children[i] == -1)
-                        break;
-                if (children[i] == child) {
-                        child_found = _gf_true;
-                        break;
-                }
-        }
-
-        if (!child_found) {
-                GF_ASSERT (i < child_count);
-                children[i] = child;
-        }
+afr_priv_need_heal_set(afr_private_t *priv, gf_boolean_t need_heal)
+{
+    LOCK(&priv->lock);
+    {
+        priv->need_heal = need_heal;
+    }
+    UNLOCK(&priv->lock);
 }
 
 void
-afr_children_rm_child (int32_t *children, int32_t child, int32_t child_count)
-{
-        int          i = 0;
-
-        GF_ASSERT ((child >= 0) && (child < child_count));
-        for (i = 0; i < child_count; i++) {
-                if (children[i] == -1)
-                        break;
-                if (children[i] == child) {
-                        if (i != (child_count - 1))
-                                memmove (children + i, children + i + 1,
-                                         sizeof (*children)*(child_count - i - 1));
-                        children[child_count - 1] = -1;
-                        break;
-                }
+afr_set_need_heal(xlator_t *this, afr_local_t *local)
+{
+    int i = 0;
+    afr_private_t *priv = this->private;
+    gf_boolean_t need_heal = _gf_false;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->replies[i].valid && local->replies[i].need_heal) {
+            need_heal = _gf_true;
+            break;
         }
+    }
+    afr_priv_need_heal_set(priv, need_heal);
+    return;
 }
 
-int
-afr_get_children_count (int32_t *children, unsigned int child_count)
+gf_boolean_t
+afr_get_need_heal(xlator_t *this)
 {
-        int count = 0;
-        int i = 0;
+    afr_private_t *priv = this->private;
+    gf_boolean_t need_heal = _gf_true;
 
-        for (i = 0; i < child_count; i++) {
-                if (children[i] == -1)
-                        break;
-                count++;
-        }
-        return count;
+    LOCK(&priv->lock);
+    {
+        need_heal = priv->need_heal;
+    }
+    UNLOCK(&priv->lock);
+    return need_heal;
 }
 
-void
-afr_set_low_priority (call_frame_t *frame)
+int
+afr_get_msg_id(char *op_type)
 {
-        frame->root->pid = LOW_PRIO_PROC_PID;
+    if (!strcmp(op_type, GF_AFR_REPLACE_BRICK))
+        return AFR_MSG_REPLACE_BRICK_STATUS;
+    else if (!strcmp(op_type, GF_AFR_ADD_BRICK))
+        return AFR_MSG_ADD_BRICK_STATUS;
+    return -1;
 }
 
 int
-afr_child_fd_ctx_set (xlator_t *this, fd_t *fd, int32_t child,
-                      int flags)
+afr_fav_child_reset_sink_xattrs_cbk(int ret, call_frame_t *heal_frame,
+                                    void *opaque)
 {
-        int             ret = 0;
-        uint64_t        ctx = 0;
-        afr_fd_ctx_t    *fd_ctx      = NULL;
+    call_frame_t *txn_frame = NULL;
+    afr_local_t *local = NULL;
+    afr_local_t *heal_local = NULL;
+    xlator_t *this = NULL;
 
-        GF_ASSERT (fd && fd->inode);
-        ret = afr_fd_ctx_set (this, fd);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "could not set fd ctx for fd=%p", fd);
-                goto out;
+    heal_local = heal_frame->local;
+    txn_frame = heal_local->heal_frame;
+    local = txn_frame->local;
+    this = txn_frame->this;
+
+    /* Refresh the inode agan and proceed with the transaction.*/
+    afr_inode_refresh(txn_frame, this, local->inode, NULL, local->refreshfn);
+
+    AFR_STACK_DESTROY(heal_frame);
+
+    return 0;
+}
+
+int
+afr_fav_child_reset_sink_xattrs(void *opaque)
+{
+    call_frame_t *heal_frame = NULL;
+    call_frame_t *txn_frame = NULL;
+    xlator_t *this = NULL;
+    gf_boolean_t d_spb = _gf_false;
+    gf_boolean_t m_spb = _gf_false;
+    afr_local_t *heal_local = NULL;
+    afr_local_t *txn_local = NULL;
+    afr_private_t *priv = NULL;
+    inode_t *inode = NULL;
+    unsigned char *locked_on = NULL;
+    unsigned char *sources = NULL;
+    unsigned char *sinks = NULL;
+    unsigned char *healed_sinks = NULL;
+    unsigned char *undid_pending = NULL;
+    struct afr_reply *locked_replies = NULL;
+    int ret = 0;
+
+    heal_frame = (call_frame_t *)opaque;
+    heal_local = heal_frame->local;
+    txn_frame = heal_local->heal_frame;
+    txn_local = txn_frame->local;
+    this = txn_frame->this;
+    inode = txn_local->inode;
+    priv = this->private;
+    locked_on = alloca0(priv->child_count);
+    sources = alloca0(priv->child_count);
+    sinks = alloca0(priv->child_count);
+    healed_sinks = alloca0(priv->child_count);
+    undid_pending = alloca0(priv->child_count);
+    locked_replies = alloca0(sizeof(*locked_replies) * priv->child_count);
+
+    ret = _afr_is_split_brain(txn_frame, this, txn_local->replies,
+                              AFR_DATA_TRANSACTION, &d_spb);
+
+    ret = _afr_is_split_brain(txn_frame, this, txn_local->replies,
+                              AFR_METADATA_TRANSACTION, &m_spb);
+
+    /* Take appropriate locks and reset sink xattrs. */
+    if (d_spb) {
+        ret = afr_selfheal_inodelk(heal_frame, this, inode, this->name, 0, 0,
+                                   locked_on);
+        {
+            if (ret < priv->child_count)
+                goto data_unlock;
+            ret = __afr_selfheal_data_prepare(
+                heal_frame, this, inode, locked_on, sources, sinks,
+                healed_sinks, undid_pending, locked_replies, NULL);
+        }
+    data_unlock:
+        afr_selfheal_uninodelk(heal_frame, this, inode, this->name, 0, 0,
+                               locked_on);
+    }
+
+    if (m_spb) {
+        memset(locked_on, 0, sizeof(*locked_on) * priv->child_count);
+        memset(undid_pending, 0, sizeof(*undid_pending) * priv->child_count);
+        ret = afr_selfheal_inodelk(heal_frame, this, inode, this->name,
+                                   LLONG_MAX - 1, 0, locked_on);
+        {
+            if (ret < priv->child_count)
+                goto mdata_unlock;
+            ret = __afr_selfheal_metadata_prepare(
+                heal_frame, this, inode, locked_on, sources, sinks,
+                healed_sinks, undid_pending, locked_replies, NULL);
         }
+    mdata_unlock:
+        afr_selfheal_uninodelk(heal_frame, this, inode, this->name,
+                               LLONG_MAX - 1, 0, locked_on);
+    }
 
-        ret = fd_ctx_get (fd, this, &ctx);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "could not get fd ctx for fd=%p", fd);
+    return ret;
+}
+
+/*
+ * Concatenates the xattrs in local->replies separated by a delimiter.
+ */
+int
+afr_serialize_xattrs_with_delimiter(call_frame_t *frame, xlator_t *this,
+                                    char *buf, const char *default_str,
+                                    int32_t *serz_len, char delimiter)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    char *xattr = NULL;
+    int i = 0;
+    int len = 0;
+    int keylen = 0;
+    size_t str_len = 0;
+    int ret = -1;
+
+    priv = this->private;
+    local = frame->local;
+
+    keylen = strlen(local->cont.getxattr.name);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid || local->replies[i].op_ret) {
+            str_len = strlen(default_str);
+            buf = strncat(buf, default_str, str_len);
+            len += str_len;
+            buf[len++] = delimiter;
+            buf[len] = '\0';
+        } else {
+            ret = dict_get_strn(local->replies[i].xattr,
+                                local->cont.getxattr.name, keylen, &xattr);
+            if (ret) {
+                gf_msg("TEST", GF_LOG_ERROR, -ret, AFR_MSG_DICT_GET_FAILED,
+                       "Failed to get the node_uuid of brick "
+                       "%d",
+                       i);
                 goto out;
-        }
+            }
+            str_len = strlen(xattr);
+            buf = strncat(buf, xattr, str_len);
+            len += str_len;
+            buf[len++] = delimiter;
+            buf[len] = '\0';
+        }
+    }
+    buf[--len] = '\0'; /*remove the last delimiter*/
+    if (serz_len)
+        *serz_len = ++len;
+    ret = 0;
 
-        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-        fd_ctx->opened_on[child] = AFR_FD_OPENED;
-        if (!IA_ISDIR (fd->inode->ia_type)) {
-                fd_ctx->flags            = flags;
-        }
-        ret = 0;
 out:
-        return ret;
+    return ret;
 }
 
-gf_boolean_t
-afr_have_quorum (char *logname, afr_private_t *priv)
+uint64_t
+afr_write_subvol_get(call_frame_t *frame, xlator_t *this)
 {
-        unsigned int        quorum = 0;
+    afr_local_t *local = NULL;
+    uint64_t write_subvol = 0;
 
-        GF_VALIDATE_OR_GOTO(logname,priv,out);
+    local = frame->local;
+    LOCK(&local->inode->lock);
+    write_subvol = local->inode_ctx->write_subvol;
+    UNLOCK(&local->inode->lock);
 
-        quorum = priv->quorum_count;
-        if (quorum != AFR_QUORUM_AUTO) {
-                return (priv->up_count >= (priv->down_count + quorum));
-        }
+    return write_subvol;
+}
 
-        quorum = priv->child_count / 2 + 1;
-        if (priv->up_count >= (priv->down_count + quorum)) {
-                return _gf_true;
-        }
+int
+afr_write_subvol_set(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    unsigned char *data_accused = NULL;
+    unsigned char *metadata_accused = NULL;
+    unsigned char *data_readable = NULL;
+    unsigned char *metadata_readable = NULL;
+    uint16_t datamap = 0;
+    uint16_t metadatamap = 0;
+    uint64_t val = 0;
+    int event = 0;
+    int i = 0;
+
+    local = frame->local;
+    priv = this->private;
+    data_accused = alloca0(priv->child_count);
+    metadata_accused = alloca0(priv->child_count);
+    data_readable = alloca0(priv->child_count);
+    metadata_readable = alloca0(priv->child_count);
+    event = local->event_generation;
+
+    afr_readables_fill(frame, this, local->inode, data_accused,
+                       metadata_accused, data_readable, metadata_readable,
+                       NULL);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (data_readable[i])
+            datamap |= (1 << i);
+        if (metadata_readable[i])
+            metadatamap |= (1 << i);
+    }
+
+    val = ((uint64_t)metadatamap) | (((uint64_t)datamap) << 16) |
+          (((uint64_t)event) << 32);
+
+    LOCK(&local->inode->lock);
+    {
+        if (local->inode_ctx->write_subvol == 0 &&
+            local->transaction.type == AFR_DATA_TRANSACTION) {
+            local->inode_ctx->write_subvol = val;
+        }
+    }
+    UNLOCK(&local->inode->lock);
+
+    return 0;
+}
 
-        /*
-         * Special case for even numbers of nodes: if we have exactly half
-         * and that includes the first ("senior-most") node, then that counts
-         * as quorum even if it wouldn't otherwise.  This supports e.g. N=2
-         * while preserving the critical property that there can only be one
-         * such group.
-         */
-        if ((priv->child_count % 2) == 0) {
-                quorum = priv->child_count / 2;
-                if (priv->up_count >= (priv->down_count + quorum)) {
-                        if (priv->child_up[0]) {
-                                return _gf_true;
-                        }
-                }
-        }
+int
+afr_write_subvol_reset(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
 
-out:
-        return _gf_false;
+    local = frame->local;
+    LOCK(&local->inode->lock);
+    {
+        GF_ASSERT(local->inode_ctx->lock_count > 0);
+        local->inode_ctx->lock_count--;
+
+        if (!local->inode_ctx->lock_count)
+            local->inode_ctx->write_subvol = 0;
+    }
+    UNLOCK(&local->inode->lock);
+
+    return 0;
 }
 
-void
-afr_priv_destroy (afr_private_t *priv)
+int
+afr_set_inode_local(xlator_t *this, afr_local_t *local, inode_t *inode)
 {
-        int            i           = 0;
+    int ret = 0;
 
-        if (!priv)
-                goto out;
-        inode_unref (priv->root_inode);
-        GF_FREE (priv->shd.pos);
-        GF_FREE (priv->shd.pending);
-        GF_FREE (priv->shd.inprogress);
-//        for (i = 0; i < priv->child_count; i++)
-//                if (priv->shd.timer && priv->shd.timer[i])
-//                        gf_timer_call_cancel (this->ctx, priv->shd.timer[i]);
-        GF_FREE (priv->shd.timer);
-
-        if (priv->shd.healed)
-                eh_destroy (priv->shd.healed);
-
-        if (priv->shd.heal_failed)
-                eh_destroy (priv->shd.heal_failed);
-
-        if (priv->shd.split_brain)
-                eh_destroy (priv->shd.split_brain);
-
-        GF_FREE (priv->last_event);
-        if (priv->pending_key) {
-                for (i = 0; i < priv->child_count; i++)
-                        GF_FREE (priv->pending_key[i]);
-        }
-        GF_FREE (priv->pending_key);
-        GF_FREE (priv->children);
-        GF_FREE (priv->child_up);
-        LOCK_DESTROY (&priv->lock);
-        LOCK_DESTROY (&priv->read_child_lock);
-        pthread_mutex_destroy (&priv->mutex);
-        GF_FREE (priv);
+    local->inode = inode_ref(inode);
+    LOCK(&local->inode->lock);
+    {
+        ret = __afr_inode_ctx_get(this, local->inode, &local->inode_ctx);
+    }
+    UNLOCK(&local->inode->lock);
+    if (ret < 0) {
+        gf_msg_callingfn(
+            this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_INODE_CTX_GET_FAILED,
+            "Error getting inode ctx %s", uuid_utoa(local->inode->gfid));
+    }
+    return ret;
+}
+
+gf_boolean_t
+afr_ta_is_fop_called_from_synctask(xlator_t *this)
+{
+    struct synctask *task = NULL;
+    gf_lkowner_t tmp_owner = {
+        0,
+    };
+
+    task = synctask_get();
+    if (!task)
+        return _gf_false;
+
+    set_lk_owner_from_ptr(&tmp_owner, (void *)this);
+
+    if (!is_same_lkowner(&tmp_owner, &task->frame->root->lk_owner))
+        return _gf_false;
+
+    return _gf_true;
+}
+
+int
+afr_ta_post_op_lock(xlator_t *this, loc_t *loc)
+{
+    int ret = 0;
+    uuid_t gfid = {
+        0,
+    };
+    afr_private_t *priv = this->private;
+    gf_boolean_t locked = _gf_false;
+    struct gf_flock flock1 = {
+        0,
+    };
+    struct gf_flock flock2 = {
+        0,
+    };
+    int32_t cmd = 0;
+
+    /* Clients must take AFR_TA_DOM_NOTIFY lock only when the previous lock
+     * has been released in afr_notify due to upcall notification from shd.
+     */
+    GF_ASSERT(priv->ta_notify_dom_lock_offset == 0);
+
+    if (!priv->shd.iamshd)
+        GF_ASSERT(afr_ta_is_fop_called_from_synctask(this));
+    flock1.l_type = F_WRLCK;
+
+    while (!locked) {
+        if (priv->shd.iamshd) {
+            cmd = F_SETLKW;
+            flock1.l_start = 0;
+            flock1.l_len = 0;
+        } else {
+            cmd = F_SETLK;
+            gf_uuid_generate(gfid);
+            flock1.l_start = gfid_to_ino(gfid);
+            if (flock1.l_start < 0)
+                flock1.l_start = -flock1.l_start;
+            flock1.l_len = 1;
+        }
+        ret = syncop_inodelk(priv->children[THIN_ARBITER_BRICK_INDEX],
+                             AFR_TA_DOM_NOTIFY, loc, cmd, &flock1, NULL, NULL);
+        if (!ret) {
+            locked = _gf_true;
+            priv->ta_notify_dom_lock_offset = flock1.l_start;
+        } else if (ret == -EAGAIN) {
+            continue;
+        } else {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+                   "Failed to get "
+                   "AFR_TA_DOM_NOTIFY lock on %s.",
+                   loc->name);
+            goto out;
+        }
+    }
+
+    flock2.l_type = F_WRLCK;
+    flock2.l_start = 0;
+    flock2.l_len = 0;
+    ret = syncop_inodelk(priv->children[THIN_ARBITER_BRICK_INDEX],
+                         AFR_TA_DOM_MODIFY, loc, F_SETLKW, &flock2, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed to get AFR_TA_DOM_MODIFY lock on %s.", loc->name);
+        flock1.l_type = F_UNLCK;
+        ret = syncop_inodelk(priv->children[THIN_ARBITER_BRICK_INDEX],
+                             AFR_TA_DOM_NOTIFY, loc, F_SETLK, &flock1, NULL,
+                             NULL);
+    }
 out:
-        return;
+    return ret;
 }
 
 int
-xlator_subvolume_count (xlator_t *this)
+afr_ta_post_op_unlock(xlator_t *this, loc_t *loc)
+{
+    afr_private_t *priv = this->private;
+    struct gf_flock flock = {
+        0,
+    };
+    int ret = 0;
+
+    if (!priv->shd.iamshd)
+        GF_ASSERT(afr_ta_is_fop_called_from_synctask(this));
+    flock.l_type = F_UNLCK;
+    flock.l_start = 0;
+    flock.l_len = 0;
+
+    ret = syncop_inodelk(priv->children[THIN_ARBITER_BRICK_INDEX],
+                         AFR_TA_DOM_MODIFY, loc, F_SETLK, &flock, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed to unlock AFR_TA_DOM_MODIFY lock.");
+        goto out;
+    }
+
+    if (!priv->shd.iamshd)
+        /* Mounts (clients) will not release the AFR_TA_DOM_NOTIFY lock
+         * in post-op as they use it as a notification mechanism. When
+         * shd sends a lock request on TA during heal, the clients will
+         * receive a lock-contention upcall notification upon which they
+         * will release the AFR_TA_DOM_NOTIFY lock after completing the
+         * in flight I/O.*/
+        goto out;
+
+    ret = syncop_inodelk(priv->children[THIN_ARBITER_BRICK_INDEX],
+                         AFR_TA_DOM_NOTIFY, loc, F_SETLK, &flock, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed to unlock AFR_TA_DOM_NOTIFY lock.");
+    }
+out:
+    return ret;
+}
+
+call_frame_t *
+afr_ta_frame_create(xlator_t *this)
 {
-        int i = 0;
-        xlator_list_t *list = NULL;
+    call_frame_t *frame = NULL;
+    void *lk_owner = NULL;
 
-        for (list = this->children; list; list = list->next)
-                i++;
-        return i;
+    frame = create_frame(this, this->ctx->pool);
+    if (!frame)
+        return NULL;
+    lk_owner = (void *)this;
+    afr_set_lk_owner(frame, this, lk_owner);
+    return frame;
 }
 
-inline gf_boolean_t
-afr_is_errno_set (int *child_errno, int child)
+gf_boolean_t
+afr_ta_has_quorum(afr_private_t *priv, afr_local_t *local)
 {
-        return child_errno[child];
+    int data_count = 0;
+
+    data_count = AFR_COUNT(local->child_up, priv->child_count);
+    if (data_count == 2) {
+        return _gf_true;
+    } else if (data_count == 1 && local->ta_child_up) {
+        return _gf_true;
+    }
+
+    return _gf_false;
 }
 
-inline gf_boolean_t
-afr_is_errno_unset (int *child_errno, int child)
+static gf_boolean_t
+afr_is_add_replica_mount_lookup_on_root(call_frame_t *frame)
 {
-        return !afr_is_errno_set (child_errno, child);
-}
+    afr_local_t *local = NULL;
 
-void
-afr_prepare_new_entry_pending_matrix (int32_t **pending,
-                                      gf_boolean_t (*is_pending) (int *, int),
-                                      int *ctx, struct iatt *buf,
-                                      unsigned int child_count)
-{
-        int midx = 0;
-        int idx  = 0;
-        int i    = 0;
-
-        midx = afr_index_for_transaction_type (AFR_METADATA_TRANSACTION);
-        if (IA_ISDIR (buf->ia_type))
-                idx = afr_index_for_transaction_type (AFR_ENTRY_TRANSACTION);
-        else if (IA_ISREG (buf->ia_type))
-                idx = afr_index_for_transaction_type (AFR_DATA_TRANSACTION);
-        else
-                idx = -1;
-        for (i = 0; i < child_count; i++) {
-                if (is_pending (ctx, i)) {
-                        pending[i][midx] = hton32 (1);
-                        if (idx == -1)
-                                continue;
-                        pending[i][idx] = hton32 (1);
-                }
-        }
+    if (frame->root->pid != GF_CLIENT_PID_ADD_REPLICA_MOUNT)
+        return _gf_false;
+
+    local = frame->local;
+
+    if (local->op != GF_FOP_LOOKUP)
+        /* TODO:If the replica count is being increased on a plain distribute
+         * volume that was never mounted, we need to allow setxattr on '/' with
+         * GF_CLIENT_PID_NO_ROOT_SQUASH to accomodate for DHT layout setting */
+        return _gf_false;
+
+    if (local->inode == NULL)
+        return _gf_false;
+
+    if (!__is_root_gfid(local->inode->gfid))
+        return _gf_false;
+
+    return _gf_true;
 }
 
 gf_boolean_t
-afr_is_fd_fixable (fd_t *fd)
+afr_lookup_has_quorum(call_frame_t *frame, const unsigned int up_children_count)
 {
-        if (!fd || !fd->inode)
-                return _gf_false;
-        else if (fd_is_anonymous (fd))
-                return _gf_false;
-        else if (uuid_is_null (fd->inode->gfid))
-                return _gf_false;
-
+    if (frame && (up_children_count > 0) &&
+        afr_is_add_replica_mount_lookup_on_root(frame))
         return _gf_true;
+
+    return _gf_false;
 }
 
 void
-afr_handle_open_fd_count (call_frame_t *frame, xlator_t *this)
+afr_handle_replies_quorum(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t     *local = NULL;
-        inode_t         *inode = NULL;
-        afr_inode_ctx_t *ctx   = NULL;
-
-        local = frame->local;
+    afr_local_t *local = frame->local;
+    afr_private_t *priv = this->private;
+    unsigned char *success_replies = NULL;
 
-        if (local->fd)
-                inode = local->fd->inode;
-        else
-                inode = local->loc.inode;
+    success_replies = alloca0(priv->child_count);
+    afr_fill_success_replies(local, priv, success_replies);
 
-        if (!inode)
-                return;
+    if (priv->quorum_count && !afr_has_quorum(success_replies, this, NULL)) {
+        local->op_errno = afr_final_errno(local, priv);
+        if (!local->op_errno)
+            local->op_errno = afr_quorum_errno(priv);
+        local->op_ret = -1;
+    }
+}
 
-        LOCK (&inode->lock);
-        {
-                ctx = __afr_inode_ctx_get (inode, this);
-                ctx->open_fd_count = local->open_fd_count;
+gf_boolean_t
+afr_ta_dict_contains_pending_xattr(dict_t *dict, afr_private_t *priv, int child)
+{
+    int *pending = NULL;
+    int ret = 0;
+    int i = 0;
+
+    ret = dict_get_ptr(dict, priv->pending_key[child], (void *)&pending);
+    if (ret == 0) {
+        for (i = 0; i < AFR_NUM_CHANGE_LOGS; i++) {
+            /* Not doing a ntoh32(pending) as we just want to check
+             * if it is non-zero or not. */
+            if (pending[i]) {
+                return _gf_true;
+            }
         }
-        UNLOCK (&inode->lock);
+    }
+
+    return _gf_false;
 }
diff --git a/xlators/cluster/afr/src/afr-dir-read.c b/xlators/cluster/afr/src/afr-dir-read.c
index 689dd84e646..f8bf8340dab 100644
--- a/xlators/cluster/afr/src/afr-dir-read.c
+++ b/xlators/cluster/afr/src/afr-dir-read.c
@@ -8,538 +8,339 @@
   cases as published by the Free Software Foundation.
 */
 
-
 #include <libgen.h>
 #include <unistd.h>
-#include <fnmatch.h>
 #include <sys/time.h>
 #include <stdlib.h>
 #include <signal.h>
 #include <string.h>
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
-#include "checksum.h"
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/dict.h>
+#include <glusterfs/list.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/compat.h>
 
 #include "afr.h"
-#include "afr-self-heal.h"
-#include "afr-self-heal-common.h"
-
-int
-afr_examine_dir_sh_unwind (call_frame_t *frame, xlator_t *this, int32_t op_ret,
-                           int32_t op_errno, int32_t sh_failed)
-{
-        afr_local_t *local  = NULL;
-
-        local = frame->local;
-
-        afr_set_opendir_done (this, local->fd->inode);
-
-        AFR_STACK_UNWIND (opendir, frame, local->op_ret,
-                          local->op_errno, local->fd, NULL);
-
-        return 0;
-}
-
-
-gf_boolean_t
-__checksums_differ (uint32_t *checksum, int child_count,
-                    unsigned char *child_up)
-{
-        int          ret            = _gf_false;
-        int          i              = 0;
-        uint32_t     cksum          = 0;
-        gf_boolean_t activate_check = _gf_false;
-
-        for (i = 0; i < child_count; i++) {
-                if (!child_up[i])
-                        continue;
-                if (_gf_false == activate_check) {
-                        cksum          = checksum[i];
-                        activate_check = _gf_true;
-                        continue;
-                }
-
-                if (cksum != checksum[i]) {
-                        ret = _gf_true;
-                        break;
-                }
-
-                cksum = checksum[i];
-        }
-
-        return ret;
-}
-
+#include "afr-transaction.h"
 
 int32_t
-afr_examine_dir_readdir_cbk (call_frame_t *frame, void *cookie,
-                             xlator_t *this, int32_t op_ret, int32_t op_errno,
-                             gf_dirent_t *entries, dict_t *xdata)
+afr_opendir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
 {
-        afr_private_t *   priv        = NULL;
-        afr_local_t *     local       = NULL;
-        afr_self_heal_t * sh          = NULL;
-        gf_dirent_t *     entry       = NULL;
-        gf_dirent_t *     tmp         = NULL;
-        char              *reason     = NULL;
-        int               child_index = 0;
-        uint32_t          entry_cksum = 0;
-        int               call_count  = 0;
-        off_t             last_offset = 0;
-        inode_t           *inode      = NULL;
-
-        priv  = this->private;
-        local = frame->local;
-        sh    = &local->self_heal;
-        inode = local->fd->inode;
-
-        child_index = (long) cookie;
-
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "%s: failed to do opendir on %s",
-                        local->loc.path, priv->children[child_index]->name);
-                local->op_ret = -1;
-                local->op_ret = op_errno;
-                goto out;
-        }
+    afr_local_t *local = NULL;
+    int call_count = -1;
+    int32_t child_index = 0;
+    afr_fd_ctx_t *fd_ctx = NULL;
 
-        if (op_ret == 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "%s: no entries found in %s",
-                        local->loc.path, priv->children[child_index]->name);
-                goto out;
-        }
+    local = frame->local;
+    fd_ctx = local->fd_ctx;
+    child_index = (long)cookie;
 
-        list_for_each_entry_safe (entry, tmp, &entries->list, list) {
-                entry_cksum = gf_rsync_weak_checksum ((unsigned char *)entry->d_name,
-                                                      strlen (entry->d_name));
-                local->cont.opendir.checksum[child_index] ^= entry_cksum;
-        }
+    local->replies[child_index].valid = 1;
+    local->replies[child_index].op_ret = op_ret;
+    local->replies[child_index].op_errno = op_errno;
 
-        list_for_each_entry (entry, &entries->list, list) {
-                last_offset = entry->d_off;
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            local->op_errno = op_errno;
+            fd_ctx->opened_on[child_index] = AFR_FD_NOT_OPENED;
+        } else {
+            local->op_ret = op_ret;
+            fd_ctx->opened_on[child_index] = AFR_FD_OPENED;
+            if (!local->xdata_rsp && xdata)
+                local->xdata_rsp = dict_ref(xdata);
         }
+        call_count = --local->call_count;
+    }
+    UNLOCK(&frame->lock);
 
-        /* read more entries */
-
-        STACK_WIND_COOKIE (frame, afr_examine_dir_readdir_cbk,
-                           (void *) (long) child_index,
-                           priv->children[child_index],
-                           priv->children[child_index]->fops->readdir,
-                           local->fd, 131072, last_offset, NULL);
+    if (call_count == 0) {
+        afr_handle_replies_quorum(frame, this);
+        AFR_STACK_UNWIND(opendir, frame, local->op_ret, local->op_errno,
+                         local->fd, NULL);
+    }
 
-        return 0;
-
-out:
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                if (__checksums_differ (local->cont.opendir.checksum,
-                                        priv->child_count,
-                                        local->child_up)) {
-
-                        sh->do_entry_self_heal  = _gf_true;
-                        sh->forced_merge          = _gf_true;
-
-                        reason = "checksums of directory differ";
-                        afr_launch_self_heal (frame, this, inode, _gf_false,
-                                              inode->ia_type, reason, NULL,
-                                              afr_examine_dir_sh_unwind);
-                } else {
-                        afr_set_opendir_done (this, inode);
-
-                        AFR_STACK_UNWIND (opendir, frame, local->op_ret,
-                                          local->op_errno, local->fd, NULL);
-                }
-        }
-
-        return 0;
+    return 0;
 }
 
-
 int
-afr_examine_dir (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t * priv       = NULL;
-        afr_local_t *   local      = NULL;
-        int             i          = 0;
-        int             call_count = 0;
-
-        local = frame->local;
-        priv  = this->private;
-
-        local->cont.opendir.checksum = GF_CALLOC (priv->child_count,
-                                                  sizeof (*local->cont.opendir.checksum),
-                                                  gf_afr_mt_int32_t);
-
-        call_count = afr_up_children_count (local->child_up, priv->child_count);
-
-        local->call_count = call_count;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND_COOKIE (frame, afr_examine_dir_readdir_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->readdir,
-                                           local->fd, 131072, 0, NULL);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-        return 0;
-}
-
-
-int32_t
-afr_opendir_cbk (call_frame_t *frame, void *cookie,
-                 xlator_t *this, int32_t op_ret, int32_t op_errno,
-                 fd_t *fd, dict_t *xdata)
+afr_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+            dict_t *xdata)
 {
-        afr_private_t *priv              = NULL;
-        afr_local_t   *local             = NULL;
-        int32_t        up_children_count = 0;
-        int            ret               = -1;
-        int            call_count        = -1;
-        int32_t        child_index       = 0;
-
-        priv  = this->private;
-        local = frame->local;
-        child_index = (long) cookie;
-
-        up_children_count = afr_up_children_count (local->child_up,
-                                                   priv->child_count);
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret >= 0) {
-                        local->op_ret = op_ret;
-                        ret = afr_child_fd_ctx_set (this, fd, child_index, 0);
-                        if (ret) {
-                                local->op_ret = -1;
-                                local->op_errno = -ret;
-                                goto unlock;
-                        }
-                }
-
-                local->op_errno = op_errno;
-        }
-unlock:
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                if (local->op_ret != 0)
-                        goto out;
-
-                if (!afr_is_opendir_done (this, local->fd->inode) &&
-                    up_children_count > 1 && priv->entry_self_heal) {
-
-                        /*
-                         * This is the first opendir on this inode. We need
-                         * to check if the directory's entries are the same
-                         * on all subvolumes. This is needed in addition
-                         * to regular entry self-heal because the readdir
-                         * call is sent only to the first subvolume, and
-                         * thus files that exist only there will never be healed
-                         * otherwise (assuming changelog shows no anomalies).
-                         */
-
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "reading contents of directory %s looking for mismatch",
-                                local->loc.path);
-
-                        afr_examine_dir (frame, this);
-
-                } else {
-                        /* do the unwind */
-                        goto out;
-                }
-        }
-
-        return 0;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int i = 0;
+    int call_count = -1;
+    int32_t op_errno = ENOMEM;
+    afr_fd_ctx_t *fd_ctx = NULL;
 
-out:
-        AFR_STACK_UNWIND (opendir, frame, local->op_ret,
-                          local->op_errno, local->fd, NULL);
-
-        return 0;
-}
+    priv = this->private;
 
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-int32_t
-afr_opendir (call_frame_t *frame, xlator_t *this,
-             loc_t *loc, fd_t *fd)
-{
-        afr_private_t * priv        = NULL;
-        afr_local_t   * local       = NULL;
-        int             child_count = 0;
-        int             i           = 0;
-        int             ret         = -1;
-        int             call_count  = -1;
-        int32_t         op_errno    = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    local->op = GF_FOP_OPENDIR;
 
-        priv = this->private;
+    if (priv->quorum_count && !afr_has_quorum(local->child_up, this, NULL)) {
+        op_errno = afr_quorum_errno(priv);
+        goto out;
+    }
 
-        child_count = priv->child_count;
+    if (!afr_is_consistent_io_possible(local, priv, &op_errno))
+        goto out;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    fd_ctx = afr_fd_ctx_get(fd, this);
+    if (!fd_ctx)
+        goto out;
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    loc_copy(&local->loc, loc);
 
-        loc_copy (&local->loc, loc);
+    local->fd = fd_ref(fd);
+    local->fd_ctx = fd_ctx;
 
-        local->fd    = fd_ref (fd);
+    call_count = local->call_count;
 
-        call_count = local->call_count;
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->child_up[i]) {
+            STACK_WIND_COOKIE(frame, afr_opendir_cbk, (void *)(long)i,
+                              priv->children[i],
+                              priv->children[i]->fops->opendir, loc, fd, NULL);
 
-        for (i = 0; i < child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND_COOKIE (frame, afr_opendir_cbk,
-                                           (void*) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->opendir,
-                                           loc, fd, NULL);
-
-                        if (!--call_count)
-                                break;
-                }
+            if (!--call_count)
+                break;
         }
+    }
 
-        ret = 0;
+    return 0;
 out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (opendir, frame, -1, op_errno, fd, NULL);
-
-        return 0;
+    AFR_STACK_UNWIND(opendir, frame, -1, op_errno, fd, NULL);
+    return 0;
 }
 
+static int
+afr_validate_read_subvol(inode_t *inode, xlator_t *this, int par_read_subvol)
+{
+    int gen = 0;
+    int entry_read_subvol = 0;
+    unsigned char *data_readable = NULL;
+    unsigned char *metadata_readable = NULL;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+    data_readable = alloca0(priv->child_count);
+    metadata_readable = alloca0(priv->child_count);
+
+    afr_inode_read_subvol_get(inode, this, data_readable, metadata_readable,
+                              &gen);
+
+    if (gen != priv->event_generation || !data_readable[par_read_subvol] ||
+        !metadata_readable[par_read_subvol])
+        return -1;
+
+    /* Once the control reaches the following statement, it means that the
+     * parent's read subvol is perfectly readable. So calling
+     * either afr_data_subvol_get() or afr_metadata_subvol_get() would
+     * yield the same result. Hence, choosing afr_data_subvol_get() below.
+     */
+
+    if (!priv->consistent_metadata)
+        return 0;
 
-/**
- * Common algorithm for directory read calls:
- *
- * - Try the fop on the first child that is up
- * - if we have failed due to ENOTCONN:
- *     try the next child
- *
- * Applicable to: readdir
- */
-
-
-struct entry_name {
-        char *name;
-        struct list_head list;
-};
+    /* For an inode fetched through readdirp which is yet to be linked,
+     * inode ctx would not be initialised (yet). So this function returns
+     * -1 above due to gen being 0, which is why it is OK to pass NULL for
+     *  read_subvol_args here.
+     */
+    entry_read_subvol = afr_data_subvol_get(inode, this, NULL, NULL, NULL,
+                                            NULL);
+    if (entry_read_subvol != par_read_subvol)
+        return -1;
+
+    return 0;
+}
 
 static void
-afr_forget_entries (fd_t *fd)
+afr_readdir_transform_entries(call_frame_t *frame, gf_dirent_t *subvol_entries,
+                              int subvol, gf_dirent_t *entries, fd_t *fd)
 {
-        struct entry_name *entry  = NULL;
-        struct entry_name *tmp    = NULL;
-        int                ret    = 0;
-        uint64_t           ctx    = 0;
-        afr_fd_ctx_t      *fd_ctx = NULL;
-
-        ret = fd_ctx_get (fd, THIS, &ctx);
-        if (ret < 0) {
-                gf_log (THIS->name, GF_LOG_INFO,
-                        "could not get fd ctx for fd=%p", fd);
-                return;
+    int ret = -1;
+    gf_dirent_t *entry = NULL;
+    gf_dirent_t *tmp = NULL;
+    xlator_t *this = NULL;
+    afr_private_t *priv = NULL;
+    gf_boolean_t need_heal = _gf_false;
+    gf_boolean_t validate_subvol = _gf_false;
+
+    this = THIS;
+    priv = this->private;
+
+    need_heal = afr_get_need_heal(this);
+    validate_subvol = need_heal | priv->consistent_metadata;
+
+    list_for_each_entry_safe(entry, tmp, &subvol_entries->list, list)
+    {
+        if (afr_is_private_directory(priv, fd->inode->gfid, entry->d_name,
+                                     frame->root->pid)) {
+            continue;
         }
 
-        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+        list_del_init(&entry->list);
+        list_add_tail(&entry->list, &entries->list);
 
-        list_for_each_entry_safe (entry, tmp, &fd_ctx->entries, list) {
-                GF_FREE (entry->name);
-                list_del (&entry->list);
-                GF_FREE (entry);
-        }
-}
+        if (!validate_subvol)
+            continue;
 
-static void
-afr_readdir_filter_trash_dir (gf_dirent_t *entries, fd_t *fd)
-{
-        gf_dirent_t *   entry       = NULL;
-        gf_dirent_t *   tmp         = NULL;
-
-        list_for_each_entry_safe (entry, tmp, &entries->list, list) {
-                if (__is_root_gfid (fd->inode->gfid) &&
-                    !strcmp (entry->d_name, GF_REPLICATE_TRASH_DIR)) {
-                        list_del_init (&entry->list);
-                        GF_FREE (entry);
-                }
+        if (entry->inode) {
+            ret = afr_validate_read_subvol(entry->inode, this, subvol);
+            if (ret == -1) {
+                inode_unref(entry->inode);
+                entry->inode = NULL;
+                continue;
+            }
         }
+    }
 }
 
 int32_t
-afr_readdir_cbk (call_frame_t *frame, void *cookie,
-                 xlator_t *this, int32_t op_ret, int32_t op_errno,
-                 gf_dirent_t *entries, dict_t *xdata)
+afr_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, gf_dirent_t *subvol_entries,
+                dict_t *xdata)
 {
-        afr_local_t     *local = NULL;
+    afr_local_t *local = NULL;
+    gf_dirent_t entries;
 
-        if (op_ret == -1)
-                goto out;
+    INIT_LIST_HEAD(&entries.list);
 
-        local = frame->local;
-        afr_readdir_filter_trash_dir (entries, local->fd);
+    local = frame->local;
 
-out:
-        AFR_STACK_UNWIND (readdir, frame, op_ret, op_errno, entries, NULL);
+    if (op_ret < 0 && !local->cont.readdir.offset) {
+        /* failover only if this was first readdir, detected
+           by offset == 0 */
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+
+        afr_read_txn_continue(frame, this, (long)cookie);
         return 0;
-}
+    }
 
+    if (op_ret >= 0)
+        afr_readdir_transform_entries(frame, subvol_entries, (long)cookie,
+                                      &entries, local->fd);
 
-int32_t
-afr_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno, gf_dirent_t *entries,
-                  dict_t *xdata)
-{
-        afr_local_t     *local = NULL;
+    AFR_STACK_UNWIND(readdir, frame, op_ret, op_errno, &entries, xdata);
 
-        if (op_ret == -1)
-                goto out;
+    gf_dirent_free(&entries);
 
-        local = frame->local;
-        afr_readdir_filter_trash_dir (entries, local->fd);
+    return 0;
+}
 
-out:
-        AFR_STACK_UNWIND (readdirp, frame, op_ret, op_errno, entries, NULL);
+int
+afr_readdir_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    afr_fd_ctx_t *fd_ctx = NULL;
+
+    priv = this->private;
+    local = frame->local;
+    fd_ctx = afr_fd_ctx_get(local->fd, this);
+    if (!fd_ctx) {
+        local->op_errno = EINVAL;
+        local->op_ret = -1;
+    }
+
+    if (subvol == -1 || !fd_ctx) {
+        AFR_STACK_UNWIND(readdir, frame, local->op_ret, local->op_errno, 0, 0);
         return 0;
+    }
+
+    fd_ctx->readdir_subvol = subvol;
+
+    if (local->op == GF_FOP_READDIR)
+        STACK_WIND_COOKIE(frame, afr_readdir_cbk, (void *)(long)subvol,
+                          priv->children[subvol],
+                          priv->children[subvol]->fops->readdir, local->fd,
+                          local->cont.readdir.size, local->cont.readdir.offset,
+                          local->xdata_req);
+    else
+        STACK_WIND_COOKIE(frame, afr_readdir_cbk, (void *)(long)subvol,
+                          priv->children[subvol],
+                          priv->children[subvol]->fops->readdirp, local->fd,
+                          local->cont.readdir.size, local->cont.readdir.offset,
+                          local->xdata_req);
+    return 0;
 }
 
-int32_t
-afr_do_readdir (call_frame_t *frame, xlator_t *this,
-                fd_t *fd, size_t size, off_t offset, int whichop, dict_t *dict)
+int
+afr_do_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+               off_t offset, int whichop, dict_t *dict)
 {
-        afr_private_t *priv      = NULL;
-        xlator_t      **children = NULL;
-        int           call_child = 0;
-        afr_local_t   *local     = NULL;
-        afr_fd_ctx_t  *fd_ctx    = NULL;
-        int           ret        = -1;
-        int32_t       op_errno   = 0;
-        uint64_t      read_child = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
-
-        priv     = this->private;
-        children = priv->children;
-
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
-
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
-
-        local->fresh_children = afr_children_create (priv->child_count);
-        if (!local->fresh_children) {
-                op_errno = ENOMEM;
-                goto out;
-        }
-
-        read_child = afr_inode_get_read_ctx (this, fd->inode,
-                                             local->fresh_children);
-        ret = afr_get_call_child (this, local->child_up, read_child,
-                                  local->fresh_children,
-                                  &call_child,
-                                  &local->cont.readdir.last_index);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
-
-        fd_ctx  = afr_fd_ctx_get (fd, this);
-        if (!fd_ctx) {
-                op_errno = EBADF;
-                goto out;
-        }
-
-        if ((offset == 0) || (fd_ctx->call_child == -1)) {
-                fd_ctx->call_child = call_child;
-        } else if ((priv->readdir_failover == _gf_false) &&
-                   (call_child != fd_ctx->call_child)) {
-                op_errno = EBADF;
-                goto out;
-        }
-
-        local->fd                  = fd_ref (fd);
-        local->cont.readdir.size   = size;
-        local->cont.readdir.dict   = (dict)? dict_ref (dict) : NULL;
-
-        if (whichop == GF_FOP_READDIR)
-                STACK_WIND_COOKIE (frame, afr_readdir_cbk,
-                                   (void *) (long) call_child,
-                                   children[call_child],
-                                   children[call_child]->fops->readdir, fd,
-                                   size, offset, dict);
-        else
-                STACK_WIND_COOKIE (frame, afr_readdirp_cbk,
-                                   (void *) (long) call_child,
-                                   children[call_child],
-                                   children[call_child]->fops->readdirp, fd,
-                                   size, offset, dict);
-
-        return 0;
+    afr_local_t *local = NULL;
+    int32_t op_errno = 0;
+    int subvol = -1;
+    afr_fd_ctx_t *fd_ctx = NULL;
+
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
+
+    fd_ctx = afr_fd_ctx_get(fd, this);
+    if (!fd_ctx) {
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    local->op = whichop;
+    local->fd = fd_ref(fd);
+    local->cont.readdir.size = size;
+    local->cont.readdir.offset = offset;
+    local->xdata_req = (dict) ? dict_ref(dict) : NULL;
+
+    subvol = fd_ctx->readdir_subvol;
+
+    if (offset == 0 || subvol == -1) {
+        /* First readdir has option of failing over and selecting
+           an appropriate read subvolume */
+        afr_read_txn(frame, this, fd->inode, afr_readdir_wind,
+                     AFR_DATA_TRANSACTION);
+    } else {
+        /* But continued readdirs MUST stick to the same subvolume
+           without an option to failover */
+        afr_readdir_wind(frame, this, subvol);
+    }
+
+    return 0;
 out:
-        AFR_STACK_UNWIND (readdir, frame, -1, op_errno, NULL, NULL);
-        return 0;
+    AFR_STACK_UNWIND(readdir, frame, -1, op_errno, NULL, NULL);
+    return 0;
 }
 
-
 int32_t
-afr_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-             off_t offset, dict_t *xdata)
+afr_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t offset, dict_t *xdata)
 {
-        afr_do_readdir (frame, this, fd, size, offset, GF_FOP_READDIR, xdata);
-        return 0;
-}
+    afr_do_readdir(frame, this, fd, size, offset, GF_FOP_READDIR, xdata);
 
+    return 0;
+}
 
 int32_t
-afr_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-              off_t offset, dict_t *dict)
+afr_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+             off_t offset, dict_t *dict)
 {
-        afr_do_readdir (frame, this, fd, size, offset, GF_FOP_READDIRP, dict);
-        return 0;
-}
+    afr_do_readdir(frame, this, fd, size, offset, GF_FOP_READDIRP, dict);
 
+    return 0;
+}
 
 int32_t
-afr_releasedir (xlator_t *this, fd_t *fd)
+afr_releasedir(xlator_t *this, fd_t *fd)
 {
-        afr_forget_entries (fd);
-        afr_cleanup_fd_ctx (this, fd);
+    afr_cleanup_fd_ctx(this, fd);
 
-        return 0;
+    return 0;
 }
diff --git a/xlators/cluster/afr/src/afr-dir-read.h b/xlators/cluster/afr/src/afr-dir-read.h
index 09456d15949..773e925ec6c 100644
--- a/xlators/cluster/afr/src/afr-dir-read.h
+++ b/xlators/cluster/afr/src/afr-dir-read.h
@@ -11,26 +11,23 @@
 #ifndef __DIR_READ_H__
 #define __DIR_READ_H__
 
-
 int32_t
-afr_opendir (call_frame_t *frame, xlator_t *this,
-	     loc_t *loc, fd_t *fd, dict_t *xdata);
+afr_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+            dict_t *xdata);
 
 int32_t
-afr_releasedir (xlator_t *this, fd_t *fd);
+afr_releasedir(xlator_t *this, fd_t *fd);
 
 int32_t
-afr_readdir (call_frame_t *frame, xlator_t *this,
-	     fd_t *fd, size_t size, off_t offset, dict_t *xdata);
-
+afr_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t offset, dict_t *xdata);
 
 int32_t
-afr_readdirp (call_frame_t *frame, xlator_t *this,
-              fd_t *fd, size_t size, off_t offset, dict_t *dict);
+afr_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+             off_t offset, dict_t *dict);
 
 int32_t
-afr_checksum (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, int32_t flags, dict_t *xdata);
-
+afr_checksum(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+             dict_t *xdata);
 
 #endif /* __DIR_READ_H__ */
diff --git a/xlators/cluster/afr/src/afr-dir-write.c b/xlators/cluster/afr/src/afr-dir-write.c
index 1943b719bb5..b7cceb79158 100644
--- a/xlators/cluster/afr/src/afr-dir-write.c
+++ b/xlators/cluster/afr/src/afr-dir-write.c
@@ -8,500 +8,493 @@
   cases as published by the Free Software Foundation.
 */
 
-
 #include <libgen.h>
 #include <unistd.h>
-#include <fnmatch.h>
 #include <sys/time.h>
 #include <stdlib.h>
 #include <signal.h>
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
+#include <glusterfs/glusterfs.h>
 #include "afr.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
+#include <glusterfs/dict.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/list.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/byte-order.h>
 
 #include "afr.h"
 #include "afr-transaction.h"
 
+void
+afr_mark_entry_pending_changelog(call_frame_t *frame, xlator_t *this);
+
 int
-afr_build_parent_loc (loc_t *parent, loc_t *child, int32_t *op_errno)
+afr_build_parent_loc(loc_t *parent, loc_t *child, int32_t *op_errno)
 {
-        int     ret = -1;
-        char    *child_path = NULL;
-
-        if (!child->parent) {
-                if (op_errno)
-                        *op_errno = EINVAL;
-                goto out;
-        }
-
-        child_path = gf_strdup (child->path);
-        if (!child_path) {
-                if (op_errno)
-                        *op_errno = ENOMEM;
-                goto out;
-        }
-        parent->path = gf_strdup( dirname (child_path) );
-	if (!parent->path) {
-                if (op_errno)
-                        *op_errno = ENOMEM;
-                goto out;
-        }
-        parent->inode  = inode_ref (child->parent);
-        uuid_copy (parent->gfid, child->pargfid);
-
-        ret = 0;
+    int ret = -1;
+    char *child_path = NULL;
+
+    if (!child->parent) {
+        if (op_errno)
+            *op_errno = EINVAL;
+        goto out;
+    }
+
+    child_path = gf_strdup(child->path);
+    if (!child_path) {
+        if (op_errno)
+            *op_errno = ENOMEM;
+        goto out;
+    }
+
+    parent->path = gf_strdup(dirname(child_path));
+    if (!parent->path) {
+        if (op_errno)
+            *op_errno = ENOMEM;
+        goto out;
+    }
+
+    parent->inode = inode_ref(child->parent);
+    gf_uuid_copy(parent->gfid, child->pargfid);
+
+    ret = 0;
 out:
-	GF_FREE(child_path);
+    GF_FREE(child_path);
 
-        return ret;
+    return ret;
 }
 
-void
-__dir_entry_fop_common_cbk (call_frame_t *frame, int child_index,
-                            xlator_t *this, int32_t op_ret,
-                            int32_t op_errno, inode_t *inode,
-                            struct iatt *buf, struct iatt *preparent,
-                            struct iatt *postparent, struct iatt *prenewparent,
-                            struct iatt *postnewparent)
+static void
+__afr_dir_write_finalize(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t     *local          = NULL;
-
-        local = frame->local;
-
-        if (afr_fop_failed (op_ret, op_errno))
-                afr_transaction_fop_failed (frame, this, child_index);
-
-        if (op_ret > -1) {
-                local->op_ret = op_ret;
-
-                if ((local->success_count == 0) ||
-                    (child_index == local->read_child_index)) {
-                        local->cont.dir_fop.preparent      = *preparent;
-                        local->cont.dir_fop.postparent     = *postparent;
-                        if (buf)
-                                local->cont.dir_fop.buf            = *buf;
-                        if (prenewparent)
-                             local->cont.dir_fop.prenewparent  = *prenewparent;
-                        if (postnewparent)
-                             local->cont.dir_fop.postnewparent = *postnewparent;
-                }
-
-                local->cont.dir_fop.inode = inode;
-
-                local->fresh_children[local->success_count] = child_index;
-                local->success_count++;
-                local->child_errno[child_index] = 0;
-        } else {
-                local->child_errno[child_index] = op_errno;
-        }
-
-        local->op_errno = op_errno;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int inode_read_subvol = -1;
+    int parent_read_subvol = -1;
+    int parent2_read_subvol = -1;
+    int i = 0;
+    afr_read_subvol_args_t args = {
+        0,
+    };
+
+    local = frame->local;
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
+        if (local->replies[i].op_ret == -1)
+            continue;
+        gf_uuid_copy(args.gfid, local->replies[i].poststat.ia_gfid);
+        args.ia_type = local->replies[i].poststat.ia_type;
+        break;
+    }
+
+    if (local->inode) {
+        if (local->op != GF_FOP_RENAME && local->op != GF_FOP_LINK)
+            afr_replies_interpret(frame, this, local->inode, NULL);
+
+        inode_read_subvol = afr_data_subvol_get(local->inode, this, NULL, NULL,
+                                                NULL, &args);
+    }
+
+    if (local->parent)
+        parent_read_subvol = afr_data_subvol_get(local->parent, this, NULL,
+                                                 local->readable, NULL, NULL);
+
+    if (local->parent2)
+        parent2_read_subvol = afr_data_subvol_get(local->parent2, this, NULL,
+                                                  local->readable2, NULL, NULL);
+
+    local->op_ret = -1;
+    local->op_errno = afr_final_errno(local, priv);
+    afr_pick_error_xdata(local, priv, local->parent, local->readable,
+                         local->parent2, local->readable2);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
+        if (local->replies[i].op_ret < 0) {
+            if (local->inode)
+                afr_inode_need_refresh_set(local->inode, this);
+            if (local->parent)
+                afr_inode_need_refresh_set(local->parent, this);
+            if (local->parent2)
+                afr_inode_need_refresh_set(local->parent2, this);
+            continue;
+        }
+
+        if (local->op_ret == -1) {
+            local->op_ret = local->replies[i].op_ret;
+            local->op_errno = local->replies[i].op_errno;
+
+            local->cont.dir_fop.buf = local->replies[i].poststat;
+            local->cont.dir_fop.preparent = local->replies[i].preparent;
+            local->cont.dir_fop.postparent = local->replies[i].postparent;
+            local->cont.dir_fop.prenewparent = local->replies[i].preparent2;
+            local->cont.dir_fop.postnewparent = local->replies[i].postparent2;
+            if (local->xdata_rsp) {
+                dict_unref(local->xdata_rsp);
+                local->xdata_rsp = NULL;
+            }
+
+            if (local->replies[i].xdata)
+                local->xdata_rsp = dict_ref(local->replies[i].xdata);
+            continue;
+        }
+
+        if (i == inode_read_subvol) {
+            local->cont.dir_fop.buf = local->replies[i].poststat;
+            if (local->replies[i].xdata) {
+                if (local->xdata_rsp)
+                    dict_unref(local->xdata_rsp);
+                local->xdata_rsp = dict_ref(local->replies[i].xdata);
+            }
+        }
+
+        if (i == parent_read_subvol) {
+            local->cont.dir_fop.preparent = local->replies[i].preparent;
+            local->cont.dir_fop.postparent = local->replies[i].postparent;
+        }
+
+        if (i == parent2_read_subvol) {
+            local->cont.dir_fop.prenewparent = local->replies[i].preparent2;
+            local->cont.dir_fop.postnewparent = local->replies[i].postparent2;
+        }
+    }
 }
 
-int
-afr_mark_new_entry_changelog_cbk (call_frame_t *frame, void *cookie,
-                                  xlator_t *this,
-                                  int32_t op_ret, int32_t op_errno,
-                                  dict_t *xattr, dict_t *xdata)
+static void
+__afr_dir_write_fill(call_frame_t *frame, xlator_t *this, int child_index,
+                     int op_ret, int op_errno, struct iatt *poststat,
+                     struct iatt *preparent, struct iatt *postparent,
+                     struct iatt *preparent2, struct iatt *postparent2,
+                     dict_t *xdata)
 {
-        int     call_count = 0;
-
-        call_count = afr_frame_return (frame);
-        if (call_count == 0) {
-                AFR_STACK_DESTROY (frame);
-        }
-        return 0;
+    afr_local_t *local = NULL;
+    afr_fd_ctx_t *fd_ctx = NULL;
+
+    local = frame->local;
+    fd_ctx = local->fd_ctx;
+
+    local->replies[child_index].valid = 1;
+    local->replies[child_index].op_ret = op_ret;
+    local->replies[child_index].op_errno = op_errno;
+    if (xdata)
+        local->replies[child_index].xdata = dict_ref(xdata);
+
+    if (op_ret >= 0) {
+        if (poststat)
+            local->replies[child_index].poststat = *poststat;
+        if (preparent)
+            local->replies[child_index].preparent = *preparent;
+        if (postparent)
+            local->replies[child_index].postparent = *postparent;
+        if (preparent2)
+            local->replies[child_index].preparent2 = *preparent2;
+        if (postparent2)
+            local->replies[child_index].postparent2 = *postparent2;
+        if (fd_ctx)
+            fd_ctx->opened_on[child_index] = AFR_FD_OPENED;
+    } else {
+        if (op_errno != ENOTEMPTY)
+            afr_transaction_fop_failed(frame, this, child_index);
+        if (fd_ctx)
+            fd_ctx->opened_on[child_index] = AFR_FD_NOT_OPENED;
+    }
+
+    return;
 }
 
-void
-afr_mark_new_entry_changelog (call_frame_t *frame, xlator_t *this)
+static int
+__afr_dir_write_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int op_ret, int op_errno, struct iatt *buf,
+                    struct iatt *preparent, struct iatt *postparent,
+                    struct iatt *preparent2, struct iatt *postparent2,
+                    dict_t *xdata)
 {
-        call_frame_t  *new_frame  = NULL;
-        afr_local_t   *local      = NULL;
-        afr_local_t   *new_local  = NULL;
-        afr_private_t *priv       = NULL;
-        dict_t        **xattr     = NULL;
-        int32_t       **changelog = NULL;
-        int           i           = 0;
-        GF_UNUSED int op_errno    = 0;
-
-        local = frame->local;
-        priv = this->private;
-
-        new_frame = copy_frame (frame);
-        if (!new_frame) {
-                goto out;
-        }
+    afr_local_t *local = NULL;
+    int child_index = (long)cookie;
+    int call_count = -1;
+    afr_private_t *priv = NULL;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (new_frame->local, out);
-        new_local = new_frame->local;
-        changelog = afr_matrix_create (priv->child_count, AFR_NUM_CHANGE_LOGS);
-        if (!changelog)
-                goto out;
-
-        xattr = GF_CALLOC (priv->child_count, sizeof (*xattr),
-                           gf_afr_mt_dict_t);
-        if (!xattr)
-                goto out;
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_errno[i])
-                        continue;
-                xattr[i] = dict_new ();
-                if (!xattr[i])
-                        goto out;
-        }
+    priv = this->private;
+    local = frame->local;
 
-        afr_prepare_new_entry_pending_matrix (changelog,
-                                              afr_is_errno_set,
-                                              local->child_errno,
-                                              &local->cont.dir_fop.buf,
-                                              priv->child_count);
-
-        new_local->pending = changelog;
-        uuid_copy (new_local->loc.gfid, local->cont.dir_fop.buf.ia_gfid);
-        new_local->loc.inode = inode_ref (local->cont.dir_fop.inode);
-        new_local->call_count = local->success_count;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_errno[i])
-                        continue;
-
-                afr_set_pending_dict (priv, xattr[i], changelog, i, LOCAL_LAST);
-                STACK_WIND_COOKIE (new_frame, afr_mark_new_entry_changelog_cbk,
-                                   (void *) (long) i, priv->children[i],
-                                   priv->children[i]->fops->xattrop,
-                                   &new_local->loc, GF_XATTROP_ADD_ARRAY,
-                                   xattr[i], NULL);
-        }
-        new_frame = NULL;
-out:
-        if (new_frame)
-                AFR_STACK_DESTROY (new_frame);
-        afr_xattr_array_destroy (xattr, priv->child_count);
-        return;
-}
+    LOCK(&frame->lock);
+    {
+        __afr_dir_write_fill(frame, this, child_index, op_ret, op_errno, buf,
+                             preparent, postparent, preparent2, postparent2,
+                             xdata);
+        call_count = --local->call_count;
+    }
+    UNLOCK(&frame->lock);
 
-gf_boolean_t
-afr_is_new_entry_changelog_needed (glusterfs_fop_t fop)
-{
-        glusterfs_fop_t fops[]   = {GF_FOP_CREATE, GF_FOP_MKNOD, GF_FOP_NULL};
-        int             i        = 0;
+    if (call_count == 0) {
+        __afr_dir_write_finalize(frame, this);
 
-        for (i = 0; fops[i] != GF_FOP_NULL; i++) {
-                if (fop == fops[i])
-                        return _gf_true;
+        if (afr_txn_nothing_failed(frame, this)) {
+            /*if it did pre-op, it will do post-op changing ctime*/
+            if (priv->consistent_metadata && afr_needs_changelog_update(local))
+                afr_zero_fill_stat(local);
+            local->transaction.unwind(frame, this);
         }
-        return _gf_false;
-}
 
-void
-afr_dir_fop_mark_entry_pending_changelog (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t   *local      = NULL;
-        afr_private_t *priv       = NULL;
+        afr_mark_entry_pending_changelog(frame, this);
 
-        local = frame->local;
-        priv  = this->private;
+        afr_transaction_resume(frame, this);
+    }
 
-        if (local->op_ret < 0)
-                goto out;
+    return 0;
+}
 
-        if (local->success_count == priv->child_count)
-                goto out;
+int
+afr_mark_new_entry_changelog_cbk(call_frame_t *frame, void *cookie,
+                                 xlator_t *this, int op_ret, int op_errno,
+                                 dict_t *xattr, dict_t *xdata)
+{
+    int call_count = 0;
 
-        if (!afr_is_new_entry_changelog_needed (local->op))
-                goto out;
+    call_count = afr_frame_return(frame);
 
-        afr_mark_new_entry_changelog (frame, this);
+    if (call_count == 0)
+        AFR_STACK_DESTROY(frame);
 
-out:
-        return;
+    return 0;
 }
 
 void
-afr_dir_fop_done (call_frame_t *frame, xlator_t *this)
+afr_mark_new_entry_changelog(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t     *local          = NULL;
-        afr_private_t   *priv           = NULL;
-
-        local = frame->local;
-        priv  = this->private;
-
-        if (local->cont.dir_fop.inode == NULL)
-                goto done;
-        afr_set_read_ctx_from_policy (this, local->cont.dir_fop.inode,
-                                      local->fresh_children,
-                                      local->read_child_index,
-                                      priv->read_child,
-                                      local->cont.dir_fop.buf.ia_gfid);
-done:
-        local->transaction.unwind (frame, this);
-        afr_dir_fop_mark_entry_pending_changelog (frame, this);
-        local->transaction.resume (frame, this);
+    call_frame_t *new_frame = NULL;
+    afr_local_t *local = NULL;
+    afr_local_t *new_local = NULL;
+    afr_private_t *priv = NULL;
+    dict_t *xattr = NULL;
+    int32_t **changelog = NULL;
+    int i = 0;
+    int op_errno = ENOMEM;
+    unsigned char *pending = NULL;
+    int call_count = 0;
+
+    local = frame->local;
+    priv = this->private;
+
+    new_frame = copy_frame(frame);
+    if (!new_frame)
+        goto out;
+
+    new_local = AFR_FRAME_INIT(new_frame, op_errno);
+    if (!new_local)
+        goto out;
+
+    xattr = dict_new();
+    if (!xattr)
+        goto out;
+
+    pending = alloca0(priv->child_count);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->transaction.pre_op[i] &&
+            !local->transaction.failed_subvols[i]) {
+            call_count++;
+            continue;
+        }
+        pending[i] = 1;
+    }
+
+    changelog = afr_mark_pending_changelog(priv, pending, xattr,
+                                           local->cont.dir_fop.buf.ia_type);
+    if (!changelog)
+        goto out;
+
+    new_local->pending = changelog;
+    gf_uuid_copy(new_local->loc.gfid, local->cont.dir_fop.buf.ia_gfid);
+    new_local->loc.inode = inode_ref(local->inode);
+
+    new_local->call_count = call_count;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (pending[i])
+            continue;
+
+        STACK_WIND_COOKIE(new_frame, afr_mark_new_entry_changelog_cbk,
+                          (void *)(long)i, priv->children[i],
+                          priv->children[i]->fops->xattrop, &new_local->loc,
+                          GF_XATTROP_ADD_ARRAY, xattr, NULL);
+        if (!--call_count)
+            break;
+    }
+
+    new_frame = NULL;
+out:
+    if (new_frame)
+        AFR_STACK_DESTROY(new_frame);
+    if (xattr)
+        dict_unref(xattr);
+    return;
 }
 
-/* {{{ create */
-
-int
-afr_create_unwind (call_frame_t *frame, xlator_t *this)
+void
+afr_mark_entry_pending_changelog(call_frame_t *frame, xlator_t *this)
 {
-        call_frame_t *main_frame = NULL;
-        afr_local_t  *local = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int pre_op_count = 0;
+    int failed_count = 0;
+    unsigned char *success_replies = NULL;
 
-        local = frame->local;
+    local = frame->local;
+    priv = this->private;
 
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame) {
-                        main_frame = local->transaction.main_frame;
-                }
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
-
-        if (main_frame) {
-                AFR_STACK_UNWIND (create, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  local->cont.create.fd,
-                                  local->cont.dir_fop.inode,
-                                  &local->cont.dir_fop.buf,
-                                  &local->cont.dir_fop.preparent,
-                                  &local->cont.dir_fop.postparent,
-                                  local->xdata_rsp);
-        }
+    if (local->op_ret < 0)
+        return;
 
-        return 0;
-}
+    if (local->op != GF_FOP_CREATE && local->op != GF_FOP_MKNOD &&
+        local->op != GF_FOP_MKDIR)
+        return;
 
+    pre_op_count = AFR_COUNT(local->transaction.pre_op, priv->child_count);
+    failed_count = AFR_COUNT(local->transaction.failed_subvols,
+                             priv->child_count);
 
-int
-afr_create_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno,
-                     fd_t *fd, inode_t *inode, struct iatt *buf,
-                     struct iatt *preparent, struct iatt *postparent,
-                     dict_t *xdata)
-{
-        afr_local_t     *local = NULL;
-        uint64_t        ctx = 0;
-        afr_fd_ctx_t    *fd_ctx = NULL;
-        int             ret = 0;
-        int             call_count = -1;
-        int             child_index = -1;
-
-        local = frame->local;
-
-        child_index = (long) cookie;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret > -1) {
-                        ret = afr_fd_ctx_set (this, fd);
-                        if (ret < 0) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "could not set ctx on fd=%p", fd);
-
-                                local->op_ret   = -1;
-                                local->op_errno = -ret;
-                                goto unlock;
-                        }
-
-                        ret = fd_ctx_get (fd, this, &ctx);
-                        if (ret < 0) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "could not get fd ctx for fd=%p", fd);
-                                local->op_ret   = -1;
-                                local->op_errno = -ret;
-                                goto unlock;
-                        }
-
-                        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
-                        fd_ctx->opened_on[child_index] = AFR_FD_OPENED;
-                        fd_ctx->flags                  = local->cont.create.flags;
-
-                        if (local->success_count == 0) {
-				if (xdata)
-					local->xdata_rsp = dict_ref(xdata);
-			}
-                }
-                __dir_entry_fop_common_cbk (frame, child_index, this,
-                                            op_ret, op_errno, inode, buf,
-                                            preparent, postparent, NULL, NULL);
-        }
+    /* FOP succeeded on all bricks. */
+    if (pre_op_count == priv->child_count && !failed_count)
+        return;
 
-unlock:
-        UNLOCK (&frame->lock);
+    /* FOP did not suceed on quorum no. of bricks. */
+    success_replies = alloca0(priv->child_count);
+    afr_fill_success_replies(local, priv, success_replies);
+    if (!afr_has_quorum(success_replies, this, NULL))
+        return;
 
-        call_count = afr_frame_return (frame);
+    if (priv->thin_arbiter_count) {
+        /*Mark new entry using ta file*/
+        local->is_new_entry = _gf_true;
+        return;
+    }
 
-        if (call_count == 0)
-                afr_dir_fop_done (frame, this);
+    afr_mark_new_entry_changelog(frame, this);
 
-        return 0;
+    return;
 }
 
+/* {{{ create */
 
 int
-afr_create_wind (call_frame_t *frame, xlator_t *this)
+afr_create_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int call_count = -1;
-        int i = 0;
-
-        local = frame->local;
-        priv = this->private;
-
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
+    call_frame_t *main_frame = NULL;
+    afr_local_t *local = NULL;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
+    local = frame->local;
 
-        local->call_count = call_count;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->transaction.pre_op[i]) {
-                        STACK_WIND_COOKIE (frame, afr_create_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->create,
-                                           &local->loc,
-                                           local->cont.create.flags,
-                                           local->cont.create.mode,
-                                           local->umask,
-                                           local->cont.create.fd,
-                                           local->xdata_req);
-                        if (!--call_count)
-                                break;
-                }
-        }
+    main_frame = afr_transaction_detach_fop_frame(frame);
 
+    if (!main_frame)
         return 0;
-}
 
+    AFR_STACK_UNWIND(create, main_frame, local->op_ret, local->op_errno,
+                     local->cont.create.fd, local->inode,
+                     &local->cont.dir_fop.buf, &local->cont.dir_fop.preparent,
+                     &local->cont.dir_fop.postparent, local->xdata_rsp);
+    return 0;
+}
 
 int
-afr_create_done (call_frame_t *frame, xlator_t *this)
+afr_create_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, fd_t *fd, inode_t *inode,
+                    struct iatt *buf, struct iatt *preparent,
+                    struct iatt *postparent, dict_t *xdata)
 {
-        afr_local_t * local = NULL;
-
-        local = frame->local;
-
-        local->transaction.unwind (frame, this);
-
-        AFR_STACK_DESTROY (frame);
-
-        return 0;
+    return __afr_dir_write_cbk(frame, cookie, this, op_ret, op_errno, buf,
+                               preparent, postparent, NULL, NULL, xdata);
 }
 
-
 int
-afr_create (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, int32_t flags, mode_t mode,
-            mode_t umask, fd_t *fd, dict_t *params)
+afr_create_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_private_t           *priv                   = NULL;
-        afr_local_t             *local                  = NULL;
-        afr_internal_lock_t     *int_lock               = NULL;
-        call_frame_t            *transaction_frame      = NULL;
-        int                     ret                     = -1;
-        int                     op_errno                = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
-
-        priv = this->private;
-
-        QUORUM_CHECK(create,out);
-
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                op_errno = ENOMEM;
-                goto out;
-        }
-
-        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
-        local = transaction_frame->local;
-
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
-
-        loc_copy (&local->loc, loc);
-
-        LOCK (&priv->read_child_lock);
-        {
-                local->read_child_index = (++priv->read_child_rr)
-                        % (priv->child_count);
-        }
-        UNLOCK (&priv->read_child_lock);
-
-        local->op                = GF_FOP_CREATE;
-        local->cont.create.flags = flags;
-        local->cont.create.mode  = mode;
-        local->cont.create.fd    = fd_ref (fd);
-        local->umask  = umask;
-        if (params)
-                local->xdata_req = dict_ref (params);
-
-        local->transaction.fop    = afr_create_wind;
-        local->transaction.done   = afr_create_done;
-        local->transaction.unwind = afr_create_unwind;
-
-        ret = afr_build_parent_loc (&local->transaction.parent_loc, loc,
-                                    &op_errno);
-        if (ret)
-                goto out;
-
-        local->transaction.main_frame = frame;
-        local->transaction.basename = AFR_BASENAME (loc->path);
-        int_lock = &local->internal_lock;
-
-        int_lock->lockee_count = 0;
-        ret = afr_init_entry_lockee (&int_lock->lockee[0], local,
-                                     &local->transaction.parent_loc,
-                                     local->transaction.basename,
-                                     priv->child_count);
-        if (ret)
-                goto out;
-
-        int_lock->lockee_count++;
-        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
-        if (ret < 0) {
-            op_errno = -ret;
-            goto out;
-        }
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    STACK_WIND_COOKIE(frame, afr_create_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->create, &local->loc,
+                      local->cont.create.flags, local->cont.create.mode,
+                      local->umask, local->cont.create.fd, local->xdata_req);
+    return 0;
+}
 
-        ret = 0;
+int
+afr_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+           mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
+
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
+
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
+
+    loc_copy(&local->loc, loc);
+
+    local->fd_ctx = afr_fd_ctx_get(fd, this);
+    if (!local->fd_ctx)
+        goto out;
+
+    local->inode = inode_ref(loc->inode);
+    local->parent = inode_ref(loc->parent);
+
+    local->op = GF_FOP_CREATE;
+    local->cont.create.flags = flags;
+    local->fd_ctx->flags = flags;
+    local->cont.create.mode = mode;
+    local->cont.create.fd = fd_ref(fd);
+    local->umask = umask;
+
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
+
+    if (!local->xdata_req)
+        goto out;
+
+    local->transaction.wind = afr_create_wind;
+    local->transaction.unwind = afr_create_unwind;
+
+    ret = afr_build_parent_loc(&local->transaction.parent_loc, loc, &op_errno);
+    if (ret)
+        goto out;
+
+    local->transaction.main_frame = frame;
+    local->transaction.basename = AFR_BASENAME(loc->path);
+    ret = afr_transaction(transaction_frame, this, AFR_ENTRY_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
+
+    return 0;
 out:
-        if (ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (create, frame, -1, op_errno,
-                                  NULL, NULL, NULL, NULL, NULL, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(create, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL,
+                     NULL);
+    return 0;
 }
 
 /* }}} */
@@ -509,814 +502,436 @@ out:
 /* {{{ mknod */
 
 int
-afr_mknod_unwind (call_frame_t *frame, xlator_t *this)
+afr_mknod_unwind(call_frame_t *frame, xlator_t *this)
 {
-        call_frame_t *main_frame = NULL;
-        afr_local_t  *local = NULL;
-
-        local = frame->local;
+    call_frame_t *main_frame = NULL;
+    afr_local_t *local = NULL;
 
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame) {
-                        main_frame = local->transaction.main_frame;
-                }
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
-
-        if (main_frame) {
-                AFR_STACK_UNWIND (mknod, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  local->cont.dir_fop.inode,
-                                  &local->cont.dir_fop.buf,
-                                  &local->cont.dir_fop.preparent,
-                                  &local->cont.dir_fop.postparent,
-                                  NULL);
-        }
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
-}
-
-
-int
-afr_mknod_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, inode_t *inode,
-                    struct iatt *buf, struct iatt *preparent,
-                    struct iatt *postparent, dict_t *xdata)
-{
-        int             call_count      = -1;
-        int             child_index     = -1;
-
-        child_index = (long) cookie;
 
-        LOCK (&frame->lock);
-        {
-                __dir_entry_fop_common_cbk (frame, child_index, this,
-                                            op_ret, op_errno, inode, buf,
-                                            preparent, postparent, NULL, NULL);
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0)
-                afr_dir_fop_done (frame, this);
-
-        return 0;
+    AFR_STACK_UNWIND(mknod, main_frame, local->op_ret, local->op_errno,
+                     local->inode, &local->cont.dir_fop.buf,
+                     &local->cont.dir_fop.preparent,
+                     &local->cont.dir_fop.postparent, local->xdata_rsp);
+    return 0;
 }
 
-
-int32_t
-afr_mknod_wind (call_frame_t *frame, xlator_t *this)
+int
+afr_mknod_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, inode_t *inode,
+                   struct iatt *buf, struct iatt *preparent,
+                   struct iatt *postparent, dict_t *xdata)
 {
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int call_count = -1;
-        int i = 0;
-
-        local = frame->local;
-        priv  = this->private;
-
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
-
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
-
-        local->call_count = call_count;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->transaction.pre_op[i]) {
-                        STACK_WIND_COOKIE (frame, afr_mknod_wind_cbk, (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->mknod,
-                                           &local->loc, local->cont.mknod.mode,
-                                           local->cont.mknod.dev,
-                                           local->umask,
-                                           local->xdata_req);
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-        return 0;
+    return __afr_dir_write_cbk(frame, cookie, this, op_ret, op_errno, buf,
+                               preparent, postparent, NULL, NULL, xdata);
 }
 
-
 int
-afr_mknod_done (call_frame_t *frame, xlator_t *this)
+afr_mknod_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_local_t * local = NULL;
-
-        local = frame->local;
-
-        local->transaction.unwind (frame, this);
-        AFR_STACK_DESTROY (frame);
-
-        return 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    STACK_WIND_COOKIE(frame, afr_mknod_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->mknod, &local->loc,
+                      local->cont.mknod.mode, local->cont.mknod.dev,
+                      local->umask, local->xdata_req);
+    return 0;
 }
 
-
 int
-afr_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
-           dev_t dev, mode_t umask, dict_t *params)
+afr_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+          dev_t dev, mode_t umask, dict_t *xdata)
 {
-        afr_private_t           *priv                   = NULL;
-        afr_local_t             *local                  = NULL;
-        afr_internal_lock_t     *int_lock               = NULL;
-        call_frame_t            *transaction_frame      = NULL;
-        int                     ret                     = -1;
-        int                     op_errno                = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
-
-        priv = this->private;
-
-        QUORUM_CHECK(mknod,out);
-
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                op_errno = ENOMEM;
-                goto out;
-        }
-
-        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
-        local = transaction_frame->local;
-
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
-
-        loc_copy (&local->loc, loc);
-
-        LOCK (&priv->read_child_lock);
-        {
-                local->read_child_index = (++priv->read_child_rr)
-                        % (priv->child_count);
-        }
-        UNLOCK (&priv->read_child_lock);
-
-        local->op               = GF_FOP_MKNOD;
-        local->cont.mknod.mode  = mode;
-        local->cont.mknod.dev   = dev;
-        local->umask = umask;
-        if (params)
-                local->xdata_req = dict_ref (params);
-
-        local->transaction.fop    = afr_mknod_wind;
-        local->transaction.done   = afr_mknod_done;
-        local->transaction.unwind = afr_mknod_unwind;
-
-        ret = afr_build_parent_loc (&local->transaction.parent_loc, loc,
-                                    &op_errno);
-        if (ret)
-                goto out;
-
-        local->transaction.main_frame = frame;
-        local->transaction.basename = AFR_BASENAME (loc->path);
-        int_lock = &local->internal_lock;
-
-        int_lock->lockee_count = 0;
-        ret = afr_init_entry_lockee (&int_lock->lockee[0], local,
-                                     &local->transaction.parent_loc,
-                                     local->transaction.basename,
-                                     priv->child_count);
-        if (ret)
-                goto out;
-
-        int_lock->lockee_count++;
-        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
-        if (ret < 0) {
-            op_errno = -ret;
-            goto out;
-        }
-
-        ret = 0;
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
+
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
+
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
+
+    loc_copy(&local->loc, loc);
+    local->inode = inode_ref(loc->inode);
+    local->parent = inode_ref(loc->parent);
+
+    local->op = GF_FOP_MKNOD;
+    local->cont.mknod.mode = mode;
+    local->cont.mknod.dev = dev;
+    local->umask = umask;
+
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
+
+    if (!local->xdata_req)
+        goto out;
+
+    local->transaction.wind = afr_mknod_wind;
+    local->transaction.unwind = afr_mknod_unwind;
+
+    ret = afr_build_parent_loc(&local->transaction.parent_loc, loc, &op_errno);
+    if (ret)
+        goto out;
+
+    local->transaction.main_frame = frame;
+    local->transaction.basename = AFR_BASENAME(loc->path);
+    ret = afr_transaction(transaction_frame, this, AFR_ENTRY_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
+
+    return 0;
 out:
-        if (ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (mknod, frame, -1, op_errno,
-                                  NULL, NULL, NULL, NULL, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+    return 0;
 }
 
 /* }}} */
 
 /* {{{ mkdir */
 
-
 int
-afr_mkdir_unwind (call_frame_t *frame, xlator_t *this)
+afr_mkdir_unwind(call_frame_t *frame, xlator_t *this)
 {
-        call_frame_t *main_frame = NULL;
-        afr_local_t  *local = NULL;
-
-        local = frame->local;
+    call_frame_t *main_frame = NULL;
+    afr_local_t *local = NULL;
 
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame) {
-                        main_frame = local->transaction.main_frame;
-                }
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
-
-        if (main_frame) {
-                AFR_STACK_UNWIND (mkdir, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  local->cont.dir_fop.inode,
-                                  &local->cont.dir_fop.buf,
-                                  &local->cont.dir_fop.preparent,
-                                  &local->cont.dir_fop.postparent,
-                                  NULL);
-        }
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
-}
-
-
-int
-afr_mkdir_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, inode_t *inode,
-                    struct iatt *buf, struct iatt *preparent,
-                    struct iatt *postparent, dict_t *xdata)
-{
-        int             call_count      = -1;
-        int             child_index     = -1;
 
-        child_index = (long) cookie;
-
-        LOCK (&frame->lock);
-        {
-                __dir_entry_fop_common_cbk (frame, child_index, this,
-                                            op_ret, op_errno, inode, buf,
-                                            preparent, postparent, NULL, NULL);
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0)
-                afr_dir_fop_done (frame, this);
-
-        return 0;
+    AFR_STACK_UNWIND(mkdir, main_frame, local->op_ret, local->op_errno,
+                     local->inode, &local->cont.dir_fop.buf,
+                     &local->cont.dir_fop.preparent,
+                     &local->cont.dir_fop.postparent, local->xdata_rsp);
+    return 0;
 }
 
-
 int
-afr_mkdir_wind (call_frame_t *frame, xlator_t *this)
+afr_mkdir_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, inode_t *inode,
+                   struct iatt *buf, struct iatt *preparent,
+                   struct iatt *postparent, dict_t *xdata)
 {
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int call_count = -1;
-        int i = 0;
-
-        local = frame->local;
-        priv  = this->private;
-
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
-
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
-
-        local->call_count = call_count;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->transaction.pre_op[i]) {
-                        STACK_WIND_COOKIE (frame, afr_mkdir_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->mkdir,
-                                           &local->loc, local->cont.mkdir.mode,
-                                           local->umask,
-                                           local->xdata_req);
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-        return 0;
+    return __afr_dir_write_cbk(frame, cookie, this, op_ret, op_errno, buf,
+                               preparent, postparent, NULL, NULL, xdata);
 }
 
-
 int
-afr_mkdir_done (call_frame_t *frame, xlator_t *this)
+afr_mkdir_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_local_t * local = NULL;
-
-        local = frame->local;
-
-        local->transaction.unwind (frame, this);
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        AFR_STACK_DESTROY (frame);
+    local = frame->local;
+    priv = this->private;
 
-        return 0;
+    STACK_WIND_COOKIE(frame, afr_mkdir_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->mkdir, &local->loc,
+                      local->cont.mkdir.mode, local->umask, local->xdata_req);
+    return 0;
 }
 
 int
-afr_mkdir (call_frame_t *frame, xlator_t *this,
-           loc_t *loc, mode_t mode, mode_t umask, dict_t *params)
+afr_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+          mode_t umask, dict_t *xdata)
 {
-        afr_private_t           *priv                   = NULL;
-        afr_local_t             *local                  = NULL;
-        afr_internal_lock_t     *int_lock               = NULL;
-        call_frame_t            *transaction_frame      = NULL;
-        int                     ret                     = -1;
-        int                     op_errno                = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
-
-        priv = this->private;
-
-        QUORUM_CHECK(mkdir,out);
-
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                op_errno = ENOMEM;
-                goto out;
-        }
-
-        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
-        local = transaction_frame->local;
-
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
-
-        loc_copy (&local->loc, loc);
-
-        LOCK (&priv->read_child_lock);
-        {
-                local->read_child_index = (++priv->read_child_rr)
-                        % (priv->child_count);
-        }
-        UNLOCK (&priv->read_child_lock);
-
-        local->cont.mkdir.mode  = mode;
-        local->umask = umask;
-        if (params)
-                local->xdata_req = dict_ref (params);
-
-        local->op = GF_FOP_MKDIR;
-        local->transaction.fop    = afr_mkdir_wind;
-        local->transaction.done   = afr_mkdir_done;
-        local->transaction.unwind = afr_mkdir_unwind;
-
-        ret = afr_build_parent_loc (&local->transaction.parent_loc, loc,
-                                    &op_errno);
-        if (ret)
-                goto out;
-
-        local->transaction.main_frame = frame;
-        local->transaction.basename = AFR_BASENAME (loc->path);
-        int_lock = &local->internal_lock;
-
-        int_lock->lockee_count = 0;
-        ret = afr_init_entry_lockee (&int_lock->lockee[0], local,
-                                     &local->transaction.parent_loc,
-                                     local->transaction.basename,
-                                     priv->child_count);
-        if (ret)
-                goto out;
-
-        int_lock->lockee_count++;
-        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
-        if (ret < 0) {
-            op_errno = -ret;
-            goto out;
-        }
-
-        ret = 0;
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
+
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
+
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
+
+    loc_copy(&local->loc, loc);
+    local->inode = inode_ref(loc->inode);
+    local->parent = inode_ref(loc->parent);
+
+    local->cont.mkdir.mode = mode;
+    local->umask = umask;
+
+    if (!xdata || !dict_get_sizen(xdata, "gfid-req")) {
+        op_errno = EPERM;
+        gf_msg_callingfn(this->name, GF_LOG_WARNING, op_errno,
+                         AFR_MSG_GFID_NULL,
+                         "mkdir: %s is received "
+                         "without gfid-req %p",
+                         loc->path, xdata);
+        goto out;
+    }
+
+    local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    if (!local->xdata_req) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    local->op = GF_FOP_MKDIR;
+    local->transaction.wind = afr_mkdir_wind;
+    local->transaction.unwind = afr_mkdir_unwind;
+
+    ret = afr_build_parent_loc(&local->transaction.parent_loc, loc, &op_errno);
+    if (ret)
+        goto out;
+
+    local->transaction.main_frame = frame;
+    local->transaction.basename = AFR_BASENAME(loc->path);
+    ret = afr_transaction(transaction_frame, this, AFR_ENTRY_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
+
+    return 0;
 out:
-        if (ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-
-                AFR_STACK_UNWIND (mkdir, frame, -1, op_errno,
-                                  NULL, NULL, NULL, NULL, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+    return 0;
 }
 
 /* }}} */
 
 /* {{{ link */
 
-
 int
-afr_link_unwind (call_frame_t *frame, xlator_t *this)
+afr_link_unwind(call_frame_t *frame, xlator_t *this)
 {
-        call_frame_t *main_frame = NULL;
-        afr_local_t  *local = NULL;
-
-        local = frame->local;
+    call_frame_t *main_frame = NULL;
+    afr_local_t *local = NULL;
 
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame) {
-                        main_frame = local->transaction.main_frame;
-                }
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
-
-        if (main_frame) {
-                AFR_STACK_UNWIND (link, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  local->cont.dir_fop.inode,
-                                  &local->cont.dir_fop.buf,
-                                  &local->cont.dir_fop.preparent,
-                                  &local->cont.dir_fop.postparent,
-                                  NULL);
-        }
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
-}
 
+    AFR_STACK_UNWIND(link, main_frame, local->op_ret, local->op_errno,
+                     local->inode, &local->cont.dir_fop.buf,
+                     &local->cont.dir_fop.preparent,
+                     &local->cont.dir_fop.postparent, local->xdata_rsp);
+    return 0;
+}
 
 int
-afr_link_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, inode_t *inode,
-                   struct iatt *buf, struct iatt *preparent,
-                   struct iatt *postparent, dict_t *xdata)
+afr_link_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, inode_t *inode,
+                  struct iatt *buf, struct iatt *preparent,
+                  struct iatt *postparent, dict_t *xdata)
 {
-        int             call_count      = -1;
-        int             child_index     = -1;
-
-        child_index = (long) cookie;
-
-        LOCK (&frame->lock);
-        {
-                __dir_entry_fop_common_cbk (frame, child_index, this,
-                                            op_ret, op_errno, inode, buf,
-                                            preparent, postparent, NULL, NULL);
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0)
-                afr_dir_fop_done (frame, this);
-
-        return 0;
+    return __afr_dir_write_cbk(frame, cookie, this, op_ret, op_errno, buf,
+                               preparent, postparent, NULL, NULL, xdata);
 }
 
-
 int
-afr_link_wind (call_frame_t *frame, xlator_t *this)
+afr_link_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int call_count = -1;
-        int i = 0;
-
-        local = frame->local;
-        priv  = this->private;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
-
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
-
-        local->call_count = call_count;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->transaction.pre_op[i]) {
-                        STACK_WIND_COOKIE (frame, afr_link_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->link,
-                                           &local->loc,
-                                           &local->newloc, local->xdata_req);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
+    local = frame->local;
+    priv = this->private;
 
-        return 0;
+    STACK_WIND_COOKIE(frame, afr_link_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->link, &local->loc,
+                      &local->newloc, local->xdata_req);
+    return 0;
 }
 
-
 int
-afr_link_done (call_frame_t *frame, xlator_t *this)
+afr_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+         dict_t *xdata)
 {
-        afr_local_t * local = frame->local;
-
-        local->transaction.unwind (frame, this);
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        AFR_STACK_DESTROY (frame);
-
-        return 0;
-}
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-int
-afr_link (call_frame_t *frame, xlator_t *this,
-          loc_t *oldloc, loc_t *newloc, dict_t *xdata)
-{
-        afr_private_t           *priv                   = NULL;
-        afr_local_t             *local                  = NULL;
-        afr_internal_lock_t     *int_lock               = NULL;
-        call_frame_t            *transaction_frame      = NULL;
-        int                     ret                     = -1;
-        int                     op_errno                = 0;
+    loc_copy(&local->loc, oldloc);
+    loc_copy(&local->newloc, newloc);
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    local->inode = inode_ref(oldloc->inode);
+    local->parent = inode_ref(newloc->parent);
 
-        priv = this->private;
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-        QUORUM_CHECK(link,out);
+    if (!local->xdata_req)
+        goto out;
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+    local->op = GF_FOP_LINK;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
-        local = transaction_frame->local;
+    local->transaction.wind = afr_link_wind;
+    local->transaction.unwind = afr_link_unwind;
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    ret = afr_build_parent_loc(&local->transaction.parent_loc, newloc,
+                               &op_errno);
+    if (ret)
+        goto out;
 
-        loc_copy (&local->loc,    oldloc);
-        loc_copy (&local->newloc, newloc);
-        if (xdata)
-                local->xdata_req = dict_ref (xdata);
+    local->transaction.main_frame = frame;
+    local->transaction.basename = AFR_BASENAME(newloc->path);
+    ret = afr_transaction(transaction_frame, this, AFR_ENTRY_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        LOCK (&priv->read_child_lock);
-        {
-                local->read_child_index = (++priv->read_child_rr)
-                        % (priv->child_count);
-        }
-        UNLOCK (&priv->read_child_lock);
-
-        local->op = GF_FOP_LINK;
-        local->transaction.fop    = afr_link_wind;
-        local->transaction.done   = afr_link_done;
-        local->transaction.unwind = afr_link_unwind;
-
-        ret = afr_build_parent_loc (&local->transaction.parent_loc, newloc,
-                                    &op_errno);
-        if (ret)
-                goto out;
-
-        local->transaction.main_frame   = frame;
-        local->transaction.basename     = AFR_BASENAME (newloc->path);
-        int_lock = &local->internal_lock;
-
-        int_lock->lockee_count = 0;
-        ret = afr_init_entry_lockee (&int_lock->lockee[0], local,
-                                     &local->transaction.parent_loc,
-                                     local->transaction.basename,
-                                     priv->child_count);
-        if (ret)
-                goto out;
-
-        int_lock->lockee_count++;
-        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
-        if (ret < 0) {
-            op_errno = -ret;
-            goto out;
-        }
-        ret = 0;
+    return 0;
 out:
-        if (ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (link, frame, -1, op_errno,
-                                  NULL, NULL, NULL, NULL, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(link, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+    return 0;
 }
 
 /* }}} */
 
 /* {{{ symlink */
 
-
 int
-afr_symlink_unwind (call_frame_t *frame, xlator_t *this)
+afr_symlink_unwind(call_frame_t *frame, xlator_t *this)
 {
-        call_frame_t *main_frame = NULL;
-        afr_local_t  *local = NULL;
-
-        local = frame->local;
+    call_frame_t *main_frame = NULL;
+    afr_local_t *local = NULL;
 
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame) {
-                        main_frame = local->transaction.main_frame;
-                }
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
-
-        if (main_frame) {
-                AFR_STACK_UNWIND (symlink, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  local->cont.dir_fop.inode,
-                                  &local->cont.dir_fop.buf,
-                                  &local->cont.dir_fop.preparent,
-                                  &local->cont.dir_fop.postparent,
-                                  NULL);
-        }
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
-}
-
-
-int
-afr_symlink_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno, inode_t *inode,
-                      struct iatt *buf, struct iatt *preparent,
-                      struct iatt *postparent, dict_t *xdata)
-{
-        int             call_count      = -1;
-        int             child_index     = -1;
-
-        child_index = (long) cookie;
-
-        LOCK (&frame->lock);
-        {
-                __dir_entry_fop_common_cbk (frame, child_index, this,
-                                            op_ret, op_errno, inode, buf,
-                                            preparent, postparent, NULL, NULL);
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0)
-                afr_dir_fop_done (frame, this);
 
-        return 0;
+    AFR_STACK_UNWIND(symlink, main_frame, local->op_ret, local->op_errno,
+                     local->inode, &local->cont.dir_fop.buf,
+                     &local->cont.dir_fop.preparent,
+                     &local->cont.dir_fop.postparent, local->xdata_rsp);
+    return 0;
 }
 
-
 int
-afr_symlink_wind (call_frame_t *frame, xlator_t *this)
+afr_symlink_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, inode_t *inode,
+                     struct iatt *buf, struct iatt *preparent,
+                     struct iatt *postparent, dict_t *xdata)
 {
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int call_count = -1;
-        int i = 0;
-
-        local = frame->local;
-        priv = this->private;
-
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
-
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
-
-        local->call_count = call_count;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->transaction.pre_op[i]) {
-                        STACK_WIND_COOKIE (frame, afr_symlink_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->symlink,
-                                           local->cont.symlink.linkpath,
-                                           &local->loc,
-                                           local->umask,
-                                           local->xdata_req);
-
-                        if (!--call_count)
-                                break;
-
-                }
-        }
-
-        return 0;
+    return __afr_dir_write_cbk(frame, cookie, this, op_ret, op_errno, buf,
+                               preparent, postparent, NULL, NULL, xdata);
 }
 
-
 int
-afr_symlink_done (call_frame_t *frame, xlator_t *this)
+afr_symlink_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_local_t * local = frame->local;
-
-        local->transaction.unwind (frame, this);
-
-        AFR_STACK_DESTROY (frame);
-
-        return 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    STACK_WIND_COOKIE(frame, afr_symlink_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->symlink,
+                      local->cont.symlink.linkpath, &local->loc, local->umask,
+                      local->xdata_req);
+    return 0;
 }
 
-
 int
-afr_symlink (call_frame_t *frame, xlator_t *this,
-             const char *linkpath, loc_t *loc, mode_t umask, dict_t *params)
+afr_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath,
+            loc_t *loc, mode_t umask, dict_t *xdata)
 {
-        afr_private_t           *priv                   = NULL;
-        afr_local_t             *local                  = NULL;
-        afr_internal_lock_t     *int_lock               = NULL;
-        call_frame_t            *transaction_frame      = NULL;
-        int                     ret                     = -1;
-        int                     op_errno                = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
-
-        priv = this->private;
-
-        QUORUM_CHECK(symlink,out);
-
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                op_errno = ENOMEM;
-                goto out;
-        }
-
-        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
-        local = transaction_frame->local;
-
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
-
-        loc_copy (&local->loc, loc);
-
-        LOCK (&priv->read_child_lock);
-        {
-                local->read_child_index = (++priv->read_child_rr)
-                        % (priv->child_count);
-        }
-        UNLOCK (&priv->read_child_lock);
-
-        local->cont.symlink.linkpath = gf_strdup (linkpath);
-        local->umask = umask;
-        if (params)
-                local->xdata_req = dict_ref (params);
-
-        local->op = GF_FOP_SYMLINK;
-        local->transaction.fop    = afr_symlink_wind;
-        local->transaction.done   = afr_symlink_done;
-        local->transaction.unwind = afr_symlink_unwind;
-
-        ret = afr_build_parent_loc (&local->transaction.parent_loc, loc,
-                                    &op_errno);
-        if (ret)
-                goto out;
-
-        local->transaction.main_frame   = frame;
-        local->transaction.basename     = AFR_BASENAME (loc->path);
-        int_lock = &local->internal_lock;
-
-        int_lock->lockee_count = 0;
-        ret = afr_init_entry_lockee (&int_lock->lockee[0], local,
-                                     &local->transaction.parent_loc,
-                                     local->transaction.basename,
-                                     priv->child_count);
-        if (ret)
-                goto out;
-
-        int_lock->lockee_count++;
-        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
-        if (ret < 0) {
-            op_errno = -ret;
-            goto out;
-        }
-
-        ret = 0;
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
+
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
+
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
+
+    loc_copy(&local->loc, loc);
+    local->inode = inode_ref(loc->inode);
+    local->parent = inode_ref(loc->parent);
+
+    local->cont.symlink.linkpath = gf_strdup(linkpath);
+    local->umask = umask;
+
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
+
+    if (!local->xdata_req)
+        goto out;
+
+    local->op = GF_FOP_SYMLINK;
+    local->transaction.wind = afr_symlink_wind;
+    local->transaction.unwind = afr_symlink_unwind;
+
+    ret = afr_build_parent_loc(&local->transaction.parent_loc, loc, &op_errno);
+    if (ret)
+        goto out;
+
+    local->transaction.main_frame = frame;
+    local->transaction.basename = AFR_BASENAME(loc->path);
+    ret = afr_transaction(transaction_frame, this, AFR_ENTRY_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
+
+    return 0;
 out:
-        if (ret  < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (symlink, frame, -1, op_errno,
-                                  NULL, NULL, NULL, NULL, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(symlink, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                     NULL);
+    return 0;
 }
 
 /* }}} */
@@ -1324,233 +939,118 @@ out:
 /* {{{ rename */
 
 int
-afr_rename_unwind (call_frame_t *frame, xlator_t *this)
+afr_rename_unwind(call_frame_t *frame, xlator_t *this)
 {
-        call_frame_t *main_frame = NULL;
-        afr_local_t  *local = NULL;
-
-        local = frame->local;
+    call_frame_t *main_frame = NULL;
+    afr_local_t *local = NULL;
 
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame) {
-                        main_frame = local->transaction.main_frame;
-                }
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
-
-        if (main_frame) {
-                AFR_STACK_UNWIND (rename, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  &local->cont.dir_fop.buf,
-                                  &local->cont.dir_fop.preparent,
-                                  &local->cont.dir_fop.postparent,
-                                  &local->cont.dir_fop.prenewparent,
-                                  &local->cont.dir_fop.postnewparent,
-                                  NULL);
-        }
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
-}
-
-
-int
-afr_rename_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, struct iatt *buf,
-                     struct iatt *preoldparent, struct iatt *postoldparent,
-                     struct iatt *prenewparent, struct iatt *postnewparent,
-                     dict_t *xdata)
-{
-        afr_local_t *   local = NULL;
-        int call_count = -1;
-        int child_index = -1;
-
-        local = frame->local;
-
-        child_index = (long) cookie;
 
-        LOCK (&frame->lock);
-        {
-                if (afr_fop_failed (op_ret, op_errno) && op_errno != ENOTEMPTY)
-                        afr_transaction_fop_failed (frame, this, child_index);
-                local->op_errno = op_errno;
-                local->child_errno[child_index] = op_errno;
-
-                if (op_ret > -1)
-                        __dir_entry_fop_common_cbk (frame, child_index, this,
-                                                   op_ret, op_errno, NULL, buf,
-                                                   preoldparent, postoldparent,
-                                                   prenewparent, postnewparent);
-
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0)
-                afr_dir_fop_done (frame, this);
-
-        return 0;
+    AFR_STACK_UNWIND(rename, main_frame, local->op_ret, local->op_errno,
+                     &local->cont.dir_fop.buf, &local->cont.dir_fop.preparent,
+                     &local->cont.dir_fop.postparent,
+                     &local->cont.dir_fop.prenewparent,
+                     &local->cont.dir_fop.postnewparent, local->xdata_rsp);
+    return 0;
 }
 
-
-int32_t
-afr_rename_wind (call_frame_t *frame, xlator_t *this)
+int
+afr_rename_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                    struct iatt *preoldparent, struct iatt *postoldparent,
+                    struct iatt *prenewparent, struct iatt *postnewparent,
+                    dict_t *xdata)
 {
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int call_count = -1;
-        int i = 0;
-
-        local = frame->local;
-        priv = this->private;
-
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
-
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
-
-        local->call_count = call_count;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->transaction.pre_op[i]) {
-                        STACK_WIND_COOKIE (frame, afr_rename_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->rename,
-                                           &local->loc,
-                                           &local->newloc, NULL);
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-        return 0;
+    return __afr_dir_write_cbk(frame, cookie, this, op_ret, op_errno, buf,
+                               preoldparent, postoldparent, prenewparent,
+                               postnewparent, xdata);
 }
 
-
 int
-afr_rename_done (call_frame_t *frame, xlator_t *this)
+afr_rename_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_local_t * local = frame->local;
-
-        local->transaction.unwind (frame, this);
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        AFR_STACK_DESTROY (frame);
+    local = frame->local;
+    priv = this->private;
 
-        return 0;
+    STACK_WIND_COOKIE(frame, afr_rename_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->rename, &local->loc,
+                      &local->newloc, local->xdata_req);
+    return 0;
 }
 
-
 int
-afr_rename (call_frame_t *frame, xlator_t *this,
-            loc_t *oldloc, loc_t *newloc, dict_t *xdata)
+afr_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+           dict_t *xdata)
 {
-        afr_private_t           *priv                   = NULL;
-        afr_local_t             *local                  = NULL;
-        afr_internal_lock_t     *int_lock               = NULL;
-        call_frame_t            *transaction_frame      = NULL;
-        int                     ret                     = -1;
-        int                     op_errno                = 0;
-        int                     nlockee                 = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
-
-        priv = this->private;
-
-        QUORUM_CHECK(rename,out);
-
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                op_errno = ENOMEM;
-                goto out;
-        }
-
-        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
-        local = transaction_frame->local;
-
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
-
-        loc_copy (&local->loc,    oldloc);
-        loc_copy (&local->newloc, newloc);
-
-        local->read_child_index = afr_inode_get_read_ctx (this, oldloc->inode, NULL);
-
-        local->op = GF_FOP_RENAME;
-        local->transaction.fop    = afr_rename_wind;
-        local->transaction.done   = afr_rename_done;
-        local->transaction.unwind = afr_rename_unwind;
-
-        ret = afr_build_parent_loc (&local->transaction.parent_loc, oldloc,
-                                    &op_errno);
-        if (ret)
-                goto out;
-        ret = afr_build_parent_loc (&local->transaction.new_parent_loc, newloc,
-                                    &op_errno);
-        if (ret)
-                goto out;
-
-        local->transaction.main_frame   = frame;
-        local->transaction.basename     = AFR_BASENAME (oldloc->path);
-        local->transaction.new_basename = AFR_BASENAME (newloc->path);
-        int_lock = &local->internal_lock;
-
-        int_lock->lockee_count = nlockee = 0;
-        ret = afr_init_entry_lockee (&int_lock->lockee[nlockee], local,
-                                     &local->transaction.new_parent_loc,
-                                     local->transaction.new_basename,
-                                     priv->child_count);
-        if (ret)
-                goto out;
-
-        nlockee++;
-        ret = afr_init_entry_lockee (&int_lock->lockee[nlockee], local,
-                                     &local->transaction.parent_loc,
-                                     local->transaction.basename,
-                                     priv->child_count);
-        if (ret)
-                goto out;
-
-        nlockee++;
-        if (local->newloc.inode && IA_ISDIR (local->newloc.inode->ia_type)) {
-                ret = afr_init_entry_lockee (&int_lock->lockee[nlockee], local,
-                                             &local->newloc,
-                                             NULL,
-                                             priv->child_count);
-                if (ret)
-                        goto out;
-
-                nlockee++;
-        }
-        qsort (int_lock->lockee, nlockee, sizeof (*int_lock->lockee),
-               afr_entry_lockee_cmp);
-        int_lock->lockee_count = nlockee;
-
-        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_RENAME_TRANSACTION);
-        if (ret < 0) {
-            op_errno = -ret;
-            goto out;
-        }
-
-        ret = 0;
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
+
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
+
+    loc_copy(&local->loc, oldloc);
+    loc_copy(&local->newloc, newloc);
+
+    local->inode = inode_ref(oldloc->inode);
+    local->parent = inode_ref(oldloc->parent);
+    local->parent2 = inode_ref(newloc->parent);
+
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
+
+    if (!local->xdata_req)
+        goto out;
+
+    local->op = GF_FOP_RENAME;
+    local->transaction.wind = afr_rename_wind;
+    local->transaction.unwind = afr_rename_unwind;
+
+    ret = afr_build_parent_loc(&local->transaction.parent_loc, oldloc,
+                               &op_errno);
+    if (ret)
+        goto out;
+    ret = afr_build_parent_loc(&local->transaction.new_parent_loc, newloc,
+                               &op_errno);
+    if (ret)
+        goto out;
+
+    local->transaction.main_frame = frame;
+    local->transaction.basename = AFR_BASENAME(oldloc->path);
+    local->transaction.new_basename = AFR_BASENAME(newloc->path);
+    ret = afr_transaction(transaction_frame, this,
+                          AFR_ENTRY_RENAME_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
+
+    return 0;
 out:
-        if (ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-
-                AFR_STACK_UNWIND (rename, frame, -1, op_errno,
-                                  NULL, NULL, NULL, NULL, NULL, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(rename, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL,
+                     NULL);
+    return 0;
 }
 
 /* }}} */
@@ -1558,405 +1058,205 @@ out:
 /* {{{ unlink */
 
 int
-afr_unlink_unwind (call_frame_t *frame, xlator_t *this)
+afr_unlink_unwind(call_frame_t *frame, xlator_t *this)
 {
-        call_frame_t *main_frame = NULL;
-        afr_local_t  *local = NULL;
-
-        local = frame->local;
+    call_frame_t *main_frame = NULL;
+    afr_local_t *local = NULL;
 
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame) {
-                        main_frame = local->transaction.main_frame;
-                }
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
-
-        if (main_frame) {
-                AFR_STACK_UNWIND (unlink, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  &local->cont.dir_fop.preparent,
-                                  &local->cont.dir_fop.postparent,
-                                  NULL);
-        }
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
-}
-
 
-int
-afr_unlink_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, struct iatt *preparent,
-                     struct iatt *postparent, dict_t *xdata)
-{
-        afr_local_t *   local = NULL;
-        int call_count  = -1;
-        int child_index = (long) cookie;
-
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                if (child_index == local->read_child_index) {
-                        local->read_child_returned = _gf_true;
-                }
-                __dir_entry_fop_common_cbk (frame, child_index, this,
-                                            op_ret, op_errno, NULL, NULL,
-                                            preparent, postparent, NULL, NULL);
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-        if (call_count == 0)
-                afr_dir_fop_done (frame, this);
-
-        return 0;
+    AFR_STACK_UNWIND(unlink, main_frame, local->op_ret, local->op_errno,
+                     &local->cont.dir_fop.preparent,
+                     &local->cont.dir_fop.postparent, local->xdata_rsp);
+    return 0;
 }
 
-
-int32_t
-afr_unlink_wind (call_frame_t *frame, xlator_t *this)
+int
+afr_unlink_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                    struct iatt *postparent, dict_t *xdata)
 {
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int call_count = -1;
-        int i = 0;
-
-        local = frame->local;
-        priv  = this->private;
-
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
-
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
-
-        local->call_count = call_count;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->transaction.pre_op[i]) {
-                        STACK_WIND_COOKIE (frame, afr_unlink_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->unlink,
-                                           &local->loc, local->xflag,
-                                           local->xdata_req);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-        return 0;
+    return __afr_dir_write_cbk(frame, cookie, this, op_ret, op_errno, NULL,
+                               preparent, postparent, NULL, NULL, xdata);
 }
 
-
-int32_t
-afr_unlink_done (call_frame_t *frame, xlator_t *this)
+int
+afr_unlink_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_local_t * local = frame->local;
-
-        local->transaction.unwind (frame, this);
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        AFR_STACK_DESTROY (frame);
+    local = frame->local;
+    priv = this->private;
 
-        return 0;
+    STACK_WIND_COOKIE(frame, afr_unlink_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->unlink, &local->loc,
+                      local->xflag, local->xdata_req);
+    return 0;
 }
 
-
-int32_t
-afr_unlink (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, int xflag, dict_t *xdata)
+int
+afr_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+           dict_t *xdata)
 {
-        afr_private_t           *priv                   = NULL;
-        afr_local_t             *local                  = NULL;
-        afr_internal_lock_t     *int_lock               = NULL;
-        call_frame_t            *transaction_frame      = NULL;
-        int                     ret                     = -1;
-        int                     op_errno                = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
-
-        priv = this->private;
-
-        QUORUM_CHECK(unlink,out);
-
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                op_errno = ENOMEM;
-                goto out;
-        }
-
-        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
-        local = transaction_frame->local;
-
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
-
-        loc_copy (&local->loc, loc);
-        local->xflag = xflag;
-        if (xdata)
-                local->xdata_req = dict_ref (xdata);
-
-        local->op = GF_FOP_UNLINK;
-        local->transaction.fop    = afr_unlink_wind;
-        local->transaction.done   = afr_unlink_done;
-        local->transaction.unwind = afr_unlink_unwind;
-
-        ret = afr_build_parent_loc (&local->transaction.parent_loc, loc,
-                                    &op_errno);
-        if (ret)
-                goto out;
-
-        local->transaction.main_frame = frame;
-        local->transaction.basename = AFR_BASENAME (loc->path);
-        int_lock = &local->internal_lock;
-
-        int_lock->lockee_count = 0;
-        ret = afr_init_entry_lockee (&int_lock->lockee[0], local,
-                                     &local->transaction.parent_loc,
-                                     local->transaction.basename,
-                                     priv->child_count);
-        if (ret)
-                goto out;
-
-        int_lock->lockee_count++;
-        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
-        if (ret < 0) {
-            op_errno = -ret;
-            goto out;
-        }
-
-        ret = 0;
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
+
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
+
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
+
+    loc_copy(&local->loc, loc);
+    local->xflag = xflag;
+
+    local->inode = inode_ref(loc->inode);
+    local->parent = inode_ref(loc->parent);
+
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
+
+    if (!local->xdata_req)
+        goto out;
+
+    local->op = GF_FOP_UNLINK;
+    local->transaction.wind = afr_unlink_wind;
+    local->transaction.unwind = afr_unlink_unwind;
+
+    ret = afr_build_parent_loc(&local->transaction.parent_loc, loc, &op_errno);
+    if (ret)
+        goto out;
+
+    local->transaction.main_frame = frame;
+    local->transaction.basename = AFR_BASENAME(loc->path);
+    ret = afr_transaction(transaction_frame, this, AFR_ENTRY_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
+
+    return 0;
 out:
-        if (ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (unlink, frame, -1, op_errno,
-                                  NULL, NULL, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(unlink, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
 /* }}} */
 
 /* {{{ rmdir */
 
-
-
 int
-afr_rmdir_unwind (call_frame_t *frame, xlator_t *this)
+afr_rmdir_unwind(call_frame_t *frame, xlator_t *this)
 {
-        call_frame_t *main_frame = NULL;
-        afr_local_t  *local = NULL;
+    call_frame_t *main_frame = NULL;
+    afr_local_t *local = NULL;
 
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame) {
-                        main_frame = local->transaction.main_frame;
-                }
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
-
-        if (main_frame) {
-                AFR_STACK_UNWIND (rmdir, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  &local->cont.dir_fop.preparent,
-                                  &local->cont.dir_fop.postparent,
-                                  NULL);
-        }
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
-}
-
-
-int
-afr_rmdir_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, struct iatt *preparent,
-                    struct iatt *postparent, dict_t *xdata)
-{
-        afr_local_t *   local = NULL;
-        int call_count  = -1;
-        int child_index = (long) cookie;
-        int read_child  = 0;
-
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
-                if (afr_fop_failed (op_ret, op_errno) && (op_errno != ENOTEMPTY))
-                        afr_transaction_fop_failed (frame, this, child_index);
-                local->op_errno = op_errno;
-                local->child_errno[child_index] = op_errno;
-                if (op_ret > -1)
-                        __dir_entry_fop_common_cbk (frame, child_index, this,
-                                                   op_ret, op_errno, NULL, NULL,
-                                                   preparent, postparent, NULL,
-                                                   NULL);
-
-        }
-        UNLOCK (&frame->lock);
 
-        call_count = afr_frame_return (frame);
-        if (call_count == 0)
-                afr_dir_fop_done (frame, this);
-
-        return 0;
+    AFR_STACK_UNWIND(rmdir, main_frame, local->op_ret, local->op_errno,
+                     &local->cont.dir_fop.preparent,
+                     &local->cont.dir_fop.postparent, local->xdata_rsp);
+    return 0;
 }
 
-
 int
-afr_rmdir_wind (call_frame_t *frame, xlator_t *this)
+afr_rmdir_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                   struct iatt *postparent, dict_t *xdata)
 {
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int call_count = -1;
-        int i = 0;
-
-        local = frame->local;
-        priv  = this->private;
-
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
-
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
-
-        local->call_count = call_count;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->transaction.pre_op[i]) {
-                        STACK_WIND_COOKIE (frame, afr_rmdir_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->rmdir,
-                                           &local->loc, local->cont.rmdir.flags,
-                                           NULL);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-        return 0;
+    return __afr_dir_write_cbk(frame, cookie, this, op_ret, op_errno, NULL,
+                               preparent, postparent, NULL, NULL, xdata);
 }
 
-
 int
-afr_rmdir_done (call_frame_t *frame, xlator_t *this)
+afr_rmdir_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_local_t * local = frame->local;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        local->transaction.unwind (frame, this);
+    local = frame->local;
+    priv = this->private;
 
-        AFR_STACK_DESTROY (frame);
-
-        return 0;
+    STACK_WIND_COOKIE(frame, afr_rmdir_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->rmdir, &local->loc,
+                      local->cont.rmdir.flags, local->xdata_req);
+    return 0;
 }
 
-
 int
-afr_rmdir (call_frame_t *frame, xlator_t *this,
-           loc_t *loc, int flags, dict_t *xdata)
+afr_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+          dict_t *xdata)
 {
-        afr_private_t           *priv                   = NULL;
-        afr_local_t             *local                  = NULL;
-        afr_internal_lock_t     *int_lock               = NULL;
-        call_frame_t            *transaction_frame      = NULL;
-        int                     ret                     = -1;
-        int                     op_errno                = 0;
-        int                     nlockee                 = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
-
-        priv = this->private;
-
-        QUORUM_CHECK(rmdir,out);
-
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                op_errno = ENOMEM;
-                goto out;
-        }
-
-        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
-        local = transaction_frame->local;
-
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
-
-        local->cont.rmdir.flags = flags;
-        loc_copy (&local->loc, loc);
-
-        local->op = GF_FOP_RMDIR;
-        local->transaction.fop    = afr_rmdir_wind;
-        local->transaction.done   = afr_rmdir_done;
-        local->transaction.unwind = afr_rmdir_unwind;
-
-        ret = afr_build_parent_loc (&local->transaction.parent_loc, loc,
-                                    &op_errno);
-        if (ret)
-                goto out;
-
-        local->transaction.main_frame = frame;
-        local->transaction.basename = AFR_BASENAME (loc->path);
-        int_lock = &local->internal_lock;
-
-        int_lock->lockee_count = nlockee = 0;
-        ret = afr_init_entry_lockee (&int_lock->lockee[nlockee], local,
-                                     &local->transaction.parent_loc,
-                                     local->transaction.basename,
-                                     priv->child_count);
-        if (ret)
-                goto out;
-
-        nlockee++;
-        ret = afr_init_entry_lockee (&int_lock->lockee[nlockee], local,
-                                     &local->loc,
-                                     NULL,
-                                     priv->child_count);
-        if (ret)
-                goto out;
-
-        nlockee++;
-        qsort (int_lock->lockee, nlockee, sizeof (*int_lock->lockee),
-               afr_entry_lockee_cmp);
-        int_lock->lockee_count = nlockee;
-
-        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
-        if (ret < 0) {
-            op_errno = -ret;
-            goto out;
-        }
-
-        ret = 0;
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
+
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
+
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
+
+    loc_copy(&local->loc, loc);
+    local->inode = inode_ref(loc->inode);
+    local->parent = inode_ref(loc->parent);
+
+    local->cont.rmdir.flags = flags;
+
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
+
+    if (!local->xdata_req)
+        goto out;
+
+    local->op = GF_FOP_RMDIR;
+    local->transaction.wind = afr_rmdir_wind;
+    local->transaction.unwind = afr_rmdir_unwind;
+
+    ret = afr_build_parent_loc(&local->transaction.parent_loc, loc, &op_errno);
+    if (ret)
+        goto out;
+
+    local->transaction.main_frame = frame;
+    local->transaction.basename = AFR_BASENAME(loc->path);
+    ret = afr_transaction(transaction_frame, this, AFR_ENTRY_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
+
+    return 0;
 out:
-        if (ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (rmdir, frame, -1, op_errno, NULL, NULL, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(rmdir, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
 /* }}} */
diff --git a/xlators/cluster/afr/src/afr-dir-write.h b/xlators/cluster/afr/src/afr-dir-write.h
index 02f0a3682d9..1d88c3b9b26 100644
--- a/xlators/cluster/afr/src/afr-dir-write.h
+++ b/xlators/cluster/afr/src/afr-dir-write.h
@@ -12,36 +12,35 @@
 #define __DIR_WRITE_H__
 
 int32_t
-afr_create (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, int32_t flags, mode_t mode,
-            mode_t umask, fd_t *fd, dict_t *xdata);
+afr_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+           mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata);
 
 int32_t
-afr_mknod (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, mode_t mode, dev_t dev, mode_t umask, dict_t *xdata);
+afr_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+          dev_t dev, mode_t umask, dict_t *xdata);
 
 int32_t
-afr_mkdir (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, mode_t mode, mode_t umask, dict_t *xdata);
+afr_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+          mode_t umask, dict_t *xdata);
 
 int32_t
-afr_unlink (call_frame_t *frame, xlator_t *this,
-	    loc_t *loc, int xflag, dict_t *xdata);
+afr_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+           dict_t *xdata);
 
 int32_t
-afr_rmdir (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, int flags, dict_t *xdata);
+afr_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+          dict_t *xdata);
 
 int32_t
-afr_link (call_frame_t *frame, xlator_t *this,
-	  loc_t *oldloc, loc_t *newloc, dict_t *xdata);
+afr_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+         dict_t *xdata);
 
 int32_t
-afr_rename (call_frame_t *frame, xlator_t *this,
-	    loc_t *oldloc, loc_t *newloc, dict_t *xdata);
+afr_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+           dict_t *xdata);
 
 int
-afr_symlink (call_frame_t *frame, xlator_t *this,
-	     const char *linkpath, loc_t *oldloc, mode_t umask, dict_t *params);
+afr_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath,
+            loc_t *oldloc, mode_t umask, dict_t *params);
 
 #endif /* __DIR_WRITE_H__ */
diff --git a/xlators/cluster/afr/src/afr-inode-read.c b/xlators/cluster/afr/src/afr-inode-read.c
index 00dd22214cb..c5521704de2 100644
--- a/xlators/cluster/afr/src/afr-inode-read.c
+++ b/xlators/cluster/afr/src/afr-inode-read.c
@@ -8,7 +8,6 @@
   cases as published by the Free Software Foundation.
 */
 
-
 #include <libgen.h>
 #include <unistd.h>
 #include <fnmatch.h>
@@ -16,1925 +15,1880 @@
 #include <stdlib.h>
 #include <signal.h>
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
+#include <glusterfs/glusterfs.h>
 #include "afr.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
+#include <glusterfs/dict.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/list.h>
+#include <glusterfs/byte-order.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/quota-common-utils.h>
+
+#include "afr-transaction.h"
+#include "afr-messages.h"
 
-/**
- * Common algorithm for inode read calls:
- *
- * - Try the fop on the first child that is up
- * - if we have failed due to ENOTCONN:
- *     try the next child
- *
- * Applicable to: access, stat, fstat, readlink, getxattr
- */
+/*
+ * Quota size xattrs are not maintained by afr. There is a
+ * possibility that they differ even when both the directory changelog xattrs
+ * suggest everything is fine. So if there is at least one 'source' check among
+ * the sources which has the maximum quota size. Otherwise check among all the
+ * available ones for maximum quota size. This way if there is a source and
+ * stale copies it always votes for the 'source'.
+ * */
+
+int
+afr_handle_quota_size(call_frame_t *frame, xlator_t *this)
+{
+    unsigned char *readable = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    struct afr_reply *replies = NULL;
+    int i = 0;
+    int ret = 0;
+    quota_meta_t size = {
+        0,
+    };
+    quota_meta_t max_size = {
+        0,
+    };
+    int readable_cnt = 0;
+    int read_subvol = -1;
+
+    local = frame->local;
+    priv = this->private;
+    replies = local->replies;
+
+    readable = alloca0(priv->child_count);
+
+    afr_inode_read_subvol_get(local->inode, this, readable, 0, 0);
+
+    readable_cnt = AFR_COUNT(readable, priv->child_count);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret == -1)
+            continue;
+        if (readable_cnt && !readable[i])
+            continue;
+        if (!replies[i].xdata)
+            continue;
+        ret = quota_dict_get_meta(replies[i].xdata, QUOTA_SIZE_KEY,
+                                  SLEN(QUOTA_SIZE_KEY), &size);
+        if (ret == -1)
+            continue;
+        if (read_subvol == -1)
+            read_subvol = i;
+        if (size.size > max_size.size ||
+            (size.file_count + size.dir_count) >
+                (max_size.file_count + max_size.dir_count))
+            read_subvol = i;
+
+        if (size.size > max_size.size)
+            max_size.size = size.size;
+        if (size.file_count > max_size.file_count)
+            max_size.file_count = size.file_count;
+        if (size.dir_count > max_size.dir_count)
+            max_size.dir_count = size.dir_count;
+    }
+
+    if (max_size.size == 0 && max_size.file_count == 0 &&
+        max_size.dir_count == 0)
+        return read_subvol;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret == -1)
+            continue;
+        if (readable_cnt && !readable[i])
+            continue;
+        if (!replies[i].xdata)
+            continue;
+        quota_dict_set_meta(replies[i].xdata, QUOTA_SIZE_KEY, &max_size,
+                            IA_IFDIR);
+    }
+
+    return read_subvol;
+}
 
 /* {{{ access */
 
-int32_t
-afr_access_cbk (call_frame_t *frame, void *cookie,
-                xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata)
+int
+afr_access_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, dict_t *xdata)
 {
-        afr_private_t * priv            = NULL;
-        afr_local_t *   local           = NULL;
-        xlator_t **     children        = NULL;
-        int             unwind          = 1;
-        int32_t         *last_index     = NULL;
-        int32_t         next_call_child = -1;
-        int32_t         read_child      = -1;
-        int32_t         *fresh_children  = NULL;
-
-        priv     = this->private;
-        children = priv->children;
+    afr_local_t *local = NULL;
 
-        local = frame->local;
+    local = frame->local;
 
-        read_child = (long) cookie;
-
-        if (op_ret == -1) {
-                last_index = &local->cont.access.last_index;
-                fresh_children = local->fresh_children;
-                next_call_child = afr_next_call_child (fresh_children,
-                                                       local->child_up,
-                                                       priv->child_count,
-                                                       last_index, read_child);
-                if (next_call_child < 0)
-                        goto out;
-
-                unwind = 0;
-
-                STACK_WIND_COOKIE (frame, afr_access_cbk,
-                                   (void *) (long) read_child,
-                                   children[next_call_child],
-                                   children[next_call_child]->fops->access,
-                                   &local->loc, local->cont.access.mask,
-                                   NULL);
-        }
-
-out:
-        if (unwind) {
-                AFR_STACK_UNWIND (access, frame, op_ret, op_errno, xdata);
-        }
+    if (op_ret < 0) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
 
+        afr_read_txn_continue(frame, this, (long)cookie);
         return 0;
-}
-
-
-int32_t
-afr_access (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
-            dict_t *xdata)
-{
-        afr_private_t   *priv      = NULL;
-        xlator_t        **children = NULL;
-        int             call_child = 0;
-        afr_local_t     *local     = NULL;
-        int32_t         op_errno   = 0;
-        int32_t         read_child = -1;
-        int             ret        = -1;
+    }
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    AFR_STACK_UNWIND(access, frame, op_ret, op_errno, xdata);
 
-        priv     = this->private;
-        VALIDATE_OR_GOTO (priv->children, out);
+    return 0;
+}
 
-        children = priv->children;
+int
+afr_access_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    priv = this->private;
+    local = frame->local;
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    if (subvol == -1) {
+        AFR_STACK_UNWIND(access, frame, local->op_ret, local->op_errno, 0);
+        return 0;
+    }
 
-        local->fresh_children = afr_children_create (priv->child_count);
-        if (!local->fresh_children) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+    STACK_WIND_COOKIE(frame, afr_access_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->access, &local->loc,
+                      local->cont.access.mask, local->xdata_req);
+    return 0;
+}
 
+int
+afr_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int mask,
+           dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int op_errno = 0;
 
-        read_child = afr_inode_get_read_ctx (this, loc->inode,
-                                             local->fresh_children);
-        ret = afr_get_call_child (this, local->child_up, read_child,
-                                     local->fresh_children,
-                                     &call_child,
-                                     &local->cont.access.last_index);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-        loc_copy (&local->loc, loc);
-        local->cont.access.mask = mask;
+    local->op = GF_FOP_ACCESS;
+    loc_copy(&local->loc, loc);
+    local->cont.access.mask = mask;
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
 
-        STACK_WIND_COOKIE (frame, afr_access_cbk,
-                           (void *) (long) call_child,
-                           children[call_child],
-                           children[call_child]->fops->access,
-                           loc, mask, xdata);
+    afr_read_txn(frame, this, loc->inode, afr_access_wind,
+                 AFR_METADATA_TRANSACTION);
 
-        ret = 0;
+    return 0;
 out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (access, frame, -1, op_errno, NULL);
-        return 0;
-}
+    AFR_STACK_UNWIND(access, frame, -1, op_errno, NULL);
 
+    return 0;
+}
 
 /* }}} */
 
 /* {{{ stat */
 
-int32_t
-afr_stat_cbk (call_frame_t *frame, void *cookie,
-              xlator_t *this, int32_t op_ret, int32_t op_errno,
-              struct iatt *buf, dict_t *xdata)
+int
+afr_stat_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, struct iatt *buf, dict_t *xdata)
 {
-        afr_private_t * priv            = NULL;
-        afr_local_t *   local           = NULL;
-        xlator_t **     children        = NULL;
-        int             unwind          = 1;
-        int32_t         *last_index     = NULL;
-        int32_t         next_call_child = -1;
-        int32_t         read_child      = -1;
-        int32_t         *fresh_children  = NULL;
+    afr_local_t *local = NULL;
 
-        priv     = this->private;
-        children = priv->children;
+    local = frame->local;
 
-        read_child = (long) cookie;
+    if (op_ret < 0) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
 
-        local = frame->local;
-
-        if (op_ret == -1) {
-                last_index = &local->cont.stat.last_index;
-                fresh_children = local->fresh_children;
-                next_call_child = afr_next_call_child (fresh_children,
-                                                       local->child_up,
-                                                       priv->child_count,
-                                                       last_index, read_child);
-                if (next_call_child < 0)
-                        goto out;
-
-                unwind = 0;
-
-                STACK_WIND_COOKIE (frame, afr_stat_cbk,
-                                   (void *) (long) read_child,
-                                   children[next_call_child],
-                                   children[next_call_child]->fops->stat,
-                                   &local->loc, NULL);
-        }
+        afr_read_txn_continue(frame, this, (long)cookie);
+        return 0;
+    }
 
-out:
-        if (unwind) {
-                AFR_STACK_UNWIND (stat, frame, op_ret, op_errno, buf, xdata);
-        }
+    AFR_STACK_UNWIND(stat, frame, op_ret, op_errno, buf, xdata);
 
-        return 0;
+    return 0;
 }
 
-
-int32_t
-afr_stat (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+int
+afr_stat_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_private_t   *priv      = NULL;
-        afr_local_t     *local     = NULL;
-        xlator_t        **children = NULL;
-        int             call_child = 0;
-        int32_t         op_errno   = 0;
-        int32_t         read_child = -1;
-        int             ret        = -1;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    priv = this->private;
+    local = frame->local;
 
-        priv     = this->private;
-        VALIDATE_OR_GOTO (priv->children, out);
-
-        children = priv->children;
+    if (subvol == -1) {
+        AFR_STACK_UNWIND(stat, frame, local->op_ret, local->op_errno, 0, 0);
+        return 0;
+    }
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    STACK_WIND_COOKIE(
+        frame, afr_stat_cbk, (void *)(long)subvol, priv->children[subvol],
+        priv->children[subvol]->fops->stat, &local->loc, local->xdata_req);
+    return 0;
+}
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+int
+afr_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int op_errno = 0;
 
-        local->fresh_children = afr_children_create (priv->child_count);
-        if (!local->fresh_children) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-        read_child = afr_inode_get_read_ctx (this, loc->inode,
-                                             local->fresh_children);
-        ret = afr_get_call_child (this, local->child_up, read_child,
-                                     local->fresh_children,
-                                     &call_child,
-                                     &local->cont.stat.last_index);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
-        loc_copy (&local->loc, loc);
+    local->op = GF_FOP_STAT;
+    loc_copy(&local->loc, loc);
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
 
-        STACK_WIND_COOKIE (frame, afr_stat_cbk, (void *) (long) call_child,
-                           children[call_child],
-                           children[call_child]->fops->stat,
-                           loc, xdata);
+    afr_read_txn(frame, this, loc->inode, afr_stat_wind, AFR_DATA_TRANSACTION);
 
-        ret = 0;
+    return 0;
 out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (stat, frame, -1, op_errno, NULL, NULL);
+    AFR_STACK_UNWIND(stat, frame, -1, op_errno, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
-
 /* }}} */
 
 /* {{{ fstat */
 
-int32_t
-afr_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-               int32_t op_ret, int32_t op_errno, struct iatt *buf,
-               dict_t *xdata)
+int
+afr_fstat_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct iatt *buf, dict_t *xdata)
 {
-        afr_private_t   *priv           = NULL;
-        afr_local_t     *local          = NULL;
-        xlator_t        **children      = NULL;
-        int             unwind          = 1;
-        int32_t         *last_index     = NULL;
-        int32_t         next_call_child = -1;
-        int32_t         read_child      = -1;
-        int32_t         *fresh_children  = NULL;
-
-        priv     = this->private;
-        children = priv->children;
+    afr_local_t *local = NULL;
 
-        local = frame->local;
+    local = frame->local;
 
-        read_child = (long) cookie;
-
-        if (op_ret == -1) {
-                last_index = &local->cont.fstat.last_index;
-                fresh_children = local->fresh_children;
-                next_call_child = afr_next_call_child (fresh_children,
-                                                       local->child_up,
-                                                       priv->child_count,
-                                                       last_index, read_child);
-                if (next_call_child < 0)
-                        goto out;
-
-                unwind = 0;
-
-                STACK_WIND_COOKIE (frame, afr_fstat_cbk,
-                                   (void *) (long) read_child,
-                                   children[next_call_child],
-                                   children[next_call_child]->fops->fstat,
-                                   local->fd, NULL);
-        }
-
-out:
-        if (unwind) {
-                AFR_STACK_UNWIND (fstat, frame, op_ret, op_errno, buf, xdata);
-        }
+    if (op_ret < 0) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
 
+        afr_read_txn_continue(frame, this, (long)cookie);
         return 0;
-}
-
-
-int32_t
-afr_fstat (call_frame_t *frame, xlator_t *this,
-           fd_t *fd, dict_t *xdata)
-{
-        afr_private_t   *priv      = NULL;
-        afr_local_t     *local     = NULL;
-        xlator_t        **children = NULL;
-        int             call_child = 0;
-        int32_t         op_errno   = 0;
-        int32_t         read_child = 0;
-        int             ret        = -1;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (fd, out);
-        VALIDATE_OR_GOTO (this->private, out);
-
-        priv     = this->private;
-        VALIDATE_OR_GOTO (priv->children, out);
+    }
 
-        children = priv->children;
-
-        VALIDATE_OR_GOTO (fd->inode, out);
-
-        if (afr_is_split_brain (this, fd->inode)) {
-                op_errno = EIO;
-                goto out;
-        }
+    AFR_STACK_UNWIND(fstat, frame, op_ret, op_errno, buf, xdata);
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    return 0;
+}
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+int
+afr_fstat_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
 
-        local->fresh_children = afr_children_create (priv->child_count);
-        if (!local->fresh_children) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+    priv = this->private;
+    local = frame->local;
 
-        read_child = afr_inode_get_read_ctx (this, fd->inode,
-                                             local->fresh_children);
+    if (subvol == -1) {
+        AFR_STACK_UNWIND(fstat, frame, local->op_ret, local->op_errno, 0, 0);
+        return 0;
+    }
 
+    STACK_WIND_COOKIE(
+        frame, afr_fstat_cbk, (void *)(long)subvol, priv->children[subvol],
+        priv->children[subvol]->fops->fstat, local->fd, local->xdata_req);
+    return 0;
+}
 
+int32_t
+afr_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int op_errno = 0;
 
-        ret = afr_get_call_child (this, local->child_up, read_child,
-                                     local->fresh_children,
-                                     &call_child,
-                                     &local->cont.fstat.last_index);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-        local->fd = fd_ref (fd);
+    local->op = GF_FOP_FSTAT;
+    local->fd = fd_ref(fd);
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
 
-        afr_open_fd_fix (fd, this);
+    afr_fix_open(fd, this);
 
-        STACK_WIND_COOKIE (frame, afr_fstat_cbk, (void *) (long) call_child,
-                           children[call_child],
-                           children[call_child]->fops->fstat,
-                           fd, xdata);
+    afr_read_txn(frame, this, fd->inode, afr_fstat_wind, AFR_DATA_TRANSACTION);
 
-        ret = 0;
+    return 0;
 out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (fstat, frame, -1, op_errno, NULL, NULL);
+    AFR_STACK_UNWIND(fstat, frame, -1, op_errno, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
 /* }}} */
 
 /* {{{ readlink */
 
-int32_t
-afr_readlink_cbk (call_frame_t *frame, void *cookie,
-                  xlator_t *this, int32_t op_ret, int32_t op_errno,
-                  const char *buf, struct iatt *sbuf, dict_t *xdata)
+int
+afr_readlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, const char *buf,
+                 struct iatt *sbuf, dict_t *xdata)
 {
-        afr_private_t * priv                  = NULL;
-        afr_local_t *   local                 = NULL;
-        xlator_t **     children              = NULL;
-        int             unwind                = 1;
-        int32_t         *last_index           = NULL;
-        int32_t         next_call_child       = -1;
-        int32_t         read_child            = -1;
-        int32_t         *fresh_children        = NULL;
-
-        priv     = this->private;
-        children = priv->children;
-
-        local = frame->local;
+    afr_local_t *local = NULL;
 
-        read_child = (long) cookie;
-
-        if (op_ret == -1) {
-                last_index = &local->cont.readlink.last_index;
-                fresh_children = local->fresh_children;
-                next_call_child = afr_next_call_child (fresh_children,
-                                                       local->child_up,
-                                                       priv->child_count,
-                                                       last_index, read_child);
-                if (next_call_child < 0)
-                        goto out;
-
-                unwind = 0;
-                STACK_WIND_COOKIE (frame, afr_readlink_cbk,
-                                   (void *) (long) read_child,
-                                   children[next_call_child],
-                                   children[next_call_child]->fops->readlink,
-                                   &local->loc,
-                                   local->cont.readlink.size, NULL);
-        }
+    local = frame->local;
 
-out:
-        if (unwind) {
-                AFR_STACK_UNWIND (readlink, frame, op_ret, op_errno, buf, sbuf,
-                                  xdata);
-        }
+    if (op_ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = op_errno;
 
+        afr_read_txn_continue(frame, this, (long)cookie);
         return 0;
-}
+    }
 
+    AFR_STACK_UNWIND(readlink, frame, op_ret, op_errno, buf, sbuf, xdata);
+    return 0;
+}
 
-int32_t
-afr_readlink (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, size_t size, dict_t *xdata)
+int
+afr_readlink_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_private_t   *priv      = NULL;
-        xlator_t        **children = NULL;
-        int             call_child = 0;
-        afr_local_t     *local     = NULL;
-        int32_t         op_errno   = 0;
-        int32_t         read_child = -1;
-        int             ret        = -1;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        priv     = this->private;
-        VALIDATE_OR_GOTO (priv->children, out);
+    local = frame->local;
+    priv = this->private;
 
-        children = priv->children;
-
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    if (subvol == -1) {
+        AFR_STACK_UNWIND(readlink, frame, local->op_ret, local->op_errno, 0, 0,
+                         0);
+        return 0;
+    }
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    STACK_WIND_COOKIE(frame, afr_readlink_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->readlink, &local->loc,
+                      local->cont.readlink.size, local->xdata_req);
+    return 0;
+}
 
-        local->fresh_children = afr_children_create (priv->child_count);
-        if (!local->fresh_children) {
-                op_errno = ENOMEM;
-                goto out;
-        }
-        read_child = afr_inode_get_read_ctx (this, loc->inode,
-                                             local->fresh_children);
-        ret = afr_get_call_child (this, local->child_up, read_child,
-                                     local->fresh_children,
-                                     &call_child,
-                                     &local->cont.readlink.last_index);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+int
+afr_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+             dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int32_t op_errno = 0;
 
-        loc_copy (&local->loc, loc);
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-        local->cont.readlink.size       = size;
+    local->op = GF_FOP_READLINK;
+    loc_copy(&local->loc, loc);
+    local->cont.readlink.size = size;
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
 
-        STACK_WIND_COOKIE (frame, afr_readlink_cbk,
-                           (void *) (long) call_child,
-                           children[call_child],
-                           children[call_child]->fops->readlink,
-                           loc, size, xdata);
+    afr_read_txn(frame, this, loc->inode, afr_readlink_wind,
+                 AFR_DATA_TRANSACTION);
 
-        ret = 0;
+    return 0;
 out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (readlink, frame, -1, op_errno, NULL, NULL, NULL);
-        return 0;
-}
+    AFR_STACK_UNWIND(readlink, frame, -1, op_errno, 0, 0, 0);
 
+    return 0;
+}
 
 /* }}} */
 
 /* {{{ getxattr */
 
 struct _xattr_key {
-        char *key;
-        struct list_head list;
+    char *key;
+    struct list_head list;
 };
 
-
 int
-__gather_xattr_keys (dict_t *dict, char *key, data_t *value,
-                     void *data)
+__gather_xattr_keys(dict_t *dict, char *key, data_t *value, void *data)
 {
-        struct list_head *  list  = data;
-        struct _xattr_key * xkey  = NULL;
+    struct list_head *list = data;
+    struct _xattr_key *xkey = NULL;
 
-        if (!strncmp (key, AFR_XATTR_PREFIX,
-                      strlen (AFR_XATTR_PREFIX))) {
+    if (!strncmp(key, AFR_XATTR_PREFIX, SLEN(AFR_XATTR_PREFIX))) {
+        xkey = GF_MALLOC(sizeof(*xkey), gf_afr_mt_xattr_key);
+        if (!xkey)
+            return -1;
 
-                xkey = GF_CALLOC (1, sizeof (*xkey), gf_afr_mt_xattr_key);
-                if (!xkey)
-                        return -1;
+        xkey->key = key;
+        INIT_LIST_HEAD(&xkey->list);
 
-                xkey->key = key;
-                INIT_LIST_HEAD (&xkey->list);
-
-                list_add_tail (&xkey->list, list);
-        }
-        return 0;
+        list_add_tail(&xkey->list, list);
+    }
+    return 0;
 }
 
-
 void
-__filter_xattrs (dict_t *dict)
+afr_filter_xattrs(dict_t *dict)
 {
-        struct list_head   keys = {0,};
-        struct _xattr_key *key  = NULL;
-        struct _xattr_key *tmp  = NULL;
+    struct list_head keys = {
+        0,
+    };
+    struct _xattr_key *key = NULL;
+    struct _xattr_key *tmp = NULL;
 
-        INIT_LIST_HEAD (&keys);
+    INIT_LIST_HEAD(&keys);
 
-        dict_foreach (dict, __gather_xattr_keys,
-                      (void *) &keys);
+    dict_foreach(dict, __gather_xattr_keys, (void *)&keys);
 
-        list_for_each_entry_safe (key, tmp, &keys, list) {
-                dict_del (dict, key->key);
+    list_for_each_entry_safe(key, tmp, &keys, list)
+    {
+        dict_del(dict, key->key);
 
-                list_del_init (&key->list);
+        list_del_init(&key->list);
 
-                GF_FREE (key);
-        }
+        GF_FREE(key);
+    }
 }
 
+static gf_boolean_t
+afr_getxattr_ignorable_errnos(int32_t op_errno)
+{
+    if (op_errno == ENODATA || op_errno == ENOTSUP || op_errno == ERANGE ||
+        op_errno == ENAMETOOLONG)
+        return _gf_true;
 
-
-int32_t
-afr_getxattr_cbk (call_frame_t *frame, void *cookie,
-                  xlator_t *this, int32_t op_ret, int32_t op_errno,
-                  dict_t *dict, dict_t *xdata)
+    return _gf_false;
+}
+int
+afr_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
 {
-        afr_private_t * priv            = NULL;
-        afr_local_t *   local           = NULL;
-        xlator_t **     children        = NULL;
-        int             unwind          = 1;
-        int32_t         *last_index     = NULL;
-        int32_t         next_call_child = -1;
-        int32_t         read_child      = -1;
-        int32_t         *fresh_children  = NULL;
-
-        priv     = this->private;
-        children = priv->children;
+    afr_local_t *local = NULL;
 
-        local = frame->local;
+    local = frame->local;
 
-        read_child = (long) cookie;
-
-        if (op_ret == -1) {
-                last_index = &local->cont.getxattr.last_index;
-                fresh_children = local->fresh_children;
-                next_call_child = afr_next_call_child (fresh_children,
-                                                       local->child_up,
-                                                       priv->child_count,
-                                                       last_index, read_child);
-                if (next_call_child < 0)
-                        goto out;
-
-                unwind = 0;
-                STACK_WIND_COOKIE (frame, afr_getxattr_cbk,
-                                   (void *) (long) read_child,
-                                   children[next_call_child],
-                                   children[next_call_child]->fops->getxattr,
-                                   &local->loc,
-                                   local->cont.getxattr.name,
-                                   NULL);
-        }
+    if (op_ret < 0 && !afr_getxattr_ignorable_errnos(op_errno)) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
 
-out:
-        if (unwind) {
-                if (op_ret >= 0 && dict)
-                        __filter_xattrs (dict);
+        afr_read_txn_continue(frame, this, (long)cookie);
+        return 0;
+    }
 
-                AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict, xdata);
-        }
+    if (dict)
+        afr_filter_xattrs(dict);
+
+    AFR_STACK_UNWIND(getxattr, frame, op_ret, op_errno, dict, xdata);
+
+    return 0;
+}
 
+int
+afr_getxattr_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    if (subvol == -1) {
+        AFR_STACK_UNWIND(getxattr, frame, local->op_ret, local->op_errno, NULL,
+                         NULL);
         return 0;
+    }
+
+    STACK_WIND_COOKIE(frame, afr_getxattr_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->getxattr, &local->loc,
+                      local->cont.getxattr.name, local->xdata_req);
+    return 0;
 }
 
 int32_t
-afr_getxattr_unwind (call_frame_t *frame, int op_ret, int op_errno,
-                     dict_t *dict, dict_t *xdata)
+afr_getxattr_unwind(call_frame_t *frame, int op_ret, int op_errno, dict_t *dict,
+                    dict_t *xdata)
 
 {
-        AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict, xdata);
-        return 0;
+    AFR_STACK_UNWIND(getxattr, frame, op_ret, op_errno, dict, xdata);
+    return 0;
 }
 
 int32_t
-afr_fgetxattr_clrlk_cbk (call_frame_t *frame, void *cookie,
-                         xlator_t *this, int32_t op_ret, int32_t op_errno,
-                         dict_t *dict, dict_t *xdata)
+afr_fgetxattr_clrlk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *dict,
+                        dict_t *xdata)
 {
-        afr_local_t     *local                  = NULL;
-        afr_private_t   *priv                   = NULL;
-        xlator_t        **children              = NULL;
-        dict_t          *xattr                  = NULL;
-        char            *tmp_report             = NULL;
-        char            lk_summary[1024]        = {0,};
-        int             serz_len                = 0;
-        int32_t         callcnt                 = 0;
-        long int        cky                     = 0;
-        int             ret                     = 0;
-
-        priv     = this->private;
-        children = priv->children;
-
-        local = frame->local;
-        cky = (long) cookie;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-                if (op_ret == -1)
-                        local->child_errno[cky] = op_errno;
-
-                if (!local->dict)
-                        local->dict = dict_new ();
-                if (local->dict) {
-                        ret = dict_get_str (dict, local->cont.getxattr.name,
-                                            &tmp_report);
-                        if (ret)
-                                goto unlock;
-                        ret = dict_set_dynstr (local->dict,
-                                               children[cky]->name,
-                                               gf_strdup (tmp_report));
-                        if (ret)
-                                goto unlock;
-                }
-        }
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    xlator_t **children = NULL;
+    dict_t *xattr = NULL;
+    char *tmp_report = NULL;
+    char lk_summary[1024] = {
+        0,
+    };
+    int serz_len = 0;
+    int32_t callcnt = 0;
+    long int cky = 0;
+    int ret = 0;
+    int keylen = 0;
+    int children_keylen = 0;
+
+    priv = this->private;
+    children = priv->children;
+
+    local = frame->local;
+    cky = (long)cookie;
+    keylen = strlen(local->cont.getxattr.name);
+    children_keylen = strlen(children[cky]->name);
+
+    LOCK(&frame->lock);
+    {
+        callcnt = --local->call_count;
+        if (op_ret == -1)
+            local->replies[cky].op_errno = op_errno;
+
+        if (!local->dict)
+            local->dict = dict_new();
+        if (local->dict) {
+            ret = dict_get_strn(dict, local->cont.getxattr.name, keylen,
+                                &tmp_report);
+            if (ret)
+                goto unlock;
+            ret = dict_set_dynstrn(local->dict, children[cky]->name,
+                                   children_keylen, gf_strdup(tmp_report));
+            if (ret)
+                goto unlock;
+        }
+    }
 unlock:
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                xattr = dict_new ();
-                if (!xattr) {
-                        op_ret = -1;
-                        op_errno = ENOMEM;
-                        goto unwind;
-                }
-                ret = dict_serialize_value_with_delim (local->dict,
-                                                       lk_summary,
-                                                       &serz_len, '\n');
-                if (ret) {
-                        op_ret = -1;
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Error serializing dictionary");
-                        goto unwind;
-                }
-                if (serz_len == -1)
-                        snprintf (lk_summary, sizeof (lk_summary),
-                                  "No locks cleared.");
-                ret = dict_set_dynstr (xattr, local->cont.getxattr.name,
-                                       gf_strdup (lk_summary));
-                if (ret) {
-                        op_ret = -1;
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Error setting dictionary");
-                        goto unwind;
-                }
-
-        unwind:
-                // Updating child_errno with more recent 'events'
-                local->child_errno[cky] = op_errno;
-                op_errno = afr_resultant_errno_get (NULL, local->child_errno,
-                                                    priv->child_count);
-                AFR_STACK_UNWIND (fgetxattr, frame, op_ret, op_errno, xattr,
-                                  xdata);
-
-                if (xattr)
-                        dict_unref (xattr);
-        }
+    UNLOCK(&frame->lock);
+
+    if (!callcnt) {
+        xattr = dict_new();
+        if (!xattr) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            goto unwind;
+        }
+        ret = dict_serialize_value_with_delim(local->dict, lk_summary,
+                                              &serz_len, '\n');
+        if (ret) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            goto unwind;
+        }
+        if (serz_len == -1)
+            snprintf(lk_summary, sizeof(lk_summary), "No locks cleared.");
+        ret = dict_set_dynstrn(xattr, local->cont.getxattr.name, keylen,
+                               gf_strdup(lk_summary));
+        if (ret) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_DICT_SET_FAILED,
+                   "Error setting dictionary");
+            goto unwind;
+        }
+
+        op_errno = afr_final_errno(local, priv);
+
+    unwind:
+        AFR_STACK_UNWIND(fgetxattr, frame, op_ret, op_errno, xattr, xdata);
+        if (xattr)
+            dict_unref(xattr);
+    }
+
+    return ret;
+}
 
-        return ret;
+int32_t
+afr_getxattr_clrlk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *dict,
+                       dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    xlator_t **children = NULL;
+    dict_t *xattr = NULL;
+    char *tmp_report = NULL;
+    char lk_summary[1024] = {
+        0,
+    };
+    int serz_len = 0;
+    int32_t callcnt = 0;
+    long int cky = 0;
+    int ret = 0;
+    int keylen = 0;
+    int children_keylen = 0;
+
+    priv = this->private;
+    children = priv->children;
+
+    local = frame->local;
+    cky = (long)cookie;
+
+    keylen = strlen(local->cont.getxattr.name);
+    children_keylen = strlen(children[cky]->name);
+
+    LOCK(&frame->lock);
+    {
+        callcnt = --local->call_count;
+        if (op_ret == -1)
+            local->replies[cky].op_errno = op_errno;
+
+        if (!local->dict)
+            local->dict = dict_new();
+        if (local->dict) {
+            ret = dict_get_strn(dict, local->cont.getxattr.name, keylen,
+                                &tmp_report);
+            if (ret)
+                goto unlock;
+            ret = dict_set_dynstrn(local->dict, children[cky]->name,
+                                   children_keylen, gf_strdup(tmp_report));
+            if (ret)
+                goto unlock;
+        }
+    }
+unlock:
+    UNLOCK(&frame->lock);
+
+    if (!callcnt) {
+        xattr = dict_new();
+        if (!xattr) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            goto unwind;
+        }
+        ret = dict_serialize_value_with_delim(local->dict, lk_summary,
+                                              &serz_len, '\n');
+        if (ret) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            goto unwind;
+        }
+        if (serz_len == -1)
+            snprintf(lk_summary, sizeof(lk_summary), "No locks cleared.");
+        ret = dict_set_dynstrn(xattr, local->cont.getxattr.name, keylen,
+                               gf_strdup(lk_summary));
+        if (ret) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_DICT_SET_FAILED,
+                   "Error setting dictionary");
+            goto unwind;
+        }
+
+        op_errno = afr_final_errno(local, priv);
+
+    unwind:
+        AFR_STACK_UNWIND(getxattr, frame, op_ret, op_errno, xattr, xdata);
+
+        if (xattr)
+            dict_unref(xattr);
+    }
+
+    return ret;
 }
 
+/**
+ * node-uuid cbk uses next child querying mechanism
+ */
 int32_t
-afr_getxattr_clrlk_cbk (call_frame_t *frame, void *cookie,
-                        xlator_t *this, int32_t op_ret, int32_t op_errno,
-                        dict_t *dict, dict_t *xdata)
+afr_getxattr_node_uuid_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int32_t op_ret, int32_t op_errno, dict_t *dict,
+                           dict_t *xdata)
 {
-        afr_local_t     *local                  = NULL;
-        afr_private_t   *priv                   = NULL;
-        xlator_t        **children              = NULL;
-        dict_t          *xattr                  = NULL;
-        char            *tmp_report             = NULL;
-        char            lk_summary[1024]        = {0,};
-        int             serz_len                = 0;
-        int32_t         callcnt                 = 0;
-        long int        cky                     = 0;
-        int             ret                     = 0;
-
-        priv     = this->private;
-        children = priv->children;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    xlator_t **children = NULL;
+    int unwind = 1;
+    int curr_call_child = 0;
 
-        local = frame->local;
-        cky = (long) cookie;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-                if (op_ret == -1)
-                        local->child_errno[cky] = op_errno;
-
-                if (!local->dict)
-                        local->dict = dict_new ();
-                if (local->dict) {
-                        ret = dict_get_str (dict, local->cont.getxattr.name,
-                                            &tmp_report);
-                        if (ret)
-                                goto unlock;
-                        ret = dict_set_dynstr (local->dict,
-                                               children[cky]->name,
-                                               gf_strdup (tmp_report));
-                        if (ret)
-                                goto unlock;
-                }
-        }
-unlock:
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                xattr = dict_new ();
-                if (!xattr) {
-                        op_ret = -1;
-                        op_errno = ENOMEM;
-                        goto unwind;
-                }
-                ret = dict_serialize_value_with_delim (local->dict,
-                                                       lk_summary,
-                                                       &serz_len, '\n');
-                if (ret) {
-                        op_ret = -1;
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Error serializing dictionary");
-                        goto unwind;
-                }
-                if (serz_len == -1)
-                        snprintf (lk_summary, sizeof (lk_summary),
-                                  "No locks cleared.");
-                ret = dict_set_dynstr (xattr, local->cont.getxattr.name,
-                                       gf_strdup (lk_summary));
-                if (ret) {
-                        op_ret = -1;
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Error setting dictionary");
-                        goto unwind;
-                }
+    priv = this->private;
+    children = priv->children;
 
-        unwind:
-                // Updating child_errno with more recent 'events'
-                local->child_errno[cky] = op_errno;
-                op_errno = afr_resultant_errno_get (NULL, local->child_errno,
-                                                    priv->child_count);
-                AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, xattr, xdata);
+    local = frame->local;
 
-                if (xattr)
-                        dict_unref (xattr);
-        }
+    if (op_ret == -1) { /** query the _next_ child */
 
-        return ret;
+        /**
+         * _current_ becomes _next_
+         * If done with all children and yet no success; give up !
+         */
+        curr_call_child = (int)((long)cookie);
+        if (++curr_call_child == priv->child_count)
+            goto unwind;
+
+        gf_msg_debug(this->name, op_errno,
+                     "op_ret (-1): Re-querying afr-child (%d/%d)",
+                     curr_call_child, priv->child_count);
+
+        unwind = 0;
+        STACK_WIND_COOKIE(
+            frame, afr_getxattr_node_uuid_cbk, (void *)(long)curr_call_child,
+            children[curr_call_child],
+            children[curr_call_child]->fops->getxattr, &local->loc,
+            local->cont.getxattr.name, local->xdata_req);
+    }
+
+unwind:
+    if (unwind)
+        AFR_STACK_UNWIND(getxattr, frame, op_ret, op_errno, dict, xdata);
+
+    return 0;
 }
 
 /**
- * node-uuid cbk uses next child querying mechanism
+ * list-node-uuids cbk returns the list of node_uuids for the subvolume.
  */
 int32_t
-afr_getxattr_node_uuid_cbk (call_frame_t *frame, void *cookie,
-                            xlator_t *this, int32_t op_ret, int32_t op_errno,
-                            dict_t *dict, dict_t *xdata)
+afr_getxattr_list_node_uuids_cbk(call_frame_t *frame, void *cookie,
+                                 xlator_t *this, int32_t op_ret,
+                                 int32_t op_errno, dict_t *dict, dict_t *xdata)
 {
-        afr_private_t  *priv            = NULL;
-        afr_local_t    *local           = NULL;
-        xlator_t      **children        = NULL;
-        int             unwind          = 1;
-        int             curr_call_child = 0;
-
-        priv = this->private;
-        children = priv->children;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int32_t callcnt = 0;
+    int ret = 0;
+    char *xattr_serz = NULL;
+    long cky = 0;
+    int32_t tlen = 0;
+
+    local = frame->local;
+    priv = this->private;
+    cky = (long)cookie;
+
+    LOCK(&frame->lock);
+    {
+        callcnt = --local->call_count;
+        local->replies[cky].valid = 1;
+        local->replies[cky].op_ret = op_ret;
+        local->replies[cky].op_errno = op_errno;
+
+        if (op_ret < 0)
+            goto unlock;
+
+        local->op_ret = 0;
+
+        if (!local->xdata_rsp && xdata)
+            local->xdata_rsp = dict_ref(xdata);
+        local->replies[cky].xattr = dict_ref(dict);
+    }
 
-        local = frame->local;
-
-        if (op_ret == -1) { /** query the _next_ child */
-
-                /**
-                 * _current_ becomes _next_
-                 * If done with all childs and yet no success; give up !
-                 */
-                curr_call_child = (int) ((long)cookie);
-                if (++curr_call_child == priv->child_count)
-                        goto unwind;
-
-                gf_log (this->name, GF_LOG_WARNING,
-                        "op_ret (-1): Re-querying afr-child (%d/%d)",
-                        curr_call_child, priv->child_count);
-
-                unwind = 0;
-                STACK_WIND_COOKIE (frame, afr_getxattr_node_uuid_cbk,
-                                   (void *) (long) curr_call_child,
-                                   children[curr_call_child],
-                                   children[curr_call_child]->fops->getxattr,
-                                   &local->loc,
-                                   local->cont.getxattr.name,
-                                   NULL);
+unlock:
+    UNLOCK(&frame->lock);
+
+    if (!callcnt) {
+        if (local->op_ret != 0) {
+            /* All bricks gave an error. */
+            local->op_errno = afr_final_errno(local, priv);
+            goto unwind;
+        }
+
+        /*Since we store the UUID0_STR as node uuid for down bricks and
+         *for non zero op_ret, assigning length to  priv->child_count
+         *number of uuids*/
+        local->cont.getxattr.xattr_len = (SLEN(UUID0_STR) + 2) *
+                                         priv->child_count;
+
+        if (!local->dict)
+            local->dict = dict_new();
+        if (!local->dict) {
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            goto unwind;
+        }
+
+        xattr_serz = GF_CALLOC(local->cont.getxattr.xattr_len, sizeof(char),
+                               gf_common_mt_char);
+
+        if (!xattr_serz) {
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            goto unwind;
+        }
+
+        ret = afr_serialize_xattrs_with_delimiter(frame, this, xattr_serz,
+                                                  UUID0_STR, &tlen, ' ');
+        if (ret) {
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            GF_FREE(xattr_serz);
+            goto unwind;
+        }
+        ret = dict_set_dynstr_sizen(local->dict, GF_XATTR_LIST_NODE_UUIDS_KEY,
+                                    xattr_serz);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+                   "Cannot set node_uuid key in dict");
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            if (ret == -EINVAL)
+                GF_FREE(xattr_serz);
+        } else {
+            local->op_ret = local->cont.getxattr.xattr_len - 1;
+            local->op_errno = 0;
         }
 
- unwind:
-        if (unwind)
-                AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict,
-                                  NULL);
+    unwind:
+        AFR_STACK_UNWIND(getxattr, frame, local->op_ret, local->op_errno,
+                         local->dict, local->xdata_rsp);
+    }
 
-        return 0;
+    return ret;
 }
 
 int32_t
-afr_getxattr_lockinfo_cbk (call_frame_t *frame, void *cookie,
-                           xlator_t *this, int32_t op_ret, int32_t op_errno,
-                           dict_t *dict, dict_t *xdata)
+afr_getxattr_quota_size_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int32_t op_ret, int32_t op_errno, dict_t *dict,
+                            dict_t *xdata)
 {
-        int          call_cnt     = 0, len = 0;
-        char        *lockinfo_buf = NULL;
-        dict_t      *lockinfo     = NULL, *newdict = NULL;
-        afr_local_t *local        = NULL;
+    int idx = (long)cookie;
+    int call_count = 0;
+    afr_local_t *local = frame->local;
+    int read_subvol = -1;
+
+    local->replies[idx].valid = 1;
+    local->replies[idx].op_ret = op_ret;
+    local->replies[idx].op_errno = op_errno;
+    if (dict)
+        local->replies[idx].xdata = dict_ref(dict);
+    call_count = afr_frame_return(frame);
+    if (call_count == 0) {
+        local->inode = inode_ref(local->loc.inode);
+        read_subvol = afr_handle_quota_size(frame, this);
+        if (read_subvol != -1) {
+            op_ret = local->replies[read_subvol].op_ret;
+            op_errno = local->replies[read_subvol].op_errno;
+            dict = local->replies[read_subvol].xdata;
+        }
+        AFR_STACK_UNWIND(getxattr, frame, op_ret, op_errno, dict, xdata);
+    }
+
+    return 0;
+}
 
-        LOCK (&frame->lock);
-        {
-                local = frame->local;
+int32_t
+afr_getxattr_lockinfo_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, dict_t *dict,
+                          dict_t *xdata)
+{
+    int call_cnt = 0, len = 0;
+    char *lockinfo_buf = NULL;
+    dict_t *lockinfo = NULL, *newdict = NULL;
+    afr_local_t *local = NULL;
 
-                call_cnt = --local->call_count;
+    LOCK(&frame->lock);
+    {
+        local = frame->local;
 
-                if ((op_ret < 0) || (!dict && !xdata)) {
-                        goto unlock;
-                }
+        call_cnt = --local->call_count;
 
-                if (xdata) {
-                        if (!local->xdata_rsp) {
-                                local->xdata_rsp = dict_new ();
-                                if (!local->xdata_rsp) {
-                                        local->op_ret = -1;
-                                        local->op_errno = ENOMEM;
-                                        goto unlock;
-                                }
-                        }
-                }
+        if ((op_ret < 0) || (!dict && !xdata)) {
+            goto unlock;
+        }
 
-                if (!dict) {
-                        goto unlock;
+        if (xdata) {
+            if (!local->xdata_rsp) {
+                local->xdata_rsp = dict_new();
+                if (!local->xdata_rsp) {
+                    local->op_ret = -1;
+                    local->op_errno = ENOMEM;
+                    goto unlock;
                 }
+            }
+        }
 
-                op_ret = dict_get_ptr_and_len (dict, GF_XATTR_LOCKINFO_KEY,
-                                               (void **)&lockinfo_buf, &len);
+        if (!dict) {
+            goto unlock;
+        }
 
-                if (!lockinfo_buf) {
-                        goto unlock;
-                }
+        op_ret = dict_get_ptr_and_len(dict, GF_XATTR_LOCKINFO_KEY,
+                                      (void **)&lockinfo_buf, &len);
 
-                if (!local->dict) {
-                        local->dict = dict_new ();
-                        if (!local->dict) {
-                                local->op_ret = -1;
-                                local->op_errno = ENOMEM;
-                                goto unlock;
-                        }
-                }
-        }
-unlock:
-        UNLOCK (&frame->lock);
-
-        if (lockinfo_buf != NULL) {
-                lockinfo = dict_new ();
-                if (lockinfo == NULL) {
-                        local->op_ret = -1;
-                        local->op_errno = ENOMEM;
-                } else {
-                        op_ret = dict_unserialize (lockinfo_buf, len,
-                                                   &lockinfo);
-
-                        if (lockinfo && local->dict) {
-                                dict_copy (lockinfo, local->dict);
-                        }
-                }
+        if (!lockinfo_buf) {
+            goto unlock;
         }
 
-        if (xdata && local->xdata_rsp) {
-                dict_copy (xdata, local->xdata_rsp);
+        if (!local->dict) {
+            local->dict = dict_new();
+            if (!local->dict) {
+                local->op_ret = -1;
+                local->op_errno = ENOMEM;
+                goto unlock;
+            }
         }
+    }
+unlock:
+    UNLOCK(&frame->lock);
 
-        if (!call_cnt) {
-                newdict = dict_new ();
-                if (!newdict) {
-                        local->op_ret = -1;
-                        local->op_errno = ENOMEM;
-                        goto unwind;
-                }
+    if (lockinfo_buf != NULL) {
+        lockinfo = dict_new();
+        if (lockinfo == NULL) {
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+        } else {
+            op_ret = dict_unserialize(lockinfo_buf, len, &lockinfo);
 
-                len = dict_serialized_length (local->dict);
-                if (len == 0) {
-                        goto unwind;
-                }
+            if (lockinfo && local->dict) {
+                dict_copy(lockinfo, local->dict);
+            }
+        }
+    }
 
-                lockinfo_buf = GF_CALLOC (1, len, gf_common_mt_char);
-                if (!lockinfo_buf) {
-                        local->op_ret = -1;
-                        local->op_errno = ENOMEM;
-                        goto unwind;
-                }
+    if (xdata && local->xdata_rsp) {
+        dict_copy(xdata, local->xdata_rsp);
+    }
 
-                op_ret = dict_serialize (local->dict, lockinfo_buf);
-                if (op_ret < 0) {
-                        local->op_ret = -1;
-                        local->op_errno = -op_ret;
-                }
+    if (!call_cnt) {
+        newdict = dict_new();
+        if (!newdict) {
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            goto unwind;
+        }
 
-                op_ret = dict_set_dynptr (newdict, GF_XATTR_LOCKINFO_KEY,
-                                          (void *)lockinfo_buf, len);
-                if (op_ret < 0) {
-                        local->op_ret = -1;
-                        local->op_errno = -op_ret;
-                        goto unwind;
-                }
+        op_ret = dict_allocate_and_serialize(
+            local->dict, (char **)&lockinfo_buf, (unsigned int *)&len);
+        if (op_ret != 0) {
+            local->op_ret = -1;
+            goto unwind;
+        }
 
-        unwind:
-                AFR_STACK_UNWIND (getxattr, frame, op_ret,
-                                  op_errno, newdict,
-                                  local->xdata_rsp);
+        op_ret = dict_set_dynptr(newdict, GF_XATTR_LOCKINFO_KEY,
+                                 (void *)lockinfo_buf, len);
+        if (op_ret < 0) {
+            local->op_ret = -1;
+            local->op_errno = -op_ret;
+            goto unwind;
         }
 
-        dict_unref (lockinfo);
+    unwind:
+        AFR_STACK_UNWIND(getxattr, frame, op_ret, op_errno, newdict,
+                         local->xdata_rsp);
+    }
 
-        return 0;
+    dict_unref(lockinfo);
+
+    return 0;
 }
 
 int32_t
-afr_fgetxattr_lockinfo_cbk (call_frame_t *frame, void *cookie,
-                            xlator_t *this, int32_t op_ret, int32_t op_errno,
-                            dict_t *dict, dict_t *xdata)
+afr_fgetxattr_lockinfo_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int32_t op_ret, int32_t op_errno, dict_t *dict,
+                           dict_t *xdata)
 {
-        int          call_cnt     = 0, len = 0;
-        char        *lockinfo_buf = NULL;
-        dict_t      *lockinfo     = NULL, *newdict = NULL;
-        afr_local_t *local        = NULL;
-
-        LOCK (&frame->lock);
-        {
-                local = frame->local;
+    int call_cnt = 0, len = 0;
+    char *lockinfo_buf = NULL;
+    dict_t *lockinfo = NULL, *newdict = NULL;
+    afr_local_t *local = NULL;
 
-                call_cnt = --local->call_count;
+    LOCK(&frame->lock);
+    {
+        local = frame->local;
 
-                if ((op_ret < 0) || (!dict && !xdata)) {
-                        goto unlock;
-                }
+        call_cnt = --local->call_count;
 
-                if (xdata) {
-                        if (!local->xdata_rsp) {
-                                local->xdata_rsp = dict_new ();
-                                if (!local->xdata_rsp) {
-                                        local->op_ret = -1;
-                                        local->op_errno = ENOMEM;
-                                        goto unlock;
-                                }
-                        }
-                }
+        if ((op_ret < 0) || (!dict && !xdata)) {
+            goto unlock;
+        }
 
-                if (!dict) {
-                        goto unlock;
+        if (xdata) {
+            if (!local->xdata_rsp) {
+                local->xdata_rsp = dict_new();
+                if (!local->xdata_rsp) {
+                    local->op_ret = -1;
+                    local->op_errno = ENOMEM;
+                    goto unlock;
                 }
+            }
+        }
 
-                op_ret = dict_get_ptr_and_len (dict, GF_XATTR_LOCKINFO_KEY,
-                                               (void **)&lockinfo_buf, &len);
+        if (!dict) {
+            goto unlock;
+        }
 
-                if (!lockinfo_buf) {
-                        goto unlock;
-                }
+        op_ret = dict_get_ptr_and_len(dict, GF_XATTR_LOCKINFO_KEY,
+                                      (void **)&lockinfo_buf, &len);
 
-                if (!local->dict) {
-                        local->dict = dict_new ();
-                        if (!local->dict) {
-                                local->op_ret = -1;
-                                local->op_errno = ENOMEM;
-                                goto unlock;
-                        }
-                }
-        }
-unlock:
-        UNLOCK (&frame->lock);
-
-        if (lockinfo_buf != NULL) {
-                lockinfo = dict_new ();
-                if (lockinfo == NULL) {
-                        local->op_ret = -1;
-                        local->op_errno = ENOMEM;
-                } else {
-                        op_ret = dict_unserialize (lockinfo_buf, len,
-                                                   &lockinfo);
-
-                        if (lockinfo && local->dict) {
-                                dict_copy (lockinfo, local->dict);
-                        }
-                }
+        if (!lockinfo_buf) {
+            goto unlock;
         }
 
-        if (xdata && local->xdata_rsp) {
-                dict_copy (xdata, local->xdata_rsp);
+        if (!local->dict) {
+            local->dict = dict_new();
+            if (!local->dict) {
+                local->op_ret = -1;
+                local->op_errno = ENOMEM;
+                goto unlock;
+            }
         }
+    }
+unlock:
+    UNLOCK(&frame->lock);
 
-        if (!call_cnt) {
-                newdict = dict_new ();
-                if (!newdict) {
-                        local->op_ret = -1;
-                        local->op_errno = ENOMEM;
-                        goto unwind;
-                }
+    if (lockinfo_buf != NULL) {
+        lockinfo = dict_new();
+        if (lockinfo == NULL) {
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+        } else {
+            op_ret = dict_unserialize(lockinfo_buf, len, &lockinfo);
 
-                len = dict_serialized_length (local->dict);
-                if (len <= 0) {
-                        goto unwind;
-                }
+            if (lockinfo && local->dict) {
+                dict_copy(lockinfo, local->dict);
+            }
+        }
+    }
 
-                lockinfo_buf = GF_CALLOC (1, len, gf_common_mt_char);
-                if (!lockinfo_buf) {
-                        local->op_ret = -1;
-                        local->op_errno = ENOMEM;
-                        goto unwind;
-                }
+    if (xdata && local->xdata_rsp) {
+        dict_copy(xdata, local->xdata_rsp);
+    }
 
-                op_ret = dict_serialize (local->dict, lockinfo_buf);
-                if (op_ret < 0) {
-                        local->op_ret = -1;
-                        local->op_errno = -op_ret;
-                }
+    if (!call_cnt) {
+        newdict = dict_new();
+        if (!newdict) {
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            goto unwind;
+        }
 
-                op_ret = dict_set_dynptr (newdict, GF_XATTR_LOCKINFO_KEY,
-                                          (void *)lockinfo_buf, len);
-                if (op_ret < 0) {
-                        local->op_ret = -1;
-                        local->op_errno = -op_ret;
-                        goto unwind;
-                }
+        op_ret = dict_allocate_and_serialize(
+            local->dict, (char **)&lockinfo_buf, (unsigned int *)&len);
+        if (op_ret != 0) {
+            local->op_ret = -1;
+            goto unwind;
+        }
 
-        unwind:
-                AFR_STACK_UNWIND (fgetxattr, frame, op_ret,
-                                  op_errno, newdict,
-                                  local->xdata_rsp);
+        op_ret = dict_set_dynptr(newdict, GF_XATTR_LOCKINFO_KEY,
+                                 (void *)lockinfo_buf, len);
+        if (op_ret < 0) {
+            local->op_ret = -1;
+            local->op_errno = -op_ret;
+            goto unwind;
         }
 
-        dict_unref (lockinfo);
+    unwind:
+        AFR_STACK_UNWIND(fgetxattr, frame, op_ret, op_errno, newdict,
+                         local->xdata_rsp);
+    }
 
-        return 0;
+    dict_unref(lockinfo);
+
+    return 0;
 }
 
 int32_t
-afr_fgetxattr_pathinfo_cbk (call_frame_t *frame, void *cookie,
-                            xlator_t *this, int32_t op_ret, int32_t op_errno,
-                            dict_t *dict, dict_t *xdata)
+afr_fgetxattr_pathinfo_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int32_t op_ret, int32_t op_errno, dict_t *dict,
+                           dict_t *xdata)
 {
-        afr_local_t *local          = NULL;
-        int32_t      callcnt        = 0;
-        int          ret            = 0;
-        char        *xattr          = NULL;
-        char        *xattr_serz     = NULL;
-        char        xattr_cky[1024] = {0,};
-        dict_t      *nxattr         = NULL;
-        long         cky            = 0;
-        int32_t      padding        = 0;
-        int32_t      tlen           = 0;
-
-        if (!frame || !frame->local || !this) {
-                gf_log ("", GF_LOG_ERROR, "possible NULL deref");
-                goto out;
+    afr_local_t *local = NULL;
+    int32_t callcnt = 0;
+    int ret = 0;
+    char *xattr = NULL;
+    char *xattr_serz = NULL;
+    int keylen = 0;
+    char xattr_cky[1024] = {
+        0,
+    };
+    int xattr_cky_len = 0;
+    dict_t *nxattr = NULL;
+    long cky = 0;
+    int32_t padding = 0;
+    int32_t tlen = 0;
+
+    if (!frame || !frame->local || !this) {
+        gf_msg("", GF_LOG_ERROR, 0, AFR_MSG_INVALID_ARG, "possible NULL deref");
+        goto out;
+    }
+
+    local = frame->local;
+    cky = (long)cookie;
+    keylen = strlen(local->cont.getxattr.name);
+    xattr_cky_len = snprintf(xattr_cky, sizeof(xattr_cky), "%s-%ld",
+                             local->cont.getxattr.name, cky);
+    LOCK(&frame->lock);
+    {
+        callcnt = --local->call_count;
+
+        if (op_ret < 0) {
+            local->op_errno = op_errno;
+        } else {
+            local->op_ret = op_ret;
+            if (!local->xdata_rsp && xdata)
+                local->xdata_rsp = dict_ref(xdata);
         }
 
-        local = frame->local;
-        cky = (long) cookie;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (!dict || (op_ret < 0))
-                        goto out;
-
-                if (!local->dict)
-                        local->dict = dict_new ();
-
-                if (local->dict) {
-                        ret = dict_get_str (dict,
-                                            local->cont.getxattr.name,
-                                            &xattr);
-                        if (ret)
-                                goto out;
-
-                        xattr = gf_strdup (xattr);
-
-                        (void)snprintf (xattr_cky, 1024, "%s-%ld",
-                                        local->cont.getxattr.name, cky);
-                        ret = dict_set_dynstr (local->dict,
-                                               xattr_cky, xattr);
-                        if (ret) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "Cannot set xattr cookie key");
-                                goto out;
-                        }
-
-                        local->cont.getxattr.xattr_len
-                                += strlen (xattr) + 1;
-                }
+        if (!dict || (op_ret < 0))
+            goto unlock;
+
+        if (!local->dict) {
+            local->dict = dict_new();
+            if (!local->dict)
+                goto unlock;
         }
-out:
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (!local->cont.getxattr.xattr_len)
-                        goto unwind;
-
-                nxattr = dict_new ();
-                if (!nxattr)
-                        goto unwind;
-
-                /* extra bytes for decorations (brackets and <>'s) */
-                padding += strlen (this->name)
-                        + strlen (AFR_PATHINFO_HEADER) + 4;
-                local->cont.getxattr.xattr_len += (padding + 2);
-
-                xattr_serz = GF_CALLOC (local->cont.getxattr.xattr_len,
-                                        sizeof (char), gf_common_mt_char);
-
-                if (!xattr_serz)
-                        goto unwind;
-
-                /* the xlator info */
-                (void) sprintf (xattr_serz, "(<"AFR_PATHINFO_HEADER"%s> ",
-                                this->name);
-
-                /* actual series of pathinfo */
-                ret = dict_serialize_value_with_delim (local->dict,
-                                                       xattr_serz
-                                                       + strlen (xattr_serz),
-                                                       &tlen, ' ');
-                if (ret) {
-                        gf_log (this->name, GF_LOG_ERROR, "Error serializing"
-                                " dictionary");
-                        goto unwind;
-                }
+        ret = dict_get_strn(dict, local->cont.getxattr.name, keylen, &xattr);
+        if (ret)
+            goto unlock;
+
+        xattr = gf_strdup(xattr);
+
+        ret = dict_set_dynstrn(local->dict, xattr_cky, xattr_cky_len, xattr);
+        if (ret) {
+            UNLOCK(&frame->lock);
+            gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+                   "Cannot set xattr cookie key");
+            goto post_unlock;
+        }
+
+        local->cont.getxattr.xattr_len += strlen(xattr) + 1;
+    }
+unlock:
+    UNLOCK(&frame->lock);
+post_unlock:
+    if (!callcnt) {
+        if (!local->cont.getxattr.xattr_len)
+            goto unwind;
+
+        nxattr = dict_new();
+        if (!nxattr)
+            goto unwind;
 
-                /* closing part */
-                *(xattr_serz + padding + tlen) = ')';
-                *(xattr_serz + padding + tlen + 1) = '\0';
+        /* extra bytes for decorations (brackets and <>'s) */
+        padding += strlen(this->name) + SLEN(AFR_PATHINFO_HEADER) + 4;
+        local->cont.getxattr.xattr_len += (padding + 2);
 
-                ret = dict_set_dynstr (nxattr, local->cont.getxattr.name,
-                                       xattr_serz);
-                if (ret)
-                        gf_log (this->name, GF_LOG_ERROR, "Cannot set pathinfo"
-                                " key in dict");
+        xattr_serz = GF_MALLOC(local->cont.getxattr.xattr_len,
+                               gf_common_mt_char);
 
-        unwind:
-                AFR_STACK_UNWIND (fgetxattr, frame, op_ret, op_errno, nxattr,
-                                  xdata);
+        if (!xattr_serz)
+            goto unwind;
 
-                if (nxattr)
-                        dict_unref (nxattr);
+        /* the xlator info */
+        int xattr_serz_len = sprintf(
+            xattr_serz, "(<" AFR_PATHINFO_HEADER "%s> ", this->name);
+
+        /* actual series of pathinfo */
+        ret = dict_serialize_value_with_delim(
+            local->dict, xattr_serz + xattr_serz_len, &tlen, ' ');
+        if (ret) {
+            GF_FREE(xattr_serz);
+            goto unwind;
+        }
+
+        /* closing part */
+        *(xattr_serz + padding + tlen) = ')';
+        *(xattr_serz + padding + tlen + 1) = '\0';
+
+        ret = dict_set_dynstrn(nxattr, local->cont.getxattr.name, keylen,
+                               xattr_serz);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+                   "Cannot set pathinfo key in dict");
+            if (ret == -EINVAL)
+                GF_FREE(xattr_serz);
         }
 
-        return ret;
+    unwind:
+        AFR_STACK_UNWIND(fgetxattr, frame, local->op_ret, local->op_errno,
+                         nxattr, local->xdata_rsp);
+
+        if (nxattr)
+            dict_unref(nxattr);
+    }
+
+out:
+    return ret;
 }
 
 int32_t
-afr_getxattr_pathinfo_cbk (call_frame_t *frame, void *cookie,
-                           xlator_t *this, int32_t op_ret, int32_t op_errno,
-                           dict_t *dict, dict_t *xdata)
+afr_getxattr_pathinfo_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, dict_t *dict,
+                          dict_t *xdata)
 {
-        afr_local_t *local          = NULL;
-        int32_t      callcnt        = 0;
-        int          ret            = 0;
-        char        *xattr          = NULL;
-        char        *xattr_serz     = NULL;
-        char        xattr_cky[1024] = {0,};
-        dict_t      *nxattr         = NULL;
-        long         cky            = 0;
-        int32_t      padding        = 0;
-        int32_t      tlen           = 0;
-
-        if (!frame || !frame->local || !this) {
-                gf_log ("", GF_LOG_ERROR, "possible NULL deref");
-                goto out;
+    afr_local_t *local = NULL;
+    int32_t callcnt = 0;
+    int ret = 0;
+    char *xattr = NULL;
+    char *xattr_serz = NULL;
+    char xattr_cky[1024] = {
+        0,
+    };
+    int keylen = 0;
+    int xattr_cky_len = 0;
+    dict_t *nxattr = NULL;
+    long cky = 0;
+    int32_t padding = 0;
+    int32_t tlen = 0;
+
+    if (!frame || !frame->local || !this) {
+        gf_msg("", GF_LOG_ERROR, 0, AFR_MSG_INVALID_ARG, "possible NULL deref");
+        goto out;
+    }
+
+    local = frame->local;
+    cky = (long)cookie;
+    keylen = strlen(local->cont.getxattr.name);
+    xattr_cky_len = snprintf(xattr_cky, sizeof(xattr_cky), "%s-%ld",
+                             local->cont.getxattr.name, cky);
+    LOCK(&frame->lock);
+    {
+        callcnt = --local->call_count;
+
+        if (op_ret < 0) {
+            local->op_errno = op_errno;
+        } else {
+            local->op_ret = op_ret;
+            if (!local->xdata_rsp && xdata)
+                local->xdata_rsp = dict_ref(xdata);
         }
 
-        local = frame->local;
-        cky = (long) cookie;
-
-        LOCK (&frame->lock);
-                {
-                        callcnt = --local->call_count;
-
-                        if (!dict || (op_ret < 0))
-                                goto out;
-
-                        if (!local->dict)
-                                local->dict = dict_new ();
-
-                        if (local->dict) {
-                                ret = dict_get_str (dict,
-                                                    local->cont.getxattr.name,
-                                                    &xattr);
-                                if (ret)
-                                        goto out;
-
-                                xattr = gf_strdup (xattr);
-
-                                (void)snprintf (xattr_cky, 1024, "%s-%ld",
-                                                local->cont.getxattr.name, cky);
-                                ret = dict_set_dynstr (local->dict,
-                                                       xattr_cky, xattr);
-                                if (ret) {
-                                        gf_log (this->name, GF_LOG_ERROR,
-                                                "Cannot set xattr cookie key");
-                                        goto out;
-                                }
-
-                                local->cont.getxattr.xattr_len += strlen (xattr) + 1;
-                        }
-                }
- out:
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (!local->cont.getxattr.xattr_len)
-                        goto unwind;
-
-                nxattr = dict_new ();
-                if (!nxattr)
-                        goto unwind;
-
-                /* extra bytes for decorations (brackets and <>'s) */
-                padding += strlen (this->name) + strlen (AFR_PATHINFO_HEADER) + 4;
-                local->cont.getxattr.xattr_len += (padding + 2);
-
-                xattr_serz = GF_CALLOC (local->cont.getxattr.xattr_len,
-                                        sizeof (char), gf_common_mt_char);
-
-                if (!xattr_serz)
-                        goto unwind;
-
-                /* the xlator info */
-                (void) sprintf (xattr_serz, "(<"AFR_PATHINFO_HEADER"%s> ",
-                                this->name);
-
-                /* actual series of pathinfo */
-                ret = dict_serialize_value_with_delim (local->dict,
-                                                       xattr_serz + strlen (xattr_serz),
-                                                       &tlen, ' ');
-                if (ret) {
-                        gf_log (this->name, GF_LOG_ERROR, "Error serializing"
-                                " dictionary");
-                        goto unwind;
-                }
+        if (!dict || (op_ret < 0))
+            goto unlock;
+
+        if (!local->dict) {
+            local->dict = dict_new();
+            if (!local->dict)
+                goto unlock;
+        }
+        ret = dict_get_strn(dict, local->cont.getxattr.name, keylen, &xattr);
+        if (ret)
+            goto unlock;
+
+        xattr = gf_strdup(xattr);
 
-                /* closing part */
-                *(xattr_serz + padding + tlen) = ')';
-                *(xattr_serz + padding + tlen + 1) = '\0';
+        ret = dict_set_dynstrn(local->dict, xattr_cky, xattr_cky_len, xattr);
+        if (ret) {
+            UNLOCK(&frame->lock);
+            gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+                   "Cannot set xattr cookie key");
+            goto post_unlock;
+        }
+
+        local->cont.getxattr.xattr_len += strlen(xattr) + 1;
+    }
+unlock:
+    UNLOCK(&frame->lock);
+post_unlock:
+    if (!callcnt) {
+        if (!local->cont.getxattr.xattr_len)
+            goto unwind;
 
-                ret = dict_set_dynstr (nxattr, local->cont.getxattr.name,
-                                       xattr_serz);
-                if (ret)
-                        gf_log (this->name, GF_LOG_ERROR, "Cannot set pathinfo"
-                                " key in dict");
+        nxattr = dict_new();
+        if (!nxattr)
+            goto unwind;
+
+        /* extra bytes for decorations (brackets and <>'s) */
+        padding += strlen(this->name) + SLEN(AFR_PATHINFO_HEADER) + 4;
+        local->cont.getxattr.xattr_len += (padding + 2);
+
+        xattr_serz = GF_MALLOC(local->cont.getxattr.xattr_len,
+                               gf_common_mt_char);
+
+        if (!xattr_serz)
+            goto unwind;
+
+        /* the xlator info */
+        int xattr_serz_len = sprintf(
+            xattr_serz, "(<" AFR_PATHINFO_HEADER "%s> ", this->name);
+
+        /* actual series of pathinfo */
+        ret = dict_serialize_value_with_delim(
+            local->dict, xattr_serz + xattr_serz_len, &tlen, ' ');
+        if (ret) {
+            GF_FREE(xattr_serz);
+            goto unwind;
+        }
 
-        unwind:
-                AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, nxattr,
-                                  xdata);
+        /* closing part */
+        *(xattr_serz + padding + tlen) = ')';
+        *(xattr_serz + padding + tlen + 1) = '\0';
 
-                if (nxattr)
-                        dict_unref (nxattr);
+        ret = dict_set_dynstrn(nxattr, local->cont.getxattr.name, keylen,
+                               xattr_serz);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+                   "Cannot set pathinfo key in dict");
+            if (ret == -EINVAL)
+                GF_FREE(xattr_serz);
         }
 
-        return ret;
+    unwind:
+        AFR_STACK_UNWIND(getxattr, frame, local->op_ret, local->op_errno,
+                         nxattr, local->xdata_rsp);
+
+        if (nxattr)
+            dict_unref(nxattr);
+    }
+
+out:
+    return ret;
 }
 
 static int
-afr_aggregate_stime_xattr (dict_t *this, char *key, data_t *value, void *data)
+afr_aggregate_stime_xattr(dict_t *this, char *key, data_t *value, void *data)
 {
-        int ret = 0;
+    int ret = 0;
 
-        if (fnmatch (GF_XATTR_STIME_PATTERN, key, FNM_NOESCAPE) == 0)
-                ret = gf_get_min_stime (THIS, data, key, value);
+    if (fnmatch(GF_XATTR_STIME_PATTERN, key, FNM_NOESCAPE) == 0)
+        ret = gf_get_max_stime(THIS, data, key, value);
 
-        return ret;
+    return ret;
 }
 
 int32_t
-afr_common_getxattr_stime_cbk (call_frame_t *frame, void *cookie,
-                               xlator_t *this, int32_t op_ret, int32_t op_errno,
-                               dict_t *dict, dict_t *xdata)
+afr_common_getxattr_stime_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                              int32_t op_ret, int32_t op_errno, dict_t *dict,
+                              dict_t *xdata)
 {
-        afr_local_t *local          = NULL;
-        int32_t      callcnt        = 0;
+    afr_local_t *local = NULL;
+    int32_t callcnt = 0;
 
-        if (!frame || !frame->local || !this) {
-                gf_log ("", GF_LOG_ERROR, "possible NULL deref");
-                goto out;
-        }
-
-        local = frame->local;
+    if (!frame || !frame->local || !this) {
+        gf_msg("", GF_LOG_ERROR, 0, AFR_MSG_INVALID_ARG, "possible NULL deref");
+        goto out;
+    }
 
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
+    local = frame->local;
 
-                if (!dict || (op_ret < 0)) {
-                        local->op_errno = op_errno;
-                        goto cleanup;
-                }
+    LOCK(&frame->lock);
+    {
+        callcnt = --local->call_count;
 
-                if (!local->dict)
-                        local->dict = dict_copy_with_ref (dict, NULL);
-                else
-                        dict_foreach (dict, afr_aggregate_stime_xattr,
-                                      local->dict);
-                local->op_ret = 0;
+        if (!dict || (op_ret < 0)) {
+            local->op_errno = op_errno;
+            goto cleanup;
         }
 
+        if (!local->dict)
+            local->dict = dict_copy_with_ref(dict, NULL);
+        else
+            dict_foreach(dict, afr_aggregate_stime_xattr, local->dict);
+        local->op_ret = 0;
+    }
+
 cleanup:
-        UNLOCK (&frame->lock);
+    UNLOCK(&frame->lock);
 
-        if (!callcnt) {
-                AFR_STACK_UNWIND (getxattr, frame, local->op_ret,
-                                  local->op_errno, local->dict, xdata);
-        }
+    if (!callcnt) {
+        AFR_STACK_UNWIND(getxattr, frame, local->op_ret, local->op_errno,
+                         local->dict, xdata);
+    }
 
 out:
-        return 0;
+    return 0;
 }
 
-
 static gf_boolean_t
-afr_is_special_xattr (const char *name, fop_getxattr_cbk_t *cbk,
-                      gf_boolean_t is_fgetxattr)
+afr_is_special_xattr(const char *name, fop_getxattr_cbk_t *cbk,
+                     gf_boolean_t is_fgetxattr)
 {
-        gf_boolean_t    is_spl = _gf_true;
-
-        GF_ASSERT (cbk);
-        if (!cbk) {
-                is_spl = _gf_false;
-                goto out;
+    gf_boolean_t is_spl = _gf_true;
+
+    GF_ASSERT(cbk);
+    if (!cbk || !name) {
+        is_spl = _gf_false;
+        goto out;
+    }
+
+    if (!strcmp(name, GF_XATTR_PATHINFO_KEY) ||
+        !strcmp(name, GF_XATTR_USER_PATHINFO_KEY)) {
+        if (is_fgetxattr) {
+            *cbk = afr_fgetxattr_pathinfo_cbk;
+        } else {
+            *cbk = afr_getxattr_pathinfo_cbk;
         }
-
-        if (!strcmp (name, GF_XATTR_PATHINFO_KEY)) {
-                if (is_fgetxattr) {
-                        *cbk = afr_fgetxattr_pathinfo_cbk;
-                } else {
-                        *cbk = afr_getxattr_pathinfo_cbk;
-                }
-        } else if (!strncmp (name, GF_XATTR_CLRLK_CMD,
-                             strlen (GF_XATTR_CLRLK_CMD))) {
-                if (is_fgetxattr) {
-                        *cbk = afr_fgetxattr_clrlk_cbk;
-                } else {
-                        *cbk = afr_getxattr_clrlk_cbk;
-                }
-        } else if (!strncmp (name, GF_XATTR_LOCKINFO_KEY,
-                             strlen (GF_XATTR_LOCKINFO_KEY))) {
-                if (is_fgetxattr) {
-                        *cbk = afr_fgetxattr_lockinfo_cbk;
-                } else {
-                        *cbk = afr_getxattr_lockinfo_cbk;
-                }
-        } else if (fnmatch (GF_XATTR_STIME_PATTERN, name, FNM_NOESCAPE) == 0) {
-                *cbk = afr_common_getxattr_stime_cbk;
+    } else if (!strncmp(name, GF_XATTR_CLRLK_CMD, SLEN(GF_XATTR_CLRLK_CMD))) {
+        if (is_fgetxattr) {
+            *cbk = afr_fgetxattr_clrlk_cbk;
         } else {
-                is_spl = _gf_false;
+            *cbk = afr_getxattr_clrlk_cbk;
         }
+    } else if (!strncmp(name, GF_XATTR_LOCKINFO_KEY,
+                        SLEN(GF_XATTR_LOCKINFO_KEY))) {
+        if (is_fgetxattr) {
+            *cbk = afr_fgetxattr_lockinfo_cbk;
+        } else {
+            *cbk = afr_getxattr_lockinfo_cbk;
+        }
+    } else if (fnmatch(GF_XATTR_STIME_PATTERN, name, FNM_NOESCAPE) == 0) {
+        *cbk = afr_common_getxattr_stime_cbk;
+    } else if (strcmp(name, QUOTA_SIZE_KEY) == 0) {
+        *cbk = afr_getxattr_quota_size_cbk;
+    } else if (!strcmp(name, GF_XATTR_LIST_NODE_UUIDS_KEY)) {
+        *cbk = afr_getxattr_list_node_uuids_cbk;
+    } else {
+        is_spl = _gf_false;
+    }
 
 out:
-        return is_spl;
+    return is_spl;
 }
 
 static void
-afr_getxattr_frm_all_children (xlator_t *this, call_frame_t *frame,
-                               const char *name, loc_t *loc,
-                               fop_getxattr_cbk_t cbk)
+afr_getxattr_all_subvols(xlator_t *this, call_frame_t *frame, const char *name,
+                         loc_t *loc, fop_getxattr_cbk_t cbk)
 {
-        afr_private_t   *priv           = NULL;
-        afr_local_t     *local          = NULL;
-        xlator_t        **children      = NULL;
-        int             i               = 0;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int i = 0;
+    int call_count = 0;
+
+    priv = this->private;
+
+    local = frame->local;
+    // local->call_count set in afr_local_init
+    call_count = local->call_count;
+
+    if (!strcmp(name, GF_XATTR_LIST_NODE_UUIDS_KEY)) {
+        GF_FREE(local->cont.getxattr.name);
+        local->cont.getxattr.name = gf_strdup(GF_XATTR_NODE_UUID_KEY);
+    }
+
+    // If up-children count is 0, afr_local_init would have failed already
+    // and the call would have unwound so not handling it here.
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->child_up[i]) {
+            STACK_WIND_COOKIE(frame, cbk, (void *)(long)i, priv->children[i],
+                              priv->children[i]->fops->getxattr, loc,
+                              local->cont.getxattr.name, NULL);
+            if (!--call_count)
+                break;
+        }
+    }
+    return;
+}
 
-        priv     = this->private;
-        children = priv->children;
+int
+afr_marker_populate_args(call_frame_t *frame, int type, int *gauge,
+                         xlator_t **subvols)
+{
+    xlator_t *this = frame->this;
+    afr_private_t *priv = this->private;
 
-        local = frame->local;
-        local->call_count = priv->child_count;
+    memcpy(subvols, priv->children, sizeof(*subvols) * priv->child_count);
 
-        for (i = 0; i < priv->child_count; i++) {
-                STACK_WIND_COOKIE (frame, cbk,
-                                   (void *) (long) i,
-                                   children[i], children[i]->fops->getxattr,
-                                   loc, name, NULL);
-        }
-        return;
+    if (type == MARKER_XTIME_TYPE) {
+        /*Don't error out on ENOENT/ENOTCONN */
+        gauge[MCNT_NOTFOUND] = 0;
+        gauge[MCNT_ENOTCONN] = 0;
+    }
+    return priv->child_count;
+}
+
+static int
+afr_handle_heal_xattrs(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                       const char *heal_op)
+{
+    int ret = -1;
+    afr_spb_status_t *data = NULL;
+
+    if (!strcmp(heal_op, GF_HEAL_INFO)) {
+        afr_get_heal_info(frame, this, loc);
+        ret = 0;
+        goto out;
+    }
+
+    if (!strcmp(heal_op, GF_AFR_HEAL_SBRAIN)) {
+        afr_heal_splitbrain_file(frame, this, loc);
+        ret = 0;
+        goto out;
+    }
+
+    if (!strcmp(heal_op, GF_AFR_SBRAIN_STATUS)) {
+        data = GF_CALLOC(1, sizeof(*data), gf_afr_mt_spb_status_t);
+        if (!data) {
+            ret = 1;
+            goto out;
+        }
+        data->frame = frame;
+        data->loc = loc;
+        ret = synctask_new(this->ctx->env, afr_get_split_brain_status,
+                           afr_get_split_brain_status_cbk, NULL, data);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN_STATUS,
+                   "Failed to create"
+                   " synctask. Unable to fetch split-brain status"
+                   " for %s.",
+                   loc->name);
+            ret = 1;
+            goto out;
+        }
+        goto out;
+    }
+
+out:
+    if (ret == 1) {
+        AFR_STACK_UNWIND(getxattr, frame, -1, ENOMEM, NULL, NULL);
+        if (data)
+            GF_FREE(data);
+        ret = 0;
+    }
+    return ret;
 }
 
 int32_t
-afr_getxattr (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, const char *name, dict_t *xdata)
+afr_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, const char *name,
+             dict_t *xdata)
 {
-        afr_private_t           *priv         = NULL;
-        xlator_t                **children    = NULL;
-        int                     call_child    = 0;
-        afr_local_t             *local        = NULL;
-        xlator_list_t           *trav         = NULL;
-        xlator_t                **sub_volumes = NULL;
-        int                     i             = 0;
-        int32_t                 op_errno      = 0;
-        int32_t                 read_child    = -1;
-        int                     ret           = -1;
-        fop_getxattr_cbk_t      cbk           = NULL;
-        int                     afr_xtime_gauge[MCNT_MAX] = {0,};
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
-
-        priv     = this->private;
-        VALIDATE_OR_GOTO (priv->children, out);
-
-        children = priv->children;
-
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    xlator_t **children = NULL;
+    int i = 0;
+    int32_t op_errno = 0;
+    int ret = -1;
+    fop_getxattr_cbk_t cbk = NULL;
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-        loc_copy (&local->loc, loc);
-        if (!name)
-                goto no_name;
+    priv = this->private;
 
-        local->cont.getxattr.name = gf_strdup (name);
+    children = priv->children;
 
-        if (!strncmp (name, AFR_XATTR_PREFIX,
-                      strlen (AFR_XATTR_PREFIX))) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "%s: no data present for key %s",
-                        loc->path, name);
-                op_errno = ENODATA;
-                goto out;
-        }
-        if ((strcmp (GF_XATTR_MARKER_KEY, name) == 0)
-            && (GF_CLIENT_PID_GSYNCD == frame->root->pid)) {
+    loc_copy(&local->loc, loc);
 
-                local->marker.call_count = priv->child_count;
+    local->op = GF_FOP_GETXATTR;
 
-                sub_volumes = alloca ( priv->child_count * sizeof (xlator_t *));
-                for (i = 0, trav = this->children; trav ;
-                     trav = trav->next, i++) {
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
 
-                        *(sub_volumes + i)  = trav->xlator;
-                }
+    if (!name)
+        goto no_name;
 
-                if (cluster_getmarkerattr (frame, this, loc, name,
-                                           local, afr_getxattr_unwind,
-                                           sub_volumes,
-                                           priv->child_count,
-                                           MARKER_UUID_TYPE,
-                                           marker_uuid_default_gauge,
-                                           priv->vol_uuid)) {
-
-                        gf_log (this->name, GF_LOG_INFO,
-                                "%s: failed to get marker attr (%s)",
-                                loc->path, name);
-                        op_errno = EINVAL;
-                        goto out;
-                }
+    local->cont.getxattr.name = gf_strdup(name);
 
-                return 0;
-        }
+    if (!local->cont.getxattr.name) {
+        op_errno = ENOMEM;
+        goto out;
+    }
 
-        /*
-         * if we are doing getxattr with pathinfo as the key then we
-         * collect information from all childs
-         */
-        if (afr_is_special_xattr (name, &cbk, 0)) {
-                afr_getxattr_frm_all_children (this, frame, name,
-                                               loc, cbk);
-                return 0;
-        }
+    if (!strncmp(name, AFR_XATTR_PREFIX, SLEN(AFR_XATTR_PREFIX))) {
+        op_errno = ENODATA;
+        goto out;
+    }
 
-        if (XATTR_IS_NODE_UUID (name)) {
-                i = 0;
-                STACK_WIND_COOKIE (frame, afr_getxattr_node_uuid_cbk,
-                                   (void *) (long) i,
-                                   children[i],
-                                   children[i]->fops->getxattr,
-                                   loc, name, xdata);
-                return 0;
-        }
+    if (cluster_handle_marker_getxattr(frame, loc, name, priv->vol_uuid,
+                                       afr_getxattr_unwind,
+                                       afr_marker_populate_args) == 0)
+        return 0;
 
-        if (*priv->vol_uuid) {
-                if ((match_uuid_local (name, priv->vol_uuid) == 0)
-                    && (GF_CLIENT_PID_GSYNCD == frame->root->pid)) {
-                        local->marker.call_count = priv->child_count;
-
-                        sub_volumes = alloca ( priv->child_count
-                                               * sizeof (xlator_t *));
-                        for (i = 0, trav = this->children; trav ;
-                             trav = trav->next, i++) {
-
-                                *(sub_volumes + i)  = trav->xlator;
-
-                        }
-
-                        /* don't err out on getting ENOTCONN (brick down)
-                         * from a subset of the bricks
-                         */
-                        memcpy (afr_xtime_gauge, marker_xtime_default_gauge,
-                                sizeof (afr_xtime_gauge));
-                        afr_xtime_gauge[MCNT_NOTFOUND] = 0;
-                        afr_xtime_gauge[MCNT_ENOTCONN] = 0;
-                        if (cluster_getmarkerattr (frame, this, loc,
-                                                   name, local,
-                                                   afr_getxattr_unwind,
-                                                   sub_volumes,
-                                                   priv->child_count,
-                                                   MARKER_XTIME_TYPE,
-                                                   afr_xtime_gauge,
-                                                   priv->vol_uuid)) {
-                                gf_log (this->name, GF_LOG_INFO,
-                                        "%s: failed to get marker attr (%s)",
-                                        loc->path, name);
-                                op_errno = EINVAL;
-                                goto out;
-                        }
-
-                        return 0;
-                }
-        }
+    ret = afr_handle_heal_xattrs(frame, this, &local->loc, name);
+    if (ret == 0)
+        return 0;
 
-no_name:
-        local->fresh_children = afr_children_create (priv->child_count);
-        if (!local->fresh_children) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+    /*
+     * Heal daemons don't have IO threads ... and as a result they
+     * send this getxattr down and eventually crash :(
+     */
+    op_errno = -1;
+    GF_CHECK_XATTR_KEY_AND_GOTO(name, IO_THREADS_QUEUE_SIZE_KEY, op_errno, out);
+
+    /*
+     * Special xattrs which need responses from all subvols
+     */
+    if (afr_is_special_xattr(name, &cbk, 0)) {
+        afr_getxattr_all_subvols(this, frame, name, loc, cbk);
+        return 0;
+    }
 
-        read_child = afr_inode_get_read_ctx (this, loc->inode,
-                                             local->fresh_children);
-        ret = afr_get_call_child (this, local->child_up, read_child,
-                                     local->fresh_children,
-                                     &call_child,
-                                     &local->cont.getxattr.last_index);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+    if (XATTR_IS_NODE_UUID(name)) {
+        i = 0;
+        STACK_WIND_COOKIE(frame, afr_getxattr_node_uuid_cbk, (void *)(long)i,
+                          children[i], children[i]->fops->getxattr, loc, name,
+                          xdata);
+        return 0;
+    }
+
+no_name:
 
-        STACK_WIND_COOKIE (frame, afr_getxattr_cbk,
-                           (void *) (long) call_child,
-                           children[call_child],
-                           children[call_child]->fops->getxattr,
-                           loc, name, xdata);
+    afr_read_txn(frame, this, local->loc.inode, afr_getxattr_wind,
+                 AFR_METADATA_TRANSACTION);
 
-        ret = 0;
+    ret = 0;
 out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (getxattr, frame, -1, op_errno, NULL, NULL);
-        return 0;
+    if (ret < 0)
+        AFR_STACK_UNWIND(getxattr, frame, -1, op_errno, NULL, NULL);
+    return 0;
 }
 
 /* {{{ fgetxattr */
 
-
 int32_t
-afr_fgetxattr_cbk (call_frame_t *frame, void *cookie,
-                   xlator_t *this, int32_t op_ret, int32_t op_errno,
-                   dict_t *dict, dict_t *xdata)
+afr_fgetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
 {
-        afr_private_t * priv            = NULL;
-        afr_local_t *   local           = NULL;
-        xlator_t **     children        = NULL;
-        int             unwind          = 1;
-        int32_t         *last_index     = NULL;
-        int32_t         next_call_child = -1;
-        int32_t         read_child      = -1;
-        int32_t         *fresh_children  = NULL;
-
-        priv     = this->private;
-        children = priv->children;
+    afr_local_t *local = NULL;
 
-        local = frame->local;
+    local = frame->local;
 
-        read_child = (long) cookie;
-
-        if (op_ret == -1) {
-                last_index = &local->cont.getxattr.last_index;
-                fresh_children = local->fresh_children;
-                next_call_child = afr_next_call_child (fresh_children,
-                                                       local->child_up,
-                                                       priv->child_count,
-                                                       last_index, read_child);
-                if (next_call_child < 0)
-                        goto out;
-
-                unwind = 0;
-                STACK_WIND_COOKIE (frame, afr_fgetxattr_cbk,
-                                   (void *) (long) read_child,
-                                   children[next_call_child],
-                                   children[next_call_child]->fops->fgetxattr,
-                                   local->fd,
-                                   local->cont.getxattr.name,
-                                   NULL);
-        }
+    if (op_ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = op_errno;
 
-out:
-        if (unwind) {
-                if (op_ret >= 0 && dict)
-                        __filter_xattrs (dict);
+        afr_read_txn_continue(frame, this, (long)cookie);
+        return 0;
+    }
 
-                AFR_STACK_UNWIND (fgetxattr, frame, op_ret, op_errno, dict,
-                                  xdata);
-        }
+    if (dict)
+        afr_filter_xattrs(dict);
 
-        return 0;
-}
+    AFR_STACK_UNWIND(fgetxattr, frame, op_ret, op_errno, dict, xdata);
 
-int32_t
-afr_fgetxattr_unwind (call_frame_t *frame,
-                      int op_ret, int op_errno, dict_t *dict, dict_t *xdata)
+    return 0;
+}
 
+int
+afr_fgetxattr_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        AFR_STACK_UNWIND (fgetxattr, frame, op_ret, op_errno, dict, xdata);
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    if (subvol == -1) {
+        AFR_STACK_UNWIND(fgetxattr, frame, local->op_ret, local->op_errno, NULL,
+                         NULL);
         return 0;
+    }
+
+    STACK_WIND_COOKIE(frame, afr_fgetxattr_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->fgetxattr, local->fd,
+                      local->cont.getxattr.name, local->xdata_req);
+    return 0;
 }
 
 static void
-afr_fgetxattr_frm_all_children (xlator_t *this, call_frame_t *frame,
-                                const char *name, fd_t *fd,
-                                fop_fgetxattr_cbk_t cbk)
+afr_fgetxattr_all_subvols(xlator_t *this, call_frame_t *frame,
+                          fop_fgetxattr_cbk_t cbk)
 {
-        afr_private_t   *priv           = NULL;
-        afr_local_t     *local          = NULL;
-        xlator_t        **children      = NULL;
-        int             i               = 0;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int i = 0;
+    int call_count = 0;
 
-        priv     = this->private;
-        children = priv->children;
+    priv = this->private;
 
-        local = frame->local;
-        local->call_count = priv->child_count;
+    local = frame->local;
+    // local->call_count set in afr_local_init
+    call_count = local->call_count;
 
-        for (i = 0; i < priv->child_count; i++) {
-                STACK_WIND_COOKIE (frame, cbk,
-                                   (void *) (long) i,
-                                   children[i], children[i]->fops->fgetxattr,
-                                   fd, name, NULL);
+    // If up-children count is 0, afr_local_init would have failed already
+    // and the call would have unwound so not handling it here.
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->child_up[i]) {
+            STACK_WIND_COOKIE(frame, cbk, (void *)(long)i, priv->children[i],
+                              priv->children[i]->fops->fgetxattr, local->fd,
+                              local->cont.getxattr.name, NULL);
+            if (!--call_count)
+                break;
         }
+    }
 
-        return;
+    return;
 }
 
-int32_t
-afr_fgetxattr (call_frame_t *frame, xlator_t *this,
-               fd_t *fd, const char *name, dict_t *xdata)
+int
+afr_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+              dict_t *xdata)
 {
-        afr_private_t        *priv       = NULL;
-        xlator_t            **children   = NULL;
-        int                   call_child = 0;
-        afr_local_t          *local      = NULL;
-        int32_t               op_ret     = -1;
-        int32_t               op_errno   = 0;
-        int32_t               read_child = -1;
-        fop_fgetxattr_cbk_t   cbk        = NULL;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
-
-        priv     = this->private;
-        VALIDATE_OR_GOTO (priv->children, out);
-
-        children = priv->children;
-
-        if (afr_is_split_brain (this, fd->inode)) {
-                op_errno = EIO;
-                goto out;
-        }
-        AFR_LOCAL_ALLOC_OR_GOTO (local, out);
-        frame->local = local;
+    afr_local_t *local = NULL;
+    int32_t op_errno = 0;
+    fop_fgetxattr_cbk_t cbk = NULL;
+
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
+
+    local->op = GF_FOP_FGETXATTR;
+    local->fd = fd_ref(fd);
+    if (name) {
+        local->cont.getxattr.name = gf_strdup(name);
+        if (!local->cont.getxattr.name) {
+            op_errno = ENOMEM;
+            goto out;
+        }
+    }
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
+
+    /* pathinfo gets handled only in getxattr(), but we need to handle
+     * lockinfo.
+     * If we are doing fgetxattr with lockinfo as the key then we
+     * collect information from all children.
+     */
+    if (afr_is_special_xattr(name, &cbk, 1)) {
+        afr_fgetxattr_all_subvols(this, frame, cbk);
+        return 0;
+    }
 
-        op_ret = afr_local_init (local, priv, &op_errno);
-        if (op_ret < 0) {
-                op_errno = -op_ret;
-                goto out;
-        }
+    afr_fix_open(fd, this);
 
-        local->fd = fd_ref (fd);
-        if (name)
-                local->cont.getxattr.name = gf_strdup (name);
+    afr_read_txn(frame, this, fd->inode, afr_fgetxattr_wind,
+                 AFR_METADATA_TRANSACTION);
 
-        /* pathinfo gets handled only in getxattr(), but we need to handle
-         * lockinfo.
-         * If we are doing fgetxattr with lockinfo as the key then we
-         * collect information from all children.
-         */
-        if (afr_is_special_xattr (name, &cbk, 1)) {
-                afr_fgetxattr_frm_all_children (this, frame, name,
-                                                fd, cbk);
-                return 0;
-        }
+    return 0;
+out:
+    AFR_STACK_UNWIND(fgetxattr, frame, -1, op_errno, NULL, NULL);
 
+    return 0;
+}
 
-        local->fresh_children = afr_children_create (priv->child_count);
-        if (!local->fresh_children) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+/* }}} */
 
-        read_child = afr_inode_get_read_ctx (this, fd->inode,
-                                             local->fresh_children);
-        op_ret = afr_get_call_child (this, local->child_up, read_child,
-                                     local->fresh_children,
-                                     &call_child,
-                                     &local->cont.getxattr.last_index);
-        if (op_ret < 0) {
-                op_errno = -op_ret;
-                op_ret = -1;
-                goto out;
-        }
+/* {{{ readv */
 
-        STACK_WIND_COOKIE (frame, afr_fgetxattr_cbk,
-                           (void *) (long) call_child,
-                           children[call_child],
-                           children[call_child]->fops->fgetxattr,
-                           fd, name, xdata);
+int
+afr_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct iovec *vector, int32_t count,
+              struct iatt *buf, struct iobref *iobref, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
 
-        op_ret = 0;
-out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (fgetxattr, frame, op_ret, op_errno, NULL,
-                                  NULL);
-        }
+    local = frame->local;
+
+    if (op_ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+
+        afr_read_txn_continue(frame, this, (long)cookie);
         return 0;
+    }
+
+    AFR_STACK_UNWIND(readv, frame, op_ret, op_errno, vector, count, buf, iobref,
+                     xdata);
+    return 0;
 }
 
+int
+afr_readv_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-/* }}} */
+    local = frame->local;
+    priv = this->private;
 
-/* {{{ readv */
+    if (subvol == -1) {
+        AFR_STACK_UNWIND(readv, frame, local->op_ret, local->op_errno, 0, 0, 0,
+                         0, 0);
+        return 0;
+    }
 
-/**
- * read algorithm:
- *
- * if the user has specified a read subvolume, use it
- * otherwise -
- *   use the inode number to hash it to one of the subvolumes, and
- *   read from there (to balance read load)
- *
- * if any of the above read's fail, try the children in sequence
- * beginning at the beginning
- */
+    STACK_WIND_COOKIE(
+        frame, afr_readv_cbk, (void *)(long)subvol, priv->children[subvol],
+        priv->children[subvol]->fops->readv, local->fd, local->cont.readv.size,
+        local->cont.readv.offset, local->cont.readv.flags, local->xdata_req);
+    return 0;
+}
 
-int32_t
-afr_readv_cbk (call_frame_t *frame, void *cookie,
-               xlator_t *this, int32_t op_ret, int32_t op_errno,
-               struct iovec *vector, int32_t count, struct iatt *buf,
-               struct iobref *iobref, dict_t *xdata)
+int
+afr_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+          off_t offset, uint32_t flags, dict_t *xdata)
 {
-        afr_private_t * priv            = NULL;
-        afr_local_t *   local           = NULL;
-        xlator_t **     children        = NULL;
-        int             unwind          = 1;
-        int32_t         *last_index     = NULL;
-        int32_t         next_call_child = -1;
-        int32_t         *fresh_children  = NULL;
-        int32_t         read_child      = -1;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    afr_local_t *local = NULL;
+    int32_t op_errno = 0;
 
-        priv     = this->private;
-        VALIDATE_OR_GOTO (priv->children, out);
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-        children = priv->children;
+    local->op = GF_FOP_READ;
+    local->fd = fd_ref(fd);
+    local->cont.readv.size = size;
+    local->cont.readv.offset = offset;
+    local->cont.readv.flags = flags;
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
 
-        local = frame->local;
+    afr_fix_open(fd, this);
 
-        read_child = (long) cookie;
-
-        if (op_ret == -1) {
-                last_index = &local->cont.readv.last_index;
-                fresh_children = local->fresh_children;
-                next_call_child = afr_next_call_child (fresh_children,
-                                                       local->child_up,
-                                                       priv->child_count,
-                                                       last_index, read_child);
-                if (next_call_child < 0)
-                        goto out;
-
-                unwind = 0;
-
-                STACK_WIND_COOKIE (frame, afr_readv_cbk,
-                                   (void *) (long) read_child,
-                                   children[next_call_child],
-                                   children[next_call_child]->fops->readv,
-                                   local->fd, local->cont.readv.size,
-                                   local->cont.readv.offset,
-                                   local->cont.readv.flags,
-                                   NULL);
-        }
+    afr_read_txn(frame, this, fd->inode, afr_readv_wind, AFR_DATA_TRANSACTION);
 
+    return 0;
 out:
-        if (unwind) {
-                AFR_STACK_UNWIND (readv, frame, op_ret, op_errno,
-                                  vector, count, buf, iobref, xdata);
-        }
+    AFR_STACK_UNWIND(readv, frame, -1, op_errno, 0, 0, 0, 0, 0);
 
-        return 0;
+    return 0;
 }
 
+/* }}} */
 
-int32_t
-afr_readv (call_frame_t *frame, xlator_t *this,
-           fd_t *fd, size_t size, off_t offset, uint32_t flags, dict_t *xdata)
+/* {{{ seek */
+
+int
+afr_seek_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, off_t offset, dict_t *xdata)
 {
-        afr_private_t * priv       = NULL;
-        afr_local_t   * local      = NULL;
-        xlator_t **     children   = NULL;
-        int             call_child = 0;
-        int32_t         op_errno   = 0;
-        int32_t         read_child = -1;
-        int             ret        = -1;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
-        VALIDATE_OR_GOTO (fd, out);
-
-        priv     = this->private;
-        children = priv->children;
-
-        if (afr_is_split_brain (this, fd->inode)) {
-                op_errno = EIO;
-                goto out;
-        }
+    afr_local_t *local = NULL;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    local = frame->local;
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    if (op_ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = op_errno;
 
-        local->fresh_children = afr_children_create (priv->child_count);
-        if (!local->fresh_children) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+        afr_read_txn_continue(frame, this, (long)cookie);
+        return 0;
+    }
 
-        read_child = afr_inode_get_read_ctx (this, fd->inode, local->fresh_children);
-        ret = afr_get_call_child (this, local->child_up, read_child,
-                                     local->fresh_children,
-                                     &call_child,
-                                     &local->cont.readv.last_index);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+    AFR_STACK_UNWIND(seek, frame, op_ret, op_errno, offset, xdata);
+    return 0;
+}
+
+int
+afr_seek_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
 
-        local->fd                    = fd_ref (fd);
+    if (subvol == -1) {
+        AFR_STACK_UNWIND(seek, frame, local->op_ret, local->op_errno, 0, NULL);
+        return 0;
+    }
+
+    STACK_WIND_COOKIE(
+        frame, afr_seek_cbk, (void *)(long)subvol, priv->children[subvol],
+        priv->children[subvol]->fops->seek, local->fd, local->cont.seek.offset,
+        local->cont.seek.what, local->xdata_req);
+    return 0;
+}
 
-        local->cont.readv.size       = size;
-        local->cont.readv.offset     = offset;
-        local->cont.readv.flags      = flags;
+int
+afr_seek(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+         gf_seek_what_t what, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int32_t op_errno = 0;
 
-        afr_open_fd_fix (fd, this);
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-        STACK_WIND_COOKIE (frame, afr_readv_cbk,
-                           (void *) (long) call_child,
-                           children[call_child],
-                           children[call_child]->fops->readv,
-                           fd, size, offset, flags, xdata);
+    local->op = GF_FOP_SEEK;
+    local->fd = fd_ref(fd);
+    local->cont.seek.offset = offset;
+    local->cont.seek.what = what;
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
 
-        ret = 0;
+    afr_fix_open(fd, this);
+
+    afr_read_txn(frame, this, fd->inode, afr_seek_wind, AFR_DATA_TRANSACTION);
+
+    return 0;
 out:
-        if (ret < 0) {
-                AFR_STACK_UNWIND (readv, frame, -1, op_errno, NULL, 0, NULL,
-                                  NULL, NULL);
-        }
-        return 0;
-}
+    AFR_STACK_UNWIND(seek, frame, -1, op_errno, 0, NULL);
 
+    return 0;
+}
 /* }}} */
diff --git a/xlators/cluster/afr/src/afr-inode-read.h b/xlators/cluster/afr/src/afr-inode-read.h
index e4091a793e0..8c982bc7e6f 100644
--- a/xlators/cluster/afr/src/afr-inode-read.h
+++ b/xlators/cluster/afr/src/afr-inode-read.h
@@ -12,31 +12,34 @@
 #define __INODE_READ_H__
 
 int32_t
-afr_access (call_frame_t *frame, xlator_t *this,
-	    loc_t *loc, int32_t mask, dict_t *xdata);
+afr_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+           dict_t *xdata);
 
 int32_t
-afr_stat (call_frame_t *frame, xlator_t *this,
-	  loc_t *loc, dict_t *xdata);
+afr_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata);
 
 int32_t
-afr_fstat (call_frame_t *frame, xlator_t *this,
-	   fd_t *fd, dict_t *xdata);
+afr_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata);
 
 int32_t
-afr_readlink (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, size_t size, dict_t *xdata);
+afr_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+             dict_t *xdata);
 
 int32_t
-afr_readv (call_frame_t *frame, xlator_t *this,
-	   fd_t *fd, size_t size, off_t offset, uint32_t flags, dict_t *xdata);
+afr_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+          off_t offset, uint32_t flags, dict_t *xdata);
 
 int32_t
-afr_getxattr (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, const char *name, dict_t *xdata);
+afr_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, const char *name,
+             dict_t *xdata);
 
 int32_t
-afr_fgetxattr (call_frame_t *frame, xlator_t *this,
-               fd_t *fd, const char *name, dict_t *xdata);
-
+afr_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+              dict_t *xdata);
+
+int
+afr_seek(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+         gf_seek_what_t what, dict_t *xdata);
+int
+afr_handle_quota_size(call_frame_t *frame, xlator_t *this);
 #endif /* __INODE_READ_H__ */
diff --git a/xlators/cluster/afr/src/afr-inode-write.c b/xlators/cluster/afr/src/afr-inode-write.c
index 5f2d3096f66..1d6e4f3570a 100644
--- a/xlators/cluster/afr/src/afr-inode-write.c
+++ b/xlators/cluster/afr/src/afr-inode-write.c
@@ -8,998 +8,779 @@
   cases as published by the Free Software Foundation.
 */
 
-
-#include <libgen.h>
 #include <unistd.h>
-#include <fnmatch.h>
 #include <sys/time.h>
 #include <stdlib.h>
 #include <signal.h>
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "afr.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
-
+#include <glusterfs/glusterfs.h>
 #include "afr.h"
+#include <glusterfs/dict.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/compat.h>
+#include "protocol-common.h"
+#include <glusterfs/byte-order.h>
 #include "afr-transaction.h"
-#include "afr-self-heal-common.h"
+#include "afr-self-heal.h"
+#include "afr-messages.h"
 
-void
-__inode_write_fop_cbk (call_frame_t *frame, int child_index, int read_child,
-                       xlator_t *this, int32_t *op_ret, int32_t *op_errno,
-                       struct iatt *prebuf, struct iatt *postbuf, dict_t *xdata)
-{
-        afr_local_t     *local = NULL;
-
-        local = frame->local;
-
-        if (afr_fop_failed (*op_ret, *op_errno)) {
-                local->child_errno[child_index] = *op_errno;
-
-                switch (local->op) {
-                case GF_FOP_TRUNCATE:
-                case GF_FOP_FTRUNCATE:
-                        if (*op_errno != EFBIG)
-                                afr_transaction_fop_failed (frame, this,
-                                                            child_index);
+static void
+__afr_inode_write_finalize(call_frame_t *frame, xlator_t *this)
+{
+    int i = 0;
+    int ret = 0;
+    int read_subvol = 0;
+    struct iatt *stbuf = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    afr_read_subvol_args_t args = {
+        0,
+    };
+
+    local = frame->local;
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, local->inode, out);
+
+    /*This code needs to stay till DHT sends fops on linked
+     * inodes*/
+    if (!inode_is_linked(local->inode)) {
+        for (i = 0; i < priv->child_count; i++) {
+            if (!local->replies[i].valid)
+                continue;
+            if (local->replies[i].op_ret == -1)
+                continue;
+            if (!gf_uuid_is_null(local->replies[i].poststat.ia_gfid)) {
+                gf_uuid_copy(args.gfid, local->replies[i].poststat.ia_gfid);
+                args.ia_type = local->replies[i].poststat.ia_type;
                 break;
-                default:
-                        afr_transaction_fop_failed (frame, this, child_index);
+            } else {
+                ret = dict_get_bin(local->replies[i].xdata,
+                                   DHT_IATT_IN_XDATA_KEY, (void **)&stbuf);
+                if (ret)
+                    continue;
+                gf_uuid_copy(args.gfid, stbuf->ia_gfid);
+                args.ia_type = stbuf->ia_type;
                 break;
-                }
-                local->op_errno = *op_errno;
-                goto out;
-        }
-
-        if ((local->success_count == 0) || (read_child == child_index)) {
-                local->op_ret              = *op_ret;
-                if (prebuf)
-                        local->cont.inode_wfop.prebuf  = *prebuf;
-                if (postbuf)
-                        local->cont.inode_wfop.postbuf = *postbuf;
-        }
-
-        local->success_count++;
+            }
+        }
+    }
+
+    if (local->transaction.type == AFR_METADATA_TRANSACTION) {
+        read_subvol = afr_metadata_subvol_get(local->inode, this, NULL,
+                                              local->readable, NULL, &args);
+    } else {
+        read_subvol = afr_data_subvol_get(local->inode, this, NULL,
+                                          local->readable, NULL, &args);
+    }
+
+    local->op_ret = -1;
+    local->op_errno = afr_final_errno(local, priv);
+    afr_pick_error_xdata(local, priv, local->inode, local->readable, NULL,
+                         NULL);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
+        if (local->replies[i].op_ret < 0)
+            continue;
+
+        /* Order of checks in the compound conditional
+           below is important.
+
+           - Highest precedence: largest op_ret
+           - Next precedence: if all op_rets are equal, read subvol
+           - Least precedence: any succeeded subvol
+        */
+        if ((local->op_ret < local->replies[i].op_ret) ||
+            ((local->op_ret == local->replies[i].op_ret) &&
+             (i == read_subvol))) {
+            local->op_ret = local->replies[i].op_ret;
+            local->op_errno = local->replies[i].op_errno;
+
+            local->cont.inode_wfop.prebuf = local->replies[i].prestat;
+            local->cont.inode_wfop.postbuf = local->replies[i].poststat;
+
+            if (local->replies[i].xdata) {
+                if (local->xdata_rsp)
+                    dict_unref(local->xdata_rsp);
+                local->xdata_rsp = dict_ref(local->replies[i].xdata);
+            }
+            if (local->replies[i].xattr) {
+                if (local->xattr_rsp)
+                    dict_unref(local->xattr_rsp);
+                local->xattr_rsp = dict_ref(local->replies[i].xattr);
+            }
+        }
+    }
+
+    afr_set_in_flight_sb_status(this, frame, local->inode);
 out:
-        return;
+    return;
 }
 
-/* {{{ writev */
-
-void
-afr_writev_copy_outvars (call_frame_t *src_frame, call_frame_t *dst_frame)
+static void
+__afr_inode_write_fill(call_frame_t *frame, xlator_t *this, int child_index,
+                       int op_ret, int op_errno, struct iatt *prebuf,
+                       struct iatt *postbuf, dict_t *xattr, dict_t *xdata)
 {
-        afr_local_t *src_local = NULL;
-        afr_local_t *dst_local = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        src_local = src_frame->local;
-        dst_local = dst_frame->local;
+    local = frame->local;
+    priv = this->private;
 
-        dst_local->op_ret = src_local->op_ret;
-        dst_local->op_errno = src_local->op_errno;
-        dst_local->cont.inode_wfop.prebuf = src_local->cont.inode_wfop.prebuf;
-        dst_local->cont.inode_wfop.postbuf = src_local->cont.inode_wfop.postbuf;
-}
+    local->replies[child_index].valid = 1;
 
-void
-afr_writev_unwind (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *   local = NULL;
-        local = frame->local;
+    if (AFR_IS_ARBITER_BRICK(priv, child_index) && op_ret == 1)
+        op_ret = iov_length(local->cont.writev.vector,
+                            local->cont.writev.count);
+
+    local->replies[child_index].op_ret = op_ret;
+    local->replies[child_index].op_errno = op_errno;
+    if (xdata)
+        local->replies[child_index].xdata = dict_ref(xdata);
 
-        AFR_STACK_UNWIND (writev, frame,
-                          local->op_ret, local->op_errno,
-                          &local->cont.inode_wfop.prebuf,
-                          &local->cont.inode_wfop.postbuf,
-                          NULL);
+    if (op_ret >= 0) {
+        if (prebuf)
+            local->replies[child_index].prestat = *prebuf;
+        if (postbuf)
+            local->replies[child_index].poststat = *postbuf;
+        if (xattr)
+            local->replies[child_index].xattr = dict_ref(xattr);
+    } else {
+        afr_transaction_fop_failed(frame, this, child_index);
+    }
+
+    return;
 }
 
-call_frame_t*
-afr_transaction_detach_fop_frame (call_frame_t *frame)
+static int
+__afr_inode_write_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                      struct iatt *postbuf, dict_t *xattr, dict_t *xdata)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t   *fop_frame = NULL;
+    afr_local_t *local = NULL;
+    int child_index = (long)cookie;
+    int call_count = -1;
+    afr_private_t *priv = NULL;
 
-        local = frame->local;
+    priv = this->private;
+    local = frame->local;
 
-        LOCK (&frame->lock);
-        {
-                fop_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
+    LOCK(&frame->lock);
+    {
+        __afr_inode_write_fill(frame, this, child_index, op_ret, op_errno,
+                               prebuf, postbuf, xattr, xdata);
+        call_count = --local->call_count;
+    }
+    UNLOCK(&frame->lock);
+
+    if (call_count == 0) {
+        __afr_inode_write_finalize(frame, this);
+
+        if (afr_txn_nothing_failed(frame, this)) {
+            /*if it did pre-op, it will do post-op changing ctime*/
+            if (priv->consistent_metadata && afr_needs_changelog_update(local))
+                afr_zero_fill_stat(local);
+            local->transaction.unwind(frame, this);
         }
-        UNLOCK (&frame->lock);
 
-        return fop_frame;
+        afr_transaction_resume(frame, this);
+    }
+
+    return 0;
 }
 
-int
-afr_transaction_writev_unwind (call_frame_t *frame, xlator_t *this)
+/* {{{ writev */
+
+void
+afr_writev_copy_outvars(call_frame_t *src_frame, call_frame_t *dst_frame)
 {
-        call_frame_t *fop_frame = NULL;
+    afr_local_t *src_local = NULL;
+    afr_local_t *dst_local = NULL;
 
-        fop_frame = afr_transaction_detach_fop_frame (frame);
+    src_local = src_frame->local;
+    dst_local = dst_frame->local;
 
-        if (fop_frame) {
-                afr_writev_copy_outvars (frame, fop_frame);
-                afr_writev_unwind (fop_frame, this);
-        }
-        return 0;
+    dst_local->op_ret = src_local->op_ret;
+    dst_local->op_errno = src_local->op_errno;
+    dst_local->cont.inode_wfop.prebuf = src_local->cont.inode_wfop.prebuf;
+    dst_local->cont.inode_wfop.postbuf = src_local->cont.inode_wfop.postbuf;
+    if (src_local->xdata_rsp)
+        dst_local->xdata_rsp = dict_ref(src_local->xdata_rsp);
 }
 
-static void
-afr_writev_handle_short_writes (call_frame_t *frame, xlator_t *this)
+void
+afr_writev_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t   *local = NULL;
-        afr_private_t *priv  = NULL;
-        int           i      = 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = this->private;
 
-        local = frame->local;
-        priv = this->private;
-        /*
-         * We already have the best case result of the writev calls staged
-         * as the return value. Any writev that returns some value less
-         * than the best case is now out of sync, so mark the fop as
-         * failed. Note that fops that have returned with errors have
-         * already been marked as failed.
-         */
-        for (i = 0; i < priv->child_count; i++) {
-                if ((!local->replies[i].valid) ||
-                    (local->replies[i].op_ret == -1))
-                        continue;
+    local = frame->local;
 
-                if (local->replies[i].op_ret < local->op_ret)
-                        afr_transaction_fop_failed(frame, this, i);
-        }
-}
+    if (priv->consistent_metadata)
+        afr_zero_fill_stat(local);
 
-int
-afr_writev_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                     struct iatt *postbuf, dict_t *xdata)
-{
-        afr_local_t *   local = NULL;
-        afr_private_t  *priv  = NULL;
-        call_frame_t    *fop_frame = NULL;
-        int child_index = (long) cookie;
-        int call_count  = -1;
-        int read_child  = 0;
-        int      ret = 0;
-        uint32_t open_fd_count = 0;
-        uint32_t write_is_append = 0;
-
-        local = frame->local;
-        priv  = this->private;
-
-        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
-
-        LOCK (&frame->lock);
-        {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
-
-                __inode_write_fop_cbk (frame, child_index, read_child, this,
-                                       &op_ret, &op_errno, prebuf, postbuf,
-                                       xdata);
-
-		local->replies[child_index].valid = 1;
-		local->replies[child_index].op_ret = op_ret;
-		local->replies[child_index].op_errno = op_errno;
-
-
-		/* stage the best case return value for unwind */
-                if ((local->success_count == 0) || (op_ret > local->op_ret)) {
-                        local->op_ret              = op_ret;
-			local->op_errno		   = op_errno;
-		}
-
-		if (op_ret != -1) {
-                        if (xdata) {
-                                ret = dict_get_uint32 (xdata,
-                                                       GLUSTERFS_OPEN_FD_COUNT,
-                                                       &open_fd_count);
-                                if ((ret == 0) &&
-                                    (open_fd_count > local->open_fd_count)) {
-                                        local->open_fd_count = open_fd_count;
-                                        local->update_open_fd_count = _gf_true;
-                                }
-
-				write_is_append = 0;
-                                ret = dict_get_uint32 (xdata,
-                                                       GLUSTERFS_WRITE_IS_APPEND,
-                                                       &write_is_append);
-                                if (ret || !write_is_append)
-					local->append_write = _gf_false;
-                        }
-
-		}
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-
-                if (local->update_open_fd_count)
-                        afr_handle_open_fd_count (frame, this);
-
-                if (!local->stable_write && !local->append_write)
-			/* An appended write removes the necessity to
-			   fsync() the file. This is because self-heal
-			   has the logic to check for larger file when
-			   the xattrs are not reliably pointing at
-			   a stale file.
-			*/
-                        afr_fd_report_unstable_write (this, local->fd);
-
-                afr_writev_handle_short_writes (frame, this);
-                if (afr_any_fops_failed (local, priv)) {
-                        //Don't unwind until post-op is complete
-                        local->transaction.resume (frame, this);
-                } else {
-                /*
-                 * Generally inode-write fops do transaction.unwind then
-                 * transaction.resume, but writev needs to make sure that
-                 * delayed post-op frame is placed in fdctx before unwind
-                 * happens. This prevents the race of flush doing the
-                 * changelog wakeup first in fuse thread and then this
-                 * writev placing its delayed post-op frame in fdctx.
-                 * This helps flush make sure all the delayed post-ops are
-                 * completed.
-                 */
-
-                        fop_frame = afr_transaction_detach_fop_frame (frame);
-                        afr_writev_copy_outvars (frame, fop_frame);
-                        local->transaction.resume (frame, this);
-                        afr_writev_unwind (fop_frame, this);
-                }
-        }
-        return 0;
+    AFR_STACK_UNWIND(writev, frame, local->op_ret, local->op_errno,
+                     &local->cont.inode_wfop.prebuf,
+                     &local->cont.inode_wfop.postbuf, local->xdata_rsp);
 }
 
 int
-afr_writev_wind (call_frame_t *frame, xlator_t *this)
+afr_transaction_writev_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int i = 0;
-        int call_count = -1;
-        dict_t *xdata = NULL;
-        GF_UNUSED int     ret = 0;
+    call_frame_t *fop_frame = NULL;
 
-        local = frame->local;
-        priv = this->private;
+    fop_frame = afr_transaction_detach_fop_frame(frame);
 
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
+    if (fop_frame) {
+        afr_writev_copy_outvars(frame, fop_frame);
+        afr_writev_unwind(fop_frame, this);
+    }
+    return 0;
+}
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
+static void
+afr_writev_handle_short_writes(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+
+    local = frame->local;
+    priv = this->private;
+    /*
+     * We already have the best case result of the writev calls staged
+     * as the return value. Any writev that returns some value less
+     * than the best case is now out of sync, so mark the fop as
+     * failed. Note that fops that have returned with errors have
+     * already been marked as failed.
+     */
+    for (i = 0; i < priv->child_count; i++) {
+        if ((!local->replies[i].valid) || (local->replies[i].op_ret == -1))
+            continue;
+
+        if (local->replies[i].op_ret < local->op_ret)
+            afr_transaction_fop_failed(frame, this, i);
+    }
+}
 
-        local->call_count = call_count;
-	local->replies = GF_CALLOC(priv->child_count, sizeof(*local->replies),
-				   gf_afr_mt_reply_t);
-	if (!local->replies) {
-		local->op_ret = -1;
-		local->op_errno = ENOMEM;
-		local->transaction.unwind(frame, this);
-		local->transaction.resume(frame, this);
-		return 0;
-	}
-
-        xdata = dict_new ();
-        if (xdata) {
-                ret = dict_set_uint32 (xdata, GLUSTERFS_OPEN_FD_COUNT,
-                                       sizeof (uint32_t));
-		ret = dict_set_uint32 (xdata, GLUSTERFS_WRITE_IS_APPEND,
-				       0);
-		/* Set append_write to be true speculatively. If on any
-		   server it turns not be true, we unset it in the
-		   callback.
-		*/
-		local->append_write = _gf_true;
+void
+afr_inode_write_fill(call_frame_t *frame, xlator_t *this, int child_index,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
+{
+    int ret = 0;
+    afr_local_t *local = frame->local;
+    uint32_t open_fd_count = 0;
+    uint32_t write_is_append = 0;
+    int32_t num_inodelks = 0;
+
+    LOCK(&frame->lock);
+    {
+        __afr_inode_write_fill(frame, this, child_index, op_ret, op_errno,
+                               prebuf, postbuf, NULL, xdata);
+        if (op_ret == -1 || !xdata)
+            goto unlock;
+
+        write_is_append = 0;
+        ret = dict_get_uint32(xdata, GLUSTERFS_WRITE_IS_APPEND,
+                              &write_is_append);
+        if (ret || !write_is_append)
+            local->append_write = _gf_false;
+
+        ret = dict_get_uint32(xdata, GLUSTERFS_ACTIVE_FD_COUNT, &open_fd_count);
+        if (ret < 0)
+            goto unlock;
+        if (open_fd_count > local->open_fd_count) {
+            local->open_fd_count = open_fd_count;
+            local->update_open_fd_count = _gf_true;
         }
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->transaction.pre_op[i]) {
-                        STACK_WIND_COOKIE (frame, afr_writev_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->writev,
-                                           local->fd,
-                                           local->cont.writev.vector,
-                                           local->cont.writev.count,
-                                           local->cont.writev.offset,
-                                           local->cont.writev.flags,
-                                           local->cont.writev.iobref,
-                                           xdata);
-
-                        if (!--call_count)
-                                break;
-                }
+        ret = dict_get_int32_sizen(xdata, GLUSTERFS_INODELK_COUNT,
+                                   &num_inodelks);
+        if (ret < 0)
+            goto unlock;
+        if (num_inodelks > local->num_inodelks) {
+            local->num_inodelks = num_inodelks;
+            local->update_num_inodelks = _gf_true;
         }
-
-        if (xdata)
-                dict_unref (xdata);
-
-        return 0;
+    }
+unlock:
+    UNLOCK(&frame->lock);
 }
 
-
-int
-afr_writev_done (call_frame_t *frame, xlator_t *this)
+void
+afr_process_post_writev(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *local = NULL;
+    afr_local_t *local = NULL;
+    afr_lock_t *lock = NULL;
 
-        local = frame->local;
+    local = frame->local;
 
-        iobref_unref (local->cont.writev.iobref);
-        local->cont.writev.iobref = NULL;
+    if (!local->stable_write && !local->append_write)
+        /* An appended write removes the necessity to
+           fsync() the file. This is because self-heal
+           has the logic to check for larger file when
+           the xattrs are not reliably pointing at
+           a stale file.
+        */
+        afr_fd_report_unstable_write(this, local);
 
-        local->transaction.unwind (frame, this);
+    __afr_inode_write_finalize(frame, this);
 
-        AFR_STACK_DESTROY (frame);
+    afr_writev_handle_short_writes(frame, this);
 
-        return 0;
+    if (local->update_open_fd_count)
+        local->inode_ctx->open_fd_count = local->open_fd_count;
+    if (local->update_num_inodelks &&
+        local->transaction.type == AFR_DATA_TRANSACTION) {
+        lock = &local->inode_ctx->lock[local->transaction.type];
+        lock->num_inodelks = local->num_inodelks;
+    }
 }
 
-
 int
-afr_do_writev (call_frame_t *frame, xlator_t *this)
+afr_writev_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                    struct iatt *postbuf, dict_t *xdata)
 {
-        call_frame_t    *transaction_frame = NULL;
-        afr_local_t     *local             = NULL;
-        int             op_ret   = -1;
-        int             op_errno = 0;
-
-        local = frame->local;
-
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                op_errno = ENOMEM;
-                goto out;
-        }
-
-        transaction_frame->local = local;
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+    call_frame_t *fop_frame = NULL;
+    int child_index = (long)cookie;
+    int call_count = -1;
 
-        local->op = GF_FOP_WRITE;
+    afr_inode_write_fill(frame, this, child_index, op_ret, op_errno, prebuf,
+                         postbuf, xdata);
 
-        local->success_count      = 0;
+    call_count = afr_frame_return(frame);
 
-        local->transaction.fop    = afr_writev_wind;
-        local->transaction.done   = afr_writev_done;
-        local->transaction.unwind = afr_transaction_writev_unwind;
+    if (call_count == 0) {
+        afr_process_post_writev(frame, this);
 
-        local->transaction.main_frame = frame;
-        if (local->fd->flags & O_APPEND) {
-               /*
-                * Backend vfs ignores the 'offset' for append mode fd so
-                * locking just the region provided for the writev does not
-                * give consistency gurantee. The actual write may happen at a
-                * completely different range than the one provided by the
-                * offset, len in the fop. So lock the entire file.
-                */
-                local->transaction.start   = 0;
-                local->transaction.len     = 0;
+        if (!afr_txn_nothing_failed(frame, this)) {
+            // Don't unwind until post-op is complete
+            afr_transaction_resume(frame, this);
         } else {
-                local->transaction.start   = local->cont.writev.offset;
-                local->transaction.len     = iov_length (local->cont.writev.vector,
-                                                         local->cont.writev.count);
-        }
-
-        op_ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
-        if (op_ret < 0) {
-            op_errno = -op_ret;
-            goto out;
-        }
-
-        op_ret = 0;
-out:
-        if (op_ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (writev, frame, op_ret, op_errno, NULL, NULL, NULL);
-        }
-
-        return 0;
+            /*
+             * Generally inode-write fops do transaction.unwind then
+             * transaction.resume, but writev needs to make sure that
+             * delayed post-op frame is placed in fdctx before unwind
+             * happens. This prevents the race of flush doing the
+             * changelog wakeup first in fuse thread and then this
+             * writev placing its delayed post-op frame in fdctx.
+             * This helps flush make sure all the delayed post-ops are
+             * completed.
+             */
+
+            fop_frame = afr_transaction_detach_fop_frame(frame);
+            afr_writev_copy_outvars(frame, fop_frame);
+            afr_transaction_resume(frame, this);
+            afr_writev_unwind(fop_frame, this);
+        }
+    }
+    return 0;
 }
 
-static void
-afr_trigger_open_fd_self_heal (fd_t *fd, xlator_t *this)
-{
-        call_frame_t    *frame   = NULL;
-        afr_local_t     *local   = NULL;
-        afr_self_heal_t *sh      = NULL;
-        char            *reason  = NULL;
-        int32_t         op_errno = 0;
-        int             ret      = 0;
-
-        if (!fd || !fd->inode || uuid_is_null (fd->inode->gfid)) {
-                gf_log_callingfn (this->name, GF_LOG_ERROR, "Invalid args: "
-                                  "fd: %p, inode: %p", fd,
-                                  fd ? fd->inode : NULL);
-                goto out;
-        }
-
-        frame = create_frame (this, this->ctx->pool);
-        if (!frame)
-                goto out;
-
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
-        ret = afr_local_init (local, this->private, &op_errno);
-        if (ret < 0)
-                goto out;
+static int
+afr_arbiter_writev_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = frame->local;
+    afr_private_t *priv = this->private;
+    static char byte = 0xFF;
+    static struct iovec vector = {&byte, 1};
+    int32_t count = 1;
 
-        local->loc.inode = inode_ref (fd->inode);
-        ret = loc_path (&local->loc, NULL);
-        if (ret < 0)
-                goto out;
+    STACK_WIND_COOKIE(
+        frame, afr_writev_wind_cbk, (void *)(long)subvol,
+        priv->children[subvol], priv->children[subvol]->fops->writev, local->fd,
+        &vector, count, local->cont.writev.offset, local->cont.writev.flags,
+        local->cont.writev.iobref, local->xdata_req);
 
-        sh    = &local->self_heal;
-        sh->do_metadata_self_heal = _gf_true;
-        if (fd->inode->ia_type == IA_IFREG)
-                sh->do_data_self_heal = _gf_true;
-        else if (fd->inode->ia_type == IA_IFDIR)
-                sh->do_entry_self_heal = _gf_true;
-
-        reason = "subvolume came online";
-        afr_launch_self_heal (frame, this, fd->inode, _gf_true,
-                              fd->inode->ia_type, reason, NULL, NULL);
-        return;
-out:
-        AFR_STACK_DESTROY (frame);
+    return 0;
 }
 
-void
-afr_open_fd_fix (fd_t *fd, xlator_t *this)
+int
+afr_writev_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        int           ret             = 0;
-        int           i               = 0;
-        afr_fd_ctx_t  *fd_ctx         = NULL;
-        gf_boolean_t  need_self_heal  = _gf_false;
-        int           *need_open      = NULL;
-        size_t        need_open_count = 0;
-        afr_private_t *priv           = NULL;
-
-        priv  = this->private;
-
-        if (!afr_is_fd_fixable (fd))
-                goto out;
-
-        fd_ctx = afr_fd_ctx_get (fd, this);
-        if (!fd_ctx)
-                goto out;
-
-        LOCK (&fd->lock);
-        {
-                if (fd_ctx->up_count < priv->up_count) {
-                        need_self_heal = _gf_true;
-                        fd_ctx->up_count   = priv->up_count;
-                        fd_ctx->down_count = priv->down_count;
-                }
-
-                need_open = alloca (priv->child_count * sizeof (*need_open));
-                for (i = 0; i < priv->child_count; i++) {
-                        need_open[i] = 0;
-                        if (fd_ctx->opened_on[i] != AFR_FD_NOT_OPENED)
-                                continue;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-                        if (!priv->child_up[i])
-                                continue;
+    local = frame->local;
+    priv = this->private;
 
-                        fd_ctx->opened_on[i] = AFR_FD_OPENING;
-
-                        need_open[i] = 1;
-                        need_open_count++;
-                }
-        }
-        UNLOCK (&fd->lock);
-        if (ret)
-                goto out;
-
-        if (need_self_heal)
-                afr_trigger_open_fd_self_heal (fd, this);
-
-        if (!need_open_count)
-                goto out;
+    if (AFR_IS_ARBITER_BRICK(priv, subvol)) {
+        afr_arbiter_writev_wind(frame, this, subvol);
+        return 0;
+    }
 
-        afr_fix_open (this, fd, need_open_count, need_open);
-out:
-        return;
+    STACK_WIND_COOKIE(frame, afr_writev_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->writev, local->fd,
+                      local->cont.writev.vector, local->cont.writev.count,
+                      local->cont.writev.offset, local->cont.writev.flags,
+                      local->cont.writev.iobref, local->xdata_req);
+    return 0;
 }
 
 int
-afr_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
-            struct iovec *vector, int32_t count, off_t offset,
-            uint32_t flags, struct iobref *iobref, dict_t *xdata)
+afr_do_writev(call_frame_t *frame, xlator_t *this)
 {
-        afr_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
-        int ret = -1;
-        int op_errno = 0;
+    call_frame_t *transaction_frame = NULL;
+    afr_local_t *local = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        priv = this->private;
+    local = frame->local;
+    transaction_frame->local = local;
+    frame->local = NULL;
 
-        if (afr_is_split_brain (this, fd->inode)) {
-                op_errno = EIO;
-                goto out;
-        }
+    if (!AFR_FRAME_INIT(frame, op_errno))
+        goto out;
 
-        QUORUM_CHECK(writev,out);
+    local->op = GF_FOP_WRITE;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    local->transaction.wind = afr_writev_wind;
+    local->transaction.unwind = afr_transaction_writev_unwind;
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
-
-        local->cont.writev.vector     = iov_dup (vector, count);
-        local->cont.writev.count      = count;
-        local->cont.writev.offset     = offset;
-        local->cont.writev.flags      = flags;
-        local->cont.writev.iobref     = iobref_ref (iobref);
-
-        local->fd                = fd_ref (fd);
+    local->transaction.main_frame = frame;
 
-	/* detect here, but set it in writev_wind_cbk *after* the unstable
-	   write is performed
-	*/
-	local->stable_write = !!((fd->flags|flags)&(O_SYNC|O_DSYNC));
-
-        afr_open_fd_fix (fd, this);
+    if (local->fd->flags & O_APPEND) {
+        /*
+         * Backend vfs ignores the 'offset' for append mode fd so
+         * locking just the region provided for the writev does not
+         * give consistency guarantee. The actual write may happen at a
+         * completely different range than the one provided by the
+         * offset, len in the fop. So lock the entire file.
+         */
+        local->transaction.start = 0;
+        local->transaction.len = 0;
+    } else {
+        local->transaction.start = local->cont.writev.offset;
+        local->transaction.len = iov_length(local->cont.writev.vector,
+                                            local->cont.writev.count);
+    }
+
+    ret = afr_transaction(transaction_frame, this, AFR_DATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
+
+    return 0;
+out:
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        afr_do_writev (frame, this);
+    AFR_STACK_UNWIND(writev, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
+}
 
-        ret = 0;
+int
+afr_writev(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,
+           int32_t count, off_t offset, uint32_t flags, struct iobref *iobref,
+           dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int op_errno = ENOMEM;
+    int ret = -1;
+
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
+
+    local->cont.writev.vector = iov_dup(vector, count);
+    if (!local->cont.writev.vector)
+        goto out;
+    local->cont.writev.count = count;
+    local->cont.writev.offset = offset;
+    local->cont.writev.flags = flags;
+    local->cont.writev.iobref = iobref_ref(iobref);
+
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
+
+    if (!local->xdata_req)
+        goto out;
+
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
+
+    if (dict_set_uint32(local->xdata_req, GLUSTERFS_ACTIVE_FD_COUNT, 4)) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    if (dict_set_str_sizen(local->xdata_req, GLUSTERFS_INODELK_DOM_COUNT,
+                           this->name)) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    if (dict_set_uint32(local->xdata_req, GLUSTERFS_WRITE_IS_APPEND, 4)) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    /* Set append_write to be true speculatively. If on any
+       server it turns not be true, we unset it in the
+       callback.
+    */
+    local->append_write = _gf_true;
+
+    /* detect here, but set it in writev_wind_cbk *after* the unstable
+       write is performed
+    */
+    local->stable_write = !!((fd->flags | flags) & (O_SYNC | O_DSYNC));
+
+    afr_fix_open(fd, this);
+
+    afr_do_writev(frame, this);
+
+    return 0;
 out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (writev, frame, -1, op_errno, NULL, NULL, NULL);
+    AFR_STACK_UNWIND(writev, frame, -1, op_errno, NULL, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
-
 /* }}} */
 
 /* {{{ truncate */
 
 int
-afr_truncate_unwind (call_frame_t *frame, xlator_t *this)
+afr_truncate_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t   *main_frame = NULL;
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame)
-                        main_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
-
-        if (main_frame) {
-                AFR_STACK_UNWIND (truncate, main_frame, local->op_ret,
-                                  local->op_errno,
-                                  &local->cont.inode_wfop.prebuf,
-                                  &local->cont.inode_wfop.postbuf,
-                                  NULL);
-        }
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
-}
-
 
-int
-afr_truncate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                       struct iatt *postbuf, dict_t *xdata)
-{
-        afr_local_t *   local = NULL;
-        int child_index = (long) cookie;
-        int read_child  = 0;
-        int call_count  = -1;
-
-        local = frame->local;
-
-        read_child = afr_inode_get_read_ctx (this, local->loc.inode, NULL);
-
-        LOCK (&frame->lock);
-        {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
-
-                if (op_ret != -1) {
-			if (prebuf->ia_size != postbuf->ia_size)
-				local->stable_write = _gf_false;
-                }
-                __inode_write_fop_cbk (frame, child_index, read_child, this,
-                                       &op_ret, &op_errno, prebuf, postbuf,
-                                       xdata);
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-		if (local->stable_write && afr_txn_nothing_failed (frame, this))
-			local->transaction.unwind (frame, this);
-
-                local->transaction.resume (frame, this);
-        }
-
-        return 0;
+    AFR_STACK_UNWIND(truncate, main_frame, local->op_ret, local->op_errno,
+                     &local->cont.inode_wfop.prebuf,
+                     &local->cont.inode_wfop.postbuf, local->xdata_rsp);
+    return 0;
 }
 
-
-int32_t
-afr_truncate_wind (call_frame_t *frame, xlator_t *this)
+int
+afr_truncate_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                      struct iatt *postbuf, dict_t *xdata)
 {
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int call_count = -1;
-        int i = 0;
-
-        local = frame->local;
-        priv = this->private;
-
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
-
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
+    afr_local_t *local = NULL;
 
-        local->call_count = call_count;
-	local->stable_write = _gf_true;
+    local = frame->local;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->transaction.pre_op[i]) {
-                        STACK_WIND_COOKIE (frame, afr_truncate_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->truncate,
-                                           &local->loc,
-                                           local->cont.truncate.offset,
-                                           NULL);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
+    if (op_ret == 0 && prebuf->ia_size != postbuf->ia_size)
+        local->stable_write = _gf_false;
 
-        return 0;
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, prebuf,
+                                 postbuf, NULL, xdata);
 }
 
-
 int
-afr_truncate_done (call_frame_t *frame, xlator_t *this)
+afr_truncate_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_local_t *local = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        local = frame->local;
+    local = frame->local;
+    priv = this->private;
 
-        local->transaction.unwind (frame, this);
-
-        AFR_STACK_DESTROY (frame);
-
-        return 0;
+    STACK_WIND_COOKIE(frame, afr_truncate_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->truncate, &local->loc,
+                      local->cont.truncate.offset, local->xdata_req);
+    return 0;
 }
 
-
 int
-afr_truncate (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, off_t offset, dict_t *xdata)
+afr_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+             dict_t *xdata)
 {
-        afr_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
-        call_frame_t   *transaction_frame = NULL;
-        int ret = -1;
-        int op_errno = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        priv = this->private;
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        QUORUM_CHECK(truncate,out);
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+    local->cont.truncate.offset = offset;
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
-        local = transaction_frame->local;
+    if (!local->xdata_req)
+        goto out;
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    local->transaction.wind = afr_truncate_wind;
+    local->transaction.unwind = afr_truncate_unwind;
 
-        local->cont.truncate.offset  = offset;
+    loc_copy(&local->loc, loc);
+    ret = afr_set_inode_local(this, local, loc->inode);
+    if (ret)
+        goto out;
 
-        local->transaction.fop    = afr_truncate_wind;
-        local->transaction.done   = afr_truncate_done;
-        local->transaction.unwind = afr_truncate_unwind;
+    local->op = GF_FOP_TRUNCATE;
 
-        loc_copy (&local->loc, loc);
+    local->transaction.main_frame = frame;
+    local->transaction.start = offset;
+    local->transaction.len = 0;
 
-        local->transaction.main_frame = frame;
-        local->transaction.start   = offset;
-        local->transaction.len     = 0;
+    /* Set it true speculatively, will get reset in afr_truncate_wind_cbk
+       if truncate was not a NOP */
+    local->stable_write = _gf_true;
 
-        ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
-        if (ret < 0) {
-            op_errno = -ret;
-            goto out;
-        }
+    ret = afr_transaction(transaction_frame, this, AFR_DATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        ret = 0;
+    return 0;
 out:
-        if (ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (truncate, frame, -1, op_errno, NULL, NULL, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(truncate, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
-
 /* }}} */
 
 /* {{{ ftruncate */
 
-
 int
-afr_ftruncate_unwind (call_frame_t *frame, xlator_t *this)
+afr_ftruncate_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t   *main_frame = NULL;
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        local = frame->local;
+    local = frame->local;
 
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame)
-                        main_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
-
-        if (main_frame) {
-                AFR_STACK_UNWIND (ftruncate, main_frame, local->op_ret,
-                                  local->op_errno,
-                                  &local->cont.inode_wfop.prebuf,
-                                  &local->cont.inode_wfop.postbuf,
-                                  NULL);
-        }
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
-}
-
-
-int
-afr_ftruncate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                        struct iatt *postbuf, dict_t *xdata)
-{
-        afr_local_t *   local = NULL;
-        int child_index = (long) cookie;
-        int call_count  = -1;
-        int read_child  = 0;
-
-        local = frame->local;
-
-        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
-
-        LOCK (&frame->lock);
-        {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
 
-                if (op_ret != -1) {
-			if (prebuf->ia_size != postbuf->ia_size)
-				local->stable_write = _gf_false;
-                }
-                __inode_write_fop_cbk (frame, child_index, read_child, this,
-                                       &op_ret, &op_errno, prebuf, postbuf,
-                                       xdata);
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-		if (local->stable_write && afr_txn_nothing_failed (frame, this))
-			local->transaction.unwind (frame, this);
-
-                local->transaction.resume (frame, this);
-        }
-
-        return 0;
+    AFR_STACK_UNWIND(ftruncate, main_frame, local->op_ret, local->op_errno,
+                     &local->cont.inode_wfop.prebuf,
+                     &local->cont.inode_wfop.postbuf, local->xdata_rsp);
+    return 0;
 }
 
-
 int
-afr_ftruncate_wind (call_frame_t *frame, xlator_t *this)
+afr_ftruncate_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                       struct iatt *postbuf, dict_t *xdata)
 {
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int call_count = -1;
-        int i = 0;
+    afr_local_t *local = NULL;
 
-        local = frame->local;
-        priv = this->private;
+    local = frame->local;
 
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
+    if (op_ret == 0 && prebuf->ia_size != postbuf->ia_size)
+        local->stable_write = _gf_false;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
-
-        local->call_count = call_count;
-	local->stable_write = _gf_true;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->transaction.pre_op[i]) {
-                        STACK_WIND_COOKIE (frame, afr_ftruncate_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->ftruncate,
-                                           local->fd,
-                                           local->cont.ftruncate.offset,
-                                           NULL);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-        return 0;
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, prebuf,
+                                 postbuf, NULL, xdata);
 }
 
-
 int
-afr_ftruncate_done (call_frame_t *frame, xlator_t *this)
+afr_ftruncate_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_local_t *local = NULL;
-
-        local = frame->local;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        local->transaction.unwind (frame, this);
+    local = frame->local;
+    priv = this->private;
 
-        AFR_STACK_DESTROY (frame);
-
-        return 0;
+    STACK_WIND_COOKIE(frame, afr_ftruncate_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->ftruncate, local->fd,
+                      local->cont.ftruncate.offset, local->xdata_req);
+    return 0;
 }
 
-
 int
-afr_do_ftruncate (call_frame_t *frame, xlator_t *this)
+afr_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+              dict_t *xdata)
 {
-        call_frame_t * transaction_frame = NULL;
-        afr_local_t *  local             = NULL;
-        int op_ret   = -1;
-        int op_errno = 0;
-
-        local = frame->local;
-
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                goto out;
-        }
-
-        transaction_frame->local = local;
-        frame->local = NULL;
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        local->op = GF_FOP_FTRUNCATE;
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        local->transaction.fop    = afr_ftruncate_wind;
-        local->transaction.done   = afr_ftruncate_done;
-        local->transaction.unwind = afr_ftruncate_unwind;
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        local->transaction.main_frame = frame;
+    local->cont.ftruncate.offset = offset;
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-        local->transaction.start   = local->cont.ftruncate.offset;
-        local->transaction.len     = 0;
+    if (!local->xdata_req)
+        goto out;
 
-        op_ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
-        if (op_ret < 0) {
-            op_errno = -op_ret;
-            goto out;
-        }
-
-        op_ret = 0;
-out:
-        if (op_ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (ftruncate, frame, op_ret, op_errno, NULL,
-                                  NULL, NULL);
-        }
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
 
-        return 0;
-}
-
-
-int
-afr_ftruncate (call_frame_t *frame, xlator_t *this,
-               fd_t *fd, off_t offset, dict_t *xdata)
-{
-        afr_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
-        call_frame_t   *transaction_frame = NULL;
-        int ret = -1;
-        int op_errno = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
-
-        priv = this->private;
-
-        if (afr_is_split_brain (this, fd->inode)) {
-                op_errno = EIO;
-                goto out;
-        }
-        QUORUM_CHECK(ftruncate,out);
+    local->op = GF_FOP_FTRUNCATE;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    local->transaction.wind = afr_ftruncate_wind;
+    local->transaction.unwind = afr_ftruncate_unwind;
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    local->transaction.main_frame = frame;
 
-        local->cont.ftruncate.offset  = offset;
+    local->transaction.start = local->cont.ftruncate.offset;
+    local->transaction.len = 0;
 
-        local->fd = fd_ref (fd);
+    afr_fix_open(fd, this);
 
-        afr_open_fd_fix (fd, this);
+    /* Set it true speculatively, will get reset in afr_ftruncate_wind_cbk
+       if truncate was not a NOP */
+    local->stable_write = _gf_true;
 
-        afr_do_ftruncate (frame, this);
+    ret = afr_transaction(transaction_frame, this, AFR_DATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        ret = 0;
+    return 0;
 out:
-        if (ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (ftruncate, frame, -1, op_errno, NULL, NULL, NULL);
-        }
+    AFR_STACK_UNWIND(ftruncate, frame, -1, op_errno, NULL, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
 /* }}} */
@@ -1007,1605 +788,1778 @@ out:
 /* {{{ setattr */
 
 int
-afr_setattr_unwind (call_frame_t *frame, xlator_t *this)
+afr_setattr_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t   *main_frame = NULL;
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame)
-                        main_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
-
-        if (main_frame) {
-                AFR_STACK_UNWIND (setattr, main_frame, local->op_ret,
-                                  local->op_errno,
-                                  &local->cont.inode_wfop.prebuf,
-                                  &local->cont.inode_wfop.postbuf,
-                                  NULL);
-        }
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
-}
-
-
-int
-afr_setattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno,
-                      struct iatt *preop, struct iatt *postop, dict_t *xdata)
-{
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
-        int child_index = (long) cookie;
-        int read_child  = 0;
-        int call_count  = -1;
-        int need_unwind = 0;
-
-        local = frame->local;
-        priv  = this->private;
-
-        read_child = afr_inode_get_read_ctx (this, local->loc.inode, NULL);
-
-        LOCK (&frame->lock);
-        {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
-
-                __inode_write_fop_cbk (frame, child_index, read_child, this,
-                                       &op_ret, &op_errno, preop, postop,
-                                       xdata);
-
-                if ((local->success_count >= priv->wait_count)
-                    && local->read_child_returned) {
-                        need_unwind = 1;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
 
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
-
-        return 0;
+    AFR_STACK_UNWIND(setattr, main_frame, local->op_ret, local->op_errno,
+                     &local->cont.inode_wfop.prebuf,
+                     &local->cont.inode_wfop.postbuf, local->xdata_rsp);
+    return 0;
 }
 
-
-int32_t
-afr_setattr_wind (call_frame_t *frame, xlator_t *this)
+int
+afr_setattr_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, struct iatt *preop,
+                     struct iatt *postop, dict_t *xdata)
 {
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int call_count = -1;
-        int i = 0;
-
-        local = frame->local;
-        priv = this->private;
-
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
-
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
-
-        local->call_count = call_count;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->transaction.pre_op[i]) {
-                        STACK_WIND_COOKIE (frame, afr_setattr_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->setattr,
-                                           &local->loc,
-                                           &local->cont.setattr.in_buf,
-                                           local->cont.setattr.valid,
-                                           NULL);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-        return 0;
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, preop,
+                                 postop, NULL, xdata);
 }
 
-
 int
-afr_setattr_done (call_frame_t *frame, xlator_t *this)
+afr_setattr_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_local_t *local = NULL;
-
-        local = frame->local;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        local->transaction.unwind (frame, this);
+    local = frame->local;
+    priv = this->private;
 
-        AFR_STACK_DESTROY (frame);
-
-        return 0;
+    STACK_WIND_COOKIE(frame, afr_setattr_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->setattr, &local->loc,
+                      &local->cont.setattr.in_buf, local->cont.setattr.valid,
+                      local->xdata_req);
+    return 0;
 }
 
-
 int
-afr_setattr (call_frame_t *frame, xlator_t *this,
-             loc_t *loc, struct iatt *buf, int32_t valid, dict_t *xdata)
+afr_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *buf,
+            int32_t valid, dict_t *xdata)
 {
-        afr_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
-        call_frame_t   *transaction_frame = NULL;
-        int ret = -1;
-        int op_errno = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        priv = this->private;
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        QUORUM_CHECK(setattr,out);
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+    local->cont.setattr.in_buf = *buf;
+    local->cont.setattr.valid = valid;
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
-        local = transaction_frame->local;
+    if (!local->xdata_req)
+        goto out;
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    local->transaction.wind = afr_setattr_wind;
+    local->transaction.unwind = afr_setattr_unwind;
 
-        local->cont.setattr.in_buf = *buf;
-        local->cont.setattr.valid  = valid;
+    loc_copy(&local->loc, loc);
+    ret = afr_set_inode_local(this, local, loc->inode);
+    if (ret)
+        goto out;
 
-        local->transaction.fop    = afr_setattr_wind;
-        local->transaction.done   = afr_setattr_done;
-        local->transaction.unwind = afr_setattr_unwind;
+    local->op = GF_FOP_SETATTR;
 
-        loc_copy (&local->loc, loc);
+    local->transaction.main_frame = frame;
+    local->transaction.start = LLONG_MAX - 1;
+    local->transaction.len = 0;
 
-        local->transaction.main_frame = frame;
-        local->transaction.start   = LLONG_MAX - 1;
-        local->transaction.len     = 0;
+    ret = afr_transaction(transaction_frame, this, AFR_METADATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        ret = afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
-        if (ret < 0) {
-            op_errno = -ret;
-            goto out;
-        }
-
-        ret = 0;
+    return 0;
 out:
-        if (ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (setattr, frame, -1, op_errno, NULL, NULL, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(setattr, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
 /* {{{ fsetattr */
 
 int
-afr_fsetattr_unwind (call_frame_t *frame, xlator_t *this)
+afr_fsetattr_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t   *main_frame = NULL;
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame)
-                        main_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
-
-        if (main_frame) {
-                AFR_STACK_UNWIND (fsetattr, main_frame, local->op_ret,
-                                  local->op_errno,
-                                  &local->cont.inode_wfop.prebuf,
-                                  &local->cont.inode_wfop.postbuf,
-                                  NULL);
-        }
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
-}
 
+    AFR_STACK_UNWIND(fsetattr, main_frame, local->op_ret, local->op_errno,
+                     &local->cont.inode_wfop.prebuf,
+                     &local->cont.inode_wfop.postbuf, local->xdata_rsp);
+    return 0;
+}
 
 int
-afr_fsetattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno,
-                       struct iatt *preop, struct iatt *postop, dict_t *xdata)
+afr_fsetattr_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *preop,
+                      struct iatt *postop, dict_t *xdata)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
-        int child_index = (long) cookie;
-        int read_child  = 0;
-        int call_count  = -1;
-        int need_unwind = 0;
-
-        local = frame->local;
-        priv  = this->private;
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, preop,
+                                 postop, NULL, xdata);
+}
 
-        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
+int
+afr_fsetattr_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        LOCK (&frame->lock);
-        {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
+    local = frame->local;
+    priv = this->private;
 
-                __inode_write_fop_cbk (frame, child_index, read_child, this,
-                                       &op_ret, &op_errno, preop, postop,
-                                       xdata);
+    STACK_WIND_COOKIE(frame, afr_fsetattr_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->fsetattr, local->fd,
+                      &local->cont.fsetattr.in_buf, local->cont.fsetattr.valid,
+                      local->xdata_req);
+    return 0;
+}
 
-                if ((local->success_count >= priv->wait_count)
-                    && local->read_child_returned) {
-                        need_unwind = 1;
-                }
-        }
-        UNLOCK (&frame->lock);
+int
+afr_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *buf,
+             int32_t valid, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        call_count = afr_frame_return (frame);
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
+    local->cont.fsetattr.in_buf = *buf;
+    local->cont.fsetattr.valid = valid;
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-        return 0;
-}
+    if (!local->xdata_req)
+        goto out;
 
+    local->transaction.wind = afr_fsetattr_wind;
+    local->transaction.unwind = afr_fsetattr_unwind;
 
-int32_t
-afr_fsetattr_wind (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int call_count = -1;
-        int i = 0;
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
 
-        local = frame->local;
-        priv = this->private;
+    local->op = GF_FOP_FSETATTR;
 
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
+    afr_fix_open(fd, this);
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
+    local->transaction.main_frame = frame;
+    local->transaction.start = LLONG_MAX - 1;
+    local->transaction.len = 0;
 
-        local->call_count = call_count;
+    ret = afr_transaction(transaction_frame, this, AFR_METADATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->transaction.pre_op[i]) {
-                        STACK_WIND_COOKIE (frame, afr_fsetattr_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->fsetattr,
-                                           local->fd,
-                                           &local->cont.fsetattr.in_buf,
-                                           local->cont.fsetattr.valid,
-                                           NULL);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
+    return 0;
+out:
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(fsetattr, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
+/* {{{ setxattr */
 
 int
-afr_fsetattr_done (call_frame_t *frame, xlator_t *this)
+afr_setxattr_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *local = NULL;
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        local = frame->local;
+    local = frame->local;
 
-        local->transaction.unwind (frame, this);
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
+        return 0;
 
-        AFR_STACK_DESTROY (frame);
+    AFR_STACK_UNWIND(setxattr, main_frame, local->op_ret, local->op_errno,
+                     local->xdata_rsp);
+    return 0;
+}
 
-        return 0;
+int
+afr_setxattr_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, NULL,
+                                 NULL, NULL, xdata);
 }
 
 int
-afr_fsetattr (call_frame_t *frame, xlator_t *this,
-              fd_t *fd, struct iatt *buf, int32_t valid, dict_t *xdata)
+afr_setxattr_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
-        call_frame_t   *transaction_frame = NULL;
-        int ret = -1;
-        int op_errno = 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    local = frame->local;
+    priv = this->private;
 
-        priv = this->private;
+    STACK_WIND_COOKIE(frame, afr_setxattr_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->setxattr, &local->loc,
+                      local->cont.setxattr.dict, local->cont.setxattr.flags,
+                      local->xdata_req);
+    return 0;
+}
 
-        if (afr_is_split_brain (this, fd->inode)) {
-                op_errno = EIO;
-                goto out;
-        }
+int
+afr_emptyb_set_pending_changelog_cbk(call_frame_t *frame, void *cookie,
+                                     xlator_t *this, int op_ret, int op_errno,
+                                     dict_t *xattr, dict_t *xdata)
 
-        QUORUM_CHECK(fsetattr,out);
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i, ret = 0;
+    char *op_type = NULL;
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+    local = frame->local;
+    priv = this->private;
+    i = (long)cookie;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
-        local = transaction_frame->local;
+    local->replies[i].valid = 1;
+    local->replies[i].op_ret = op_ret;
+    local->replies[i].op_errno = op_errno;
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    ret = dict_get_str_sizen(local->xdata_req, "replicate-brick-op", &op_type);
+    if (ret)
+        goto out;
 
-        local->cont.fsetattr.in_buf = *buf;
-        local->cont.fsetattr.valid  = valid;
+    gf_smsg(this->name, op_ret ? GF_LOG_ERROR : GF_LOG_INFO,
+            op_ret ? op_errno : 0, AFR_MSG_SET_PEND_XATTR, "name=%s",
+            priv->children[i]->name, "op_ret=%s",
+            op_ret ? "failed" : "succeeded", NULL);
+
+out:
+    syncbarrier_wake(&local->barrier);
+    return 0;
+}
 
-        local->transaction.fop    = afr_fsetattr_wind;
-        local->transaction.done   = afr_fsetattr_done;
-        local->transaction.unwind = afr_fsetattr_unwind;
+int
+afr_emptyb_set_pending_changelog(call_frame_t *frame, xlator_t *this,
+                                 unsigned char *locked_nodes)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int ret = 0, i = 0;
 
-        local->fd                 = fd_ref (fd);
+    local = frame->local;
+    priv = this->private;
 
-        afr_open_fd_fix (fd, this);
+    AFR_ONLIST(locked_nodes, frame, afr_emptyb_set_pending_changelog_cbk,
+               xattrop, &local->loc, GF_XATTROP_ADD_ARRAY, local->xattr_req,
+               NULL);
 
-        local->transaction.main_frame = frame;
-        local->transaction.start   = LLONG_MAX - 1;
-        local->transaction.len     = 0;
+    /* It is sufficient if xattrop was successful on one child */
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
 
-        ret = afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
-        if (ret < 0) {
-            op_errno = -ret;
+        if (local->replies[i].op_ret == 0) {
+            ret = 0;
             goto out;
+        } else {
+            ret = afr_higher_errno(ret, local->replies[i].op_errno);
         }
-
-        ret = 0;
+    }
 out:
-        if (ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (fsetattr, frame, -1, op_errno, NULL, NULL, NULL);
-        }
-
-        return 0;
+    return -ret;
 }
 
+static int
+_afr_handle_empty_brick_type(xlator_t *this, call_frame_t *frame, loc_t *loc,
+                             int empty_index, afr_transaction_type type,
+                             char *op_type, const int op_type_len)
+{
+    int count = 0;
+    int ret = -ENOMEM;
+    int idx = -1;
+    int d_idx = -1;
+    unsigned char *locked_nodes = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    locked_nodes = alloca0(priv->child_count);
+
+    idx = afr_index_for_transaction_type(type);
+    d_idx = afr_index_for_transaction_type(AFR_DATA_TRANSACTION);
+
+    local->pending = afr_matrix_create(priv->child_count, AFR_NUM_CHANGE_LOGS);
+    if (!local->pending)
+        goto out;
+
+    local->pending[empty_index][idx] = hton32(1);
+
+    if ((priv->esh_granular) && (type == AFR_ENTRY_TRANSACTION))
+        local->pending[empty_index][d_idx] = hton32(1);
+
+    local->xdata_req = dict_new();
+    if (!local->xdata_req)
+        goto out;
+
+    ret = dict_set_nstrn(local->xdata_req, "replicate-brick-op",
+                         SLEN("replicate-brick-op"), op_type, op_type_len);
+    if (ret)
+        goto out;
+
+    local->xattr_req = dict_new();
+    if (!local->xattr_req)
+        goto out;
+
+    ret = afr_set_pending_dict(priv, local->xattr_req, local->pending);
+    if (ret < 0)
+        goto out;
+
+    if (AFR_ENTRY_TRANSACTION == type) {
+        count = afr_selfheal_entrylk(frame, this, loc->inode, this->name, NULL,
+                                     locked_nodes);
+    } else {
+        count = afr_selfheal_inodelk(frame, this, loc->inode, this->name,
+                                     LLONG_MAX - 1, 0, locked_nodes);
+    }
+
+    if (!count) {
+        gf_smsg(this->name, GF_LOG_ERROR, EAGAIN, AFR_MSG_REPLACE_BRICK_STATUS,
+                NULL);
+        ret = -EAGAIN;
+        goto unlock;
+    }
+
+    ret = afr_emptyb_set_pending_changelog(frame, this, locked_nodes);
+    if (ret)
+        goto unlock;
+    ret = 0;
+unlock:
+    if (AFR_ENTRY_TRANSACTION == type) {
+        afr_selfheal_unentrylk(frame, this, loc->inode, this->name, NULL,
+                               locked_nodes, NULL);
+    } else {
+        afr_selfheal_uninodelk(frame, this, loc->inode, this->name,
+                               LLONG_MAX - 1, 0, locked_nodes);
+    }
+out:
+    return ret;
+}
 
-/* {{{ setxattr */
+void
+afr_brick_args_cleanup(void *opaque)
+{
+    afr_empty_brick_args_t *data = NULL;
 
+    data = opaque;
+    loc_wipe(&data->loc);
+    GF_FREE(data);
+}
 
 int
-afr_setxattr_unwind (call_frame_t *frame, xlator_t *this)
+_afr_handle_empty_brick_cbk(int ret, call_frame_t *frame, void *opaque)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t   *main_frame = NULL;
-
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame)
-                        main_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
+    afr_brick_args_cleanup(opaque);
+    return 0;
+}
 
-        if (main_frame) {
-                AFR_STACK_UNWIND (setxattr, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  NULL);
-        }
-        return 0;
+int
+_afr_handle_empty_brick(void *opaque)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int empty_index = -1;
+    int ret = -1;
+    int op_errno = ENOMEM;
+    call_frame_t *frame = NULL;
+    xlator_t *this = NULL;
+    char *op_type = NULL;
+    int op_type_len = 0;
+    afr_empty_brick_args_t *data = NULL;
+    call_frame_t *op_frame = NULL;
+
+    data = opaque;
+    frame = data->frame;
+    empty_index = data->empty_index;
+    if (!data->op_type)
+        goto out;
+
+    op_frame = copy_frame(frame);
+    if (!op_frame) {
+        ret = -1;
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    op_type = data->op_type;
+    op_type_len = strlen(op_type);
+    this = op_frame->this;
+    priv = this->private;
+
+    afr_set_lk_owner(op_frame, this, op_frame->root);
+    local = AFR_FRAME_INIT(op_frame, op_errno);
+    if (!local)
+        goto out;
+
+    loc_copy(&local->loc, &data->loc);
+
+    gf_smsg(this->name, GF_LOG_INFO, 0, AFR_MSG_NEW_BRICK, "name=%s",
+            priv->children[empty_index]->name, NULL);
+
+    ret = _afr_handle_empty_brick_type(this, op_frame, &local->loc, empty_index,
+                                       AFR_METADATA_TRANSACTION, op_type,
+                                       op_type_len);
+    if (ret) {
+        op_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    dict_unref(local->xdata_req);
+    dict_unref(local->xattr_req);
+    afr_matrix_cleanup(local->pending, priv->child_count);
+    local->pending = NULL;
+    local->xattr_req = NULL;
+    local->xdata_req = NULL;
+
+    ret = _afr_handle_empty_brick_type(this, op_frame, &local->loc, empty_index,
+                                       AFR_ENTRY_TRANSACTION, op_type,
+                                       op_type_len);
+    if (ret) {
+        op_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+    ret = 0;
+out:
+    if (op_frame) {
+        AFR_STACK_DESTROY(op_frame);
+    }
+    AFR_STACK_UNWIND(setxattr, frame, ret, op_errno, NULL);
+    return 0;
 }
 
+int
+afr_split_brain_resolve_do(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                           char *data)
+{
+    afr_local_t *local = NULL;
+    int ret = -1;
+    int op_errno = EINVAL;
+
+    local = frame->local;
+    local->xdata_req = dict_new();
+
+    if (!local->xdata_req) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    ret = dict_set_int32_sizen(local->xdata_req, "heal-op",
+                               GF_SHD_OP_SBRAIN_HEAL_FROM_BRICK);
+    if (ret) {
+        op_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+    ret = dict_set_str_sizen(local->xdata_req, "child-name", data);
+    if (ret) {
+        op_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+    /* set spb choice to -1 whether heal succeeds or not:
+     * If heal succeeds : spb-choice should be set to -1 as
+     *                    it is no longer valid; file is not
+     *                    in split-brain anymore.
+     * If heal doesn't succeed:
+     *                    spb-choice should be set to -1
+     *                    otherwise reads will be served
+     *                    from spb-choice which is misleading.
+     */
+    ret = afr_inode_split_brain_choice_set(loc->inode, this, -1);
+    if (ret)
+        gf_smsg(this->name, GF_LOG_WARNING, 0, AFR_MSG_SPLIT_BRAIN_SET_FAILED,
+                NULL);
+    afr_heal_splitbrain_file(frame, this, loc);
+    ret = 0;
+out:
+    if (ret < 0)
+        AFR_STACK_UNWIND(setxattr, frame, -1, op_errno, NULL);
+    return 0;
+}
 
 int
-afr_setxattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
+afr_get_split_brain_child_index(xlator_t *this, void *value, size_t len)
 {
-        afr_local_t   *local      = NULL;
-        afr_private_t *priv       = NULL;
-        int           call_count  = -1;
-        int           need_unwind = 0;
-        int           child_index = (long) cookie;
-
-        local = frame->local;
-        priv = this->private;
-
-        LOCK (&frame->lock);
-        {
-                __inode_write_fop_cbk (frame, child_index, -1, this,
-                                       &op_ret, &op_errno, NULL, NULL,
-                                       xdata);
-                if (local->success_count == priv->child_count) {
-                        need_unwind = 1;
-                }
-        }
-        UNLOCK (&frame->lock);
+    int spb_child_index = -1;
+    char *spb_child_str = NULL;
 
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
+    spb_child_str = alloca0(len + 1);
+    memcpy(spb_child_str, value, len);
 
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
+    if (!strcmp(spb_child_str, "none"))
+        return -2;
 
-        return 0;
+    spb_child_index = afr_get_child_index_from_name(this, spb_child_str);
+    if (spb_child_index < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, AFR_MSG_INVALID_SUBVOL,
+                "subvol=%s", spb_child_str, NULL);
+    }
+    return spb_child_index;
 }
 
-
 int
-afr_setxattr_wind (call_frame_t *frame, xlator_t *this)
+afr_can_set_split_brain_choice(void *opaque)
 {
-        afr_local_t      *local         = NULL;
-        afr_private_t    *priv          = NULL;
-        int               call_count    = -1;
-        int               i             = 0;
+    afr_spbc_timeout_t *data = opaque;
+    call_frame_t *frame = NULL;
+    xlator_t *this = NULL;
+    loc_t *loc = NULL;
+    int ret = -1;
 
-        local = frame->local;
-        priv = this->private;
+    frame = data->frame;
+    loc = data->loc;
+    this = frame->this;
 
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
+    ret = afr_is_split_brain(frame, this, loc->inode, loc->gfid, &data->d_spb,
+                             &data->m_spb);
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
+    if (ret)
+        gf_smsg(this->name, GF_LOG_ERROR, 0,
+                AFR_MSG_SPLIT_BRAIN_DETERMINE_FAILED, "gfid=%s",
+                uuid_utoa(loc->gfid), NULL);
+    return ret;
+}
 
-        local->call_count = call_count;
+int
+afr_handle_split_brain_commands(xlator_t *this, call_frame_t *frame, loc_t *loc,
+                                dict_t *dict)
+{
+    void *choice_value = NULL;
+    void *resolve_value = NULL;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    afr_spbc_timeout_t *data = NULL;
+    int len = 0;
+    int spb_child_index = -1;
+    int ret = -1;
+    int op_errno = EINVAL;
+
+    priv = this->private;
+
+    ret = dict_get_ptr_and_len(dict, GF_AFR_SBRAIN_CHOICE, &choice_value, &len);
+    ret = dict_get_ptr_and_len(dict, GF_AFR_SBRAIN_RESOLVE, &resolve_value,
+                               &len);
+    if (!choice_value && !resolve_value) {
+        ret = -1;
+        goto out;
+    }
+
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local) {
+        ret = 1;
+        goto out;
+    }
+
+    local->op = GF_FOP_SETXATTR;
+
+    if (choice_value) {
+        spb_child_index = afr_get_split_brain_child_index(this, choice_value,
+                                                          len);
+        if (spb_child_index < 0) {
+            /* Case where value was "none" */
+            if (spb_child_index == -2)
+                spb_child_index = -1;
+            else {
+                ret = 1;
+                op_errno = EINVAL;
+                goto out;
+            }
+        }
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->transaction.pre_op[i]) {
-                        STACK_WIND_COOKIE (frame, afr_setxattr_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->setxattr,
-                                           &local->loc,
-                                           local->cont.setxattr.dict,
-                                           local->cont.setxattr.flags,
-                                           NULL);
-
-                        if (!--call_count)
-                                break;
-                }
+        data = GF_CALLOC(1, sizeof(*data), gf_afr_mt_spbc_timeout_t);
+        if (!data) {
+            ret = 1;
+            goto out;
+        }
+        data->spb_child_index = spb_child_index;
+        data->frame = frame;
+        loc_copy(&local->loc, loc);
+        data->loc = &local->loc;
+        ret = synctask_new(this->ctx->env, afr_can_set_split_brain_choice,
+                           afr_set_split_brain_choice, NULL, data);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN_STATUS,
+                    "name=%s", loc->name, NULL);
+            ret = 1;
+            op_errno = ENOMEM;
+            goto out;
+        }
+        ret = 0;
+        goto out;
+    }
+
+    if (resolve_value) {
+        spb_child_index = afr_get_split_brain_child_index(this, resolve_value,
+                                                          len);
+        if (spb_child_index < 0) {
+            ret = 1;
+            goto out;
         }
 
-        return 0;
+        afr_split_brain_resolve_do(frame, this, loc,
+                                   priv->children[spb_child_index]->name);
+        ret = 0;
+    }
+out:
+    /* key was correct but value was invalid when ret == 1 */
+    if (ret == 1) {
+        AFR_STACK_UNWIND(setxattr, frame, -1, op_errno, NULL);
+        if (data)
+            GF_FREE(data);
+        ret = 0;
+    }
+    return ret;
 }
 
-
 int
-afr_setxattr_done (call_frame_t *frame, xlator_t *this)
+afr_handle_spb_choice_timeout(xlator_t *this, call_frame_t *frame, dict_t *dict)
 {
-        afr_local_t   *local    = frame->local;
+    int ret = -1;
+    int op_errno = 0;
+    uint64_t timeout = 0;
+    afr_private_t *priv = NULL;
 
-        local->transaction.unwind (frame, this);
+    priv = this->private;
 
-        AFR_STACK_DESTROY (frame);
+    ret = dict_get_uint64(dict, GF_AFR_SPB_CHOICE_TIMEOUT, &timeout);
+    if (!ret) {
+        priv->spb_choice_timeout = timeout * 60;
+        AFR_STACK_UNWIND(setxattr, frame, ret, op_errno, NULL);
+    }
 
-        return 0;
+    return ret;
 }
 
 int
-afr_setxattr (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, dict_t *dict, int32_t flags, dict_t *xdata)
-{
-        afr_private_t  *priv              = NULL;
-        afr_local_t    *local             = NULL;
-        call_frame_t   *transaction_frame = NULL;
-        int             ret               = -1;
-        int             op_errno          = EINVAL;
-
-        VALIDATE_OR_GOTO (this, out);
-
-        GF_IF_INTERNAL_XATTR_GOTO ("trusted.afr.*", dict,
-                                   op_errno, out);
-
-        GF_IF_INTERNAL_XATTR_GOTO ("trusted.glusterfs.afr.*", dict,
-                                   op_errno, out);
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this->private, out);
-
-        priv = this->private;
-
-        QUORUM_CHECK(setxattr,out);
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                op_errno = ENOMEM;
-                goto out;
+afr_handle_empty_brick(xlator_t *this, call_frame_t *frame, loc_t *loc,
+                       dict_t *dict)
+{
+    int ret = -1;
+    int ab_ret = -1;
+    int empty_index = -1;
+    int op_errno = EPERM;
+    char *empty_brick = NULL;
+    char *op_type = NULL;
+    afr_empty_brick_args_t *data = NULL;
+
+    ret = dict_get_str_sizen(dict, GF_AFR_REPLACE_BRICK, &empty_brick);
+    if (!ret)
+        op_type = GF_AFR_REPLACE_BRICK;
+
+    ab_ret = dict_get_str_sizen(dict, GF_AFR_ADD_BRICK, &empty_brick);
+    if (!ab_ret)
+        op_type = GF_AFR_ADD_BRICK;
+
+    if (ret && ab_ret)
+        goto out;
+
+    if (frame->root->pid != GF_CLIENT_PID_ADD_REPLICA_MOUNT) {
+        gf_smsg(this->name, GF_LOG_ERROR, EPERM, AFR_MSG_INTERNAL_ATTR,
+                "op_type=%s", op_type, NULL);
+        ret = 1;
+        goto out;
+    }
+    empty_index = afr_get_child_index_from_name(this, empty_brick);
+
+    if (empty_index < 0) {
+        /* Didn't belong to this replica pair
+         * Just do a no-op
+         */
+        AFR_STACK_UNWIND(setxattr, frame, 0, 0, NULL);
+        return 0;
+    } else {
+        data = GF_CALLOC(1, sizeof(*data), gf_afr_mt_empty_brick_t);
+        if (!data) {
+            ret = 1;
+            op_errno = ENOMEM;
+            goto out;
         }
-
-        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
-        local = transaction_frame->local;
-
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
-
-        local->cont.setxattr.dict  = dict_ref (dict);
-        local->cont.setxattr.flags = flags;
-
-        local->transaction.fop    = afr_setxattr_wind;
-        local->transaction.done   = afr_setxattr_done;
-        local->transaction.unwind = afr_setxattr_unwind;
-
-        loc_copy (&local->loc, loc);
-
-        local->transaction.main_frame = frame;
-        local->transaction.start   = LLONG_MAX - 1;
-        local->transaction.len     = 0;
-
-        ret = afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
-        if (ret < 0) {
-            op_errno = -ret;
+        data->frame = frame;
+        loc_copy(&data->loc, loc);
+        data->empty_index = empty_index;
+        data->op_type = op_type;
+        ret = synctask_new(this->ctx->env, _afr_handle_empty_brick,
+                           _afr_handle_empty_brick_cbk, NULL, data);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN_STATUS,
+                    NULL);
+            ret = 1;
+            op_errno = ENOMEM;
+            afr_brick_args_cleanup(data);
             goto out;
         }
-
-        ret = 0;
+    }
+    ret = 0;
 out:
-        if (ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (setxattr, frame, -1, op_errno, NULL);
-        }
-
-        return 0;
+    if (ret == 1) {
+        AFR_STACK_UNWIND(setxattr, frame, -1, op_errno, NULL);
+        ret = 0;
+    }
+    return ret;
 }
 
-/* {{{ fsetxattr */
-
-
-int
-afr_fsetxattr_unwind (call_frame_t *frame, xlator_t *this)
+static int
+afr_handle_special_xattr(xlator_t *this, call_frame_t *frame, loc_t *loc,
+                         dict_t *dict)
 {
-        afr_local_t    *local         = NULL;
-        call_frame_t   *main_frame    = NULL;
+    int ret = -1;
 
-        local = frame->local;
+    ret = afr_handle_split_brain_commands(this, frame, loc, dict);
+    if (ret == 0)
+        goto out;
 
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame)
-                        main_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
+    ret = afr_handle_spb_choice_timeout(this, frame, dict);
+    if (ret == 0)
+        goto out;
 
-        if (main_frame) {
-                AFR_STACK_UNWIND (fsetxattr, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  NULL);
-        }
-        return 0;
+    /* Applicable for replace-brick and add-brick commands */
+    ret = afr_handle_empty_brick(this, frame, loc, dict);
+out:
+    return ret;
 }
 
-
 int
-afr_fsetxattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+afr_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+             int32_t flags, dict_t *xdata)
 {
-        afr_local_t   *local      = NULL;
-        afr_private_t *priv       = NULL;
-        int           call_count  = -1;
-        int           need_unwind = 0;
-        int           child_index = (long) cookie;
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = EINVAL;
 
-        local = frame->local;
-        priv = this->private;
+    GF_IF_INTERNAL_XATTR_GOTO("trusted.afr.*", dict, op_errno, out);
 
-        LOCK (&frame->lock);
-        {
+    GF_IF_INTERNAL_XATTR_GOTO("trusted.glusterfs.afr.*", dict, op_errno, out);
 
-                __inode_write_fop_cbk (frame, child_index, -1, this,
-                                       &op_ret, &op_errno, NULL, NULL,
-                                       xdata);
-                if (local->success_count == priv->child_count) {
-                        need_unwind = 1;
-                }
-        }
-        UNLOCK (&frame->lock);
+    ret = afr_handle_special_xattr(this, frame, loc, dict);
+    if (ret == 0)
+        return 0;
 
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        call_count = afr_frame_return (frame);
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
+    local->cont.setxattr.dict = dict_ref(dict);
+    local->cont.setxattr.flags = flags;
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-        return 0;
-}
+    if (!local->xdata_req)
+        goto out;
 
+    local->transaction.wind = afr_setxattr_wind;
+    local->transaction.unwind = afr_setxattr_unwind;
 
-int
-afr_fsetxattr_wind (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t        *local       = NULL;
-        afr_private_t      *priv        = NULL;
-        int                 call_count  = -1;
-        int                 i           = 0;
+    loc_copy(&local->loc, loc);
+    ret = afr_set_inode_local(this, local, loc->inode);
+    if (ret)
+        goto out;
 
-        local = frame->local;
-        priv = this->private;
+    local->transaction.main_frame = frame;
+    local->transaction.start = LLONG_MAX - 1;
+    local->transaction.len = 0;
 
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
+    local->op = GF_FOP_SETXATTR;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
+    ret = afr_transaction(transaction_frame, this, AFR_METADATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        local->call_count = call_count;
+    return 0;
+out:
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->transaction.pre_op[i]) {
-                        STACK_WIND_COOKIE (frame, afr_fsetxattr_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->fsetxattr,
-                                           local->fd,
-                                           local->cont.fsetxattr.dict,
-                                           local->cont.fsetxattr.flags,
-                                           NULL);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
+    AFR_STACK_UNWIND(setxattr, frame, -1, op_errno, NULL);
 
-        return 0;
+    return 0;
 }
 
+/* {{{ fsetxattr */
 
 int
-afr_fsetxattr_done (call_frame_t *frame, xlator_t *this)
+afr_fsetxattr_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t   *local   = frame->local;
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        local->transaction.unwind (frame, this);
-
-        AFR_STACK_DESTROY (frame);
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
+
+    AFR_STACK_UNWIND(fsetxattr, main_frame, local->op_ret, local->op_errno,
+                     local->xdata_rsp);
+    return 0;
 }
 
 int
-afr_fsetxattr (call_frame_t *frame, xlator_t *this,
-               fd_t *fd, dict_t *dict, int32_t flags, dict_t *xdata)
+afr_fsetxattr_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        afr_private_t    *priv              = NULL;
-        afr_local_t      *local             = NULL;
-        call_frame_t     *transaction_frame = NULL;
-        int               ret               = -1;
-        int               op_errno          = EINVAL;
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, NULL,
+                                 NULL, NULL, xdata);
+}
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+int
+afr_fsetxattr_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        GF_IF_INTERNAL_XATTR_GOTO ("trusted.afr.*", dict,
-                                   op_errno, out);
+    local = frame->local;
+    priv = this->private;
 
-        GF_IF_INTERNAL_XATTR_GOTO ("trusted.glusterfs.afr.*", dict,
-                                   op_errno, out);
+    STACK_WIND_COOKIE(frame, afr_fsetxattr_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->fsetxattr, local->fd,
+                      local->cont.fsetxattr.dict, local->cont.fsetxattr.flags,
+                      local->xdata_req);
+    return 0;
+}
 
-        priv = this->private;
+int
+afr_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+              int32_t flags, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        if (afr_is_split_brain (this, fd->inode)) {
-                op_errno = EIO;
-                goto out;
-        }
+    GF_IF_INTERNAL_XATTR_GOTO("trusted.afr.*", dict, op_errno, out);
 
-        QUORUM_CHECK(fsetxattr,out);
+    GF_IF_INTERNAL_XATTR_GOTO("trusted.glusterfs.afr.*", dict, op_errno, out);
 
-        AFR_LOCAL_ALLOC_OR_GOTO (local, out);
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                goto out;
-        }
+    local->cont.fsetxattr.dict = dict_ref(dict);
+    local->cont.fsetxattr.flags = flags;
 
-        transaction_frame->local = local;
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-        local->op_ret = -1;
+    if (!local->xdata_req)
+        goto out;
 
-        local->cont.fsetxattr.dict  = dict_ref (dict);
-        local->cont.fsetxattr.flags = flags;
+    local->transaction.wind = afr_fsetxattr_wind;
+    local->transaction.unwind = afr_fsetxattr_unwind;
 
-        local->transaction.fop    = afr_fsetxattr_wind;
-        local->transaction.done   = afr_fsetxattr_done;
-        local->transaction.unwind = afr_fsetxattr_unwind;
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
 
-        local->fd                 = fd_ref (fd);
+    local->op = GF_FOP_FSETXATTR;
 
-        local->transaction.main_frame = frame;
-        local->transaction.start  = LLONG_MAX - 1;
-        local->transaction.len    = 0;
+    local->transaction.main_frame = frame;
+    local->transaction.start = LLONG_MAX - 1;
+    local->transaction.len = 0;
 
-        ret = afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
-        if (ret < 0) {
-            op_errno = -ret;
-            goto out;
-        }
+    ret = afr_transaction(transaction_frame, this, AFR_METADATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        ret = 0;
+    return 0;
 out:
-        if (ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (fsetxattr, frame, -1, op_errno, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(fsetxattr, frame, -1, op_errno, NULL);
+    return 0;
 }
 
 /* }}} */
 
-
 /* {{{ removexattr */
 
-
 int
-afr_removexattr_unwind (call_frame_t *frame, xlator_t *this)
+afr_removexattr_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t   *main_frame = NULL;
-
-        local = frame->local;
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame)
-                        main_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
+    local = frame->local;
 
-        if (main_frame) {
-                AFR_STACK_UNWIND (removexattr, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  NULL);
-        }
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
-}
-
-
-int
-afr_removexattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                          int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-        afr_local_t   *local      = NULL;
-        afr_private_t *priv       = NULL;
-        int           call_count  = -1;
-        int           need_unwind = 0;
-        int           child_index = (long) cookie;
-
-        local = frame->local;
-        priv = this->private;
-
-        LOCK (&frame->lock);
-        {
-                __inode_write_fop_cbk (frame, child_index, -1, this,
-                                       &op_ret, &op_errno, NULL, NULL,
-                                       xdata);
-                if (local->success_count == priv->wait_count) {
-                        need_unwind = 1;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
 
-        return 0;
+    AFR_STACK_UNWIND(removexattr, main_frame, local->op_ret, local->op_errno,
+                     local->xdata_rsp);
+    return 0;
 }
 
-
-int32_t
-afr_removexattr_wind (call_frame_t *frame, xlator_t *this)
+int
+afr_removexattr_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int call_count = -1;
-        int i = 0;
-
-        local = frame->local;
-        priv = this->private;
-
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
-
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
-
-        local->call_count = call_count;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->transaction.pre_op[i]) {
-                        STACK_WIND_COOKIE (frame, afr_removexattr_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->removexattr,
-                                           &local->loc,
-                                           local->cont.removexattr.name,
-                                           NULL);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-        return 0;
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, NULL,
+                                 NULL, NULL, xdata);
 }
 
-
 int
-afr_removexattr_done (call_frame_t *frame, xlator_t *this)
+afr_removexattr_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_local_t * local = frame->local;
-
-        local->transaction.unwind (frame, this);
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        AFR_STACK_DESTROY (frame);
+    local = frame->local;
+    priv = this->private;
 
-        return 0;
+    STACK_WIND_COOKIE(frame, afr_removexattr_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->removexattr, &local->loc,
+                      local->cont.removexattr.name, local->xdata_req);
+    return 0;
 }
 
-
 int
-afr_removexattr (call_frame_t *frame, xlator_t *this,
-                 loc_t *loc, const char *name, dict_t *xdata)
+afr_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                const char *name, dict_t *xdata)
 {
-        afr_private_t   *priv              = NULL;
-        afr_local_t     *local             = NULL;
-        call_frame_t    *transaction_frame = NULL;
-        int              ret               = -1;
-        int              op_errno          = 0;
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        VALIDATE_OR_GOTO (this, out);
+    GF_IF_NATIVE_XATTR_GOTO("trusted.afr.*", name, op_errno, out);
 
-        GF_IF_NATIVE_XATTR_GOTO ("trusted.afr.*",
-                                 name, op_errno, out);
+    GF_IF_NATIVE_XATTR_GOTO("trusted.glusterfs.afr.*", name, op_errno, out);
 
-        GF_IF_NATIVE_XATTR_GOTO ("trusted.glusterfs.afr.*",
-                                 name, op_errno, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this->private, out);
-        VALIDATE_OR_GOTO (loc, out);
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        priv = this->private;
+    local->cont.removexattr.name = gf_strdup(name);
 
-        QUORUM_CHECK(removexattr,out);
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+    if (!local->xdata_req)
+        goto out;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
-        local = transaction_frame->local;
+    local->transaction.wind = afr_removexattr_wind;
+    local->transaction.unwind = afr_removexattr_unwind;
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
-
-        local->cont.removexattr.name = gf_strdup (name);
-
-        local->transaction.fop    = afr_removexattr_wind;
-        local->transaction.done   = afr_removexattr_done;
-        local->transaction.unwind = afr_removexattr_unwind;
+    loc_copy(&local->loc, loc);
+    ret = afr_set_inode_local(this, local, loc->inode);
+    if (ret)
+        goto out;
 
-        loc_copy (&local->loc, loc);
+    local->op = GF_FOP_REMOVEXATTR;
 
-        local->transaction.main_frame = frame;
-        local->transaction.start   = LLONG_MAX - 1;
-        local->transaction.len     = 0;
+    local->transaction.main_frame = frame;
+    local->transaction.start = LLONG_MAX - 1;
+    local->transaction.len = 0;
 
-        ret = afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
-        if (ret < 0) {
-            op_errno = -ret;
-            goto out;
-        }
+    ret = afr_transaction(transaction_frame, this, AFR_METADATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        ret = 0;
+    return 0;
 out:
-        if (ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (removexattr, frame, -1, op_errno, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(removexattr, frame, -1, op_errno, NULL);
+    return 0;
 }
 
 /* ffremovexattr */
 int
-afr_fremovexattr_unwind (call_frame_t *frame, xlator_t *this)
+afr_fremovexattr_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t   *main_frame = NULL;
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame)
-                        main_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
+    local = frame->local;
 
-        if (main_frame) {
-                AFR_STACK_UNWIND (fremovexattr, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  NULL);
-        }
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
-}
 
+    AFR_STACK_UNWIND(fremovexattr, main_frame, local->op_ret, local->op_errno,
+                     local->xdata_rsp);
+    return 0;
+}
 
 int
-afr_fremovexattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+afr_fremovexattr_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
                           int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        afr_local_t *   local       = NULL;
-        afr_private_t * priv        = NULL;
-        int             call_count  = -1;
-        int             need_unwind = 0;
-        int             child_index = (long) cookie;
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, NULL,
+                                 NULL, NULL, xdata);
+}
+
+int
+afr_fremovexattr_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        local = frame->local;
-        priv = this->private;
+    local = frame->local;
+    priv = this->private;
 
-        LOCK (&frame->lock);
-        {
-                __inode_write_fop_cbk (frame, child_index, -1, this,
-                                       &op_ret, &op_errno, NULL, NULL,
-                                       xdata);
+    STACK_WIND_COOKIE(frame, afr_fremovexattr_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->fremovexattr, local->fd,
+                      local->cont.removexattr.name, local->xdata_req);
+    return 0;
+}
 
-                if (local->success_count == priv->wait_count) {
-                        need_unwind = 1;
-                }
-        }
-        UNLOCK (&frame->lock);
+int
+afr_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                 const char *name, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
+    GF_IF_NATIVE_XATTR_GOTO("trusted.afr.*", name, op_errno, out);
 
-        call_count = afr_frame_return (frame);
+    GF_IF_NATIVE_XATTR_GOTO("trusted.glusterfs.afr.*", name, op_errno, out);
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        return 0;
-}
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
+    local->cont.removexattr.name = gf_strdup(name);
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-int32_t
-afr_fremovexattr_wind (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int call_count = -1;
-        int i = 0;
+    if (!local->xdata_req)
+        goto out;
 
-        local = frame->local;
-        priv = this->private;
+    local->transaction.wind = afr_fremovexattr_wind;
+    local->transaction.unwind = afr_fremovexattr_unwind;
 
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
+    local->op = GF_FOP_FREMOVEXATTR;
 
-        local->call_count = call_count;
+    local->transaction.main_frame = frame;
+    local->transaction.start = LLONG_MAX - 1;
+    local->transaction.len = 0;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->transaction.pre_op[i]) {
-                        STACK_WIND_COOKIE (frame, afr_fremovexattr_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->fremovexattr,
-                                           local->fd,
-                                           local->cont.removexattr.name,
-                                           NULL);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
+    ret = afr_transaction(transaction_frame, this, AFR_METADATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        return 0;
-}
+    return 0;
+out:
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
+
+    AFR_STACK_UNWIND(fremovexattr, frame, -1, op_errno, NULL);
 
+    return 0;
+}
 
 int
-afr_fremovexattr_done (call_frame_t *frame, xlator_t *this)
+afr_fallocate_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t * local = frame->local;
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        local->transaction.unwind (frame, this);
-
-        AFR_STACK_DESTROY (frame);
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
-}
 
+    AFR_STACK_UNWIND(fallocate, main_frame, local->op_ret, local->op_errno,
+                     &local->cont.inode_wfop.prebuf,
+                     &local->cont.inode_wfop.postbuf, local->xdata_rsp);
+    return 0;
+}
 
 int
-afr_fremovexattr (call_frame_t *frame, xlator_t *this,
-                  fd_t *fd, const char *name, dict_t *xdata)
+afr_fallocate_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                       struct iatt *postbuf, dict_t *xdata)
 {
-        afr_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
-        call_frame_t   *transaction_frame = NULL;
-        int ret = -1;
-        int op_ret   = -1;
-        int op_errno = 0;
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, prebuf,
+                                 postbuf, NULL, xdata);
+}
 
-        VALIDATE_OR_GOTO (this, out);
+int
+afr_fallocate_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        GF_IF_NATIVE_XATTR_GOTO ("trusted.afr.*",
-                                 name, op_errno, out);
+    local = frame->local;
+    priv = this->private;
 
-        GF_IF_NATIVE_XATTR_GOTO ("trusted.glusterfs.afr.*",
-                                 name, op_errno, out);
+    STACK_WIND_COOKIE(frame, afr_fallocate_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->fallocate, local->fd,
+                      local->cont.fallocate.mode, local->cont.fallocate.offset,
+                      local->cont.fallocate.len, local->xdata_req);
+    return 0;
+}
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this->private, out);
+int
+afr_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+              off_t offset, size_t len, dict_t *xdata)
+{
+    call_frame_t *transaction_frame = NULL;
+    afr_local_t *local = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        priv = this->private;
-        if (afr_is_split_brain (this, fd->inode)) {
-                op_errno = EIO;
-                goto out;
-        }
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        QUORUM_CHECK(fremovexattr, out);
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                goto out;
-        }
+    local->cont.fallocate.mode = mode;
+    local->cont.fallocate.offset = offset;
+    local->cont.fallocate.len = len;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (local, out);
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-        transaction_frame->local = local;
+    if (!local->xdata_req)
+        goto out;
 
-        local->op_ret = -1;
+    local->op = GF_FOP_FALLOCATE;
 
-        local->cont.removexattr.name = gf_strdup (name);
+    local->transaction.wind = afr_fallocate_wind;
+    local->transaction.unwind = afr_fallocate_unwind;
 
-        local->transaction.fop    = afr_fremovexattr_wind;
-        local->transaction.done   = afr_fremovexattr_done;
-        local->transaction.unwind = afr_fremovexattr_unwind;
+    local->transaction.main_frame = frame;
 
-        local->fd = fd_ref (fd);
+    local->transaction.start = local->cont.fallocate.offset;
+    local->transaction.len = 0;
 
-        local->transaction.main_frame = frame;
-        local->transaction.start   = LLONG_MAX - 1;
-        local->transaction.len     = 0;
+    afr_fix_open(fd, this);
 
-        op_ret = afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
-        if (op_ret < 0) {
-            op_errno = -op_ret;
-            goto out;
-        }
+    ret = afr_transaction(transaction_frame, this, AFR_DATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        op_ret = 0;
+    return 0;
 out:
-        if (op_ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (fremovexattr, frame, op_ret, op_errno, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(fallocate, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
-static int
-afr_fallocate_unwind (call_frame_t *frame, xlator_t *this)
+/* }}} */
+
+/* {{{ discard */
+
+int
+afr_discard_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t   *main_frame = NULL;
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        local = frame->local;
+    local = frame->local;
 
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame)
-                        main_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
-
-        if (main_frame) {
-                AFR_STACK_UNWIND (fallocate, main_frame, local->op_ret,
-                                  local->op_errno,
-                                  &local->cont.inode_wfop.prebuf,
-                                  &local->cont.inode_wfop.postbuf,
-                                  NULL);
-        }
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
-}
 
-static int
-afr_fallocate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                        struct iatt *postbuf, dict_t *xdata)
-{
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
-        int child_index = (long) cookie;
-        int call_count  = -1;
-        int need_unwind = 0;
-        int read_child  = 0;
-
-        local = frame->local;
-        priv  = this->private;
-
-        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
-
-        LOCK (&frame->lock);
-        {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
-
-                __inode_write_fop_cbk (frame, child_index, read_child, this,
-                                       &op_ret, &op_errno, prebuf, postbuf,
-                                       xdata);
-
-                if ((local->success_count >= priv->wait_count)
-                    && local->read_child_returned) {
-                        need_unwind = 1;
-                }
-        }
-        UNLOCK (&frame->lock);
+    AFR_STACK_UNWIND(discard, main_frame, local->op_ret, local->op_errno,
+                     &local->cont.inode_wfop.prebuf,
+                     &local->cont.inode_wfop.postbuf, local->xdata_rsp);
+    return 0;
+}
 
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
+int
+afr_discard_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
+{
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, prebuf,
+                                 postbuf, NULL, xdata);
+}
 
-        call_count = afr_frame_return (frame);
+int
+afr_discard_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
+    local = frame->local;
+    priv = this->private;
 
-        return 0;
+    STACK_WIND_COOKIE(frame, afr_discard_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->discard, local->fd,
+                      local->cont.discard.offset, local->cont.discard.len,
+                      local->xdata_req);
+    return 0;
 }
 
-static int
-afr_fallocate_wind (call_frame_t *frame, xlator_t *this)
+int
+afr_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+            size_t len, dict_t *xdata)
 {
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int call_count = -1;
-        int i = 0;
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        local = frame->local;
-        priv = this->private;
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
+    local->cont.discard.offset = offset;
+    local->cont.discard.len = len;
 
-        local->call_count = call_count;
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->transaction.pre_op[i]) {
-                        STACK_WIND_COOKIE (frame, afr_fallocate_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->fallocate,
-                                           local->fd,
-                                           local->cont.fallocate.mode,
-                                           local->cont.fallocate.offset,
-                                           local->cont.fallocate.len,
-                                           NULL);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-        return 0;
-}
+    if (!local->xdata_req)
+        goto out;
 
-static int
-afr_fallocate_done (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local = NULL;
+    local->op = GF_FOP_DISCARD;
 
-        local = frame->local;
+    local->transaction.wind = afr_discard_wind;
+    local->transaction.unwind = afr_discard_unwind;
 
-        local->transaction.unwind (frame, this);
+    local->transaction.main_frame = frame;
 
-        AFR_STACK_DESTROY (frame);
+    local->transaction.start = local->cont.discard.offset;
+    local->transaction.len = 0;
 
-        return 0;
-}
+    afr_fix_open(fd, this);
 
-static int
-afr_do_fallocate (call_frame_t *frame, xlator_t *this)
-{
-        call_frame_t * transaction_frame = NULL;
-        afr_local_t *  local             = NULL;
-        int op_ret   = -1;
-        int op_errno = 0;
+    ret = afr_transaction(transaction_frame, this, AFR_DATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        local = frame->local;
+    return 0;
+out:
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                goto out;
-        }
+    AFR_STACK_UNWIND(discard, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
+}
 
-        transaction_frame->local = local;
-        frame->local = NULL;
+/* {{{ zerofill */
 
-        local->op = GF_FOP_FALLOCATE;
+int
+afr_zerofill_unwind(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        local->transaction.fop    = afr_fallocate_wind;
-        local->transaction.done   = afr_fallocate_done;
-        local->transaction.unwind = afr_fallocate_unwind;
+    local = frame->local;
 
-        local->transaction.main_frame = frame;
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
+        return 0;
 
-        local->transaction.start   = local->cont.fallocate.offset;
-        local->transaction.len     = 0;
+    AFR_STACK_UNWIND(discard, main_frame, local->op_ret, local->op_errno,
+                     &local->cont.inode_wfop.prebuf,
+                     &local->cont.inode_wfop.postbuf, local->xdata_rsp);
+    return 0;
+}
 
-        /* fallocate can modify the file size */
-        op_ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
-        if (op_ret < 0) {
-            op_errno = -op_ret;
-            goto out;
-        }
+int
+afr_zerofill_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                      struct iatt *postbuf, dict_t *xdata)
+{
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, prebuf,
+                                 postbuf, NULL, xdata);
+}
 
-        op_ret = 0;
-out:
-        if (op_ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (fallocate, frame, op_ret, op_errno, NULL,
-                                  NULL, NULL);
-        }
+int
+afr_zerofill_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        return 0;
+    local = frame->local;
+    priv = this->private;
+
+    STACK_WIND_COOKIE(frame, afr_zerofill_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->zerofill, local->fd,
+                      local->cont.zerofill.offset, local->cont.zerofill.len,
+                      local->xdata_req);
+    return 0;
 }
 
 int
-afr_fallocate (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
-               off_t offset, size_t len, dict_t *xdata)
+afr_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             size_t len, dict_t *xdata)
 {
-        afr_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
-        call_frame_t   *transaction_frame = NULL;
-        int ret = -1;
-        int op_errno = 0;
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        priv = this->private;
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        if (afr_is_split_brain (this, fd->inode)) {
-                op_errno = EIO;
-                goto out;
-        }
-        QUORUM_CHECK(fallocate,out);
+    local->cont.zerofill.offset = offset;
+    local->cont.zerofill.len = len;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-        local->cont.fallocate.mode = mode;
-        local->cont.fallocate.offset  = offset;
-        local->cont.fallocate.len = len;
+    if (!local->xdata_req)
+        goto out;
 
-        local->fd = fd_ref (fd);
+    local->op = GF_FOP_ZEROFILL;
 
-        afr_open_fd_fix (fd, this);
+    local->transaction.wind = afr_zerofill_wind;
+    local->transaction.unwind = afr_zerofill_unwind;
 
-        afr_do_fallocate (frame, this);
+    local->transaction.main_frame = frame;
 
-        ret = 0;
+    local->transaction.start = local->cont.zerofill.offset;
+    local->transaction.len = len;
+
+    afr_fix_open(fd, this);
+
+    ret = afr_transaction(transaction_frame, this, AFR_DATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
+
+    return 0;
 out:
-        if (ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (fallocate, frame, -1, op_errno, NULL, NULL, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(zerofill, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
 /* }}} */
 
-/* {{{ discard */
+int32_t
+afr_xattrop_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, dict_t *xattr,
+                     dict_t *xdata)
+{
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, NULL,
+                                 NULL, xattr, xdata);
+}
 
-static int
-afr_discard_unwind (call_frame_t *frame, xlator_t *this)
+int
+afr_xattrop_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t   *main_frame = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        local = frame->local;
+    local = frame->local;
+    priv = this->private;
 
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame)
-                        main_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
-
-        if (main_frame) {
-                AFR_STACK_UNWIND (discard, main_frame, local->op_ret,
-                                  local->op_errno,
-                                  &local->cont.inode_wfop.prebuf,
-                                  &local->cont.inode_wfop.postbuf,
-                                  NULL);
-        }
-        return 0;
+    STACK_WIND_COOKIE(frame, afr_xattrop_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->xattrop, &local->loc,
+                      local->cont.xattrop.optype, local->cont.xattrop.xattr,
+                      local->xdata_req);
+    return 0;
 }
 
-static int
-afr_discard_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                      struct iatt *postbuf, dict_t *xdata)
+int
+afr_xattrop_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
-        int child_index = (long) cookie;
-        int call_count  = -1;
-        int need_unwind = 0;
-        int read_child  = 0;
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        local = frame->local;
-        priv  = this->private;
+    local = frame->local;
 
-        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
-
-        LOCK (&frame->lock);
-        {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
+        return 0;
 
-                __inode_write_fop_cbk (frame, child_index, read_child, this,
-                                       &op_ret, &op_errno, prebuf, postbuf,
-                                       xdata);
+    AFR_STACK_UNWIND(xattrop, main_frame, local->op_ret, local->op_errno,
+                     local->xattr_rsp, local->xdata_rsp);
+    return 0;
+}
 
-                if ((local->success_count >= priv->wait_count)
-                    && local->read_child_returned) {
-                        need_unwind = 1;
-                }
-        }
-        UNLOCK (&frame->lock);
+int32_t
+afr_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+            gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        call_count = afr_frame_return (frame);
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
+    local->cont.xattrop.xattr = dict_ref(xattr);
+    local->cont.xattrop.optype = optype;
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
 
-        return 0;
-}
+    local->transaction.wind = afr_xattrop_wind;
+    local->transaction.unwind = afr_xattrop_unwind;
 
-static int
-afr_discard_wind (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-        int call_count = -1;
-        int i = 0;
+    loc_copy(&local->loc, loc);
+    ret = afr_set_inode_local(this, local, loc->inode);
+    if (ret)
+        goto out;
 
-        local = frame->local;
-        priv = this->private;
+    local->op = GF_FOP_XATTROP;
 
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
+    local->transaction.main_frame = frame;
+    local->transaction.start = LLONG_MAX - 1;
+    local->transaction.len = 0;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
+    ret = afr_transaction(transaction_frame, this, AFR_METADATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        local->call_count = call_count;
+    return 0;
+out:
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->transaction.pre_op[i]) {
-                        STACK_WIND_COOKIE (frame, afr_discard_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->discard,
-                                           local->fd,
-                                           local->cont.discard.offset,
-                                           local->cont.discard.len,
-                                           NULL);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
+    AFR_STACK_UNWIND(xattrop, frame, -1, op_errno, NULL, NULL);
+    return 0;
+}
 
-        return 0;
+int32_t
+afr_fxattrop_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xattr,
+                      dict_t *xdata)
+{
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, NULL,
+                                 NULL, xattr, xdata);
 }
 
-static int
-afr_discard_done (call_frame_t *frame, xlator_t *this)
+int
+afr_fxattrop_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_local_t *local = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        local = frame->local;
+    local = frame->local;
+    priv = this->private;
 
-        local->transaction.unwind (frame, this);
+    STACK_WIND_COOKIE(frame, afr_fxattrop_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->fxattrop, local->fd,
+                      local->cont.xattrop.optype, local->cont.xattrop.xattr,
+                      local->xdata_req);
+    return 0;
+}
 
-        AFR_STACK_DESTROY (frame);
+int
+afr_fxattrop_unwind(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
+
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
+
+    AFR_STACK_UNWIND(fxattrop, main_frame, local->op_ret, local->op_errno,
+                     local->xattr_rsp, local->xdata_rsp);
+    return 0;
 }
 
-static int
-afr_do_discard (call_frame_t *frame, xlator_t *this)
+int32_t
+afr_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+             gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
 {
-        call_frame_t * transaction_frame = NULL;
-        afr_local_t *  local             = NULL;
-        int op_ret   = -1;
-        int op_errno = 0;
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        local = frame->local;
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                goto out;
-        }
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        transaction_frame->local = local;
-        frame->local = NULL;
+    local->cont.xattrop.xattr = dict_ref(xattr);
+    local->cont.xattrop.optype = optype;
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
 
-        local->op = GF_FOP_DISCARD;
+    local->transaction.wind = afr_fxattrop_wind;
+    local->transaction.unwind = afr_fxattrop_unwind;
 
-        local->transaction.fop    = afr_discard_wind;
-        local->transaction.done   = afr_discard_done;
-        local->transaction.unwind = afr_discard_unwind;
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
 
-        local->transaction.main_frame = frame;
+    local->op = GF_FOP_FXATTROP;
 
-        local->transaction.start   = local->cont.discard.offset;
-        local->transaction.len     = 0;
+    local->transaction.main_frame = frame;
+    local->transaction.start = LLONG_MAX - 1;
+    local->transaction.len = 0;
 
-        op_ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
-        if (op_ret < 0) {
-            op_errno = -op_ret;
-            goto out;
-        }
+    ret = afr_transaction(transaction_frame, this, AFR_METADATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        op_ret = 0;
+    return 0;
 out:
-        if (op_ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (discard, frame, op_ret, op_errno, NULL,
-                                  NULL, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(fxattrop, frame, -1, op_errno, NULL, NULL);
+    return 0;
 }
 
 int
-afr_discard (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
-             size_t len, dict_t *xdata)
+afr_fsync_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
-        call_frame_t   *transaction_frame = NULL;
-        int ret = -1;
-        int op_errno = 0;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        priv = this->private;
+    local = frame->local;
 
-        if (afr_is_split_brain (this, fd->inode)) {
-                op_errno = EIO;
-                goto out;
-        }
-        QUORUM_CHECK(discard, out);
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
+        return 0;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
+    AFR_STACK_UNWIND(fsync, main_frame, local->op_ret, local->op_errno,
+                     &local->cont.inode_wfop.prebuf,
+                     &local->cont.inode_wfop.postbuf, local->xdata_rsp);
 
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    return 0;
+}
 
-        local->cont.discard.offset  = offset;
-        local->cont.discard.len = len;
+int
+afr_fsync_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                   struct iatt *postbuf, dict_t *xdata)
+{
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, prebuf,
+                                 postbuf, NULL, xdata);
+}
 
-        local->fd = fd_ref (fd);
+int
+afr_fsync_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        afr_open_fd_fix (fd, this);
+    local = frame->local;
+    priv = this->private;
 
-        afr_do_discard(frame, this);
+    STACK_WIND_COOKIE(frame, afr_fsync_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->fsync, local->fd,
+                      local->cont.fsync.datasync, local->xdata_req);
+    return 0;
+}
 
-        ret = 0;
+int
+afr_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+          dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int32_t op_errno = ENOMEM;
+    int8_t last_fsync = 0;
+
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
+
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
+
+    if (xdata) {
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+        if (dict_get_int8(xdata, "last-fsync", &last_fsync) == 0) {
+            if (last_fsync) {
+                local->transaction.disable_delayed_post_op = _gf_true;
+            }
+        }
+    } else {
+        local->xdata_req = dict_new();
+    }
+
+    if (!local->xdata_req)
+        goto out;
+
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
+
+    local->op = GF_FOP_FSYNC;
+    local->cont.fsync.datasync = datasync;
+
+    if (afr_fd_has_witnessed_unstable_write(this, fd->inode)) {
+        /* don't care. we only wanted to CLEAR the bit */
+    }
+
+    local->transaction.wind = afr_fsync_wind;
+    local->transaction.unwind = afr_fsync_unwind;
+
+    local->transaction.main_frame = frame;
+
+    ret = afr_transaction(transaction_frame, this, AFR_DATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
+
+    return 0;
 out:
-        if (ret < 0) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (discard, frame, -1, op_errno, NULL, NULL, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(fsync, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
 }
diff --git a/xlators/cluster/afr/src/afr-inode-write.h b/xlators/cluster/afr/src/afr-inode-write.h
index 883faae6cb5..a787069b7a1 100644
--- a/xlators/cluster/afr/src/afr-inode-write.h
+++ b/xlators/cluster/afr/src/afr-inode-write.h
@@ -12,67 +12,83 @@
 #define __INODE_WRITE_H__
 
 int32_t
-afr_chmod (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, mode_t mode, dict_t *xdata);
+afr_chmod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+          dict_t *xdata);
 
 int32_t
-afr_chown (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, uid_t uid, gid_t gid, dict_t *xdata);
+afr_chown(call_frame_t *frame, xlator_t *this, loc_t *loc, uid_t uid, gid_t gid,
+          dict_t *xdata);
 
 int
-afr_fchown (call_frame_t *frame, xlator_t *this,
-	    fd_t *fd, uid_t uid, gid_t gid, dict_t *xdata);
+afr_fchown(call_frame_t *frame, xlator_t *this, fd_t *fd, uid_t uid, gid_t gid,
+           dict_t *xdata);
 
 int32_t
-afr_fchmod (call_frame_t *frame, xlator_t *this,
-	    fd_t *fd, mode_t mode, dict_t *xdata);
+afr_fchmod(call_frame_t *frame, xlator_t *this, fd_t *fd, mode_t mode,
+           dict_t *xdata);
 
 int32_t
-afr_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
-	    struct iovec *vector, int32_t count, off_t offset,
-            uint32_t flags, struct iobref *iobref, dict_t *xdata);
+afr_writev(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,
+           int32_t count, off_t offset, uint32_t flags, struct iobref *iobref,
+           dict_t *xdata);
 
 int32_t
-afr_truncate (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, off_t offset, dict_t *xdata);
+afr_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+             dict_t *xdata);
 
 int32_t
-afr_ftruncate (call_frame_t *frame, xlator_t *this,
-	       fd_t *fd, off_t offset, dict_t *xdata);
+afr_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+              dict_t *xdata);
 
 int32_t
-afr_utimens (call_frame_t *frame, xlator_t *this,
-	     loc_t *loc, struct timespec tv[2], dict_t *xdata);
+afr_utimens(call_frame_t *frame, xlator_t *this, loc_t *loc,
+            struct timespec tv[2], dict_t *xdata);
 
 int
-afr_setattr (call_frame_t *frame, xlator_t *this,
-             loc_t *loc, struct iatt *buf, int32_t valid, dict_t *xdata);
+afr_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *buf,
+            int32_t valid, dict_t *xdata);
 
 int
-afr_fsetattr (call_frame_t *frame, xlator_t *this,
-              fd_t *fd, struct iatt *buf, int32_t valid, dict_t *xdata);
+afr_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *buf,
+             int32_t valid, dict_t *xdata);
 
 int32_t
-afr_setxattr (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, dict_t *dict, int32_t flags, dict_t *xdata);
+afr_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+             int32_t flags, dict_t *xdata);
 
 int32_t
-afr_fsetxattr (call_frame_t *frame, xlator_t *this,
-               fd_t *fd, dict_t *dict, int32_t flags, dict_t *xdata);
+afr_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+              int32_t flags, dict_t *xdata);
 
 int32_t
-afr_removexattr (call_frame_t *frame, xlator_t *this,
-		 loc_t *loc, const char *name, dict_t *xdata);
+afr_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                const char *name, dict_t *xdata);
 
 int32_t
-afr_fremovexattr (call_frame_t *frame, xlator_t *this,
-                  fd_t *fd, const char *name, dict_t *xdata);
+afr_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                 const char *name, dict_t *xdata);
 
 int
-afr_discard (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
-             size_t len, dict_t *xdata);
+afr_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+            size_t len, dict_t *xdata);
 
 int
-afr_fallocate (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
-               off_t offset, size_t len, dict_t *xdata);
+afr_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+              off_t offset, size_t len, dict_t *xdata);
+
+int
+afr_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             off_t len, dict_t *xdata);
+
+int32_t
+afr_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+            gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata);
+
+int32_t
+afr_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+             gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata);
+
+int
+afr_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+          dict_t *xdata);
 #endif /* __INODE_WRITE_H__ */
diff --git a/xlators/cluster/afr/src/afr-lk-common.c b/xlators/cluster/afr/src/afr-lk-common.c
index e2136089bdc..bc8eabe0f43 100644
--- a/xlators/cluster/afr/src/afr-lk-common.c
+++ b/xlators/cluster/afr/src/afr-lk-common.c
@@ -8,2163 +8,784 @@
   cases as published by the Free Software Foundation.
 */
 
-#include "dict.h"
-#include "byte-order.h"
-#include "common-utils.h"
+#include <glusterfs/dict.h>
+#include <glusterfs/byte-order.h>
+#include <glusterfs/common-utils.h>
 
 #include "afr.h"
 #include "afr-transaction.h"
+#include "afr-messages.h"
 
 #include <signal.h>
 
-
-#define LOCKED_NO       0x0        /* no lock held */
-#define LOCKED_YES      0x1        /* for DATA, METADATA, ENTRY and higher_path */
-#define LOCKED_LOWER    0x2        /* for lower path */
-
-#define AFR_TRACE_INODELK_IN(frame, this, params ...)           \
-        do {                                                    \
-                afr_private_t *_priv = this->private;           \
-                if (!_priv->inodelk_trace)                      \
-                        break;                                  \
-                afr_trace_inodelk_in (frame, this, params);     \
-        } while (0);
-
-#define AFR_TRACE_INODELK_OUT(frame, this, params ...)          \
-        do {                                                    \
-                afr_private_t *_priv = this->private;           \
-                if (!_priv->inodelk_trace)                      \
-                        break;                                  \
-                afr_trace_inodelk_out (frame, this, params);    \
-        } while (0);
-
-#define AFR_TRACE_ENTRYLK_IN(frame, this, params ...)           \
-        do {                                                    \
-                afr_private_t *_priv = this->private;           \
-                if (!_priv->entrylk_trace)                      \
-                        break;                                  \
-                afr_trace_entrylk_in (frame, this, params);     \
-        } while (0);
-
-#define AFR_TRACE_ENTRYLK_OUT(frame, this, params ...)          \
-        do {                                                    \
-                afr_private_t *_priv = this->private;           \
-                if (!_priv->entrylk_trace)                      \
-                        break;                                  \
-                afr_trace_entrylk_out (frame, this, params);    \
-        } while (0);
-
-int
-afr_entry_lockee_cmp (const void *l1, const void *l2)
-{
-        const afr_entry_lockee_t       *r1 = l1;
-        const afr_entry_lockee_t       *r2 = l2;
-        int                            ret = 0;
-        uuid_t                         gfid1 = {0};
-        uuid_t                         gfid2 = {0};
-
-        loc_gfid ((loc_t*)&r1->loc, gfid1);
-        loc_gfid ((loc_t*)&r2->loc, gfid2);
-        ret = uuid_compare (gfid1, gfid2);
-        /*Entrylks with NULL basename are the 'smallest'*/
-        if (ret == 0) {
-                if (!r1->basename)
-                        return -1;
-                if (!r2->basename)
-                        return 1;
-                ret = strcmp (r1->basename, r2->basename);
-        }
-
-        if (ret <= 0)
-                return -1;
-        else
-                return 1;
-}
-
-int afr_lock_blocking (call_frame_t *frame, xlator_t *this, int child_index);
-
-static int
-afr_copy_locked_nodes (call_frame_t *frame, xlator_t *this);
-
-static uint64_t afr_lock_number = 1;
-
-static uint64_t
-get_afr_lock_number ()
-{
-        return (++afr_lock_number);
-}
-
-int
-afr_set_lock_number (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t         *local    = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        int_lock->lock_number = get_afr_lock_number ();
-
-        return 0;
-}
+#define LOCKED_NO 0x0    /* no lock held */
+#define LOCKED_YES 0x1   /* for DATA, METADATA, ENTRY and higher_path */
+#define LOCKED_LOWER 0x2 /* for lower path */
 
 void
-afr_set_lk_owner (call_frame_t *frame, xlator_t *this, void *lk_owner)
-{
-        gf_log (this->name, GF_LOG_TRACE,
-                "Setting lk-owner=%llu",
-                (unsigned long long) (unsigned long)lk_owner);
-
-        set_lk_owner_from_ptr (&frame->root->lk_owner, lk_owner);
-}
-
-static int
-is_afr_lock_selfheal (afr_local_t *local)
+afr_lockee_cleanup(afr_lockee_t *lockee)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        int                  ret      = -1;
-
-        int_lock = &local->internal_lock;
-
-        switch (int_lock->selfheal_lk_type) {
-        case AFR_DATA_SELF_HEAL_LK:
-        case AFR_METADATA_SELF_HEAL_LK:
-                ret = 1;
-                break;
-        case AFR_ENTRY_SELF_HEAL_LK:
-                ret = 0;
-                break;
-        }
+    if (lockee->fd) {
+        fd_unref(lockee->fd);
+        lockee->fd = NULL;
+    } else {
+        loc_wipe(&lockee->loc);
+    }
 
-        return ret;
+    GF_FREE(lockee->basename);
+    lockee->basename = NULL;
+    GF_FREE(lockee->locked_nodes);
+    lockee->locked_nodes = NULL;
 
-}
-
-int32_t
-internal_lock_count (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t   *local = NULL;
-        afr_private_t *priv  = NULL;
-        int32_t call_count = 0;
-        int i = 0;
-
-        local = frame->local;
-        priv  = this->private;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i])
-                        ++call_count;
-        }
-
-        return call_count;
-}
-
-static void
-afr_print_inodelk (char *str, int size, int cmd,
-                   struct gf_flock *flock, gf_lkowner_t *owner)
-{
-        char *cmd_str = NULL;
-        char *type_str = NULL;
-
-        switch (cmd) {
-#if F_GETLK != F_GETLK64
-        case F_GETLK64:
-#endif
-        case F_GETLK:
-                cmd_str = "GETLK";
-                break;
-
-#if F_SETLK != F_SETLK64
-        case F_SETLK64:
-#endif
-        case F_SETLK:
-                cmd_str = "SETLK";
-                break;
-
-#if F_SETLKW != F_SETLKW64
-        case F_SETLKW64:
-#endif
-        case F_SETLKW:
-                cmd_str = "SETLKW";
-                break;
-
-        default:
-                cmd_str = "<null>";
-                break;
-        }
-
-        switch (flock->l_type) {
-        case F_RDLCK:
-                type_str = "READ";
-                break;
-        case F_WRLCK:
-                type_str = "WRITE";
-                break;
-        case F_UNLCK:
-                type_str = "UNLOCK";
-                break;
-        default:
-                type_str = "UNKNOWN";
-                break;
-        }
-
-        snprintf (str, size, "lock=INODELK, cmd=%s, type=%s, "
-                  "start=%llu, len=%llu, pid=%llu, lk-owner=%s",
-                  cmd_str, type_str, (unsigned long long) flock->l_start,
-                  (unsigned long long) flock->l_len,
-                  (unsigned long long) flock->l_pid,
-                  lkowner_utoa (owner));
-
-}
-
-static void
-afr_print_lockee (char *str, int size, loc_t *loc, fd_t *fd,
-                  int child_index)
-{
-        snprintf (str, size, "path=%s, fd=%p, child=%d",
-                  loc->path ? loc->path : "<nul>",
-                  fd ? fd : NULL,
-                  child_index);
+    return;
 }
 
 void
-afr_print_entrylk (char *str, int size, const char *basename,
-                   gf_lkowner_t *owner)
-{
-        snprintf (str, size, "Basename=%s, lk-owner=%s",
-                  basename ? basename : "<nul>",
-                  lkowner_utoa (owner));
-}
-
-static void
-afr_print_verdict (int op_ret, int op_errno, char *str)
+afr_lockees_cleanup(afr_internal_lock_t *int_lock)
 {
-        if (op_ret < 0) {
-                if (op_errno == EAGAIN)
-                        strcpy (str, "EAGAIN");
-                else
-                        strcpy (str, "FAILED");
-        }
-        else
-                strcpy (str, "GRANTED");
-}
+    int i = 0;
 
-static void
-afr_set_lock_call_type (afr_lock_call_type_t lock_call_type,
-                        char *lock_call_type_str,
-                        afr_internal_lock_t *int_lock)
-{
-        switch (lock_call_type) {
-        case AFR_INODELK_TRANSACTION:
-                if (int_lock->transaction_lk_type == AFR_TRANSACTION_LK)
-                        strcpy (lock_call_type_str, "AFR_INODELK_TRANSACTION");
-                else
-                        strcpy (lock_call_type_str, "AFR_INODELK_SELFHEAL");
-                break;
-        case AFR_INODELK_NB_TRANSACTION:
-                if (int_lock->transaction_lk_type == AFR_TRANSACTION_LK)
-                        strcpy (lock_call_type_str, "AFR_INODELK_NB_TRANSACTION");
-                else
-                        strcpy (lock_call_type_str, "AFR_INODELK_NB_SELFHEAL");
-                break;
-        case AFR_ENTRYLK_TRANSACTION:
-                if (int_lock->transaction_lk_type == AFR_TRANSACTION_LK)
-                        strcpy (lock_call_type_str, "AFR_ENTRYLK_TRANSACTION");
-                else
-                        strcpy (lock_call_type_str, "AFR_ENTRYLK_SELFHEAL");
-                break;
-        case AFR_ENTRYLK_NB_TRANSACTION:
-                if (int_lock->transaction_lk_type == AFR_TRANSACTION_LK)
-                        strcpy (lock_call_type_str, "AFR_ENTRYLK_NB_TRANSACTION");
-                else
-                        strcpy (lock_call_type_str, "AFR_ENTRYLK_NB_SELFHEAL");
-                break;
-        default:
-                strcpy (lock_call_type_str, "UNKNOWN");
-                break;
-        }
+    for (i = 0; i < int_lock->lockee_count; i++) {
+        afr_lockee_cleanup(&int_lock->lockee[i]);
+    }
 
+    return;
 }
-
-static void
-afr_trace_inodelk_out (call_frame_t *frame, xlator_t *this,
-                       afr_lock_call_type_t lock_call_type,
-                       afr_lock_op_type_t lk_op_type, struct gf_flock *flock,
-                       int op_ret, int op_errno, int32_t child_index)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-
-        char lockee[256];
-        char lock_call_type_str[256];
-        char verdict[16];
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        afr_print_lockee (lockee, 256, &local->loc, local->fd, child_index);
-
-        afr_set_lock_call_type (lock_call_type, lock_call_type_str, int_lock);
-
-        afr_print_verdict (op_ret, op_errno, verdict);
-
-        gf_log (this->name, GF_LOG_INFO,
-                "[%s %s] [%s] lk-owner=%s Lockee={%s} Number={%llu}",
-                lock_call_type_str,
-                lk_op_type == AFR_LOCK_OP ? "LOCK REPLY" : "UNLOCK REPLY",
-                verdict, lkowner_utoa (&frame->root->lk_owner), lockee,
-                (unsigned long long) int_lock->lock_number);
-
-}
-
-static void
-afr_trace_inodelk_in (call_frame_t *frame, xlator_t *this,
-                      afr_lock_call_type_t lock_call_type,
-                      afr_lock_op_type_t lk_op_type, struct gf_flock *flock,
-                      int32_t cmd, int32_t child_index)
-{
-        afr_local_t         *local    = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-
-        char lock[256];
-        char lockee[256];
-        char lock_call_type_str[256];
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        afr_print_inodelk (lock, 256, cmd, flock, &frame->root->lk_owner);
-        afr_print_lockee (lockee, 256, &local->loc, local->fd, child_index);
-
-        afr_set_lock_call_type (lock_call_type, lock_call_type_str, int_lock);
-
-        gf_log (this->name, GF_LOG_INFO,
-                "[%s %s] Lock={%s} Lockee={%s} Number={%llu}",
-                lock_call_type_str,
-                lk_op_type == AFR_LOCK_OP ? "LOCK REQUEST" : "UNLOCK REQUEST",
-                lock, lockee,
-                (unsigned long long) int_lock->lock_number);
-
+int
+afr_entry_lockee_cmp(const void *l1, const void *l2)
+{
+    const afr_lockee_t *r1 = l1;
+    const afr_lockee_t *r2 = l2;
+    int ret = 0;
+    uuid_t gfid1 = {0};
+    uuid_t gfid2 = {0};
+
+    loc_gfid((loc_t *)&r1->loc, gfid1);
+    loc_gfid((loc_t *)&r2->loc, gfid2);
+    ret = gf_uuid_compare(gfid1, gfid2);
+    /*Entrylks with NULL basename are the 'smallest'*/
+    if (ret == 0) {
+        if (!r1->basename)
+            return -1;
+        if (!r2->basename)
+            return 1;
+        ret = strcmp(r1->basename, r2->basename);
+    }
+
+    if (ret <= 0)
+        return -1;
+    else
+        return 1;
 }
 
-static void
-afr_trace_entrylk_in (call_frame_t *frame, xlator_t *this,
-                      afr_lock_call_type_t lock_call_type,
-                      afr_lock_op_type_t lk_op_type, const char *basename,
-                      int32_t cookie)
-{
-        afr_local_t         *local    = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-        afr_private_t       *priv     = NULL;
-        int                 child_index = 0;
-        int                 lockee_no = 0;
-
-        char lock[256];
-        char lockee[256];
-        char lock_call_type_str[256];
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        priv     = this->private;
-
-        if (!priv->entrylk_trace) {
-                return;
-        }
-        lockee_no = cookie / priv->child_count;
-        child_index = cookie % priv->child_count;
-
-        afr_print_entrylk (lock, 256, basename, &frame->root->lk_owner);
-        afr_print_lockee (lockee, 256, &int_lock->lockee[lockee_no].loc, local->fd,
-                          child_index);
-
-        afr_set_lock_call_type (lock_call_type, lock_call_type_str, int_lock);
-
-        gf_log (this->name, GF_LOG_INFO,
-                "[%s %s] Lock={%s} Lockee={%s} Number={%llu}, Cookie={%d}",
-                lock_call_type_str,
-                lk_op_type == AFR_LOCK_OP ? "LOCK REQUEST" : "UNLOCK REQUEST",
-                lock, lockee,
-                (unsigned long long) int_lock->lock_number,
-                cookie);
-}
+int
+afr_lock_blocking(call_frame_t *frame, xlator_t *this, int child_index);
 
-static void
-afr_trace_entrylk_out (call_frame_t *frame, xlator_t *this,
-                       afr_lock_call_type_t lock_call_type,
-                       afr_lock_op_type_t lk_op_type, const char *basename,
-                       int op_ret, int op_errno, int32_t cookie)
+void
+afr_set_lk_owner(call_frame_t *frame, xlator_t *this, void *lk_owner)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_private_t       *priv     = NULL;
-        int                 lockee_no = 0;
-        int                 child_index = 0;
-
-        char lock[256];
-        char lockee[256];
-        char lock_call_type_str[256];
-        char verdict[16];
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        priv     = this->private;
-
-        if (!priv->entrylk_trace) {
-                return;
-        }
-        lockee_no = cookie / priv->child_count;
-        child_index = cookie % priv->child_count;
-
-        afr_print_entrylk (lock, 256, basename, &frame->root->lk_owner);
-        afr_print_lockee (lockee, 256, &int_lock->lockee[lockee_no].loc, local->fd,
-                          child_index);
-
-        afr_set_lock_call_type (lock_call_type, lock_call_type_str, int_lock);
-
-        afr_print_verdict (op_ret, op_errno, verdict);
-
-        gf_log (this->name, GF_LOG_INFO,
-                "[%s %s] [%s] Lock={%s} Lockee={%s} Number={%llu} Cookie={%d}",
-                lock_call_type_str,
-                lk_op_type == AFR_LOCK_OP ? "LOCK REPLY" : "UNLOCK REPLY",
-                verdict,
-                lock, lockee,
-                (unsigned long long) int_lock->lock_number,
-                cookie);
+    gf_msg_trace(this->name, 0, "Setting lk-owner=%llu",
+                 (unsigned long long)(unsigned long)lk_owner);
 
+    set_lk_owner_from_ptr(&frame->root->lk_owner, lk_owner);
 }
 
-static int
-transaction_lk_op (afr_local_t *local)
+int32_t
+internal_lock_count(call_frame_t *frame, xlator_t *this)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        int ret = -1;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int32_t call_count = 0;
+    int i = 0;
 
-        int_lock = &local->internal_lock;
+    local = frame->local;
+    priv = this->private;
 
-        if (int_lock->transaction_lk_type == AFR_TRANSACTION_LK) {
-                gf_log (THIS->name, GF_LOG_DEBUG,
-                        "lk op is for a transaction");
-                ret = 1;
-        }
-        else if (int_lock->transaction_lk_type == AFR_SELFHEAL_LK) {
-                gf_log (THIS->name, GF_LOG_DEBUG,
-                        "lk op is for a self heal");
-
-                ret = 0;
-        }
-
-        if (ret == -1)
-                gf_log (THIS->name, GF_LOG_DEBUG,
-                        "lk op is not set");
-
-        return ret;
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->child_up[i])
+            ++call_count;
+    }
 
+    return call_count;
 }
 
-static int
-is_afr_lock_transaction (afr_local_t *local)
+int
+afr_add_entry_lockee(afr_local_t *local, loc_t *loc, char *basename,
+                     int child_count)
 {
-        int ret = 0;
+    int ret = -ENOMEM;
+    afr_internal_lock_t *int_lock = &local->internal_lock;
+    afr_lockee_t *lockee = &int_lock->lockee[int_lock->lockee_count];
 
-        switch (local->transaction.type) {
-        case AFR_DATA_TRANSACTION:
-        case AFR_METADATA_TRANSACTION:
-                ret = 1;
-                break;
+    GF_ASSERT(int_lock->lockee_count < AFR_LOCKEE_COUNT_MAX);
+    loc_copy(&lockee->loc, loc);
+    lockee->basename = (basename) ? gf_strdup(basename) : NULL;
+    if (basename && !lockee->basename)
+        goto out;
 
-        case AFR_ENTRY_RENAME_TRANSACTION:
-        case AFR_ENTRY_TRANSACTION:
-                ret = 0;
-                break;
+    lockee->locked_count = 0;
+    lockee->locked_nodes = GF_CALLOC(child_count, sizeof(*lockee->locked_nodes),
+                                     gf_afr_mt_afr_node_character);
 
-        }
+    if (!lockee->locked_nodes)
+        goto out;
 
-        return ret;
+    ret = 0;
+    int_lock->lockee_count++;
+out:
+    if (ret) {
+        afr_lockee_cleanup(lockee);
+    }
+    return ret;
 }
 
 int
-afr_init_entry_lockee (afr_entry_lockee_t *lockee, afr_local_t *local,
-                       loc_t *loc, char *basename, int child_count)
+afr_add_inode_lockee(afr_local_t *local, int child_count)
 {
-        int     ret     = -1;
+    int ret = -ENOMEM;
+    afr_internal_lock_t *int_lock = &local->internal_lock;
+    afr_lockee_t *lockee = &int_lock->lockee[int_lock->lockee_count];
 
-        loc_copy (&lockee->loc, loc);
-        lockee->basename        = (basename)? gf_strdup (basename): NULL;
-        if (basename && !lockee->basename)
-                goto out;
+    if (local->fd) {
+        lockee->fd = fd_ref(local->fd);
+    } else {
+        loc_copy(&lockee->loc, &local->loc);
+    }
 
-        lockee->locked_count    = 0;
-        lockee->locked_nodes    = GF_CALLOC (child_count,
-                                             sizeof (*lockee->locked_nodes),
-                                             gf_afr_mt_afr_node_character);
+    lockee->locked_count = 0;
+    lockee->locked_nodes = GF_CALLOC(child_count, sizeof(*lockee->locked_nodes),
+                                     gf_afr_mt_afr_node_character);
 
-        if (!lockee->locked_nodes)
-                goto out;
+    if (!lockee->locked_nodes)
+        goto out;
 
-        ret = 0;
+    ret = 0;
+    int_lock->lockee_count++;
 out:
-        return ret;
-
-}
-
-void
-afr_entry_lockee_cleanup (afr_internal_lock_t *int_lock)
-{
-        int     i   = 0;
-
-        for (i = 0; i < int_lock->lockee_count; i++) {
-                loc_wipe (&int_lock->lockee[i].loc);
-                if (int_lock->lockee[i].basename)
-                        GF_FREE (int_lock->lockee[i].basename);
-                if (int_lock->lockee[i].locked_nodes)
-                       GF_FREE (int_lock->lockee[i].locked_nodes);
-        }
-
-        return;
+    if (ret) {
+        afr_lockee_cleanup(lockee);
+    }
+    return ret;
 }
 
 static int
-initialize_entrylk_variables (call_frame_t *frame, xlator_t *this)
+initialize_internal_lock_variables(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t         *local    = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-        afr_private_t       *priv     = NULL;
-
-        int i = 0;
-
-        priv     = this->private;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        int_lock->entrylk_lock_count = 0;
-        int_lock->lock_op_ret        = -1;
-        int_lock->lock_op_errno      = 0;
-
-        for (i = 0; i < AFR_LOCKEE_COUNT_MAX; i++) {
-                if (!int_lock->lockee[i].locked_nodes)
-                        break;
-                int_lock->lockee[i].locked_count = 0;
-                memset (int_lock->lockee[i].locked_nodes, 0,
-                        sizeof (*int_lock->lockee[i].locked_nodes) *
-                        priv->child_count);
-        }
+    afr_local_t *local = NULL;
+    afr_internal_lock_t *int_lock = NULL;
+    afr_private_t *priv = NULL;
 
-        return 0;
-}
+    int i = 0;
 
-static int
-initialize_inodelk_variables (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t         *local    = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-        afr_private_t       *priv     = NULL;
-        afr_inodelk_t       *inodelk  = NULL;
-
-        priv     = this->private;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
+    priv = this->private;
+    local = frame->local;
+    int_lock = &local->internal_lock;
 
-        inodelk->lock_count    = 0;
-        int_lock->lk_attempted_count = 0;
-        int_lock->lock_op_ret   = -1;
-        int_lock->lock_op_errno = 0;
+    int_lock->lock_count = 0;
+    int_lock->lock_op_ret = -1;
+    int_lock->lock_op_errno = 0;
+    int_lock->lk_attempted_count = 0;
 
-        memset (inodelk->locked_nodes, 0,
-                sizeof (*inodelk->locked_nodes) * priv->child_count);
-        memset (int_lock->locked_nodes, 0,
-                sizeof (*int_lock->locked_nodes) * priv->child_count);
+    for (i = 0; i < AFR_LOCKEE_COUNT_MAX; i++) {
+        if (!int_lock->lockee[i].locked_nodes)
+            break;
+        int_lock->lockee[i].locked_count = 0;
+        memset(int_lock->lockee[i].locked_nodes, 0,
+               sizeof(*int_lock->lockee[i].locked_nodes) * priv->child_count);
+    }
 
-        return 0;
-}
-
-loc_t *
-lower_path (loc_t *l1, const char *b1, loc_t *l2, const char *b2)
-{
-        int ret = 0;
-
-        ret = uuid_compare (l1->inode->gfid, l2->inode->gfid);
-
-        if (ret == 0)
-                ret = strcmp (b1, b2);
-
-        if (ret <= 0)
-                return l1;
-        else
-                return l2;
+    return 0;
 }
 
 int
-afr_lockee_locked_nodes_count (afr_internal_lock_t *int_lock)
+afr_lockee_locked_nodes_count(afr_internal_lock_t *int_lock)
 {
-        int call_count  = 0;
-        int i           = 0;
+    int call_count = 0;
+    int i = 0;
 
-        for (i = 0; i < int_lock->lockee_count; i++)
-                call_count += int_lock->lockee[i].locked_count;
+    for (i = 0; i < int_lock->lockee_count; i++)
+        call_count += int_lock->lockee[i].locked_count;
 
-        return call_count;
+    return call_count;
 }
 
 int
-afr_locked_nodes_count (unsigned char *locked_nodes, int child_count)
-
-{
-        int i = 0;
-        int call_count = 0;
-
-        for (i = 0; i < child_count; i++) {
-                if (locked_nodes[i] & LOCKED_YES)
-                        call_count++;
-        }
-
-        return call_count;
-}
-
-/* FIXME: What if UNLOCK fails */
-static int32_t
-afr_unlock_common_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-        afr_local_t         *local    = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-        int call_count = 0;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        LOCK (&frame->lock);
-        {
-                call_count = --int_lock->lk_call_count;
-        }
-        UNLOCK (&frame->lock);
-
-        if (call_count == 0) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "All internal locks unlocked");
-                int_lock->lock_cbk (frame, this);
-        }
-
-        return 0;
-}
-
-static int32_t
-afr_unlock_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-        afr_local_t         *local = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-        afr_inodelk_t       *inodelk = NULL;
-        int32_t             child_index = (long)cookie;
-
-        local = frame->local;
-        int_lock = &local->internal_lock;
-
-        AFR_TRACE_INODELK_OUT (frame, this, AFR_INODELK_TRANSACTION,
-                               AFR_UNLOCK_OP, NULL, op_ret,
-                               op_errno, child_index);
-
-        if (op_ret < 0 && op_errno != ENOTCONN && op_errno != EBADFD) {
-                gf_log (this->name, GF_LOG_INFO, "%s: unlock failed on %d "
-                        "unlock by %s", local->loc.path, child_index,
-                        lkowner_utoa (&frame->root->lk_owner));
-        }
-
-
-        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
-        inodelk->locked_nodes[child_index] &= LOCKED_NO;
-        if (local->transaction.eager_lock)
-                local->transaction.eager_lock[child_index] = 0;
-
-        afr_unlock_common_cbk (frame, cookie, this, op_ret, op_errno, xdata);
-
-        return 0;
-
-}
-
-static int
-afr_unlock_inodelk (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_inodelk_t       *inodelk  = NULL;
-        afr_local_t         *local    = NULL;
-        afr_private_t       *priv     = NULL;
-        struct gf_flock flock = {0,};
-        struct gf_flock full_flock = {0,};
-        struct gf_flock *flock_use = NULL;
-        int call_count = 0;
-        int i = 0;
-        int piggyback = 0;
-        afr_fd_ctx_t        *fd_ctx      = NULL;
-
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        priv     = this->private;
-
-        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
-
-        flock.l_start = inodelk->flock.l_start;
-        flock.l_len   = inodelk->flock.l_len;
-        flock.l_type  = F_UNLCK;
-
-        full_flock.l_type = F_UNLCK;
-        call_count = afr_locked_nodes_count (inodelk->locked_nodes,
-                                             priv->child_count);
-
-        int_lock->lk_call_count = call_count;
-
-        if (!call_count) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "No internal locks unlocked");
-                int_lock->lock_cbk (frame, this);
-                goto out;
-        }
-
-        if (local->fd)
-                fd_ctx = afr_fd_ctx_get (local->fd, this);
-
-        for (i = 0; i < priv->child_count; i++) {
-                if ((inodelk->locked_nodes[i] & LOCKED_YES) != LOCKED_YES)
-                        continue;
-
-                if (local->fd) {
-                        flock_use = &flock;
-                        if (!local->transaction.eager_lock[i]) {
-                                goto wind;
-                        }
-
-                        piggyback = 0;
-
-                        LOCK (&local->fd->lock);
-                        {
-                                if (fd_ctx->lock_piggyback[i]) {
-                                        fd_ctx->lock_piggyback[i]--;
-                                        piggyback = 1;
-                                } else {
-                                        fd_ctx->lock_acquired[i]--;
-                                }
-                        }
-                        UNLOCK (&local->fd->lock);
-
-                        if (piggyback) {
-                                afr_unlock_inodelk_cbk (frame, (void *) (long) i,
-                                                        this, 1, 0, NULL);
-                                if (!--call_count)
-                                        break;
-                                continue;
-                        }
-
-                        flock_use = &full_flock;
-                wind:
-                        AFR_TRACE_INODELK_IN (frame, this,
-                                              AFR_INODELK_TRANSACTION,
-                                              AFR_UNLOCK_OP, flock_use, F_SETLK,
-                                              i);
-
-                        STACK_WIND_COOKIE (frame, afr_unlock_inodelk_cbk,
-                                           (void *) (long)i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->finodelk,
-                                           int_lock->domain, local->fd,
-                                           F_SETLK, flock_use, NULL);
-
-                        if (!--call_count)
-                                break;
-
-                } else {
-                        AFR_TRACE_INODELK_IN (frame, this,
-                                              AFR_INODELK_TRANSACTION,
-                                              AFR_UNLOCK_OP, &flock, F_SETLK, i);
-
-                        STACK_WIND_COOKIE (frame, afr_unlock_inodelk_cbk,
-                                           (void *) (long)i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->inodelk,
-                                           int_lock->domain, &local->loc,
-                                           F_SETLK, &flock, NULL);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-out:
-        return 0;
-}
-
-static int32_t
-afr_unlock_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-        afr_local_t         *local = NULL;
-        afr_private_t       *priv  = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-        int32_t             child_index = 0;
-        int                 lockee_no   = 0;
-
-        priv = this->private;
-        lockee_no = (int)((long) cookie) / priv->child_count;
-        child_index = (int) ((long) cookie) % priv->child_count;
-
-        local = frame->local;
-        int_lock = &local->internal_lock;
-
-        AFR_TRACE_ENTRYLK_OUT (frame, this, AFR_ENTRYLK_TRANSACTION,
-                               AFR_UNLOCK_OP,
-                               int_lock->lockee[lockee_no].basename, op_ret,
-                               op_errno, (int) ((long)cookie));
+afr_locked_nodes_count(unsigned char *locked_nodes, int child_count)
 
-        if (op_ret < 0) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s: unlock failed on %d, reason: %s",
-                        local->loc.path, child_index, strerror (op_errno));
-        }
-
-        int_lock->lockee[lockee_no].locked_nodes[child_index] &= LOCKED_NO;
-        afr_unlock_common_cbk (frame, cookie, this, op_ret, op_errno, NULL);
-
-        return 0;
-}
-
-static int
-afr_unlock_entrylk (call_frame_t *frame, xlator_t *this)
 {
-        afr_internal_lock_t     *int_lock       = NULL;
-        afr_local_t             *local          = NULL;
-        afr_private_t           *priv           = NULL;
-        int                     call_count      = 0;
-        int                     index           = 0;
-        int                     lockee_no       = 0;
-        int                     copies          = 0;
-        int                     i               = -1;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        priv     = this->private;
-        copies   = priv->child_count;
-
-        call_count = afr_lockee_locked_nodes_count (int_lock);
-
-        int_lock->lk_call_count = call_count;
-
-        if (!call_count){
-                gf_log (this->name, GF_LOG_TRACE,
-                        "No internal locks unlocked");
-                int_lock->lock_cbk (frame, this);
-                goto out;
-        }
+    int i = 0;
+    int call_count = 0;
 
-        for (i = 0; i < int_lock->lockee_count * priv->child_count; i++) {
-                lockee_no = i / copies;
-                index     = i % copies;
-                if (int_lock->lockee[lockee_no].locked_nodes[index] & LOCKED_YES) {
-                        AFR_TRACE_ENTRYLK_IN (frame, this, AFR_ENTRYLK_NB_TRANSACTION,
-                                              AFR_UNLOCK_OP,
-                                              int_lock->lockee[lockee_no].basename,
-                                              i);
-
-                        STACK_WIND_COOKIE (frame, afr_unlock_entrylk_cbk,
-                                           (void *) (long) i,
-                                           priv->children[index],
-                                           priv->children[index]->fops->entrylk,
-                                           int_lock->domain,
-                                           &int_lock->lockee[lockee_no].loc,
-                                           int_lock->lockee[lockee_no].basename,
-                                           ENTRYLK_UNLOCK, ENTRYLK_WRLCK, NULL);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-out:
-        return 0;
+    for (i = 0; i < child_count; i++) {
+        if (locked_nodes[i] & LOCKED_YES)
+            call_count++;
+    }
 
+    return call_count;
 }
 
-static int32_t
-afr_lock_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-              int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-        afr_internal_lock_t     *int_lock       = NULL;
-        afr_local_t             *local          = NULL;
-        afr_private_t           *priv           = NULL;
-        int                     cky             = (long) cookie;
-        int                     child_index     = 0;
-        int                     lockee_no       = 0;
-
-        priv     = this->private;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        child_index = ((int)cky) % priv->child_count;
-        lockee_no   = ((int)cky) / priv->child_count;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        if (op_errno == ENOSYS) {
-                                /* return ENOTSUP */
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "subvolume does not support locking. "
-                                        "please load features/locks xlator on server");
-                                local->op_ret = op_ret;
-                                int_lock->lock_op_ret = op_ret;
-                        }
-
-                        local->op_errno              = op_errno;
-                        int_lock->lock_op_errno      = op_errno;
-                }
-
-		int_lock->lk_attempted_count++;
-        }
-        UNLOCK (&frame->lock);
-
-        if ((op_ret == -1) &&
-            (op_errno == ENOSYS)) {
-                afr_unlock (frame, this);
-        } else {
-                if (op_ret == 0) {
-                        if (local->transaction.type == AFR_ENTRY_TRANSACTION ||
-                            local->transaction.type == AFR_ENTRY_RENAME_TRANSACTION) {
-                                int_lock->lockee[lockee_no].locked_nodes[child_index] |= LOCKED_YES;
-                                int_lock->lockee[lockee_no].locked_count++;
-                                int_lock->entrylk_lock_count++;
-                        } else {
-                                int_lock->locked_nodes[child_index] |= LOCKED_YES;
-                                int_lock->lock_count++;
-                        }
-                }
-                afr_lock_blocking (frame, this, cky + 1);
-        }
-
-        return 0;
-}
-
-static int32_t
-afr_blocking_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                          int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-        AFR_TRACE_INODELK_OUT (frame, this, AFR_INODELK_TRANSACTION,
-                               AFR_LOCK_OP, NULL, op_ret,
-                               op_errno, (long) cookie);
-
-        afr_lock_cbk (frame, cookie, this, op_ret, op_errno, xdata);
-        return 0;
-
-}
-
-static int32_t
-afr_blocking_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                          int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-        AFR_TRACE_ENTRYLK_OUT (frame, this, AFR_ENTRYLK_TRANSACTION,
-                               AFR_LOCK_OP, NULL, op_ret,
-                               op_errno, (long)cookie);
-
-        afr_lock_cbk (frame, cookie, this, op_ret, op_errno, xdata);
-        return 0;
-}
-
-static int
-afr_copy_locked_nodes (call_frame_t *frame, xlator_t *this)
+static void
+afr_log_locks_failure(call_frame_t *frame, char *where, char *what,
+                      int op_errno)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_inodelk_t       *inodelk  = NULL;
-        afr_local_t         *local    = NULL;
-        afr_private_t       *priv     = NULL;
-
-        priv     = this->private;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+    xlator_t *this = frame->this;
+    gf_lkowner_t *lk_owner = &frame->root->lk_owner;
+    afr_local_t *local = frame->local;
+    const char *fop = NULL;
+    char *gfid = NULL;
+    const char *name = NULL;
 
-        switch (local->transaction.type) {
-        case AFR_DATA_TRANSACTION:
-        case AFR_METADATA_TRANSACTION:
-                inodelk = afr_get_inodelk (int_lock, int_lock->domain);
-                memcpy (inodelk->locked_nodes, int_lock->locked_nodes,
-                        sizeof (*inodelk->locked_nodes) * priv->child_count);
-                inodelk->lock_count = int_lock->lock_count;
-                break;
+    fop = gf_fop_list[local->op];
 
+    switch (local->transaction.type) {
         case AFR_ENTRY_RENAME_TRANSACTION:
         case AFR_ENTRY_TRANSACTION:
-                /*entrylk_count is being used in both non-blocking and blocking
-                 * modes */
-                break;
-        }
-
-        return 0;
-
+            switch (local->op) {
+                case GF_FOP_LINK:
+                    gfid = uuid_utoa(local->newloc.pargfid);
+                    name = local->newloc.name;
+                    break;
+                default:
+                    gfid = uuid_utoa(local->loc.pargfid);
+                    name = local->loc.name;
+                    break;
+            }
+            gf_msg(this->name, GF_LOG_WARNING, op_errno,
+                   AFR_MSG_INTERNAL_LKS_FAILED,
+                   "Unable to do entry %s with lk-owner:%s on %s "
+                   "while attempting %s on {pgfid:%s, name:%s}.",
+                   what, lkowner_utoa(lk_owner), where, fop, gfid, name);
+            break;
+        case AFR_DATA_TRANSACTION:
+        case AFR_METADATA_TRANSACTION:
+            gfid = uuid_utoa(local->inode->gfid);
+            gf_msg(this->name, GF_LOG_WARNING, op_errno,
+                   AFR_MSG_INTERNAL_LKS_FAILED,
+                   "Unable to do inode %s with lk-owner:%s on %s "
+                   "while attempting %s on gfid:%s.",
+                   what, lkowner_utoa(lk_owner), where, fop, gfid);
+            break;
+    }
 }
 
-static inline gf_boolean_t
-afr_is_entrylk (afr_internal_lock_t *int_lock,
-                afr_transaction_type trans_type)
+static int32_t
+afr_unlock_common_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        gf_boolean_t is_entrylk = _gf_false;
-
-        if ((int_lock->transaction_lk_type == AFR_SELFHEAL_LK) &&
-            int_lock->selfheal_lk_type == AFR_ENTRY_SELF_HEAL_LK) {
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    afr_internal_lock_t *int_lock = NULL;
+    int lockee_num = 0;
+    int call_count = 0;
+    int child_index = 0;
+    int ret = 0;
 
-                is_entrylk = _gf_true;
+    local = frame->local;
+    int_lock = &local->internal_lock;
+    priv = this->private;
+    lockee_num = (int)((long)cookie) / priv->child_count;
+    child_index = (int)((long)cookie) % priv->child_count;
 
-        } else if ((int_lock->transaction_lk_type == AFR_TRANSACTION_LK) &&
-                 (trans_type == AFR_ENTRY_TRANSACTION ||
-                  trans_type == AFR_ENTRY_RENAME_TRANSACTION)) {
+    if (op_ret < 0 && op_errno != ENOTCONN && op_errno != EBADFD) {
+        afr_log_locks_failure(frame, priv->children[child_index]->name,
+                              "unlock", op_errno);
+    }
 
-                is_entrylk = _gf_true;
+    int_lock->lockee[lockee_num].locked_nodes[child_index] &= LOCKED_NO;
+    if (local->transaction.type == AFR_DATA_TRANSACTION && op_ret != 1)
+        ret = afr_write_subvol_reset(frame, this);
 
-        } else {
-                is_entrylk = _gf_false;
-        }
-
-        return is_entrylk;
-}
+    LOCK(&frame->lock);
+    {
+        call_count = --int_lock->lk_call_count;
+    }
+    UNLOCK(&frame->lock);
 
-static gf_boolean_t
-_is_lock_wind_needed (afr_local_t *local, int child_index)
-{
-        if (!local->child_up[child_index])
-                return _gf_false;
+    if (call_count == 0) {
+        int_lock->lock_cbk(frame, this);
+    }
 
-        return _gf_true;
+    return ret;
 }
 
-int
-afr_lock_blocking (call_frame_t *frame, xlator_t *this, int cookie)
-{
-        afr_internal_lock_t *int_lock    = NULL;
-        afr_inodelk_t       *inodelk     = NULL;
-        afr_local_t         *local       = NULL;
-        afr_private_t       *priv        = NULL;
-        struct gf_flock flock = {0,};
-        uint64_t ctx = 0;
-        int ret = 0;
-        int child_index = 0;
-        int lockee_no   = 0;
-        gf_boolean_t is_entrylk = _gf_false;
-
-        local         = frame->local;
-        int_lock      = &local->internal_lock;
-        priv          = this->private;
-        child_index   = cookie % priv->child_count;
-        lockee_no     = cookie / priv->child_count;
-        is_entrylk    = afr_is_entrylk (int_lock, local->transaction.type);
-
-
-        if (!is_entrylk) {
-                inodelk = afr_get_inodelk (int_lock, int_lock->domain);
-                flock.l_start = inodelk->flock.l_start;
-                flock.l_len   = inodelk->flock.l_len;
-                flock.l_type  = inodelk->flock.l_type;
-        }
-
-        if (local->fd) {
-                ret = fd_ctx_get (local->fd, this, &ctx);
-
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "unable to get fd ctx for fd=%p",
-                                local->fd);
-
-                        local->op_ret           = -1;
-                        int_lock->lock_op_ret   = -1;
-
-                        afr_copy_locked_nodes (frame, this);
-
-                        afr_unlock (frame, this);
-
-                        return 0;
-                }
-        }
-
-        if (int_lock->lk_expected_count == int_lock->lk_attempted_count) {
-                if ((is_entrylk && int_lock->entrylk_lock_count == 0) ||
-                    (!is_entrylk && int_lock->lock_count == 0)) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "unable to lock on even one child");
-
-                        local->op_ret           = -1;
-                        int_lock->lock_op_ret   = -1;
-
-                        afr_copy_locked_nodes (frame, this);
-
-                        afr_unlock(frame, this);
-
-                        return 0;
-                }
-        }
-
-        if (int_lock->lk_expected_count == int_lock->lk_attempted_count) {
-                /* we're done locking */
-
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "we're done locking");
-
-                afr_copy_locked_nodes (frame, this);
-
-                int_lock->lock_op_ret = 0;
-                int_lock->lock_cbk (frame, this);
-                return 0;
-        }
-
-        if (!_is_lock_wind_needed (local, child_index)) {
-                afr_lock_blocking (frame, this, cookie + 1);
-                return 0;
-        }
-
-        switch (local->transaction.type) {
-        case AFR_DATA_TRANSACTION:
-        case AFR_METADATA_TRANSACTION:
-
-                if (local->fd) {
-                        AFR_TRACE_INODELK_IN (frame, this,
-                                              AFR_INODELK_TRANSACTION,
-                                              AFR_LOCK_OP, &flock, F_SETLKW,
-                                              child_index);
-
-                        STACK_WIND_COOKIE (frame, afr_blocking_inodelk_cbk,
-                                           (void *) (long) child_index,
-                                           priv->children[child_index],
-                                           priv->children[child_index]->fops->finodelk,
-                                           int_lock->domain, local->fd,
-                                           F_SETLKW, &flock, NULL);
-
-                } else {
-                        AFR_TRACE_INODELK_IN (frame, this,
-                                              AFR_INODELK_TRANSACTION,
-                                              AFR_LOCK_OP, &flock, F_SETLKW,
-                                              child_index);
-
-                        STACK_WIND_COOKIE (frame, afr_blocking_inodelk_cbk,
-                                           (void *) (long) child_index,
-                                           priv->children[child_index],
-                                           priv->children[child_index]->fops->inodelk,
-                                           int_lock->domain, &local->loc,
-                                           F_SETLKW, &flock, NULL);
-                }
-
-                break;
-
-        case AFR_ENTRY_RENAME_TRANSACTION:
+void
+afr_internal_lock_wind(call_frame_t *frame,
+                       int32_t (*cbk)(call_frame_t *, void *, xlator_t *,
+                                      int32_t, int32_t, dict_t *),
+                       void *cookie, int child, int lockee_num,
+                       gf_boolean_t blocking, gf_boolean_t unlock)
+{
+    afr_local_t *local = frame->local;
+    xlator_t *this = frame->this;
+    afr_private_t *priv = this->private;
+    afr_internal_lock_t *int_lock = &local->internal_lock;
+    entrylk_cmd cmd = ENTRYLK_LOCK_NB;
+    int32_t cmd1 = F_SETLK;
+    struct gf_flock flock = {
+        0,
+    };
+
+    switch (local->transaction.type) {
         case AFR_ENTRY_TRANSACTION:
-                /*Accounting for child_index increments on 'down'
-                 *and 'fd-less' children */
-
-                if (local->fd) {
-                        AFR_TRACE_ENTRYLK_IN (frame, this, AFR_ENTRYLK_TRANSACTION,
-                                              AFR_LOCK_OP,
-                                              int_lock->lockee[lockee_no].basename,
-                                              cookie);
-
-                        STACK_WIND_COOKIE (frame, afr_blocking_entrylk_cbk,
-                                           (void *) (long) cookie,
-                                           priv->children[child_index],
-                                           priv->children[child_index]->fops->fentrylk,
-                                           int_lock->domain, local->fd,
-                                           int_lock->lockee[lockee_no].basename,
-                                           ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL);
-                } else {
-                        AFR_TRACE_ENTRYLK_IN (frame, this,
-                                              AFR_ENTRYLK_TRANSACTION,
-                                              AFR_LOCK_OP, local->transaction.basename,
-                                              child_index);
-
-                        STACK_WIND_COOKIE (frame, afr_blocking_entrylk_cbk,
-                                           (void *) (long) cookie,
-                                           priv->children[child_index],
-                                           priv->children[child_index]->fops->entrylk,
-                                           int_lock->domain,
-                                           &int_lock->lockee[lockee_no].loc,
-                                           int_lock->lockee[lockee_no].basename,
-                                           ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL);
-                }
-
-                break;
-        }
-
-        return 0;
-}
-
-int32_t
-afr_blocking_lock (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_private_t       *priv     = NULL;
-        int                  up_count = 0;
-
-        priv     = this->private;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+        case AFR_ENTRY_RENAME_TRANSACTION:
+            if (unlock) {
+                cmd = ENTRYLK_UNLOCK;
+            } else if (blocking) { /*Doesn't make sense to have blocking
+                                      unlock*/
+                cmd = ENTRYLK_LOCK;
+            }
+
+            if (local->fd) {
+                STACK_WIND_COOKIE(frame, cbk, cookie, priv->children[child],
+                                  priv->children[child]->fops->fentrylk,
+                                  int_lock->domain,
+                                  int_lock->lockee[lockee_num].fd,
+                                  int_lock->lockee[lockee_num].basename, cmd,
+                                  ENTRYLK_WRLCK, NULL);
+            } else {
+                STACK_WIND_COOKIE(frame, cbk, cookie, priv->children[child],
+                                  priv->children[child]->fops->entrylk,
+                                  int_lock->domain,
+                                  &int_lock->lockee[lockee_num].loc,
+                                  int_lock->lockee[lockee_num].basename, cmd,
+                                  ENTRYLK_WRLCK, NULL);
+            }
+            break;
 
-        switch (local->transaction.type) {
         case AFR_DATA_TRANSACTION:
         case AFR_METADATA_TRANSACTION:
-                initialize_inodelk_variables (frame, this);
-                break;
-
-        case AFR_ENTRY_RENAME_TRANSACTION:
-        case AFR_ENTRY_TRANSACTION:
-                up_count = afr_up_children_count (local->child_up,
-                                                  priv->child_count);
-                int_lock->lk_call_count = int_lock->lk_expected_count
-                                        = (int_lock->lockee_count *
-                                           up_count);
-                initialize_entrylk_variables (frame, this);
-                break;
-        }
-
-        afr_lock_blocking (frame, this, 0);
-
-        return 0;
+            flock = int_lock->lockee[lockee_num].flock;
+            if (unlock) {
+                flock.l_type = F_UNLCK;
+            } else if (blocking) { /*Doesn't make sense to have blocking
+                                      unlock*/
+                cmd1 = F_SETLKW;
+            }
+
+            if (local->fd) {
+                STACK_WIND_COOKIE(
+                    frame, cbk, cookie, priv->children[child],
+                    priv->children[child]->fops->finodelk, int_lock->domain,
+                    int_lock->lockee[lockee_num].fd, cmd1, &flock, NULL);
+            } else {
+                STACK_WIND_COOKIE(
+                    frame, cbk, cookie, priv->children[child],
+                    priv->children[child]->fops->inodelk, int_lock->domain,
+                    &int_lock->lockee[lockee_num].loc, cmd1, &flock, NULL);
+            }
+            break;
+    }
 }
 
-static int32_t
-afr_nonblocking_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                             int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        int call_count          = 0;
-        int child_index         = (long) cookie;
-        int copies              = 0;
-        int index               = 0;
-        int lockee_no           = 0;
-        afr_private_t       *priv = NULL;
-
-        priv = this->private;
-
-        copies = priv->child_count;
-        index = child_index % copies;
-        lockee_no = child_index / copies;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        AFR_TRACE_ENTRYLK_OUT (frame, this, AFR_ENTRYLK_TRANSACTION,
-                               AFR_LOCK_OP,
-                               int_lock->lockee[lockee_no].basename, op_ret,
-                               op_errno, (long) cookie);
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret < 0 ) {
-			if (op_errno == ENOSYS) {
-                        /* return ENOTSUP */
-				gf_log (this->name, GF_LOG_ERROR,
-					"subvolume does not support locking. "
-					"please load features/locks xlator on server");
-				local->op_ret         = op_ret;
-				int_lock->lock_op_ret = op_ret;
-
-				int_lock->lock_op_errno      = op_errno;
-				local->op_errno              = op_errno;
-			}
-		} else if (op_ret == 0) {
-			int_lock->lockee[lockee_no].locked_nodes[index] |= \
-				LOCKED_YES;
-			int_lock->lockee[lockee_no].locked_count++;
-			int_lock->entrylk_lock_count++;
-		}
-
-                call_count = --int_lock->lk_call_count;
-        }
-        UNLOCK (&frame->lock);
-
-        if (call_count == 0) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "Last locking reply received");
-                /* all locks successful. Proceed to call FOP */
-                if (int_lock->entrylk_lock_count ==
-                                int_lock->lk_expected_count) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "All servers locked. Calling the cbk");
-                        int_lock->lock_op_ret = 0;
-                        int_lock->lock_cbk (frame, this);
-                }
-                /* Not all locks were successful. Unlock and try locking
-                   again, this time with serially blocking locks */
-                else {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "%d servers locked. Trying again with blocking calls",
-                                int_lock->lock_count);
-
-                        afr_unlock(frame, this);
-                }
+static int
+afr_unlock_now(call_frame_t *frame, xlator_t *this)
+{
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int call_count = 0;
+    int child_index = 0;
+    int lockee_num = 0;
+    int i = -1;
+
+    local = frame->local;
+    int_lock = &local->internal_lock;
+    priv = this->private;
+
+    call_count = afr_lockee_locked_nodes_count(int_lock);
+
+    int_lock->lk_call_count = call_count;
+
+    if (!call_count) {
+        gf_msg_trace(this->name, 0, "No internal locks unlocked");
+        int_lock->lock_cbk(frame, this);
+        goto out;
+    }
+
+    for (i = 0; i < int_lock->lockee_count * priv->child_count; i++) {
+        lockee_num = i / priv->child_count;
+        child_index = i % priv->child_count;
+        if (int_lock->lockee[lockee_num].locked_nodes[child_index] &
+            LOCKED_YES) {
+            afr_internal_lock_wind(frame, afr_unlock_common_cbk,
+                                   (void *)(long)i, child_index, lockee_num,
+                                   _gf_false, _gf_true);
+            if (!--call_count)
+                break;
         }
+    }
 
-        return 0;
-}
-
-int
-afr_nonblocking_entrylk (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock   = NULL;
-        afr_local_t         *local      = NULL;
-        afr_private_t       *priv       = NULL;
-        afr_fd_ctx_t        *fd_ctx     = NULL;
-        int                 copies      = 0;
-        int                 index       = 0;
-        int                 lockee_no   = 0;
-        int32_t             call_count  = 0;
-        int i = 0;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        priv     = this->private;
-
-        copies = priv->child_count;
-        initialize_entrylk_variables (frame, this);
-
-        if (local->fd) {
-                fd_ctx = afr_fd_ctx_get (local->fd, this);
-                if (!fd_ctx) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "unable to get fd ctx for fd=%p",
-                                local->fd);
-
-                        local->op_ret           = -1;
-                        int_lock->lock_op_ret   = -1;
-                        local->op_errno         = EINVAL;
-                        int_lock->lock_op_errno = EINVAL;
-
-			afr_unlock (frame, this);
-                        return -1;
-                }
-
-                call_count = int_lock->lockee_count * internal_lock_count (frame, this);
-                int_lock->lk_call_count = call_count;
-                int_lock->lk_expected_count = call_count;
-
-                if (!call_count) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "fd not open on any subvolumes. aborting.");
-                        afr_unlock (frame, this);
-                        goto out;
-                }
-
-                /* Send non-blocking entrylk calls only on up children
-                   and where the fd has been opened */
-                for (i = 0; i < int_lock->lockee_count*priv->child_count; i++) {
-                        index = i%copies;
-                        lockee_no = i/copies;
-                        if (local->child_up[index]) {
-                                AFR_TRACE_ENTRYLK_IN (frame, this, AFR_ENTRYLK_NB_TRANSACTION,
-                                                      AFR_LOCK_OP,
-                                                      int_lock->lockee[lockee_no].basename,
-                                                      i);
-
-                                STACK_WIND_COOKIE (frame, afr_nonblocking_entrylk_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[index],
-                                                   priv->children[index]->fops->fentrylk,
-                                                   this->name, local->fd,
-                                                   int_lock->lockee[lockee_no].basename,
-                                                   ENTRYLK_LOCK_NB, ENTRYLK_WRLCK,
-                                                   NULL);
-                                if (!--call_count)
-                                        break;
-                        }
-                }
-        } else {
-                call_count = int_lock->lockee_count * internal_lock_count (frame, this);
-                int_lock->lk_call_count = call_count;
-                int_lock->lk_expected_count = call_count;
-
-                for (i = 0; i < int_lock->lockee_count*priv->child_count; i++) {
-                        index = i%copies;
-                        lockee_no = i/copies;
-                        if (local->child_up[index]) {
-                                AFR_TRACE_ENTRYLK_IN (frame, this, AFR_ENTRYLK_NB_TRANSACTION,
-                                                      AFR_LOCK_OP,
-                                                      int_lock->lockee[lockee_no].basename,
-                                                      i);
-
-                                STACK_WIND_COOKIE (frame, afr_nonblocking_entrylk_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[index],
-                                                   priv->children[index]->fops->entrylk,
-                                                   this->name, &int_lock->lockee[lockee_no].loc,
-                                                   int_lock->lockee[lockee_no].basename,
-                                                   ENTRYLK_LOCK_NB, ENTRYLK_WRLCK,
-                                                   NULL);
-
-                                if (!--call_count)
-                                        break;
-                        }
-                }
-        }
 out:
-        return 0;
+    return 0;
 }
 
-int32_t
-afr_nonblocking_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                             int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_inodelk_t       *inodelk  = NULL;
-        afr_local_t         *local    = NULL;
-        int call_count  = 0;
-        int child_index = (long) cookie;
-        afr_fd_ctx_t        *fd_ctx = NULL;
-
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
-
-        AFR_TRACE_INODELK_OUT (frame, this, AFR_INODELK_NB_TRANSACTION,
-                               AFR_LOCK_OP, NULL, op_ret,
-                               op_errno, (long) cookie);
-
-	if (local->fd)
-		fd_ctx = afr_fd_ctx_get (local->fd, this);
-
-        LOCK (&frame->lock);
-        {
-		if (op_ret < 0) {
-			if (op_errno == ENOSYS) {
-				/* return ENOTSUP */
-				gf_log (this->name, GF_LOG_ERROR,
-					"subvolume does not support locking. "
-					"please load features/locks xlator on "
-					"server");
-				local->op_ret                = op_ret;
-				int_lock->lock_op_ret        = op_ret;
-				int_lock->lock_op_errno      = op_errno;
-				local->op_errno              = op_errno;
-			}
-			if (local->transaction.eager_lock)
-				local->transaction.eager_lock[child_index] = 0;
-		} else {
-			inodelk->locked_nodes[child_index] |= LOCKED_YES;
-			inodelk->lock_count++;
-
-			if (local->transaction.eager_lock &&
-			    local->transaction.eager_lock[child_index] &&
-			    local->fd) {
-				/* piggybacked */
-				if (op_ret == 1) {
-					/* piggybacked */
-				} else if (op_ret == 0) {
-					/* lock acquired from server */
-                                        fd_ctx->lock_acquired[child_index]++;
-				}
-			}
-		}
-
-                call_count = --int_lock->lk_call_count;
-        }
-        UNLOCK (&frame->lock);
-
-        if (call_count == 0) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "Last inode locking reply received");
-                /* all locks successful. Proceed to call FOP */
-                if (inodelk->lock_count == int_lock->lk_expected_count) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "All servers locked. Calling the cbk");
-                        int_lock->lock_op_ret = 0;
-                        int_lock->lock_cbk (frame, this);
-                }
-                /* Not all locks were successful. Unlock and try locking
-                   again, this time with serially blocking locks */
-                else {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "%d servers locked. Trying again with blocking calls",
-                                int_lock->lock_count);
-
-                        afr_unlock(frame, this);
+static int32_t
+afr_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, dict_t *xdata)
+{
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int cky = (long)cookie;
+    int child_index = 0;
+    int lockee_num = 0;
+
+    priv = this->private;
+    local = frame->local;
+    int_lock = &local->internal_lock;
+
+    child_index = ((int)cky) % priv->child_count;
+    lockee_num = ((int)cky) / priv->child_count;
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            if (op_errno == ENOSYS) {
+                /* return ENOTSUP */
+                gf_msg(this->name, GF_LOG_ERROR, ENOSYS,
+                       AFR_MSG_LOCK_XLATOR_NOT_LOADED,
+                       "subvolume does not support locking. "
+                       "please load features/locks xlator on server");
+                local->op_ret = op_ret;
+                int_lock->lock_op_ret = op_ret;
+            }
+
+            local->op_errno = op_errno;
+            int_lock->lock_op_errno = op_errno;
+        }
+
+        int_lock->lk_attempted_count++;
+    }
+    UNLOCK(&frame->lock);
+
+    if ((op_ret == -1) && (op_errno == ENOSYS)) {
+        afr_unlock_now(frame, this);
+    } else {
+        if (op_ret == 0) {
+            int_lock->lockee[lockee_num]
+                .locked_nodes[child_index] |= LOCKED_YES;
+            int_lock->lockee[lockee_num].locked_count++;
+            int_lock->lock_count++;
+            if (local->transaction.type == AFR_DATA_TRANSACTION) {
+                LOCK(&local->inode->lock);
+                {
+                    local->inode_ctx->lock_count++;
                 }
+                UNLOCK(&local->inode->lock);
+            }
         }
+        afr_lock_blocking(frame, this, cky + 1);
+    }
 
-        return 0;
+    return 0;
 }
 
-int
-afr_nonblocking_inodelk (call_frame_t *frame, xlator_t *this)
+static gf_boolean_t
+_is_lock_wind_needed(afr_local_t *local, int child_index)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_inodelk_t       *inodelk  = NULL;
-        afr_local_t         *local    = NULL;
-        afr_private_t       *priv     = NULL;
-        afr_fd_ctx_t        *fd_ctx   = NULL;
-        int32_t             call_count = 0;
-        int                 i          = 0;
-        int                 ret        = 0;
-        struct              gf_flock flock = {0,};
-        struct              gf_flock full_flock = {0,};
-        struct              gf_flock *flock_use = NULL;
-        int                 piggyback = 0;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        priv     = this->private;
-
-        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
-
-        flock.l_start = inodelk->flock.l_start;
-        flock.l_len   = inodelk->flock.l_len;
-        flock.l_type  = inodelk->flock.l_type;
-
-        full_flock.l_type = inodelk->flock.l_type;
-
-        initialize_inodelk_variables (frame, this);
-
-        if (local->fd) {
-                fd_ctx = afr_fd_ctx_get (local->fd, this);
-                if (!fd_ctx) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "unable to get fd ctx for fd=%p",
-                                local->fd);
-
-                        local->op_ret           = -1;
-                        int_lock->lock_op_ret   = -1;
-                        local->op_errno         = EINVAL;
-                        int_lock->lock_op_errno = EINVAL;
-
-			afr_unlock (frame, this);
-                        ret = -1;
-                        goto out;
-                }
-
-                call_count = internal_lock_count (frame, this);
-                int_lock->lk_call_count = call_count;
-                int_lock->lk_expected_count = call_count;
+    if (!local->child_up[child_index])
+        return _gf_false;
 
-                if (!call_count) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "fd not open on any subvolumes. aborting.");
-                        afr_unlock (frame, this);
-                        goto out;
-                }
-
-                /* Send non-blocking inodelk calls only on up children
-                   and where the fd has been opened */
-                for (i = 0; i < priv->child_count; i++) {
-                        if (!local->child_up[i])
-                                continue;
-
-                        flock_use = &flock;
-                        if (!local->transaction.eager_lock_on) {
-                                goto wind;
-                        }
-
-                        piggyback = 0;
-                        local->transaction.eager_lock[i] = 1;
-
-			afr_set_delayed_post_op (frame, this);
-
-                        LOCK (&local->fd->lock);
-                        {
-                                if (fd_ctx->lock_acquired[i]) {
-                                        fd_ctx->lock_piggyback[i]++;
-                                        piggyback = 1;
-                                }
-                        }
-                        UNLOCK (&local->fd->lock);
-
-                        if (piggyback) {
-                                /* (op_ret == 1) => indicate piggybacked lock */
-                                afr_nonblocking_inodelk_cbk (frame, (void *) (long) i,
-                                                             this, 1, 0, NULL);
-                                if (!--call_count)
-                                        break;
-                                continue;
-                        }
-                        flock_use = &full_flock;
-                wind:
-                        AFR_TRACE_INODELK_IN (frame, this,
-                                              AFR_INODELK_NB_TRANSACTION,
-                                              AFR_LOCK_OP, flock_use, F_SETLK, i);
-
-                        STACK_WIND_COOKIE (frame, afr_nonblocking_inodelk_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->finodelk,
-                                           int_lock->domain, local->fd,
-                                           F_SETLK, flock_use, NULL);
-
-                        if (!--call_count)
-                                break;
-                }
-        } else {
-                call_count = internal_lock_count (frame, this);
-                int_lock->lk_call_count = call_count;
-                int_lock->lk_expected_count = call_count;
-
-                for (i = 0; i < priv->child_count; i++) {
-                        if (!local->child_up[i])
-                                continue;
-                        AFR_TRACE_INODELK_IN (frame, this,
-                                              AFR_INODELK_NB_TRANSACTION,
-                                              AFR_LOCK_OP, &flock, F_SETLK, i);
-
-                        STACK_WIND_COOKIE (frame, afr_nonblocking_inodelk_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->inodelk,
-                                           int_lock->domain, &local->loc,
-                                           F_SETLK, &flock, NULL);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-out:
-        return ret;
+    return _gf_true;
 }
 
-int32_t
-afr_unlock (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local = NULL;
-
-        local = frame->local;
-
-        if (transaction_lk_op (local)) {
-                if (is_afr_lock_transaction (local))
-                        afr_unlock_inodelk (frame, this);
-                else
-                        afr_unlock_entrylk (frame, this);
-
-        } else {
-                if (is_afr_lock_selfheal (local))
-                        afr_unlock_inodelk (frame, this);
-                else
-                        afr_unlock_entrylk (frame, this);
+static gf_boolean_t
+is_blocking_locks_count_sufficient(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    afr_internal_lock_t *int_lock = NULL;
+    int child = 0;
+    int nlockee = 0;
+    int lockee_count = 0;
+    gf_boolean_t ret = _gf_true;
+
+    local = frame->local;
+    priv = this->private;
+    int_lock = &local->internal_lock;
+    lockee_count = int_lock->lockee_count;
+
+    if (int_lock->lock_count == 0) {
+        afr_log_locks_failure(frame, "any subvolume", "lock",
+                              int_lock->lock_op_errno);
+        return _gf_false;
+    }
+    /* For FOPS that take multiple sets of locks (mkdir, rename),
+     * there must be at least one brick on which the locks from
+     * all lock sets were successful. */
+    for (child = 0; child < priv->child_count; child++) {
+        ret = _gf_true;
+        for (nlockee = 0; nlockee < lockee_count; nlockee++) {
+            if (!(int_lock->lockee[nlockee].locked_nodes[child] & LOCKED_YES))
+                ret = _gf_false;
         }
-
-        return 0;
-}
-
-int
-afr_mark_locked_nodes (xlator_t *this, fd_t *fd,
-                       unsigned char *locked_nodes)
-{
-        afr_private_t *priv  = NULL;
-        afr_fd_ctx_t  *fdctx = NULL;
-        uint64_t       tmp   = 0;
-        int            ret   = 0;
-
-        priv = this->private;
-
-        ret = afr_fd_ctx_set (this, fd);
         if (ret)
-                goto out;
-
-        ret = fd_ctx_get (fd, this, &tmp);
-        if (ret) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "failed to get the fd ctx");
-                goto out;
-        }
-        fdctx = (afr_fd_ctx_t *) (long) tmp;
-
-        GF_ASSERT (fdctx->locked_on);
-
-        memcpy (fdctx->locked_on, locked_nodes,
-                priv->child_count);
-
-out:
-        return ret;
-}
-
-static int
-__is_fd_saved (xlator_t *this, fd_t *fd)
-{
-        afr_locked_fd_t *locked_fd = NULL;
-        afr_private_t   *priv      = NULL;
-        int              found     = 0;
-
-        priv = this->private;
-
-        list_for_each_entry (locked_fd, &priv->saved_fds, list) {
-                if (locked_fd->fd == fd) {
-                        found = 1;
-                        break;
-                }
-        }
-
-        return found;
-}
-
-static int
-__afr_save_locked_fd (xlator_t *this, fd_t *fd)
-{
-        afr_private_t   *priv      = NULL;
-        afr_locked_fd_t *locked_fd = NULL;
-        int              ret       = 0;
-
-        priv = this->private;
+            return ret;
+    }
+    if (!ret)
+        afr_log_locks_failure(frame, "all", "lock", int_lock->lock_op_errno);
 
-        locked_fd = GF_CALLOC (1, sizeof (*locked_fd),
-                               gf_afr_mt_locked_fd);
-        if (!locked_fd) {
-                ret = -1;
-                goto out;
-        }
-
-        locked_fd->fd = fd;
-        INIT_LIST_HEAD (&locked_fd->list);
-
-        list_add_tail (&locked_fd->list, &priv->saved_fds);
-
-out:
-        return ret;
+    return ret;
 }
 
 int
-afr_save_locked_fd (xlator_t *this, fd_t *fd)
-{
-        afr_private_t   *priv      = NULL;
-        int              ret       = 0;
-
-        priv = this->private;
-
-        pthread_mutex_lock (&priv->mutex);
-        {
-                if (__is_fd_saved (this, fd)) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "fd=%p already saved", fd);
-                        goto unlock;
-                }
-
-                ret = __afr_save_locked_fd (this, fd);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "fd=%p could not be saved", fd);
-                        goto unlock;
-                }
-        }
-unlock:
-        pthread_mutex_unlock (&priv->mutex);
-
-        return ret;
-}
-
-static int
-afr_lock_recovery_cleanup (call_frame_t *frame, xlator_t *this)
+afr_lock_blocking(call_frame_t *frame, xlator_t *this, int cookie)
 {
-        afr_local_t     *local     = NULL;
-        afr_locked_fd_t *locked_fd = NULL;
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    uint64_t ctx = 0;
+    int ret = 0;
+    int child_index = 0;
+    int lockee_num = 0;
 
-        local = frame->local;
+    local = frame->local;
+    int_lock = &local->internal_lock;
+    priv = this->private;
+    child_index = cookie % priv->child_count;
+    lockee_num = cookie / priv->child_count;
 
-        locked_fd = local->locked_fd;
+    if (local->fd) {
+        ret = fd_ctx_get(local->fd, this, &ctx);
 
-        STACK_DESTROY (frame->root);
-        afr_local_cleanup (local, this);
-
-        afr_save_locked_fd (this, locked_fd->fd);
-
-        return 0;
-
-}
-
-static int
-afr_get_source_lock_recovery (xlator_t *this, fd_t *fd)
-{
-        afr_fd_ctx_t  *fdctx        = NULL;
-        afr_private_t *priv         = NULL;
-        uint64_t      tmp           = 0;
-        int           i             = 0;
-        int           source_child  = -1;
-        int           ret           = 0;
-
-        priv = this->private;
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_FD_CTX_GET_FAILED,
+                   "unable to get fd ctx for fd=%p", local->fd);
 
-        ret = fd_ctx_get (fd, this, &tmp);
-        if (ret)
-                goto out;
+            local->op_ret = -1;
+            int_lock->lock_op_ret = -1;
 
-        fdctx = (afr_fd_ctx_t *) (long) tmp;
+            afr_unlock_now(frame, this);
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (fdctx->locked_on[i]) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Found lock recovery source=%d", i);
-                        source_child = i;
-                        break;
-                }
+            return 0;
         }
+    }
 
-out:
-        return source_child;
+    if (int_lock->lk_expected_count == int_lock->lk_attempted_count) {
+        if (!is_blocking_locks_count_sufficient(frame, this)) {
+            local->op_ret = -1;
+            int_lock->lock_op_ret = -1;
 
-}
+            afr_unlock_now(frame, this);
 
-int32_t
-afr_get_locks_fd_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno, struct gf_flock *lock,
-                      dict_t *xdata);
-int32_t
-afr_recover_lock_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno, struct gf_flock *lock,
-                      dict_t *xdata)
-{
-        afr_local_t   *local = NULL;
-        afr_private_t *priv  = NULL;
-        int32_t source_child = 0;
-        struct gf_flock flock   = {0,};
-
-        local = frame->local;
-        priv  = this->private;
-
-        if (op_ret) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "lock recovery failed");
-                goto cleanup;
+            return 0;
         }
+    }
 
-        source_child = local->source_child;
-
-        memcpy (&flock, lock, sizeof (*lock));
+    if (int_lock->lk_expected_count == int_lock->lk_attempted_count) {
+        /* we're done locking */
 
-        STACK_WIND_COOKIE (frame, afr_get_locks_fd_cbk,
-                           (void *) (long) source_child,
-                           priv->children[source_child],
-                           priv->children[source_child]->fops->lk,
-                           local->fd, F_GETLK_FD, &flock, NULL);
-
-        return 0;
+        gf_msg_debug(this->name, 0, "we're done locking");
 
-cleanup:
-        afr_lock_recovery_cleanup (frame, this);
+        int_lock->lock_op_ret = 0;
+        int_lock->lock_cbk(frame, this);
         return 0;
-}
-
-int
-afr_recover_lock (call_frame_t *frame, xlator_t *this,
-                  struct gf_flock *flock)
-{
-        afr_local_t   *local             = NULL;
-        afr_private_t *priv              = NULL;
-        int32_t      lock_recovery_child = 0;
-
-        priv  = this->private;
-        local = frame->local;
-
-        lock_recovery_child = local->lock_recovery_child;
-
-        frame->root->lk_owner = flock->l_owner;
-
-        STACK_WIND_COOKIE (frame, afr_recover_lock_cbk,
-                           (void *) (long) lock_recovery_child,
-                           priv->children[lock_recovery_child],
-                           priv->children[lock_recovery_child]->fops->lk,
-                           local->fd, F_SETLK, flock, NULL);
+    }
 
+    if (!_is_lock_wind_needed(local, child_index)) {
+        afr_lock_blocking(frame, this, cookie + 1);
         return 0;
-}
-
-static int
-is_afr_lock_eol (struct gf_flock *lock)
-{
-        int ret = 0;
+    }
 
-        if ((lock->l_type == GF_LK_EOL))
-                ret = 1;
+    afr_internal_lock_wind(frame, afr_lock_cbk, (void *)(long)cookie,
+                           child_index, lockee_num, _gf_true, _gf_false);
 
-        return ret;
+    return 0;
 }
 
 int32_t
-afr_get_locks_fd_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno, struct gf_flock *lock,
-                      dict_t *xdata)
-{
-        if (op_ret) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "Failed to get locks on fd");
-                goto cleanup;
-        }
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "Got a lock on fd");
-
-        if (is_afr_lock_eol (lock)) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "Reached EOL on locks on fd");
-                goto cleanup;
-        }
-
-        afr_recover_lock (frame, this, lock);
-
-        return 0;
-
-cleanup:
-        afr_lock_recovery_cleanup (frame, this);
-
-        return 0;
-}
-
-static int
-afr_lock_recovery (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t   *local        = NULL;
-        afr_private_t *priv         = NULL;
-        fd_t          *fd           = NULL;
-        int            ret          = 0;
-        int32_t        source_child = 0;
-        struct gf_flock   flock        = {0,};
-
-        priv  = this->private;
-        local = frame->local;
-
-        fd = local->fd;
-
-        source_child = afr_get_source_lock_recovery (this, fd);
-        if (source_child < 0) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Could not recover locks due to lock "
-                        "split brain");
-                ret = -1;
-                goto out;
-        }
-
-        local->source_child = source_child;
-
-        /* the flock can be zero filled as we're querying incrementally
-           the locks held on the fd.
-        */
-        STACK_WIND_COOKIE (frame, afr_get_locks_fd_cbk,
-                           (void *) (long) source_child,
-                           priv->children[source_child],
-                           priv->children[source_child]->fops->lk,
-                           local->fd, F_GETLK_FD, &flock, NULL);
-
-out:
-        return ret;
-}
-
-
-static int
-afr_mark_fd_opened (xlator_t *this, fd_t *fd, int32_t child_index)
+afr_blocking_lock(call_frame_t *frame, xlator_t *this)
 {
-        afr_fd_ctx_t *fdctx = NULL;
-        uint64_t      tmp   = 0;
-        int           ret   = 0;
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int up_count = 0;
 
-        ret = fd_ctx_get (fd, this, &tmp);
-        if (ret)
-                goto out;
+    priv = this->private;
+    local = frame->local;
+    int_lock = &local->internal_lock;
 
-        fdctx = (afr_fd_ctx_t *) (long) tmp;
+    up_count = AFR_COUNT(local->child_up, priv->child_count);
+    int_lock->lk_call_count = int_lock->lk_expected_count =
+        (int_lock->lockee_count * up_count);
+    initialize_internal_lock_variables(frame, this);
 
-        fdctx->opened_on[child_index] = AFR_FD_OPENED;
+    afr_lock_blocking(frame, this, 0);
 
-out:
-        return ret;
+    return 0;
 }
 
-int32_t
-afr_lock_recovery_preopen_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                               int32_t op_ret, int32_t op_errno, fd_t *fd,
-                               dict_t *xdata)
+static int32_t
+afr_nb_internal_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        int32_t child_index = (long )cookie;
-        int ret = 0;
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
+    int call_count = 0;
+    int child_index = 0;
+    int lockee_num = 0;
+    afr_private_t *priv = NULL;
 
-        if (op_ret) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "Reopen during lock-recovery failed");
-                goto cleanup;
-        }
+    priv = this->private;
 
-        gf_log (this->name, GF_LOG_DEBUG,
-                "Open succeeded => proceed to recover locks");
+    child_index = ((long)cookie) % priv->child_count;
+    lockee_num = ((long)cookie) / priv->child_count;
 
-        ret = afr_lock_recovery (frame, this);
-        if (ret) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "Lock recovery failed");
-                goto cleanup;
-        }
+    local = frame->local;
+    int_lock = &local->internal_lock;
 
-        ret = afr_mark_fd_opened (this, fd, child_index);
-        if (ret) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "Marking fd open failed");
-                goto cleanup;
+    if (op_ret == 0 && local->transaction.type == AFR_DATA_TRANSACTION) {
+        LOCK(&local->inode->lock);
+        {
+            local->inode_ctx->lock_count++;
         }
+        UNLOCK(&local->inode->lock);
+    }
 
-        return 0;
-
-cleanup:
-        afr_lock_recovery_cleanup (frame, this);
-        return 0;
-}
-
-static int
-afr_lock_recovery_preopen (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t *priv  = NULL;
-        afr_local_t   *local = NULL;
-        uint64_t       tmp   = 0;
-        afr_fd_ctx_t  *fdctx = NULL;
-        loc_t          loc   = {0,};
-        int32_t        child_index = 0;
-        int            ret   = 0;
-
-        priv  = this->private;
-        local = frame->local;
-
-        GF_ASSERT (local && local->fd);
-
-        ret = fd_ctx_get (local->fd, this, &tmp);
-        if (ret)
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to get the context of fd",
-                        uuid_utoa (local->fd->inode->gfid));
-        fdctx = (afr_fd_ctx_t *) (long) tmp;
-        /* TODO: instead we should return from the function */
-        GF_ASSERT (fdctx);
-
-        child_index = local->lock_recovery_child;
-
-        inode_path (local->fd->inode, NULL, (char **)&loc.path);
-        loc.name   = strrchr (loc.path, '/');
-        loc.inode  = inode_ref (local->fd->inode);
-        loc.parent = inode_parent (local->fd->inode, 0, NULL);
-
-
-        STACK_WIND_COOKIE (frame, afr_lock_recovery_preopen_cbk,
-                           (void *)(long) child_index,
-                           priv->children[child_index],
-                           priv->children[child_index]->fops->open,
-                           &loc, fdctx->flags, local->fd, NULL);
-
-        return 0;
-}
-
-static int
-is_fd_opened (fd_t *fd, int32_t child_index)
-{
-        afr_fd_ctx_t *fdctx = NULL;
-        uint64_t      tmp = 0;
-        int           ret = 0;
-
-        ret = fd_ctx_get (fd, THIS, &tmp);
-        if (ret)
-                goto out;
-
-        fdctx = (afr_fd_ctx_t *) (long) tmp;
-
-        if (fdctx->opened_on[child_index] == AFR_FD_OPENED)
-                ret = 1;
-
-out:
-        return ret;
+    LOCK(&frame->lock);
+    {
+        if (op_ret < 0) {
+            if (op_errno == ENOSYS) {
+                /* return ENOTSUP */
+                gf_msg(this->name, GF_LOG_ERROR, ENOSYS,
+                       AFR_MSG_LOCK_XLATOR_NOT_LOADED,
+                       "subvolume does not support "
+                       "locking. please load features/locks"
+                       " xlator on server");
+                local->op_ret = op_ret;
+                int_lock->lock_op_ret = op_ret;
+
+                int_lock->lock_op_errno = op_errno;
+                local->op_errno = op_errno;
+            }
+        } else if (op_ret == 0) {
+            int_lock->lockee[lockee_num]
+                .locked_nodes[child_index] |= LOCKED_YES;
+            int_lock->lockee[lockee_num].locked_count++;
+            int_lock->lock_count++;
+        }
+
+        call_count = --int_lock->lk_call_count;
+    }
+    UNLOCK(&frame->lock);
+
+    if (call_count == 0) {
+        gf_msg_trace(this->name, 0, "Last locking reply received");
+        /* all locks successful. Proceed to call FOP */
+        if (int_lock->lock_count == int_lock->lk_expected_count) {
+            gf_msg_trace(this->name, 0, "All servers locked. Calling the cbk");
+            int_lock->lock_op_ret = 0;
+            int_lock->lock_cbk(frame, this);
+        }
+        /* Not all locks were successful. Unlock and try locking
+           again, this time with serially blocking locks */
+        else {
+            gf_msg_trace(this->name, 0,
+                         "%d servers locked. Trying again "
+                         "with blocking calls",
+                         int_lock->lock_count);
+
+            afr_unlock_now(frame, this);
+        }
+    }
+
+    return 0;
 }
 
 int
-afr_attempt_lock_recovery (xlator_t *this, int32_t child_index)
-{
-        call_frame_t    *frame      = NULL;
-        afr_private_t   *priv       = NULL;
-        afr_local_t     *local      = NULL;
-        afr_locked_fd_t *locked_fd  = NULL;
-        afr_locked_fd_t  *tmp       = NULL;
-        int              ret        = -1;
-        struct list_head locks_list = {0,};
-        int32_t          op_errno   = 0;
-
-
-        priv = this->private;
-
-        if (list_empty (&priv->saved_fds))
-                goto out;
-
-        frame = create_frame (this, this->ctx->pool);
-        if (!frame) {
-                ret = -1;
-                goto out;
-        }
-
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0) {
-                ret = -1;
-                goto out;
-        }
-
-        frame->local = local;
-
-        INIT_LIST_HEAD (&locks_list);
-
-        pthread_mutex_lock (&priv->mutex);
-        {
-                list_splice_init (&priv->saved_fds, &locks_list);
-        }
-        pthread_mutex_unlock (&priv->mutex);
-
-        list_for_each_entry_safe (locked_fd, tmp,
-                                  &locks_list, list) {
-
-                list_del_init (&locked_fd->list);
-
-                local->fd                  = fd_ref (locked_fd->fd);
-                local->lock_recovery_child = child_index;
-                local->locked_fd           = locked_fd;
-
-                if (!is_fd_opened (locked_fd->fd, child_index)) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "attempting open before lock "
-                                "recovery");
-                        afr_lock_recovery_preopen (frame, this);
-                } else {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "attempting lock recovery "
-                                "without a preopen");
-                        afr_lock_recovery (frame, this);
-                }
+afr_lock_nonblocking(call_frame_t *frame, xlator_t *this)
+{
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    afr_fd_ctx_t *fd_ctx = NULL;
+    int child = 0;
+    int lockee_num = 0;
+    int32_t call_count = 0;
+    int i = 0;
+    int ret = 0;
+
+    local = frame->local;
+    int_lock = &local->internal_lock;
+    priv = this->private;
+
+    initialize_internal_lock_variables(frame, this);
+
+    if (local->fd) {
+        fd_ctx = afr_fd_ctx_get(local->fd, this);
+        if (!fd_ctx) {
+            gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_FD_CTX_GET_FAILED,
+                   "unable to get fd ctx for fd=%p", local->fd);
+
+            local->op_ret = -1;
+            int_lock->lock_op_ret = -1;
+            local->op_errno = EINVAL;
+            int_lock->lock_op_errno = EINVAL;
+
+            afr_unlock_now(frame, this);
+            ret = -1;
+            goto out;
+        }
+    }
+
+    call_count = int_lock->lockee_count * internal_lock_count(frame, this);
+    int_lock->lk_call_count = call_count;
+    int_lock->lk_expected_count = call_count;
+
+    if (!call_count) {
+        gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_INFO_COMMON,
+               "fd not open on any subvolumes. aborting.");
+        afr_unlock_now(frame, this);
+        goto out;
+    }
+
+    /* Send non-blocking lock calls only on up children
+       and where the fd has been opened */
+    for (i = 0; i < int_lock->lockee_count * priv->child_count; i++) {
+        child = i % priv->child_count;
+        lockee_num = i / priv->child_count;
+        if (local->child_up[child]) {
+            afr_internal_lock_wind(frame, afr_nb_internal_lock_cbk,
+                                   (void *)(long)i, child, lockee_num,
+                                   _gf_false, _gf_false);
+            if (!--call_count)
+                break;
         }
-
+    }
 out:
-        if ((ret < 0) && frame)
-                AFR_STACK_DESTROY (frame);
-        return ret;
+    return ret;
 }
 
-int
-afr_lk_transfer_datalock (call_frame_t *dst, call_frame_t *src, char *dom,
-                          unsigned int child_count)
-{
-        afr_local_t         *dst_local   = NULL;
-        afr_local_t         *src_local   = NULL;
-        afr_internal_lock_t *dst_lock    = NULL;
-        afr_internal_lock_t *src_lock    = NULL;
-        afr_inodelk_t       *dst_inodelk = NULL;
-        afr_inodelk_t       *src_inodelk = NULL;
-        int                 ret = -1;
-
-        src_local = src->local;
-        src_lock  = &src_local->internal_lock;
-        src_inodelk = afr_get_inodelk (src_lock, dom);
-        dst_local = dst->local;
-        dst_lock  = &dst_local->internal_lock;
-        dst_inodelk = afr_get_inodelk (dst_lock, dom);
-        if (!dst_inodelk || !src_inodelk)
-                goto out;
-        if (src_inodelk->locked_nodes) {
-                memcpy (dst_inodelk->locked_nodes, src_inodelk->locked_nodes,
-                        sizeof (*dst_inodelk->locked_nodes) * child_count);
-                memset (src_inodelk->locked_nodes, 0,
-                        sizeof (*src_inodelk->locked_nodes) * child_count);
-        }
+int32_t
+afr_unlock(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_lock_t *lock = NULL;
+
+    local = frame->local;
+
+    if (!local->transaction.eager_lock_on)
+        goto out;
+    lock = &local->inode_ctx->lock[local->transaction.type];
+    LOCK(&local->inode->lock);
+    {
+        list_del_init(&local->transaction.owner_list);
+        if (list_empty(&lock->owners) && list_empty(&lock->post_op)) {
+            local->transaction.do_eager_unlock = _gf_true;
+            /*TODO: Need to get metadata use on_disk and inherit/uninherit
+             *GF_ASSERT (!local->inode_ctx->on_disk[local->transaction.type]);
+             *GF_ASSERT (!local->inode_ctx->inherited[local->transaction.type]);
+             */
+            GF_ASSERT(lock->release);
+        }
+    }
+    UNLOCK(&local->inode->lock);
+    if (!local->transaction.do_eager_unlock) {
+        local->internal_lock.lock_cbk(frame, this);
+        return 0;
+    }
 
-        dst_lock->transaction_lk_type = src_lock->transaction_lk_type;
-        dst_lock->selfheal_lk_type    = src_lock->selfheal_lk_type;
-        dst_inodelk->lock_count = src_inodelk->lock_count;
-        src_inodelk->lock_count = 0;
-        ret = 0;
 out:
-        return ret;
+    afr_unlock_now(frame, this);
+    return 0;
 }
diff --git a/xlators/cluster/afr/src/afr-mem-types.h b/xlators/cluster/afr/src/afr-mem-types.h
index e01ab366f4f..816065fb57a 100644
--- a/xlators/cluster/afr/src/afr-mem-types.h
+++ b/xlators/cluster/afr/src/afr-mem-types.h
@@ -8,41 +8,31 @@
   cases as published by the Free Software Foundation.
 */
 
-
 #ifndef __AFR_MEM_TYPES_H__
 #define __AFR_MEM_TYPES_H__
 
-#include "mem-types.h"
+#include <glusterfs/mem-types.h>
 
 enum gf_afr_mem_types_ {
-        gf_afr_mt_iovec  = gf_common_mt_end + 1,
-        gf_afr_mt_afr_fd_ctx_t,
-        gf_afr_mt_afr_private_t,
-        gf_afr_mt_int32_t,
-        gf_afr_mt_char,
-        gf_afr_mt_xattr_key,
-        gf_afr_mt_dict_t,
-        gf_afr_mt_xlator_t,
-        gf_afr_mt_iatt,
-        gf_afr_mt_int,
-        gf_afr_mt_afr_node_character,
-        gf_afr_mt_sh_diff_loop_state,
-        gf_afr_mt_uint8_t,
-        gf_afr_mt_loc_t,
-        gf_afr_mt_entry_name,
-        gf_afr_mt_pump_priv,
-        gf_afr_mt_locked_fd,
-        gf_afr_mt_inode_ctx_t,
-        gf_afr_fd_paused_call_t,
-        gf_afr_mt_crawl_data_t,
-        gf_afr_mt_brick_pos_t,
-        gf_afr_mt_shd_bool_t,
-        gf_afr_mt_shd_timer_t,
-        gf_afr_mt_shd_event_t,
-        gf_afr_mt_time_t,
-        gf_afr_mt_pos_data_t,
-	gf_afr_mt_reply_t,
-        gf_afr_mt_end
+    gf_afr_mt_afr_fd_ctx_t = gf_common_mt_end + 1,
+    gf_afr_mt_afr_private_t,
+    gf_afr_mt_int32_t,
+    gf_afr_mt_char,
+    gf_afr_mt_xattr_key,
+    gf_afr_mt_dict_t,
+    gf_afr_mt_xlator_t,
+    gf_afr_mt_afr_node_character,
+    gf_afr_mt_inode_ctx_t,
+    gf_afr_mt_shd_event_t,
+    gf_afr_mt_reply_t,
+    gf_afr_mt_subvol_healer_t,
+    gf_afr_mt_spbc_timeout_t,
+    gf_afr_mt_spb_status_t,
+    gf_afr_mt_empty_brick_t,
+    gf_afr_mt_child_latency_t,
+    gf_afr_mt_atomic_t,
+    gf_afr_mt_lk_heal_info_t,
+    gf_afr_mt_gf_lock,
+    gf_afr_mt_end
 };
 #endif
-
diff --git a/xlators/cluster/afr/src/afr-messages.h b/xlators/cluster/afr/src/afr-messages.h
new file mode 100644
index 00000000000..e73fd997765
--- /dev/null
+++ b/xlators/cluster/afr/src/afr-messages.h
@@ -0,0 +1,167 @@
+/*
+ Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+ This file is part of GlusterFS.
+
+ This file is licensed to you under your choice of the GNU Lesser
+ General Public License, version 3 or any later version (LGPLv3 or
+ later), or the GNU General Public License, version 2 (GPLv2), in all
+ cases as published by the Free Software Foundation.
+ */
+
+#ifndef _AFR_MESSAGES_H_
+#define _AFR_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(
+    AFR, AFR_MSG_QUORUM_FAIL, AFR_MSG_QUORUM_MET, AFR_MSG_QUORUM_OVERRIDE,
+    AFR_MSG_INVALID_CHILD_UP, AFR_MSG_SUBVOL_UP, AFR_MSG_SUBVOLS_DOWN,
+    AFR_MSG_ENTRY_UNLOCK_FAIL, AFR_MSG_SPLIT_BRAIN, AFR_MSG_OPEN_FAIL,
+    AFR_MSG_UNLOCK_FAIL, AFR_MSG_REPLACE_BRICK_STATUS, AFR_MSG_GFID_NULL,
+    AFR_MSG_FD_CREATE_FAILED, AFR_MSG_DICT_SET_FAILED,
+    AFR_MSG_EXPUNGING_FILE_OR_DIR, AFR_MSG_MIGRATION_IN_PROGRESS,
+    AFR_MSG_CHILD_MISCONFIGURED, AFR_MSG_VOL_MISCONFIGURED,
+    AFR_MSG_INTERNAL_LKS_FAILED, AFR_MSG_INVALID_FD, AFR_MSG_LOCK_INFO,
+    AFR_MSG_LOCK_XLATOR_NOT_LOADED, AFR_MSG_FD_CTX_GET_FAILED,
+    AFR_MSG_INVALID_SUBVOL, AFR_MSG_PUMP_XLATOR_ERROR, AFR_MSG_SELF_HEAL_INFO,
+    AFR_MSG_READ_SUBVOL_ERROR, AFR_MSG_DICT_GET_FAILED, AFR_MSG_INFO_COMMON,
+    AFR_MSG_SPLIT_BRAIN_CHOICE_ERROR, AFR_MSG_LOCAL_CHILD, AFR_MSG_INVALID_DATA,
+    AFR_MSG_INVALID_ARG, AFR_MSG_INDEX_DIR_GET_FAILED, AFR_MSG_FSYNC_FAILED,
+    AFR_MSG_FAVORITE_CHILD, AFR_MSG_SELF_HEAL_FAILED,
+    AFR_MSG_SPLIT_BRAIN_STATUS, AFR_MSG_ADD_BRICK_STATUS, AFR_MSG_NO_CHANGELOG,
+    AFR_MSG_TIMER_CREATE_FAIL, AFR_MSG_SBRAIN_FAV_CHILD_POLICY,
+    AFR_MSG_INODE_CTX_GET_FAILED, AFR_MSG_THIN_ARB,
+    AFR_MSG_THIN_ARB_XATTROP_FAILED, AFR_MSG_THIN_ARB_LOC_POP_FAILED,
+    AFR_MSG_GET_PEND_VAL, AFR_MSG_THIN_ARB_SKIP_SHD, AFR_MSG_UNKNOWN_SET,
+    AFR_MSG_NO_XL_ID, AFR_MSG_SELF_HEAL_INFO_START,
+    AFR_MSG_SELF_HEAL_INFO_FINISH, AFR_MSG_INCRE_COUNT,
+    AFR_MSG_ADD_TO_OUTPUT_FAILED, AFR_MSG_SET_TIME_FAILED,
+    AFR_MSG_GFID_MISMATCH_DETECTED, AFR_MSG_GFID_HEAL_MSG,
+    AFR_MSG_THIN_ARB_LOOKUP_FAILED, AFR_MSG_DICT_CREATE_FAILED,
+    AFR_MSG_NO_MAJORITY_TO_RESOLVE, AFR_MSG_TYPE_MISMATCH,
+    AFR_MSG_SIZE_POLICY_NOT_APPLICABLE, AFR_MSG_NO_CHILD_SELECTED,
+    AFR_MSG_INVALID_CHILD, AFR_MSG_RESOLVE_CONFLICTING_DATA,
+    SERROR_GETTING_SRC_BRICK, SNO_DIFF_IN_MTIME, SNO_BIGGER_FILE,
+    SALL_BRICKS_UP_TO_RESOLVE, AFR_MSG_UNLOCK_FAILED, AFR_MSG_POST_OP_FAILED,
+    AFR_MSG_TA_FRAME_CREATE_FAILED, AFR_MSG_SET_KEY_XATTROP_FAILED,
+    AFR_MSG_BLOCKING_ENTRYLKS_FAILED, AFR_MSG_FOP_FAILED,
+    AFR_MSG_CLEAN_UP_FAILED, AFR_MSG_UNABLE_TO_FETCH, AFR_MSG_XATTR_SET_FAILED,
+    AFR_MSG_SPLIT_BRAIN_REPLICA, AFR_MSG_INODE_CTX_FAILED,
+    AFR_MSG_LOOKUP_FAILED, AFR_MSG_ALL_SUBVOLS_DOWN,
+    AFR_MSG_RELEASE_LOCK_FAILED, AFR_MSG_CLEAR_TIME_SPLIT_BRAIN,
+    AFR_MSG_READ_FAILED, AFR_MSG_LAUNCH_FAILED, AFR_MSG_READ_SUBVOL_NOT_UP,
+    AFR_MSG_LK_HEAL_DOM, AFR_MSG_NEW_BRICK, AFR_MSG_SPLIT_BRAIN_SET_FAILED,
+    AFR_MSG_SPLIT_BRAIN_DETERMINE_FAILED, AFR_MSG_HEALER_SPAWN_FAILED,
+    AFR_MSG_ADD_CRAWL_EVENT_FAILED, AFR_MSG_NULL_DEREF, AFR_MSG_SET_PEND_XATTR,
+    AFR_MSG_INTERNAL_ATTR);
+
+#define AFR_MSG_DICT_GET_FAILED_STR "Dict get failed"
+#define AFR_MSG_DICT_SET_FAILED_STR "Dict set failed"
+#define AFR_MSG_HEALER_SPAWN_FAILED_STR "Healer spawn failed"
+#define AFR_MSG_ADD_CRAWL_EVENT_FAILED_STR "Adding crawl event failed"
+#define AFR_MSG_INVALID_ARG_STR "Invalid argument"
+#define AFR_MSG_INDEX_DIR_GET_FAILED_STR "unable to get index-dir on "
+#define AFR_MSG_THIN_ARB_LOOKUP_FAILED_STR "Failed lookup on file"
+#define AFR_MSG_DICT_CREATE_FAILED_STR "Failed to create dict."
+#define AFR_MSG_THIN_ARB_XATTROP_FAILED_STR "Xattrop failed."
+#define AFR_MSG_THIN_ARB_LOC_POP_FAILED_STR                                    \
+    "Failed to populate loc for thin-arbiter"
+#define AFR_MSG_GET_PEND_VAL_STR "Error getting value of pending"
+#define AFR_MSG_THIN_ARB_SKIP_SHD_STR "I am not the god shd. skipping."
+#define AFR_MSG_UNKNOWN_SET_STR "Unknown set"
+#define AFR_MSG_NO_XL_ID_STR "xl does not have id"
+#define AFR_MSG_SELF_HEAL_INFO_START_STR "starting full sweep on"
+#define AFR_MSG_SELF_HEAL_INFO_FINISH_STR "finished full sweep on"
+#define AFR_MSG_INCRE_COUNT_STR "Could not increment the counter."
+#define AFR_MSG_ADD_TO_OUTPUT_FAILED_STR "Could not add to output"
+#define AFR_MSG_SET_TIME_FAILED_STR "Could not set time"
+#define AFR_MSG_GFID_HEAL_MSG_STR "Error setting gfid-heal-msg dict"
+#define AFR_MSG_NO_MAJORITY_TO_RESOLVE_STR                                     \
+    "No majority to resolve gfid split brain"
+#define AFR_MSG_GFID_MISMATCH_DETECTED_STR "Gfid mismatch dectected"
+#define AFR_MSG_SELF_HEAL_INFO_STR "performing selfheal"
+#define AFR_MSG_TYPE_MISMATCH_STR "TYPE mismatch"
+#define AFR_MSG_SIZE_POLICY_NOT_APPLICABLE_STR                                 \
+    "Size policy is not applicable to directories."
+#define AFR_MSG_NO_CHILD_SELECTED_STR                                          \
+    "No child selected by favorite-child policy"
+#define AFR_MSG_INVALID_CHILD_STR "Invalid child"
+#define AFR_MSG_RESOLVE_CONFLICTING_DATA_STR                                   \
+    "selected as authentic to resolve conflicting data"
+#define SERROR_GETTING_SRC_BRICK_STR "Error getting the source brick"
+#define SNO_DIFF_IN_MTIME_STR "No difference in mtime"
+#define SNO_BIGGER_FILE_STR "No bigger file"
+#define SALL_BRICKS_UP_TO_RESOLVE_STR                                          \
+    "All the bricks should be up to resolve the gfid split brain"
+#define AFR_MSG_UNLOCK_FAILED_STR "Failed to unlock"
+#define AFR_MSG_POST_OP_FAILED_STR "Post-op on thin-arbiter failed"
+#define AFR_MSG_TA_FRAME_CREATE_FAILED_STR "Failed to create ta_frame"
+#define AFR_MSG_SET_KEY_XATTROP_FAILED_STR "Could not set key during xattrop"
+#define AFR_MSG_BLOCKING_ENTRYLKS_FAILED_STR "Blocking entrylks failed"
+#define AFR_MSG_FSYNC_FAILED_STR "fsync failed"
+#define AFR_MSG_QUORUM_FAIL_STR "quorum is not met"
+#define AFR_MSG_FOP_FAILED_STR "Failing Fop"
+#define AFR_MSG_INVALID_SUBVOL_STR "not a subvolume"
+#define AFR_MSG_VOL_MISCONFIGURED_STR "Volume is dangling"
+#define AFR_MSG_CHILD_MISCONFIGURED_STR                                        \
+    "replicate translator needs more than one subvolume defined"
+#define AFR_MSG_CLEAN_UP_FAILED_STR "Failed to clean up healer threads"
+#define AFR_MSG_QUORUM_OVERRIDE_STR "overriding quorum-count"
+#define AFR_MSG_UNABLE_TO_FETCH_STR                                            \
+    "Unable to fetch afr-pending-xattr option from volfile. Falling back to "  \
+    "using client translator names"
+#define AFR_MSG_NULL_DEREF_STR "possible NULL deref"
+#define AFR_MSG_XATTR_SET_FAILED_STR "Cannot set xattr cookie key"
+#define AFR_MSG_SPLIT_BRAIN_STATUS_STR "Failed to create synctask"
+#define AFR_MSG_SUBVOLS_DOWN_STR "All subvolumes are not up"
+#define AFR_MSG_SPLIT_BRAIN_CHOICE_ERROR_STR                                   \
+    "Failed to cancel split-brain choice"
+#define AFR_MSG_SPLIT_BRAIN_REPLICA_STR                                        \
+    "Cannot set replica. File is not in data/metadata split-brain"
+#define AFR_MSG_INODE_CTX_FAILED_STR "Failed to get inode_ctx"
+#define AFR_MSG_READ_SUBVOL_ERROR_STR "no read subvols"
+#define AFR_MSG_LOCAL_CHILD_STR "selecting local read-child"
+#define AFR_MSG_LOOKUP_FAILED_STR "Failed to lookup/create thin-arbiter id file"
+#define AFR_MSG_TIMER_CREATE_FAIL_STR                                          \
+    "Cannot create timer for delayed initialization"
+#define AFR_MSG_SUBVOL_UP_STR "Subvolume came back up; going online"
+#define AFR_MSG_ALL_SUBVOLS_DOWN_STR                                           \
+    "All subvolumes are down. Going offline until atleast one of them is up"
+#define AFR_MSG_RELEASE_LOCK_FAILED_STR "Failed to release lock"
+#define AFR_MSG_INVALID_CHILD_UP_STR "Received child_up from invalid subvolume"
+#define AFR_MSG_QUORUM_MET_STR "Client-quorum is met"
+#define AFR_MSG_EXPUNGING_FILE_OR_DIR_STR "expunging file or dir"
+#define AFR_MSG_SELF_HEAL_FAILED_STR "Invalid"
+#define AFR_MSG_SPLIT_BRAIN_STR "Skipping conservative mergeon the file"
+#define AFR_MSG_CLEAR_TIME_SPLIT_BRAIN_STR "clear time split brain"
+#define AFR_MSG_READ_FAILED_STR "Failing read since good brick is down"
+#define AFR_MSG_LAUNCH_FAILED_STR "Failed to launch synctask"
+#define AFR_MSG_READ_SUBVOL_NOT_UP_STR                                         \
+    "read subvolume in this generation is not up"
+#define AFR_MSG_INTERNAL_LKS_FAILED_STR                                        \
+    "Unable to work with lk-owner while attempting fop"
+#define AFR_MSG_LOCK_XLATOR_NOT_LOADED_STR                                     \
+    "subvolume does not support locking. please load features/locks xlator "   \
+    "on server."
+#define AFR_MSG_FD_CTX_GET_FAILED_STR "unable to get fd ctx"
+#define AFR_MSG_INFO_COMMON_STR "fd not open on any subvolumes, aborting."
+#define AFR_MSG_REPLACE_BRICK_STATUS_STR "Couldn't acquire lock on any child."
+#define AFR_MSG_NEW_BRICK_STR "New brick"
+#define AFR_MSG_SPLIT_BRAIN_SET_FAILED_STR                                     \
+    "Failed to set split-brain choice to -1"
+#define AFR_MSG_SPLIT_BRAIN_DETERMINE_FAILED_STR                               \
+    "Failed to determine split-brain. Aborting split-brain-choice set"
+#define AFR_MSG_OPEN_FAIL_STR "Failed to open subvolume"
+#define AFR_MSG_SET_PEND_XATTR_STR "Set of pending xattr"
+#define AFR_MSG_INTERNAL_ATTR_STR "is an internal extended attribute"
+#endif /* !_AFR_MESSAGES_H_ */
diff --git a/xlators/cluster/afr/src/afr-open.c b/xlators/cluster/afr/src/afr-open.c
index 643a5d692df..64856042b65 100644
--- a/xlators/cluster/afr/src/afr-open.c
+++ b/xlators/cluster/afr/src/afr-open.c
@@ -8,375 +8,346 @@
   cases as published by the Free Software Foundation.
 */
 
-#include <libgen.h>
 #include <unistd.h>
-#include <fnmatch.h>
 #include <sys/time.h>
 #include <stdlib.h>
 #include <signal.h>
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
+#include <glusterfs/glusterfs.h>
 #include "afr.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
-#include "byte-order.h"
-#include "statedump.h"
-
-#include "fd.h"
-
-#include "afr-inode-read.h"
-#include "afr-inode-write.h"
-#include "afr-dir-read.h"
-#include "afr-dir-write.h"
-#include "afr-transaction.h"
-#include "afr-self-heal.h"
-#include "afr-self-heal-common.h"
-
-int
-afr_stale_child_up (afr_local_t *local, xlator_t *this)
-{
-        int             i = 0;
-        afr_private_t   *priv = NULL;
-        int             up = -1;
-
-        priv = this->private;
-
-        if (!local->fresh_children)
-                local->fresh_children = afr_children_create (priv->child_count);
-        if (!local->fresh_children)
-                goto out;
-
-        afr_inode_get_read_ctx (this, local->fd->inode, local->fresh_children);
-        if (priv->child_count == afr_get_children_count (local->fresh_children,
-                                                         priv->child_count))
-                goto out;
+#include <glusterfs/dict.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/byte-order.h>
+#include <glusterfs/statedump.h>
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (!local->child_up[i])
-                        continue;
-                if (afr_is_child_present (local->fresh_children,
-                                          priv->child_count, i))
-                        continue;
-                up = i;
-                break;
-        }
-out:
-        return up;
-}
+#include "afr-transaction.h"
 
-void
-afr_perform_data_self_heal (call_frame_t *frame, xlator_t *this)
+gf_boolean_t
+afr_is_fd_fixable(fd_t *fd)
 {
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        inode_t         *inode = NULL;
-        int             st_child = -1;
-        char            reason[64] = {0};
-
-        local = frame->local;
-        sh = &local->self_heal;
-        inode = local->fd->inode;
-
-        if (!IA_ISREG (inode->ia_type))
-                goto out;
-
-        st_child = afr_stale_child_up (local, this);
-        if (st_child < 0)
-                goto out;
-
-        sh->do_data_self_heal          = _gf_true;
-        sh->do_metadata_self_heal      = _gf_true;
-        sh->do_gfid_self_heal          = _gf_true;
-        sh->do_missing_entry_self_heal = _gf_true;
-
-        snprintf (reason, sizeof (reason), "stale subvolume %d detected",
-                  st_child);
-        afr_launch_self_heal (frame, this, inode, _gf_true, inode->ia_type,
-                              reason, NULL, NULL);
-out:
-        return;
+    if (!fd || !fd->inode)
+        return _gf_false;
+    else if (fd_is_anonymous(fd))
+        return _gf_false;
+    else if (gf_uuid_is_null(fd->inode->gfid))
+        return _gf_false;
+
+    return _gf_true;
 }
 
 int
-afr_open_ftruncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                        struct iatt *postbuf, dict_t *xdata)
+afr_open_ftruncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                       struct iatt *postbuf, dict_t *xdata)
 {
-        afr_local_t * local = frame->local;
-        afr_private_t *priv = NULL;
-
-        priv = this->private;
-        if (afr_open_only_data_self_heal (priv->data_self_heal))
-                afr_perform_data_self_heal (frame, this);
-        AFR_STACK_UNWIND (open, frame, local->op_ret, local->op_errno,
-                          local->fd, xdata);
-        return 0;
-}
+    afr_local_t *local = frame->local;
 
+    AFR_STACK_UNWIND(open, frame, local->op_ret, local->op_errno,
+                     local->cont.open.fd, xdata);
+    return 0;
+}
 
 int
-afr_open_cbk (call_frame_t *frame, void *cookie,
-              xlator_t *this, int32_t op_ret, int32_t op_errno,
-              fd_t *fd, dict_t *xdata)
+afr_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, fd_t *fd, dict_t *xdata)
 {
-        afr_local_t *  local       = NULL;
-        int            ret         = 0;
-        int            call_count  = -1;
-        int            child_index = (long) cookie;
-        afr_private_t *priv        = NULL;
-
-        priv = this->private;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        local->op_errno = op_errno;
-                }
-
-                if (op_ret >= 0) {
-                        local->op_ret = op_ret;
-                        local->success_count++;
-
-                        ret = afr_child_fd_ctx_set (this, fd, child_index,
-                                                    local->cont.open.flags);
-                        if (ret) {
-                                local->op_ret = -1;
-                                local->op_errno = -ret;
-                                goto unlock;
-                        }
-                }
+    afr_local_t *local = NULL;
+    int call_count = -1;
+    int child_index = (long)cookie;
+    afr_fd_ctx_t *fd_ctx = NULL;
+
+    local = frame->local;
+    fd_ctx = local->fd_ctx;
+
+    local->replies[child_index].valid = 1;
+    local->replies[child_index].op_ret = op_ret;
+    local->replies[child_index].op_errno = op_errno;
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            local->op_errno = op_errno;
+            fd_ctx->opened_on[child_index] = AFR_FD_NOT_OPENED;
+        } else {
+            local->op_ret = op_ret;
+            fd_ctx->opened_on[child_index] = AFR_FD_OPENED;
+            if (!local->xdata_rsp && xdata)
+                local->xdata_rsp = dict_ref(xdata);
         }
-unlock:
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                if ((local->cont.open.flags & O_TRUNC)
-                    && (local->op_ret >= 0)) {
-                        STACK_WIND (frame, afr_open_ftruncate_cbk,
-                                    this, this->fops->ftruncate,
-                                    fd, 0, NULL);
-                } else {
-                        if (afr_open_only_data_self_heal (priv->data_self_heal))
-                                afr_perform_data_self_heal (frame, this);
-                        AFR_STACK_UNWIND (open, frame, local->op_ret,
-                                          local->op_errno, local->fd, xdata);
-                }
+        call_count = --local->call_count;
+    }
+    UNLOCK(&frame->lock);
+
+    if (call_count == 0) {
+        afr_handle_replies_quorum(frame, this);
+        if (local->op_ret == -1) {
+            AFR_STACK_UNWIND(open, frame, local->op_ret, local->op_errno, NULL,
+                             NULL);
+        } else if (fd_ctx->flags & O_TRUNC) {
+            STACK_WIND(frame, afr_open_ftruncate_cbk, this,
+                       this->fops->ftruncate, fd, 0, NULL);
+        } else {
+            AFR_STACK_UNWIND(open, frame, local->op_ret, local->op_errno,
+                             local->cont.open.fd, local->xdata_rsp);
         }
+    }
 
-        return 0;
+    return 0;
 }
 
 int
-afr_open (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
-          fd_t *fd, dict_t *xdata)
+afr_open_continue(call_frame_t *frame, xlator_t *this, int err)
 {
-        afr_private_t * priv       = NULL;
-        afr_local_t *   local      = NULL;
-        int             i          = 0;
-        int             ret        = -1;
-        int32_t         call_count = 0;
-        int32_t         op_errno   = 0;
-        int32_t         wind_flags = flags & (~O_TRUNC);
-        //We can't let truncation to happen outside transaction.
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
-        VALIDATE_OR_GOTO (loc, out);
-
-        priv = this->private;
-
-        if (flags & (O_CREAT|O_TRUNC)) {
-                QUORUM_CHECK(open,out);
-        }
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int call_count = 0;
+    int i = 0;
 
-        if (afr_is_split_brain (this, loc->inode)) {
-                /* self-heal failed */
-                gf_log (this->name, GF_LOG_WARNING,
-                        "failed to open as split brain seen, returning EIO");
-                op_errno = EIO;
-                goto out;
-        }
-
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
-
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
-
-        call_count   = local->call_count;
-        loc_copy (&local->loc, loc);
+    local = frame->local;
+    priv = this->private;
 
-        local->cont.open.flags   = flags;
-
-        local->fd = fd_ref (fd);
+    if (err) {
+        AFR_STACK_UNWIND(open, frame, -1, err, NULL, NULL);
+    } else {
+        local->call_count = AFR_COUNT(local->child_up, priv->child_count);
+        call_count = local->call_count;
 
         for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND_COOKIE (frame, afr_open_cbk, (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->open,
-                                           loc, wind_flags, fd, xdata);
-
-                        if (!--call_count)
-                                break;
-                }
+            if (local->child_up[i]) {
+                STACK_WIND_COOKIE(frame, afr_open_cbk, (void *)(long)i,
+                                  priv->children[i],
+                                  priv->children[i]->fops->open, &local->loc,
+                                  (local->cont.open.flags & ~O_TRUNC),
+                                  local->cont.open.fd, local->xdata_req);
+                if (!--call_count)
+                    break;
+            }
         }
+    }
+    return 0;
+}
 
-        ret = 0;
+int
+afr_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+         fd_t *fd, dict_t *xdata)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int spb_subvol = 0;
+    int event_generation = 0;
+    int ret = 0;
+    int32_t op_errno = 0;
+    afr_fd_ctx_t *fd_ctx = NULL;
+
+    // We can't let truncation to happen outside transaction.
+
+    priv = this->private;
+
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
+
+    local->op = GF_FOP_OPEN;
+    fd_ctx = afr_fd_ctx_get(fd, this);
+    if (!fd_ctx) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    if (priv->quorum_count && !afr_has_quorum(local->child_up, this, NULL)) {
+        op_errno = afr_quorum_errno(priv);
+        goto out;
+    }
+
+    if (!afr_is_consistent_io_possible(local, priv, &op_errno))
+        goto out;
+
+    local->inode = inode_ref(loc->inode);
+    loc_copy(&local->loc, loc);
+    local->fd_ctx = fd_ctx;
+    fd_ctx->flags = flags;
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
+
+    local->cont.open.flags = flags;
+    local->cont.open.fd = fd_ref(fd);
+
+    ret = afr_inode_get_readable(frame, local->inode, this, NULL,
+                                 &event_generation, AFR_DATA_TRANSACTION);
+    if ((ret < 0) &&
+        (afr_split_brain_read_subvol_get(local->inode, this, NULL,
+                                         &spb_subvol) == 0) &&
+        spb_subvol < 0) {
+        afr_inode_refresh(frame, this, local->inode, local->inode->gfid,
+                          afr_open_continue);
+    } else {
+        afr_open_continue(frame, this, 0);
+    }
+
+    return 0;
 out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (open, frame, -1, op_errno, fd, xdata);
+    AFR_STACK_UNWIND(open, frame, -1, op_errno, fd, NULL);
 
-        return 0;
+    return 0;
 }
 
 int
-afr_openfd_fix_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, fd_t *fd,
-                         dict_t *xdata)
+afr_openfd_fix_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, fd_t *fd,
+                        dict_t *xdata)
 {
-        afr_local_t   *local      = NULL;
-        afr_private_t *priv       = NULL;
-        afr_fd_ctx_t  *fd_ctx     = NULL;
-        int           call_count  = 0;
-        int           child_index = (long) cookie;
-
-        priv     = this->private;
-        local    = frame->local;
-
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    afr_fd_ctx_t *fd_ctx = NULL;
+    int call_count = 0;
+    int child_index = (long)cookie;
+
+    priv = this->private;
+    local = frame->local;
+
+    if (op_ret >= 0) {
+        gf_msg_debug(this->name, 0,
+                     "fd for %s opened "
+                     "successfully on subvolume %s",
+                     local->loc.path, priv->children[child_index]->name);
+    } else {
+        gf_smsg(this->name, fop_log_level(GF_FOP_OPEN, op_errno), op_errno,
+                AFR_MSG_OPEN_FAIL, "path=%s", local->loc.path, "subvolume=%s",
+                priv->children[child_index]->name, NULL);
+    }
+
+    fd_ctx = local->fd_ctx;
+
+    LOCK(&local->fd->lock);
+    {
         if (op_ret >= 0) {
-                gf_log (this->name, GF_LOG_DEBUG, "fd for %s opened "
-                        "successfully on subvolume %s", local->loc.path,
-                        priv->children[child_index]->name);
+            fd_ctx->opened_on[child_index] = AFR_FD_OPENED;
         } else {
-                gf_log (this->name, GF_LOG_ERROR, "Failed to open %s "
-                        "on subvolume %s", local->loc.path,
-                        priv->children[child_index]->name);
+            fd_ctx->opened_on[child_index] = AFR_FD_NOT_OPENED;
         }
+    }
+    UNLOCK(&local->fd->lock);
 
-        fd_ctx = afr_fd_ctx_get (local->fd, this);
-        if (!fd_ctx) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "failed to get fd context, %p", local->fd);
-                goto out;
-        }
+    call_count = afr_frame_return(frame);
+    if (call_count == 0)
+        AFR_STACK_DESTROY(frame);
 
-        LOCK (&local->fd->lock);
-        {
-                if (op_ret >= 0) {
-                        fd_ctx->opened_on[child_index] = AFR_FD_OPENED;
-                } else {
-                        fd_ctx->opened_on[child_index] = AFR_FD_NOT_OPENED;
-                }
-        }
-        UNLOCK (&local->fd->lock);
-out:
-        call_count = afr_frame_return (frame);
-        if (call_count == 0)
-                AFR_STACK_DESTROY (frame);
+    return 0;
+}
+
+static int
+afr_fd_ctx_need_open(fd_t *fd, xlator_t *this, unsigned char *need_open)
+{
+    afr_fd_ctx_t *fd_ctx = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int count = 0;
+
+    priv = this->private;
 
+    fd_ctx = afr_fd_ctx_get(fd, this);
+    if (!fd_ctx)
         return 0;
+
+    LOCK(&fd->lock);
+    {
+        for (i = 0; i < priv->child_count; i++) {
+            if (fd_ctx->opened_on[i] == AFR_FD_NOT_OPENED &&
+                priv->child_up[i]) {
+                fd_ctx->opened_on[i] = AFR_FD_OPENING;
+                need_open[i] = 1;
+                count++;
+            } else {
+                need_open[i] = 0;
+            }
+        }
+    }
+    UNLOCK(&fd->lock);
+
+    return count;
 }
 
 void
-afr_fix_open (xlator_t *this, fd_t *fd, size_t need_open_count, int *need_open)
+afr_fix_open(fd_t *fd, xlator_t *this)
 {
-        afr_private_t *priv    = NULL;
-        int           i        = 0;
-        call_frame_t  *frame   = NULL;
-        afr_local_t   *local   = NULL;
-        int           ret      = -1;
-        int32_t       op_errno = 0;
-        afr_fd_ctx_t  *fd_ctx  = NULL;
-
-        priv  = this->private;
-
-        if (!afr_is_fd_fixable (fd) || !need_open || !need_open_count)
-                goto out;
-
-        fd_ctx = afr_fd_ctx_get (fd, this);
-        if (!fd_ctx) {
-                ret = -1;
-                goto out;
-        }
+    afr_private_t *priv = NULL;
+    int i = 0;
+    call_frame_t *frame = NULL;
+    afr_local_t *local = NULL;
+    int ret = -1;
+    int32_t op_errno = 0;
+    afr_fd_ctx_t *fd_ctx = NULL;
+    unsigned char *need_open = NULL;
+    int call_count = 0;
 
-        frame = create_frame (this, this->ctx->pool);
-        if (!frame) {
-                ret = -1;
-                goto out;
-        }
+    priv = this->private;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-        local = frame->local;
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
+    if (!afr_is_fd_fixable(fd))
+        goto out;
 
-        local->loc.inode = inode_ref (fd->inode);
-        ret = loc_path (&local->loc, NULL);
-        if (ret < 0)
-                goto out;
+    fd_ctx = afr_fd_ctx_get(fd, this);
+    if (!fd_ctx)
+        goto out;
 
-        local->fd = fd_ref (fd);
-        local->call_count = need_open_count;
+    need_open = alloca0(priv->child_count);
 
-        gf_log (this->name, GF_LOG_DEBUG, "need open count: %zd",
-                need_open_count);
+    call_count = afr_fd_ctx_need_open(fd, this, need_open);
+    if (!call_count)
+        goto out;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (!need_open[i])
-                        continue;
-
-                if (IA_IFDIR == fd->inode->ia_type) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "opening fd for dir %s on subvolume %s",
-                                local->loc.path, priv->children[i]->name);
-
-                        STACK_WIND_COOKIE (frame, afr_openfd_fix_open_cbk,
-                                           (void*) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->opendir,
-                                           &local->loc, local->fd,
-                                           NULL);
-                } else {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "opening fd for file %s on subvolume %s",
-                                local->loc.path, priv->children[i]->name);
-
-                        STACK_WIND_COOKIE (frame, afr_openfd_fix_open_cbk,
-                                           (void *)(long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->open,
-                                           &local->loc,
-                                           fd_ctx->flags & (~O_TRUNC),
-                                           local->fd, NULL);
-                }
+    frame = create_frame(this, this->ctx->pool);
+    if (!frame)
+        goto out;
+
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
+
+    local->loc.inode = inode_ref(fd->inode);
+    ret = loc_path(&local->loc, NULL);
+    if (ret < 0)
+        goto out;
+
+    local->fd = fd_ref(fd);
+    local->fd_ctx = fd_ctx;
 
+    local->call_count = call_count;
+
+    gf_msg_debug(this->name, 0, "need open count: %d", call_count);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!need_open[i])
+            continue;
+
+        if (IA_IFDIR == fd->inode->ia_type) {
+            gf_msg_debug(this->name, 0, "opening fd for dir %s on subvolume %s",
+                         local->loc.path, priv->children[i]->name);
+
+            STACK_WIND_COOKIE(frame, afr_openfd_fix_open_cbk, (void *)(long)i,
+                              priv->children[i],
+                              priv->children[i]->fops->opendir, &local->loc,
+                              local->fd, NULL);
+        } else {
+            gf_msg_debug(this->name, 0,
+                         "opening fd for file %s on subvolume %s",
+                         local->loc.path, priv->children[i]->name);
+
+            STACK_WIND_COOKIE(frame, afr_openfd_fix_open_cbk, (void *)(long)i,
+                              priv->children[i], priv->children[i]->fops->open,
+                              &local->loc, fd_ctx->flags & (~O_TRUNC),
+                              local->fd, NULL);
         }
-        op_errno = 0;
-        ret = 0;
+
+        if (!--call_count)
+            break;
+    }
+
+    return;
 out:
-        if (op_errno)
-                ret = -1; //For handling ALLOC_OR_GOTO
-        if (ret && frame)
-                AFR_STACK_DESTROY (frame);
+    if (frame)
+        AFR_STACK_DESTROY(frame);
 }
diff --git a/xlators/cluster/afr/src/afr-read-txn.c b/xlators/cluster/afr/src/afr-read-txn.c
new file mode 100644
index 00000000000..6fc2c75145c
--- /dev/null
+++ b/xlators/cluster/afr/src/afr-read-txn.c
@@ -0,0 +1,494 @@
+/*
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "afr.h"
+#include "afr-transaction.h"
+#include "afr-messages.h"
+
+void
+afr_pending_read_increment(afr_private_t *priv, int child_index)
+{
+    if (child_index < 0 || child_index > priv->child_count)
+        return;
+
+    GF_ATOMIC_INC(priv->pending_reads[child_index]);
+}
+
+void
+afr_pending_read_decrement(afr_private_t *priv, int child_index)
+{
+    if (child_index < 0 || child_index > priv->child_count)
+        return;
+
+    GF_ATOMIC_DEC(priv->pending_reads[child_index]);
+}
+
+void
+afr_read_txn_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    afr_pending_read_decrement(priv, local->read_subvol);
+    local->read_subvol = subvol;
+    afr_pending_read_increment(priv, subvol);
+    local->readfn(frame, this, subvol);
+}
+
+int
+afr_read_txn_next_subvol(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int subvol = -1;
+
+    local = frame->local;
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->readable[i]) {
+            /* don't even bother trying here.
+               just mark as attempted and move on. */
+            local->read_attempted[i] = 1;
+            continue;
+        }
+
+        if (!local->read_attempted[i]) {
+            subvol = i;
+            break;
+        }
+    }
+
+    /* If no more subvols were available for reading, we leave
+       @subvol as -1, which is an indication we have run out of
+       readable subvols. */
+    if (subvol != -1)
+        local->read_attempted[subvol] = 1;
+    afr_read_txn_wind(frame, this, subvol);
+
+    return 0;
+}
+
+static int
+afr_ta_read_txn_done(int ret, call_frame_t *ta_frame, void *opaque)
+{
+    STACK_DESTROY(ta_frame->root);
+    return 0;
+}
+
+static int
+afr_ta_read_txn(void *opaque)
+{
+    call_frame_t *frame = NULL;
+    xlator_t *this = NULL;
+    int read_subvol = -1;
+    int query_child = AFR_CHILD_UNKNOWN;
+    int possible_bad_child = AFR_CHILD_UNKNOWN;
+    int ret = 0;
+    int op_errno = ENOMEM;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    struct gf_flock flock = {
+        0,
+    };
+    dict_t *xdata_req = NULL;
+    dict_t *xdata_rsp = NULL;
+    int **pending = NULL;
+    loc_t loc = {
+        0,
+    };
+
+    frame = (call_frame_t *)opaque;
+    this = frame->this;
+    local = frame->local;
+    priv = this->private;
+    query_child = local->read_txn_query_child;
+
+    if (query_child == AFR_CHILD_ZERO) {
+        possible_bad_child = AFR_CHILD_ONE;
+    } else if (query_child == AFR_CHILD_ONE) {
+        possible_bad_child = AFR_CHILD_ZERO;
+    } else {
+        /*read_txn_query_child is AFR_CHILD_UNKNOWN*/
+        goto out;
+    }
+
+    /* Ask the query_child to see if it blames the possibly bad one. */
+    xdata_req = dict_new();
+    if (!xdata_req)
+        goto out;
+
+    pending = afr_matrix_create(priv->child_count, AFR_NUM_CHANGE_LOGS);
+    if (!pending)
+        goto out;
+
+    ret = afr_set_pending_dict(priv, xdata_req, pending);
+    if (ret < 0)
+        goto out;
+
+    if (local->fd) {
+        ret = syncop_fxattrop(priv->children[query_child], local->fd,
+                              GF_XATTROP_ADD_ARRAY, xdata_req, NULL, &xdata_rsp,
+                              NULL);
+    } else {
+        ret = syncop_xattrop(priv->children[query_child], &local->loc,
+                             GF_XATTROP_ADD_ARRAY, xdata_req, NULL, &xdata_rsp,
+                             NULL);
+    }
+    if (ret || !xdata_rsp) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed xattrop for gfid %s on %s",
+               uuid_utoa(local->inode->gfid),
+               priv->children[query_child]->name);
+        op_errno = -ret;
+        goto out;
+    }
+
+    if (afr_ta_dict_contains_pending_xattr(xdata_rsp, priv,
+                                           possible_bad_child)) {
+        read_subvol = query_child;
+        goto out;
+    }
+    dict_unref(xdata_rsp);
+    xdata_rsp = NULL;
+
+    /* It doesn't. So query thin-arbiter to see if it blames any data brick. */
+    ret = afr_fill_ta_loc(this, &loc, _gf_true);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed to populate thin-arbiter loc for: %s.", loc.name);
+        goto out;
+    }
+    flock.l_type = F_WRLCK; /*start and length are already zero. */
+    ret = syncop_inodelk(priv->children[THIN_ARBITER_BRICK_INDEX],
+                         AFR_TA_DOM_MODIFY, &loc, F_SETLKW, &flock, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "gfid:%s: Failed to get AFR_TA_DOM_MODIFY lock on %s.",
+               uuid_utoa(local->inode->gfid),
+               priv->pending_key[THIN_ARBITER_BRICK_INDEX]);
+        op_errno = -ret;
+        goto out;
+    }
+
+    ret = syncop_xattrop(priv->children[THIN_ARBITER_BRICK_INDEX], &loc,
+                         GF_XATTROP_ADD_ARRAY, xdata_req, NULL, &xdata_rsp,
+                         NULL);
+    if (ret || !xdata_rsp) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "gfid:%s: Failed xattrop on %s.", uuid_utoa(local->inode->gfid),
+               priv->pending_key[THIN_ARBITER_BRICK_INDEX]);
+        op_errno = -ret;
+        goto unlock;
+    }
+
+    if (!afr_ta_dict_contains_pending_xattr(xdata_rsp, priv, query_child)) {
+        read_subvol = query_child;
+    } else {
+        gf_msg(this->name, GF_LOG_ERROR, EIO, AFR_MSG_THIN_ARB,
+               "Failing read for gfid %s since good brick %s is down",
+               uuid_utoa(local->inode->gfid),
+               priv->children[possible_bad_child]->name);
+        op_errno = EIO;
+    }
+
+unlock:
+    flock.l_type = F_UNLCK;
+    ret = syncop_inodelk(priv->children[THIN_ARBITER_BRICK_INDEX],
+                         AFR_TA_DOM_MODIFY, &loc, F_SETLK, &flock, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "gfid:%s: Failed to unlock AFR_TA_DOM_MODIFY lock on "
+               "%s.",
+               uuid_utoa(local->inode->gfid),
+               priv->pending_key[THIN_ARBITER_BRICK_INDEX]);
+    }
+out:
+    if (xdata_req)
+        dict_unref(xdata_req);
+    if (xdata_rsp)
+        dict_unref(xdata_rsp);
+    if (pending)
+        afr_matrix_cleanup(pending, priv->child_count);
+    loc_wipe(&loc);
+
+    if (read_subvol == -1) {
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+    }
+    afr_read_txn_wind(frame, this, read_subvol);
+    return ret;
+}
+
+void
+afr_ta_read_txn_synctask(call_frame_t *frame, xlator_t *this)
+{
+    call_frame_t *ta_frame = NULL;
+    afr_local_t *local = NULL;
+    int ret = 0;
+
+    local = frame->local;
+    ta_frame = afr_ta_frame_create(this);
+    if (!ta_frame) {
+        local->op_ret = -1;
+        local->op_errno = ENOMEM;
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_THIN_ARB,
+               "Failed to create ta_frame");
+        goto out;
+    }
+    ret = synctask_new(this->ctx->env, afr_ta_read_txn, afr_ta_read_txn_done,
+                       ta_frame, frame);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_THIN_ARB,
+               "Failed to launch "
+               "afr_ta_read_txn synctask for gfid %s.",
+               uuid_utoa(local->inode->gfid));
+        local->op_ret = -1;
+        local->op_errno = ENOMEM;
+        STACK_DESTROY(ta_frame->root);
+        goto out;
+    }
+    return;
+out:
+    afr_read_txn_wind(frame, this, -1);
+}
+
+int
+afr_read_txn_refresh_done(call_frame_t *frame, xlator_t *this, int err)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int read_subvol = -1;
+    inode_t *inode = NULL;
+    int ret = -1;
+    int spb_subvol = -1;
+
+    local = frame->local;
+    inode = local->inode;
+    priv = this->private;
+
+    if (err) {
+        if (!priv->thin_arbiter_count)
+            goto readfn;
+        if (err != EINVAL)
+            goto readfn;
+        /* We need to query the good bricks and/or thin-arbiter.*/
+        afr_ta_read_txn_synctask(frame, this);
+        return 0;
+    }
+
+    read_subvol = afr_read_subvol_select_by_policy(inode, this, local->readable,
+                                                   NULL);
+    if (read_subvol == -1) {
+        err = EIO;
+        goto readfn;
+    }
+
+    if (local->read_attempted[read_subvol]) {
+        afr_read_txn_next_subvol(frame, this);
+        return 0;
+    }
+
+    local->read_attempted[read_subvol] = 1;
+readfn:
+    if (read_subvol == -1) {
+        ret = afr_split_brain_read_subvol_get(inode, this, frame, &spb_subvol);
+        if ((ret == 0) && spb_subvol >= 0)
+            read_subvol = spb_subvol;
+    }
+
+    if (read_subvol == -1) {
+        AFR_SET_ERROR_AND_CHECK_SPLIT_BRAIN(-1, err);
+    }
+    afr_read_txn_wind(frame, this, read_subvol);
+
+    return 0;
+}
+
+int
+afr_read_txn_continue(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (!local->refreshed) {
+        local->refreshed = _gf_true;
+        afr_inode_refresh(frame, this, local->inode, NULL,
+                          afr_read_txn_refresh_done);
+    } else {
+        afr_read_txn_next_subvol(frame, this);
+    }
+
+    return 0;
+}
+
+/* afr_read_txn_wipe:
+
+   clean internal variables in @local in order to make
+   it possible to call afr_read_txn() multiple times from
+   the same frame
+*/
+
+void
+afr_read_txn_wipe(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+
+    local = frame->local;
+    priv = this->private;
+
+    local->readfn = NULL;
+
+    if (local->inode)
+        inode_unref(local->inode);
+
+    for (i = 0; i < priv->child_count; i++) {
+        local->read_attempted[i] = 0;
+        local->readable[i] = 0;
+    }
+}
+
+/*
+  afr_read_txn:
+
+  This is the read transaction function. The way it works:
+
+  - Determine read-subvolume from inode ctx.
+
+  - If read-subvolume's generation was stale, refresh ctx once by
+    calling afr_inode_refresh()
+
+    Else make an attempt to read on read-subvolume.
+
+  - If attempted read on read-subvolume fails, refresh ctx once
+    by calling afr_inode_refresh()
+
+  - After ctx refresh, query read-subvolume freshly and attempt
+    read once.
+
+  - If read fails, try every other readable[] subvolume before
+    finally giving up. readable[] elements are set by afr_inode_refresh()
+    based on dirty and pending flags.
+
+  - If file is in split brain in the backend, generation will be
+    kept 0 by afr_inode_refresh() and readable[] will be set 0 for
+    all elements. Therefore reads always fail.
+*/
+
+int
+afr_read_txn(call_frame_t *frame, xlator_t *this, inode_t *inode,
+             afr_read_txn_wind_t readfn, afr_transaction_type type)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    unsigned char *data = NULL;
+    unsigned char *metadata = NULL;
+    int read_subvol = -1;
+    int event_generation = 0;
+    int ret = -1;
+
+    priv = this->private;
+    local = frame->local;
+    data = alloca0(priv->child_count);
+    metadata = alloca0(priv->child_count);
+
+    afr_read_txn_wipe(frame, this);
+
+    local->readfn = readfn;
+    local->inode = inode_ref(inode);
+    local->is_read_txn = _gf_true;
+    local->transaction.type = type;
+
+    if (priv->quorum_count && !afr_has_quorum(local->child_up, this, NULL)) {
+        local->op_ret = -1;
+        local->op_errno = afr_quorum_errno(priv);
+        goto read;
+    }
+
+    if (!afr_is_consistent_io_possible(local, priv, &local->op_errno)) {
+        local->op_ret = -1;
+        goto read;
+    }
+
+    if (priv->thin_arbiter_count && !afr_ta_has_quorum(priv, local)) {
+        local->op_ret = -1;
+        local->op_errno = -afr_quorum_errno(priv);
+        goto read;
+    }
+
+    if (priv->thin_arbiter_count &&
+        AFR_COUNT(local->child_up, priv->child_count) != priv->child_count) {
+        if (local->child_up[0]) {
+            local->read_txn_query_child = AFR_CHILD_ZERO;
+        } else if (local->child_up[1]) {
+            local->read_txn_query_child = AFR_CHILD_ONE;
+        }
+        afr_ta_read_txn_synctask(frame, this);
+        return 0;
+    }
+
+    ret = afr_inode_read_subvol_get(inode, this, data, metadata,
+                                    &event_generation);
+    if (ret == -1)
+        /* very first transaction on this inode */
+        goto refresh;
+    AFR_INTERSECT(local->readable, data, metadata, priv->child_count);
+
+    gf_msg_debug(this->name, 0,
+                 "%s: generation now vs cached: %d, "
+                 "%d",
+                 uuid_utoa(inode->gfid), local->event_generation,
+                 event_generation);
+    if (afr_is_inode_refresh_reqd(inode, this, local->event_generation,
+                                  event_generation))
+        /* servers have disconnected / reconnected, and possibly
+           rebooted, very likely changing the state of freshness
+           of copies */
+        goto refresh;
+
+    read_subvol = afr_read_subvol_select_by_policy(inode, this, local->readable,
+                                                   NULL);
+
+    if (read_subvol < 0 || read_subvol > priv->child_count) {
+        gf_msg_debug(this->name, 0,
+                     "Unreadable subvolume %d found "
+                     "with event generation %d for gfid %s.",
+                     read_subvol, event_generation, uuid_utoa(inode->gfid));
+        goto refresh;
+    }
+
+    if (!local->child_up[read_subvol]) {
+        /* should never happen, just in case */
+        gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_READ_SUBVOL_ERROR,
+               "subvolume %d is the "
+               "read subvolume in this generation, but is not up",
+               read_subvol);
+        goto refresh;
+    }
+
+    local->read_attempted[read_subvol] = 1;
+
+read:
+    afr_read_txn_wind(frame, this, read_subvol);
+
+    return 0;
+
+refresh:
+    afr_inode_refresh(frame, this, inode, NULL, afr_read_txn_refresh_done);
+
+    return 0;
+}
diff --git a/xlators/cluster/afr/src/afr-self-heal-algorithm.c b/xlators/cluster/afr/src/afr-self-heal-algorithm.c
deleted file mode 100644
index 83846f152d2..00000000000
--- a/xlators/cluster/afr/src/afr-self-heal-algorithm.c
+++ /dev/null
@@ -1,837 +0,0 @@
-/*
-  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
-  This file is part of GlusterFS.
-
-  This file is licensed to you under your choice of the GNU Lesser
-  General Public License, version 3 or any later version (LGPLv3 or
-  later), or the GNU General Public License, version 2 (GPLv2), in all
-  cases as published by the Free Software Foundation.
-*/
-
-
-#include <openssl/md5.h>
-#include "glusterfs.h"
-#include "afr.h"
-#include "xlator.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
-#include "byte-order.h"
-
-#include "afr-transaction.h"
-#include "afr-self-heal.h"
-#include "afr-self-heal-common.h"
-#include "afr-self-heal-algorithm.h"
-
-/*
-  This file contains the various self-heal algorithms
-*/
-
-static int
-sh_loop_driver (call_frame_t *sh_frame, xlator_t *this,
-                gf_boolean_t is_first_call, call_frame_t *old_loop_frame);
-static int
-sh_loop_return (call_frame_t *sh_frame, xlator_t *this, call_frame_t *loop_frame,
-                int32_t op_ret, int32_t op_errno);
-static int
-sh_destroy_frame (call_frame_t *frame, xlator_t *this)
-{
-        if (!frame)
-                goto out;
-
-        AFR_STACK_DESTROY (frame);
-out:
-        return 0;
-}
-
-static void
-sh_private_cleanup (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t             *local   = NULL;
-        afr_self_heal_t         *sh      = NULL;
-        afr_sh_algo_private_t   *sh_priv = NULL;
-
-        local = frame->local;
-        sh    = &local->self_heal;
-
-        sh_priv = sh->private;
-        GF_FREE (sh_priv);
-}
-
-static int
-sh_number_of_writes_needed (unsigned char *write_needed, int child_count)
-{
-        int writes = 0;
-        int i      = 0;
-
-        for (i = 0; i < child_count; i++) {
-                if (write_needed[i])
-                        writes++;
-        }
-
-        return writes;
-}
-
-
-static int
-sh_loop_driver_done (call_frame_t *sh_frame, xlator_t *this,
-                     call_frame_t *last_loop_frame)
-{
-        afr_local_t             *local        = NULL;
-        afr_self_heal_t         *sh           = NULL;
-        afr_sh_algo_private_t   *sh_priv      = NULL;
-        int32_t                 total_blocks = 0;
-        int32_t                 diff_blocks  = 0;
-
-        local        = sh_frame->local;
-        sh           = &local->self_heal;
-        sh_priv      = sh->private;
-        if (sh_priv) {
-                total_blocks = sh_priv->total_blocks;
-                diff_blocks  = sh_priv->diff_blocks;
-        }
-
-        sh_private_cleanup (sh_frame, this);
-        if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
-                GF_ASSERT (!last_loop_frame);
-                //loop_finish should have happened and the old_loop should be NULL
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "self-heal aborting on %s",
-                        local->loc.path);
-
-                local->self_heal.algo_abort_cbk (sh_frame, this);
-        } else {
-                GF_ASSERT (last_loop_frame);
-                if (diff_blocks == total_blocks) {
-                        gf_log (this->name, GF_LOG_DEBUG, "full self-heal "
-                                "completed on %s",local->loc.path);
-                } else {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "diff self-heal on %s: completed. "
-                                "(%d blocks of %d were different (%.2f%%))",
-                                local->loc.path, diff_blocks, total_blocks,
-                                ((diff_blocks * 1.0)/total_blocks) * 100);
-                }
-
-                sh->old_loop_frame = last_loop_frame;
-                local->self_heal.algo_completion_cbk (sh_frame, this);
-        }
-
-        return 0;
-}
-
-int
-sh_loop_finish (call_frame_t *loop_frame, xlator_t *this)
-{
-        afr_local_t             *loop_local = NULL;
-        afr_self_heal_t         *loop_sh = NULL;
-
-        if (!loop_frame)
-                goto out;
-
-        loop_local = loop_frame->local;
-        if (loop_local) {
-                loop_sh = &loop_local->self_heal;
-        }
-
-        if (loop_sh && loop_sh->data_lock_held) {
-                afr_sh_data_unlock (loop_frame, this, this->name,
-                                    sh_destroy_frame);
-        } else {
-                sh_destroy_frame (loop_frame, this);
-        }
-out:
-        return 0;
-}
-
-static int
-sh_loop_lock_success (call_frame_t *loop_frame, xlator_t *this)
-{
-        afr_local_t                 *loop_local = NULL;
-        afr_self_heal_t             *loop_sh    = NULL;
-
-        loop_local = loop_frame->local;
-        loop_sh = &loop_local->self_heal;
-
-        sh_loop_finish (loop_sh->old_loop_frame, this);
-        loop_sh->old_loop_frame = NULL;
-
-        gf_log (this->name, GF_LOG_DEBUG, "Acquired lock for range %"PRIu64
-                " %"PRIu64, loop_sh->offset, loop_sh->block_size);
-        loop_sh->data_lock_held = _gf_true;
-        loop_sh->sh_data_algo_start (loop_frame, this);
-        return 0;
-}
-
-static int
-sh_loop_lock_failure (call_frame_t *loop_frame, xlator_t *this)
-{
-        call_frame_t                *sh_frame = NULL;
-        afr_local_t                 *loop_local = NULL;
-        afr_self_heal_t             *loop_sh    = NULL;
-
-        loop_local = loop_frame->local;
-        loop_sh = &loop_local->self_heal;
-        sh_frame = loop_sh->sh_frame;
-
-        gf_log (this->name, GF_LOG_ERROR, "failed lock for range %"PRIu64
-                " %"PRIu64, loop_sh->offset, loop_sh->block_size);
-        sh_loop_finish (loop_sh->old_loop_frame, this);
-        loop_sh->old_loop_frame = NULL;
-        sh_loop_return (sh_frame, this, loop_frame, -1, ENOTCONN);
-        return 0;
-}
-
-static int
-sh_loop_frame_create (call_frame_t *sh_frame, xlator_t *this,
-                      call_frame_t *old_loop_frame, call_frame_t **loop_frame)
-{
-        call_frame_t                *new_loop_frame = NULL;
-        afr_local_t                 *local          = NULL;
-        afr_self_heal_t             *sh             = NULL;
-        afr_local_t                 *new_loop_local = NULL;
-        afr_self_heal_t             *new_loop_sh    = NULL;
-        afr_private_t               *priv           = NULL;
-
-        GF_ASSERT (sh_frame);
-        GF_ASSERT (loop_frame);
-
-        *loop_frame = NULL;
-        local   = sh_frame->local;
-        sh      = &local->self_heal;
-        priv    = this->private;
-
-        new_loop_frame = copy_frame (sh_frame);
-        if (!new_loop_frame)
-                goto out;
-        //We want the frame to have same lk_owner as sh_frame
-        //so that locks translator allows conflicting locks
-        new_loop_local = afr_self_heal_local_init (local, this);
-        if (!new_loop_local)
-                goto out;
-        new_loop_frame->local = new_loop_local;
-
-        new_loop_sh = &new_loop_local->self_heal;
-        new_loop_sh->sources = memdup (sh->sources,
-                                       priv->child_count * sizeof (*sh->sources));
-        if (!new_loop_sh->sources)
-                goto out;
-        new_loop_sh->write_needed = GF_CALLOC (priv->child_count,
-                                               sizeof (*new_loop_sh->write_needed),
-                                               gf_afr_mt_char);
-        if (!new_loop_sh->write_needed)
-                goto out;
-        new_loop_sh->checksum = GF_CALLOC (priv->child_count, MD5_DIGEST_LENGTH,
-                                           gf_afr_mt_uint8_t);
-        if (!new_loop_sh->checksum)
-                goto out;
-        new_loop_sh->inode      = inode_ref (sh->inode);
-        new_loop_sh->sh_data_algo_start = sh->sh_data_algo_start;
-        new_loop_sh->source = sh->source;
-        new_loop_sh->active_sinks = sh->active_sinks;
-        new_loop_sh->healing_fd = fd_ref (sh->healing_fd);
-        new_loop_sh->file_has_holes = sh->file_has_holes;
-        new_loop_sh->old_loop_frame = old_loop_frame;
-        new_loop_sh->sh_frame = sh_frame;
-        *loop_frame = new_loop_frame;
-        return 0;
-out:
-        sh_destroy_frame (new_loop_frame, this);
-        return -ENOMEM;
-}
-
-static int
-sh_loop_start (call_frame_t *sh_frame, xlator_t *this, off_t offset,
-               call_frame_t *old_loop_frame)
-{
-        call_frame_t                *new_loop_frame = NULL;
-        afr_local_t                 *local          = NULL;
-        afr_self_heal_t             *sh             = NULL;
-        afr_local_t                 *new_loop_local = NULL;
-        afr_self_heal_t             *new_loop_sh    = NULL;
-        int                         ret             = 0;
-
-        GF_ASSERT (sh_frame);
-
-        local   = sh_frame->local;
-        sh      = &local->self_heal;
-
-        ret = sh_loop_frame_create (sh_frame, this, old_loop_frame,
-                                    &new_loop_frame);
-        if (ret)
-                goto out;
-        new_loop_local = new_loop_frame->local;
-        new_loop_sh = &new_loop_local->self_heal;
-        new_loop_sh->offset = offset;
-        new_loop_sh->block_size = sh->block_size;
-        afr_sh_data_lock (new_loop_frame, this, offset, new_loop_sh->block_size,
-                          _gf_true, this->name, sh_loop_lock_success, sh_loop_lock_failure);
-        return 0;
-out:
-        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-        if (old_loop_frame)
-                sh_loop_finish (old_loop_frame, this);
-        sh_loop_return (sh_frame, this, new_loop_frame, -1, ENOMEM);
-        return 0;
-}
-
-static int
-sh_loop_driver (call_frame_t *sh_frame, xlator_t *this,
-                gf_boolean_t is_first_call, call_frame_t *old_loop_frame)
-{
-        afr_local_t *               local          = NULL;
-        afr_self_heal_t *           sh             = NULL;
-        afr_sh_algo_private_t       *sh_priv        = NULL;
-        gf_boolean_t                is_driver_done = _gf_false;
-        blksize_t                   block_size     = 0;
-        int                         loop           = 0;
-        off_t                       offset         = 0;
-        afr_private_t               *priv          = NULL;
-
-        priv    = this->private;
-        local   = sh_frame->local;
-        sh      = &local->self_heal;
-        sh_priv = sh->private;
-
-        LOCK (&sh_priv->lock);
-        {
-                if (!is_first_call)
-                        sh_priv->loops_running--;
-                offset = sh_priv->offset;
-                block_size = sh->block_size;
-                while ((!sh->eof_reached) &&
-                       (!is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) &&
-                     (sh_priv->loops_running < priv->data_self_heal_window_size)
-                       && (sh_priv->offset < sh->file_size)) {
-
-                        loop++;
-                        sh_priv->offset += block_size;
-                        sh_priv->loops_running++;
-
-                        if (!is_first_call)
-                                break;
-                }
-                if (0 == sh_priv->loops_running) {
-                        is_driver_done = _gf_true;
-                }
-        }
-        UNLOCK (&sh_priv->lock);
-
-        if (0 == loop) {
-                //loop finish does unlock, but the erasing of the pending
-                //xattrs needs to happen before that so do not finish the loop
-                if (is_driver_done &&
-                    !is_self_heal_failed (sh, AFR_CHECK_SPECIFIC))
-                        goto driver_done;
-                if (old_loop_frame) {
-                        sh_loop_finish (old_loop_frame, this);
-                        old_loop_frame = NULL;
-                }
-        }
-
-        //If we have more loops to form we should finish previous loop after
-        //the next loop lock
-        while (loop--) {
-                if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
-                        // op failed in other loop, stop spawning more loops
-                        if (old_loop_frame) {
-                                sh_loop_finish (old_loop_frame, this);
-                                old_loop_frame = NULL;
-                        }
-                        sh_loop_driver (sh_frame, this, _gf_false, NULL);
-                } else {
-                        gf_log (this->name, GF_LOG_TRACE, "spawning a loop "
-                                "for offset %"PRId64, offset);
-
-                        sh_loop_start (sh_frame, this, offset, old_loop_frame);
-                        old_loop_frame = NULL;
-                        offset += block_size;
-                }
-        }
-
-driver_done:
-        if (is_driver_done) {
-                sh_loop_driver_done (sh_frame, this, old_loop_frame);
-        }
-        return 0;
-}
-
-static int
-sh_loop_return (call_frame_t *sh_frame, xlator_t *this, call_frame_t *loop_frame,
-                int32_t op_ret, int32_t op_errno)
-{
-        afr_local_t *               loop_local = NULL;
-        afr_self_heal_t *           loop_sh    = NULL;
-        afr_local_t *               sh_local = NULL;
-        afr_self_heal_t            *sh       = NULL;
-
-        sh_local = sh_frame->local;
-        sh       = &sh_local->self_heal;
-
-        if (loop_frame) {
-                loop_local = loop_frame->local;
-                if (loop_local)
-                        loop_sh    = &loop_local->self_heal;
-                if (loop_sh)
-                        gf_log (this->name, GF_LOG_TRACE, "loop for offset "
-                                "%"PRId64" returned", loop_sh->offset);
-        }
-
-        if (op_ret == -1) {
-                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-                afr_sh_set_error (sh, op_errno);
-                if (loop_frame) {
-                        sh_loop_finish (loop_frame, this);
-                        loop_frame = NULL;
-                }
-        }
-
-        sh_loop_driver (sh_frame, this, _gf_false, loop_frame);
-
-        return 0;
-}
-
-static int
-sh_loop_write_cbk (call_frame_t *loop_frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct iatt *buf,
-                   struct iatt *postbuf, dict_t *xdata)
-{
-        afr_private_t *             priv        = NULL;
-        afr_local_t *               loop_local    = NULL;
-        afr_self_heal_t *           loop_sh       = NULL;
-        call_frame_t               *sh_frame    = NULL;
-        afr_local_t *               sh_local    = NULL;
-        afr_self_heal_t            *sh          = NULL;
-        int                         call_count  = 0;
-        int                         child_index = 0;
-
-        priv     = this->private;
-        loop_local = loop_frame->local;
-        loop_sh    = &loop_local->self_heal;
-
-        sh_frame = loop_sh->sh_frame;
-        sh_local = sh_frame->local;
-        sh       = &sh_local->self_heal;
-
-        child_index =  (long) cookie;
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "wrote %d bytes of data from %s to child %d, offset %"PRId64"",
-                op_ret, sh_local->loc.path, child_index, loop_sh->offset);
-
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "write to %s failed on subvolume %s (%s)",
-                        sh_local->loc.path,
-                        priv->children[child_index]->name,
-                        strerror (op_errno));
-
-                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-                afr_sh_set_error (loop_sh, op_errno);
-        } else if (op_ret < loop_local->cont.writev.vector->iov_len) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "incomplete write to %s on subvolume %s "
-                        "(expected %lu, returned %d)", sh_local->loc.path,
-                        priv->children[child_index]->name,
-                        loop_local->cont.writev.vector->iov_len, op_ret);
-                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-        }
-
-        call_count = afr_frame_return (loop_frame);
-
-        if (call_count == 0) {
-		iobref_unref(loop_local->cont.writev.iobref);
-
-                sh_loop_return (sh_frame, this, loop_frame,
-                                loop_sh->op_ret, loop_sh->op_errno);
-        }
-
-        return 0;
-}
-
-static void
-sh_prune_writes_needed (call_frame_t *sh_frame, call_frame_t *loop_frame,
-                        afr_private_t *priv)
-{
-        afr_local_t     *sh_local     = NULL;
-        afr_self_heal_t *sh           = NULL;
-        afr_local_t     *loop_local   = NULL;
-        afr_self_heal_t *loop_sh      = NULL;
-        int             i             = 0;
-
-        sh_local   = sh_frame->local;
-        sh         = &sh_local->self_heal;
-
-        if (!strcmp (sh->algo->name, "diff"))
-                return;
-
-        loop_local = loop_frame->local;
-        loop_sh    = &loop_local->self_heal;
-
-        /* full self-heal guarantees there exists atleast 1 file with size 0
-         * That means for other files we can preserve holes that come after
-         * its size before 'trim'
-         */
-        for (i = 0; i < priv->child_count; i++) {
-                if (loop_sh->write_needed[i] &&
-                    ((loop_sh->offset + 1) > sh->buf[i].ia_size))
-                        loop_sh->write_needed[i] = 0;
-        }
-}
-
-static int
-sh_loop_read_cbk (call_frame_t *loop_frame, void *cookie,
-                  xlator_t *this, int32_t op_ret, int32_t op_errno,
-                  struct iovec *vector, int32_t count, struct iatt *buf,
-                  struct iobref *iobref, dict_t *xdata)
-{
-        afr_private_t *               priv       = NULL;
-        afr_local_t *                 loop_local   = NULL;
-        afr_self_heal_t *             loop_sh      = NULL;
-        call_frame_t                 *sh_frame   = NULL;
-        int                           i          = 0;
-        int                           call_count = 0;
-        afr_local_t *                 sh_local   = NULL;
-        afr_self_heal_t *             sh      = NULL;
-
-        priv       = this->private;
-        loop_local = loop_frame->local;
-        loop_sh    = &loop_local->self_heal;
-
-        sh_frame = loop_sh->sh_frame;
-        sh_local = sh_frame->local;
-        sh       = &sh_local->self_heal;
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "read %d bytes of data from %s, offset %"PRId64"",
-                op_ret, loop_local->loc.path, loop_sh->offset);
-
-        if (op_ret <= 0) {
-                if (op_ret < 0) {
-                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-                        gf_log (this->name, GF_LOG_ERROR, "read failed on %d "
-                                "for %s reason :%s", sh->source,
-                                sh_local->loc.path, strerror (errno));
-                } else {
-                        sh->eof_reached = _gf_true;
-                        gf_log (this->name, GF_LOG_DEBUG, "Eof reached for %s",
-                                sh_local->loc.path);
-                }
-                sh_loop_return (sh_frame, this, loop_frame, op_ret, op_errno);
-                goto out;
-        }
-
-        if (loop_sh->file_has_holes && iov_0filled (vector, count) == 0)
-                sh_prune_writes_needed (sh_frame, loop_frame, priv);
-
-        call_count = sh_number_of_writes_needed (loop_sh->write_needed,
-                                                 priv->child_count);
-        if (call_count == 0) {
-                sh_loop_return (sh_frame, this, loop_frame, 0, 0);
-                goto out;
-        }
-
-        loop_local->call_count = call_count;
-
-	/*
-	 * We only really need the request size at the moment, but the buffer
-	 * is required if we want to issue a retry in the event of a short write.
-	 * Therefore, we duplicate the vector and ref the iobref here...
-	 */
-	loop_local->cont.writev.vector = iov_dup(vector, count);
-	loop_local->cont.writev.iobref = iobref_ref(iobref);
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (!loop_sh->write_needed[i])
-                        continue;
-                STACK_WIND_COOKIE (loop_frame, sh_loop_write_cbk,
-                                   (void *) (long) i,
-                                   priv->children[i],
-                                   priv->children[i]->fops->writev,
-                                   loop_sh->healing_fd, vector, count,
-                                   loop_sh->offset, 0, iobref, NULL);
-
-                if (!--call_count)
-                        break;
-        }
-
-out:
-        return 0;
-}
-
-
-static int
-sh_loop_read (call_frame_t *loop_frame, xlator_t *this)
-{
-        afr_private_t           *priv       = NULL;
-        afr_local_t             *loop_local   = NULL;
-        afr_self_heal_t         *loop_sh      = NULL;
-
-        priv     = this->private;
-        loop_local = loop_frame->local;
-        loop_sh    = &loop_local->self_heal;
-
-        STACK_WIND_COOKIE (loop_frame, sh_loop_read_cbk,
-                           (void *) (long) loop_sh->source,
-                           priv->children[loop_sh->source],
-                           priv->children[loop_sh->source]->fops->readv,
-                           loop_sh->healing_fd, loop_sh->block_size,
-                           loop_sh->offset, 0, NULL);
-
-        return 0;
-}
-
-
-static int
-sh_diff_checksum_cbk (call_frame_t *loop_frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno,
-                      uint32_t weak_checksum, uint8_t *strong_checksum,
-                      dict_t *xdata)
-{
-        afr_private_t                 *priv         = NULL;
-        afr_local_t                   *loop_local   = NULL;
-        afr_self_heal_t               *loop_sh      = NULL;
-        call_frame_t                  *sh_frame     = NULL;
-        afr_local_t                   *sh_local     = NULL;
-        afr_self_heal_t               *sh           = NULL;
-        afr_sh_algo_private_t         *sh_priv      = NULL;
-        int                           child_index  = 0;
-        int                           call_count   = 0;
-        int                           i            = 0;
-        int                           write_needed = 0;
-
-        priv  = this->private;
-
-        loop_local = loop_frame->local;
-        loop_sh    = &loop_local->self_heal;
-
-        sh_frame = loop_sh->sh_frame;
-        sh_local = sh_frame->local;
-        sh       = &sh_local->self_heal;
-
-        sh_priv = sh->private;
-
-        child_index = (long) cookie;
-
-        if (op_ret < 0) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "checksum on %s failed on subvolume %s (%s)",
-                        sh_local->loc.path, priv->children[child_index]->name,
-                        strerror (op_errno));
-                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-        } else {
-                memcpy (loop_sh->checksum + child_index * MD5_DIGEST_LENGTH,
-                        strong_checksum, MD5_DIGEST_LENGTH);
-        }
-
-        call_count = afr_frame_return (loop_frame);
-
-        if (call_count == 0) {
-                for (i = 0; i < priv->child_count; i++) {
-                        if (sh->sources[i] || !sh_local->child_up[i])
-                                continue;
-
-                        if (memcmp (loop_sh->checksum + (i * MD5_DIGEST_LENGTH),
-                                    loop_sh->checksum + (sh->source * MD5_DIGEST_LENGTH),
-                                    MD5_DIGEST_LENGTH)) {
-                                /*
-                                  Checksums differ, so this block
-                                  must be written to this sink
-                                */
-
-                                gf_log (this->name, GF_LOG_DEBUG,
-                                        "checksum on subvolume %s at offset %"
-                                        PRId64" differs from that on source",
-                                        priv->children[i]->name, loop_sh->offset);
-
-                                write_needed = loop_sh->write_needed[i] = 1;
-                        }
-                }
-
-                LOCK (&sh_priv->lock);
-                {
-                        sh_priv->total_blocks++;
-                        if (write_needed)
-                                sh_priv->diff_blocks++;
-                }
-                UNLOCK (&sh_priv->lock);
-
-                if (write_needed &&
-                    !is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
-                        sh_loop_read (loop_frame, this);
-                } else {
-                        sh_loop_return (sh_frame, this, loop_frame,
-                                        op_ret, op_errno);
-                }
-        }
-
-        return 0;
-}
-
-static int
-sh_diff_checksum (call_frame_t *loop_frame, xlator_t *this)
-{
-        afr_private_t           *priv         = NULL;
-        afr_local_t             *loop_local   = NULL;
-        afr_self_heal_t         *loop_sh      = NULL;
-        int                     call_count    = 0;
-        int                     i             = 0;
-
-        priv         = this->private;
-        loop_local   = loop_frame->local;
-        loop_sh      = &loop_local->self_heal;
-
-        call_count = loop_sh->active_sinks + 1;  /* sinks and source */
-
-        loop_local->call_count = call_count;
-
-        STACK_WIND_COOKIE (loop_frame, sh_diff_checksum_cbk,
-                           (void *) (long) loop_sh->source,
-                           priv->children[loop_sh->source],
-                           priv->children[loop_sh->source]->fops->rchecksum,
-                           loop_sh->healing_fd,
-                           loop_sh->offset, loop_sh->block_size, NULL);
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (loop_sh->sources[i] || !loop_local->child_up[i])
-                        continue;
-
-                STACK_WIND_COOKIE (loop_frame, sh_diff_checksum_cbk,
-                                   (void *) (long) i,
-                                   priv->children[i],
-                                   priv->children[i]->fops->rchecksum,
-                                   loop_sh->healing_fd,
-                                   loop_sh->offset, loop_sh->block_size, NULL);
-
-                if (!--call_count)
-                        break;
-        }
-
-        return 0;
-}
-
-static int
-sh_full_read_write_to_sinks (call_frame_t *loop_frame, xlator_t *this)
-{
-        afr_private_t           *priv         = NULL;
-        afr_local_t             *loop_local   = NULL;
-        afr_self_heal_t         *loop_sh      = NULL;
-        int                     i             = 0;
-
-        priv         = this->private;
-        loop_local   = loop_frame->local;
-        loop_sh      = &loop_local->self_heal;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (loop_sh->sources[i] || !loop_local->child_up[i])
-                        continue;
-                loop_sh->write_needed[i] = 1;
-        }
-        sh_loop_read (loop_frame, this);
-        return 0;
-}
-
-afr_sh_algo_private_t*
-afr_sh_priv_init ()
-{
-        afr_sh_algo_private_t   *sh_priv = NULL;
-
-        sh_priv = GF_CALLOC (1, sizeof (*sh_priv),
-                             gf_afr_mt_afr_private_t);
-        if (!sh_priv)
-                goto out;
-
-        LOCK_INIT (&sh_priv->lock);
-out:
-        return sh_priv;
-}
-
-int
-afr_sh_transfer_lock (call_frame_t *dst, call_frame_t *src, char *dom,
-                      unsigned int child_count)
-{
-        afr_local_t             *dst_local   = NULL;
-        afr_self_heal_t         *dst_sh      = NULL;
-        afr_local_t             *src_local   = NULL;
-        afr_self_heal_t         *src_sh      = NULL;
-        int                     ret          = -1;
-
-        dst_local = dst->local;
-        dst_sh = &dst_local->self_heal;
-        src_local = src->local;
-        src_sh = &src_local->self_heal;
-        GF_ASSERT (src_sh->data_lock_held);
-        GF_ASSERT (!dst_sh->data_lock_held);
-        ret = afr_lk_transfer_datalock (dst, src, dom, child_count);
-        if (ret)
-                return ret;
-        src_sh->data_lock_held = _gf_false;
-        dst_sh->data_lock_held = _gf_true;
-        return 0;
-}
-
-int
-afr_sh_start_loops (call_frame_t *sh_frame, xlator_t *this,
-                    afr_sh_algo_fn sh_data_algo_start)
-{
-        call_frame_t            *first_loop_frame = NULL;
-        afr_local_t             *local   = NULL;
-        afr_self_heal_t         *sh      = NULL;
-        int                     ret      = 0;
-        afr_private_t           *priv    = NULL;
-
-        local = sh_frame->local;
-        sh    = &local->self_heal;
-        priv  = this->private;
-
-        sh->sh_data_algo_start = sh_data_algo_start;
-        local->call_count = 0;
-        ret = sh_loop_frame_create (sh_frame, this, NULL, &first_loop_frame);
-        if (ret)
-                goto out;
-        ret = afr_sh_transfer_lock (first_loop_frame, sh_frame, this->name,
-                                    priv->child_count);
-        if (ret)
-                goto out;
-        sh->private = afr_sh_priv_init ();
-        if (!sh->private) {
-                ret = -1;
-                goto out;
-        }
-        sh_loop_driver (sh_frame, this, _gf_true, first_loop_frame);
-        ret = 0;
-out:
-        if (ret) {
-                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-                sh_loop_driver_done (sh_frame, this, NULL);
-        }
-        return 0;
-}
-
-int
-afr_sh_algo_diff (call_frame_t *sh_frame, xlator_t *this)
-{
-        afr_sh_start_loops (sh_frame, this, sh_diff_checksum);
-        return 0;
-}
-
-int
-afr_sh_algo_full (call_frame_t *sh_frame, xlator_t *this)
-{
-        afr_sh_start_loops (sh_frame, this, sh_full_read_write_to_sinks);
-        return 0;
-}
-
-struct afr_sh_algorithm afr_self_heal_algorithms[] = {
-        {.name = "full",  .fn = afr_sh_algo_full},
-        {.name = "diff",  .fn = afr_sh_algo_diff},
-        {0, 0},
-};
diff --git a/xlators/cluster/afr/src/afr-self-heal-algorithm.h b/xlators/cluster/afr/src/afr-self-heal-algorithm.h
deleted file mode 100644
index 6b20789b1bb..00000000000
--- a/xlators/cluster/afr/src/afr-self-heal-algorithm.h
+++ /dev/null
@@ -1,32 +0,0 @@
-/*
-  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
-  This file is part of GlusterFS.
-
-  This file is licensed to you under your choice of the GNU Lesser
-  General Public License, version 3 or any later version (LGPLv3 or
-  later), or the GNU General Public License, version 2 (GPLv2), in all
-  cases as published by the Free Software Foundation.
-*/
-
-#ifndef __AFR_SELF_HEAL_ALGORITHM_H__
-#define __AFR_SELF_HEAL_ALGORITHM_H__
-
-typedef int (*afr_sh_algo_fn) (call_frame_t *frame,
-                               xlator_t *this);
-
-struct afr_sh_algorithm {
-        const char *name;
-        afr_sh_algo_fn fn;
-};
-
-extern struct afr_sh_algorithm afr_self_heal_algorithms[3];
-typedef struct {
-        gf_lock_t lock;
-        unsigned int loops_running;
-        off_t offset;
-
-        int32_t total_blocks;
-        int32_t diff_blocks;
-} afr_sh_algo_private_t;
-
-#endif /* __AFR_SELF_HEAL_ALGORITHM_H__ */
diff --git a/xlators/cluster/afr/src/afr-self-heal-common.c b/xlators/cluster/afr/src/afr-self-heal-common.c
index ef92b420551..a580a1584cc 100644
--- a/xlators/cluster/afr/src/afr-self-heal-common.c
+++ b/xlators/cluster/afr/src/afr-self-heal-common.c
@@ -1,5 +1,5 @@
 /*
-  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
   This file is licensed to you under your choice of the GNU Lesser
@@ -8,2805 +8,2927 @@
   cases as published by the Free Software Foundation.
 */
 
-#include "glusterfs.h"
-#include "xlator.h"
-#include "byte-order.h"
-
 #include "afr.h"
-#include "afr-transaction.h"
-#include "afr-self-heal-common.h"
 #include "afr-self-heal.h"
-#include "pump.h"
-
-#define ADD_FMT_STRING(msg, off, sh_str, status, print_log)                 \
-        do {                                                                \
-                if (AFR_SELF_HEAL_NOT_ATTEMPTED != status) {                \
-                        off += snprintf (msg + off, sizeof (msg) - off,     \
-                                         " "sh_str" self heal %s,",         \
-                                         get_sh_completion_status (status));\
-                        print_log = 1;                                      \
-                }                                                           \
-        } while (0)
-
-#define ADD_FMT_STRING_SYNC(msg, off, sh_str, status, print_log)            \
-        do {                                                                \
-                if (AFR_SELF_HEAL_SYNC_BEGIN == status ||                   \
-                    AFR_SELF_HEAL_FAILED == status)  {                      \
-                        off += snprintf (msg + off, sizeof (msg) - off,     \
-                                         " "sh_str" self heal %s,",         \
-                                         get_sh_completion_status (status));\
-                        print_log = 1;                                      \
-                }                                                           \
-        } while (0)
-
+#include <glusterfs/byte-order.h>
+#include "protocol-common.h"
+#include "afr-messages.h"
+#include <glusterfs/events.h>
 
 void
-afr_sh_reset (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
+afr_heal_synctask(xlator_t *this, afr_local_t *local);
 
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
+int
+afr_lookup_and_heal_gfid(xlator_t *this, inode_t *parent, const char *name,
+                         inode_t *inode, struct afr_reply *replies, int source,
+                         unsigned char *sources, void *gfid, int *gfid_idx)
+{
+    afr_private_t *priv = NULL;
+    call_frame_t *frame = NULL;
+    afr_local_t *local = NULL;
+    unsigned char *wind_on = NULL;
+    ia_type_t ia_type = IA_INVAL;
+    dict_t *xdata = NULL;
+    loc_t loc = {
+        0,
+    };
+    int ret = 0;
+    int i = 0;
+
+    priv = this->private;
+    wind_on = alloca0(priv->child_count);
+    if (source >= 0 && replies[source].valid && replies[source].op_ret == 0)
+        ia_type = replies[source].poststat.ia_type;
+
+    if (ia_type != IA_INVAL)
+        goto heal;
+
+    /* If ia_type is still invalid, it means either
+     * (a)'source' was -1, i.e. parent dir pending xattrs are in split-brain
+     * (or) (b) The parent dir pending xattrs are all zeroes (i.e. all bricks
+     * are sources) and the 'source' we selected earlier might be the one where
+     * the file is not actually present.
+     *
+     * In both cases, let us pick a brick with a successful reply and use its
+     * ia_type.
+     * */
+    for (i = 0; i < priv->child_count; i++) {
+        if (source == -1) {
+            /* case (a) above. */
+            if (replies[i].valid && replies[i].op_ret == 0 &&
+                replies[i].poststat.ia_type != IA_INVAL) {
+                ia_type = replies[i].poststat.ia_type;
+                break;
+            }
+        } else {
+            /* case (b) above. */
+            if (i == source)
+                continue;
+            if (sources[i] && replies[i].valid && replies[i].op_ret == 0 &&
+                replies[i].poststat.ia_type != IA_INVAL) {
+                ia_type = replies[i].poststat.ia_type;
+                break;
+            }
+        }
+    }
 
-        memset (sh->child_errno, 0,
-                sizeof (*sh->child_errno) * priv->child_count);
-        memset (sh->buf, 0, sizeof (*sh->buf) * priv->child_count);
-        memset (sh->parentbufs, 0,
-                sizeof (*sh->parentbufs) * priv->child_count);
-        memset (sh->success, 0, sizeof (*sh->success) * priv->child_count);
-        memset (sh->locked_nodes, 0,
-                sizeof (*sh->locked_nodes) * priv->child_count);
-        sh->active_sinks = 0;
+heal:
+    /* gfid heal on those subvolumes that do not have gfid associated
+     * with the inode and update those replies.
+     */
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret != 0)
+            continue;
 
-        afr_reset_xattr (sh->xattr, priv->child_count);
-}
+        if (gf_uuid_is_null(gfid) &&
+            !gf_uuid_is_null(replies[i].poststat.ia_gfid) &&
+            replies[i].poststat.ia_type == ia_type)
+            gfid = replies[i].poststat.ia_gfid;
 
-//Intersection[child]=1 if child is part of intersection
-void
-afr_children_intersection_get (int32_t *set1, int32_t *set2,
-                               int *intersection, unsigned int child_count)
-{
-        int                      i = 0;
+        if (!gf_uuid_is_null(replies[i].poststat.ia_gfid) ||
+            replies[i].poststat.ia_type != ia_type)
+            continue;
+
+        wind_on[i] = 1;
+    }
 
-        memset (intersection, 0, sizeof (*intersection) * child_count);
-        for (i = 0; i < child_count; i++) {
-                intersection[i] = afr_is_child_present (set1, child_count, i)
-                                     && afr_is_child_present (set2, child_count,
-                                                              i);
+    if (AFR_COUNT(wind_on, priv->child_count) == 0)
+        return 0;
+
+    xdata = dict_new();
+    if (!xdata) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    ret = dict_set_gfuuid(xdata, "gfid-req", gfid, true);
+    if (ret) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    frame = afr_frame_create(this, &ret);
+    if (!frame) {
+        ret = -ret;
+        goto out;
+    }
+
+    local = frame->local;
+    loc.parent = inode_ref(parent);
+    gf_uuid_copy(loc.pargfid, parent->gfid);
+    loc.name = name;
+    loc.inode = inode_ref(inode);
+
+    AFR_ONLIST(wind_on, frame, afr_selfheal_discover_cbk, lookup, &loc, xdata);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!wind_on[i])
+            continue;
+        afr_reply_wipe(&replies[i]);
+        afr_reply_copy(&replies[i], &local->replies[i]);
+    }
+    if (gfid_idx && (*gfid_idx == -1)) {
+        /*Pick a brick where the gifd heal was successful.*/
+        for (i = 0; i < priv->child_count; i++) {
+            if (!wind_on[i])
+                continue;
+            if (replies[i].valid && replies[i].op_ret == 0 &&
+                !gf_uuid_is_null(replies[i].poststat.ia_gfid)) {
+                *gfid_idx = i;
+                break;
+            }
         }
-}
+    }
+out:
+    if (gfid_idx && (*gfid_idx == -1) && (ret == 0) && local) {
+        ret = -afr_final_errno(local, priv);
+    }
+    loc_wipe(&loc);
+    if (frame)
+        AFR_STACK_DESTROY(frame);
+    if (xdata)
+        dict_unref(xdata);
 
-/**
- * select_source - select a source and return it
- */
+    return ret;
+}
 
 int
-afr_sh_select_source (int sources[], int child_count)
+afr_gfid_sbrain_source_from_src_brick(xlator_t *this, struct afr_reply *replies,
+                                      char *src_brick)
 {
-        int i = 0;
-        for (i = 0; i < child_count; i++)
-                if (sources[i])
-                        return i;
+    int i = 0;
+    afr_private_t *priv = NULL;
 
-        return -1;
+    priv = this->private;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret == -1)
+            continue;
+        if (strcmp(priv->children[i]->name, src_brick) == 0)
+            return i;
+    }
+    return -1;
 }
 
-void
-afr_sh_mark_source_sinks (call_frame_t *frame, xlator_t *this)
+int
+afr_selfheal_gfid_mismatch_by_majority(struct afr_reply *replies,
+                                       int child_count)
 {
-        int              i = 0;
-        afr_local_t     *local      = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int              active_sinks = 0;
+    int j = 0;
+    int i = 0;
+    int votes;
 
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
+    for (i = 0; i < child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret == -1)
+            continue;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (sh->sources[i] == 0 && local->child_up[i] == 1) {
-                        active_sinks++;
-                        sh->success[i] = 1;
-                } else if (sh->sources[i] == 1 && local->child_up[i] == 1) {
-                        sh->success[i] = 1;
-                }
+        votes = 1;
+        for (j = i + 1; j < child_count; j++) {
+            if ((!gf_uuid_compare(replies[i].poststat.ia_gfid,
+                                  replies[j].poststat.ia_gfid)))
+                votes++;
+            if (votes > child_count / 2)
+                return i;
         }
-        sh->active_sinks = active_sinks;
+    }
+
+    return -1;
 }
 
 int
-afr_sh_source_count (int sources[], int child_count)
+afr_gfid_sbrain_source_from_bigger_file(struct afr_reply *replies,
+                                        int child_count)
 {
-        int i = 0;
-        int nsource = 0;
+    int i = 0;
+    int src = -1;
+    uint64_t size = 0;
 
-        for (i = 0; i < child_count; i++)
-                if (sources[i])
-                        nsource++;
-        return nsource;
+    for (i = 0; i < child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret == -1)
+            continue;
+        if (size < replies[i].poststat.ia_size) {
+            src = i;
+            size = replies[i].poststat.ia_size;
+        } else if (replies[i].poststat.ia_size == size) {
+            src = -1;
+        }
+    }
+    return src;
 }
 
-void
-afr_sh_set_error (afr_self_heal_t *sh, int32_t op_errno)
-{
-        sh->op_ret = -1;
-	sh->op_errno = afr_most_important_error(sh->op_errno, op_errno,
-						_gf_false);
+int
+afr_gfid_sbrain_source_from_latest_mtime(struct afr_reply *replies,
+                                         int child_count)
+{
+    int i = 0;
+    int src = -1;
+    uint32_t mtime = 0;
+    uint32_t mtime_nsec = 0;
+
+    for (i = 0; i < child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret != 0)
+            continue;
+        if ((mtime < replies[i].poststat.ia_mtime) ||
+            ((mtime == replies[i].poststat.ia_mtime) &&
+             (mtime_nsec < replies[i].poststat.ia_mtime_nsec))) {
+            src = i;
+            mtime = replies[i].poststat.ia_mtime;
+            mtime_nsec = replies[i].poststat.ia_mtime_nsec;
+        } else if ((mtime == replies[i].poststat.ia_mtime) &&
+                   (mtime_nsec == replies[i].poststat.ia_mtime_nsec)) {
+            src = -1;
+        }
+    }
+    return src;
 }
 
-void
-afr_sh_print_pending_matrix (int32_t *pending_matrix[], xlator_t *this)
-{
-        afr_private_t *  priv = this->private;
-        char            *buf  = NULL;
-        char            *ptr  = NULL;
-        int              i    = 0;
-        int              j    = 0;
-
-        /* 10 digits per entry + 1 space + '[' and ']' */
-        buf = GF_MALLOC (priv->child_count * 11 + 8, gf_afr_mt_char);
-
-        for (i = 0; i < priv->child_count; i++) {
-                ptr = buf;
-                ptr += sprintf (ptr, "[ ");
-                for (j = 0; j < priv->child_count; j++) {
-                        ptr += sprintf (ptr, "%d ", pending_matrix[i][j]);
+int
+afr_gfid_split_brain_source(xlator_t *this, struct afr_reply *replies,
+                            inode_t *inode, uuid_t pargfid, const char *bname,
+                            int src_idx, int child_idx,
+                            unsigned char *locked_on, int *src, dict_t *xdata)
+{
+    afr_private_t *priv = NULL;
+    char g1[64] = {
+        0,
+    };
+    char g2[64] = {
+        0,
+    };
+    int up_count = 0;
+    int heal_op = -1;
+    int ret = -1;
+    char *src_brick = NULL;
+
+    *src = -1;
+    priv = this->private;
+    up_count = AFR_COUNT(locked_on, priv->child_count);
+    if (up_count != priv->child_count) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN,
+               "All the bricks should be up to resolve the gfid split "
+               "barin");
+        if (xdata) {
+            ret = dict_set_sizen_str_sizen(xdata, "gfid-heal-msg",
+                                           SALL_BRICKS_UP_TO_RESOLVE);
+            if (ret)
+                gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_DICT_SET_FAILED,
+                       "Error setting"
+                       " gfid-heal-msg dict");
+        }
+        goto out;
+    }
+
+    if (xdata) {
+        ret = dict_get_int32_sizen(xdata, "heal-op", &heal_op);
+        if (ret)
+            goto fav_child;
+    } else {
+        goto fav_child;
+    }
+
+    switch (heal_op) {
+        case GF_SHD_OP_SBRAIN_HEAL_FROM_BIGGER_FILE:
+            *src = afr_gfid_sbrain_source_from_bigger_file(replies,
+                                                           priv->child_count);
+            if (*src == -1) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN,
+                       SNO_BIGGER_FILE);
+                if (xdata) {
+                    ret = dict_set_sizen_str_sizen(xdata, "gfid-heal-msg",
+                                                   SNO_BIGGER_FILE);
+                    if (ret)
+                        gf_msg(this->name, GF_LOG_ERROR, 0,
+                               AFR_MSG_DICT_SET_FAILED,
+                               "Error"
+                               " setting gfid-heal-msg dict");
                 }
-                sprintf (ptr, "]");
-                gf_log (this->name, GF_LOG_DEBUG, "pending_matrix: %s", buf);
-        }
-
-        GF_FREE (buf);
-}
-
-char*
-afr_get_pending_matrix_str (int32_t *pending_matrix[], xlator_t *this)
-{
-        afr_private_t *  priv = this->private;
-        char            *buf  = NULL;
-        char            *ptr  = NULL;
-        int              i    = 0;
-        int              j    = 0;
-        int             child_count = priv->child_count;
-        char            *matrix_begin = "[ [ ";
-        char            *matrix_end = "] ]";
-        char            *seperator = "] [ ";
-        int             pending_entry_strlen = 12; //Including space after entry
-        int             matrix_begin_strlen = 0;
-        int             matrix_end_strlen = 0;
-        int             seperator_strlen = 0;
-        int             string_length = 0;
-        char            *msg = "- Pending matrix:  ";
-
-        /*
-         *  for a list of lists of [ [ a b ] [ c d ] ]
-         * */
-
-        matrix_begin_strlen = strlen (matrix_begin);
-        matrix_end_strlen = strlen (matrix_end);
-        seperator_strlen = strlen (seperator);
-        string_length = matrix_begin_strlen + matrix_end_strlen
-                        + (child_count -1) * seperator_strlen
-                        + (child_count * child_count * pending_entry_strlen);
-
-        buf = GF_CALLOC (1, 1 + strlen (msg) + string_length , gf_afr_mt_char);
-        if (!buf)
-                goto out;
-
-        ptr = buf;
-        ptr += sprintf (ptr, "%s", msg);
-        ptr += sprintf (ptr, "%s", matrix_begin);
-        for (i = 0; i < priv->child_count; i++) {
-                for (j = 0; j < priv->child_count; j++) {
-                        ptr += sprintf (ptr, "%d ", pending_matrix[i][j]);
+            }
+            break;
+
+        case GF_SHD_OP_SBRAIN_HEAL_FROM_LATEST_MTIME:
+            *src = afr_gfid_sbrain_source_from_latest_mtime(replies,
+                                                            priv->child_count);
+            if (*src == -1) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN,
+                       SNO_DIFF_IN_MTIME);
+                if (xdata) {
+                    ret = dict_set_sizen_str_sizen(xdata, "gfid-heal-msg",
+                                                   SNO_DIFF_IN_MTIME);
+                    if (ret)
+                        gf_msg(this->name, GF_LOG_ERROR, 0,
+                               AFR_MSG_DICT_SET_FAILED,
+                               "Error"
+                               "setting gfid-heal-msg dict");
                 }
-                if (i < priv->child_count -1)
-                        ptr += sprintf (ptr, "%s", seperator);
-        }
+            }
+            break;
+
+        case GF_SHD_OP_SBRAIN_HEAL_FROM_BRICK:
+            ret = dict_get_str_sizen(xdata, "child-name", &src_brick);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN,
+                       "Error getting the source "
+                       "brick");
+                break;
+            }
+            *src = afr_gfid_sbrain_source_from_src_brick(this, replies,
+                                                         src_brick);
+            if (*src == -1) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN,
+                       SERROR_GETTING_SRC_BRICK);
+                if (xdata) {
+                    ret = dict_set_sizen_str_sizen(xdata, "gfid-heal-msg",
+                                                   SERROR_GETTING_SRC_BRICK);
+                    if (ret)
+                        gf_msg(this->name, GF_LOG_ERROR, 0,
+                               AFR_MSG_DICT_SET_FAILED,
+                               "Error"
+                               " setting gfid-heal-msg dict");
+                }
+            }
+            break;
 
-        ptr += sprintf (ptr, "%s", matrix_end);
+        default:
+            break;
+    }
+    goto out;
+
+fav_child:
+    switch (priv->fav_child_policy) {
+        case AFR_FAV_CHILD_BY_SIZE:
+            *src = afr_sh_fav_by_size(this, replies, inode);
+            break;
+        case AFR_FAV_CHILD_BY_MTIME:
+            *src = afr_sh_fav_by_mtime(this, replies, inode);
+            break;
+        case AFR_FAV_CHILD_BY_CTIME:
+            *src = afr_sh_fav_by_ctime(this, replies, inode);
+            break;
+        case AFR_FAV_CHILD_BY_MAJORITY:
+            if (priv->child_count != 2)
+                *src = afr_selfheal_gfid_mismatch_by_majority(
+                    replies, priv->child_count);
+            else
+                *src = -1;
+
+            if (*src == -1) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN,
+                       "No majority to resolve "
+                       "gfid split brain");
+            }
+            break;
+        default:
+            break;
+    }
 
 out:
-        return buf;
+    if (*src == -1) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN,
+               "Gfid mismatch detected for <gfid:%s>/%s>, %s on %s and"
+               " %s on %s.",
+               uuid_utoa(pargfid), bname,
+               uuid_utoa_r(replies[child_idx].poststat.ia_gfid, g1),
+               priv->children[child_idx]->name,
+               uuid_utoa_r(replies[src_idx].poststat.ia_gfid, g2),
+               priv->children[src_idx]->name);
+        gf_event(EVENT_AFR_SPLIT_BRAIN,
+                 "client-pid=%d;"
+                 "subvol=%s;type=gfid;file="
+                 "<gfid:%s>/%s>;count=2;child-%d=%s;gfid-%d=%s;"
+                 "child-%d=%s;gfid-%d=%s",
+                 this->ctx->cmd_args.client_pid, this->name, uuid_utoa(pargfid),
+                 bname, child_idx, priv->children[child_idx]->name, child_idx,
+                 uuid_utoa_r(replies[child_idx].poststat.ia_gfid, g1), src_idx,
+                 priv->children[src_idx]->name, src_idx,
+                 uuid_utoa_r(replies[src_idx].poststat.ia_gfid, g2));
+        return -1;
+    }
+    return 0;
 }
 
-void
-afr_sh_print_split_brain_log (int32_t *pending_matrix[], xlator_t *this,
-                              const char *loc)
+int
+afr_selfheal_post_op_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
 {
-        char *buf      = NULL;
-        char *free_ptr = NULL;
+    afr_local_t *local = NULL;
 
-        buf = afr_get_pending_matrix_str (pending_matrix, this);
-        if (buf)
-                free_ptr = buf;
-        else
-                buf = "";
+    local = frame->local;
 
+    local->op_ret = op_ret;
+    local->op_errno = op_errno;
+    syncbarrier_wake(&local->barrier);
 
-        gf_log (this->name, GF_LOG_ERROR, "Unable to self-heal contents of '%s'"
-                " (possible split-brain). Please delete the file from all but "
-                "the preferred subvolume.%s", loc, buf);
-        GF_FREE (free_ptr);
-        return;
+    return 0;
 }
 
-
-void
-afr_init_pending_matrix (int32_t **pending_matrix, size_t child_count)
+int
+afr_selfheal_post_op(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                     int subvol, dict_t *xattr, dict_t *xdata)
 {
-        int             i   = 0;
-        int             j   = 0;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    loc_t loc = {
+        0,
+    };
+    int ret = 0;
 
-        GF_ASSERT (pending_matrix);
+    priv = this->private;
+    local = frame->local;
 
-        for (i = 0; i < child_count; i++) {
-                for (j = 0; j < child_count; j++) {
-                        pending_matrix[i][j] = 0;
-                }
-        }
-}
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-void
-afr_mark_ignorant_subvols_as_pending (int32_t **pending_matrix,
-                                      unsigned char *ignorant_subvols,
-                                      size_t  child_count)
-{
-        int            i                = 0;
-        int            j                = 0;
-
-        GF_ASSERT (pending_matrix);
-        GF_ASSERT (ignorant_subvols);
-
-        for (i = 0; i < child_count; i++) {
-                if (ignorant_subvols[i]) {
-                        for (j = 0; j < child_count; j++) {
-                                if (!ignorant_subvols[j])
-                                        pending_matrix[j][i] += 1;
-                        }
-                }
-        }
-}
+    local->op_ret = 0;
 
-int
-afr_build_pending_matrix (char **pending_key, int32_t **pending_matrix,
-                          unsigned char *ignorant_subvols,
-                          dict_t *xattr[], afr_transaction_type type,
-                          size_t child_count)
-{
-        /* Indexable by result of afr_index_for_transaction_type(): 0 -- 2. */
-        int32_t        pending[3]       = {0,};
-        void          *pending_raw      = NULL;
-        int            ret              = -1;
-        int            i                = 0;
-        int            j                = 0;
-        int            k                = 0;
-
-        afr_init_pending_matrix (pending_matrix, child_count);
-
-        for (i = 0; i < child_count; i++) {
-                pending_raw = NULL;
-
-                for (j = 0; j < child_count; j++) {
-                        ret = dict_get_ptr (xattr[i], pending_key[j],
-                                            &pending_raw);
-
-                        if (ret != 0) {
-                                /*
-                                 * There is no xattr present. This means this
-                                 * subvolume should be considered an 'ignorant'
-                                 * subvolume.
-                                 */
-
-                                if (ignorant_subvols)
-                                        ignorant_subvols[i] = 1;
-                                continue;
-                        }
-
-                        memcpy (pending, pending_raw, sizeof(pending));
-                        k = afr_index_for_transaction_type (type);
-
-                        pending_matrix[i][j] = ntoh32 (pending[k]);
-                }
-        }
+    STACK_WIND(frame, afr_selfheal_post_op_cbk, priv->children[subvol],
+               priv->children[subvol]->fops->xattrop, &loc,
+               GF_XATTROP_ADD_ARRAY, xattr, xdata);
 
-        return ret;
-}
+    syncbarrier_wait(&local->barrier, 1);
+    if (local->op_ret < 0)
+        ret = -local->op_errno;
 
-typedef enum {
-        AFR_NODE_INVALID,
-        AFR_NODE_INNOCENT,
-        AFR_NODE_FOOL,
-        AFR_NODE_WISE,
-} afr_node_type;
+    loc_wipe(&loc);
+    local->op_ret = 0;
 
-typedef struct {
-        afr_node_type type;
-        int           wisdom;
-} afr_node_character;
+    return ret;
+}
 
+int
+afr_check_stale_error(struct afr_reply *replies, afr_private_t *priv)
+{
+    int i = 0;
+    int op_errno = 0;
+    int tmp_errno = 0;
+    int stale_count = 0;
+
+    for (i = 0; i < priv->child_count; i++) {
+        tmp_errno = replies[i].op_errno;
+        if (tmp_errno == ENOENT || tmp_errno == ESTALE) {
+            op_errno = afr_higher_errno(op_errno, tmp_errno);
+            stale_count++;
+        }
+    }
+    if (stale_count != priv->child_count)
+        return -ENOTCONN;
+    else
+        return -op_errno;
+}
 
-static int
-afr_sh_is_innocent (int32_t *array, int child_count)
+int
+afr_sh_generic_fop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno, struct iatt *pre,
+                       struct iatt *post, dict_t *xdata)
 {
-        int i   = 0;
-        int ret = 1;   /* innocent until proven guilty */
+    int i = (long)cookie;
+    afr_local_t *local = NULL;
 
-        for (i = 0; i < child_count; i++) {
-                if (array[i]) {
-                        ret = 0;
-                        break;
-                }
-        }
+    local = frame->local;
 
-        return ret;
-}
+    local->replies[i].valid = 1;
+    local->replies[i].op_ret = op_ret;
+    local->replies[i].op_errno = op_errno;
+    if (pre)
+        local->replies[i].prestat = *pre;
+    if (post)
+        local->replies[i].poststat = *post;
+    if (xdata)
+        local->replies[i].xdata = dict_ref(xdata);
 
+    syncbarrier_wake(&local->barrier);
 
-static int
-afr_sh_is_fool (int32_t *array, int i, int child_count)
-{
-        return array[i];   /* fool if accuses itself */
+    return 0;
 }
 
+int
+afr_selfheal_restore_time(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                          int source, unsigned char *healed_sinks,
+                          struct afr_reply *replies)
+{
+    loc_t loc = {
+        0,
+    };
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+
+    AFR_ONLIST(healed_sinks, frame, afr_sh_generic_fop_cbk, setattr, &loc,
+               &replies[source].poststat,
+               (GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME | GF_SET_ATTR_CTIME),
+               NULL);
+
+    loc_wipe(&loc);
+
+    return 0;
+}
+
+dict_t *
+afr_selfheal_output_xattr(xlator_t *this, gf_boolean_t is_full_crawl,
+                          afr_transaction_type type, int *output_dirty,
+                          int **output_matrix, int subvol,
+                          int **full_heal_mtx_out)
+{
+    int j = 0;
+    int idx = 0;
+    int d_idx = 0;
+    int ret = 0;
+    int *raw = 0;
+    dict_t *xattr = NULL;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+    idx = afr_index_for_transaction_type(type);
+    d_idx = afr_index_for_transaction_type(AFR_DATA_TRANSACTION);
+
+    xattr = dict_new();
+    if (!xattr)
+        return NULL;
+
+    /* clear dirty */
+    raw = GF_CALLOC(sizeof(int), AFR_NUM_CHANGE_LOGS, gf_afr_mt_int32_t);
+    if (!raw)
+        goto err;
+
+    raw[idx] = hton32(output_dirty[subvol]);
+    ret = dict_set_bin(xattr, AFR_DIRTY, raw,
+                       sizeof(int) * AFR_NUM_CHANGE_LOGS);
+    if (ret) {
+        GF_FREE(raw);
+        goto err;
+    }
+
+    /* clear/set pending */
+    for (j = 0; j < priv->child_count; j++) {
+        raw = GF_CALLOC(sizeof(int), AFR_NUM_CHANGE_LOGS, gf_afr_mt_int32_t);
+        if (!raw)
+            goto err;
+
+        raw[idx] = hton32(output_matrix[subvol][j]);
+        if (is_full_crawl)
+            raw[d_idx] = hton32(full_heal_mtx_out[subvol][j]);
+
+        ret = dict_set_bin(xattr, priv->pending_key[j], raw,
+                           sizeof(int) * AFR_NUM_CHANGE_LOGS);
+        if (ret) {
+            GF_FREE(raw);
+            goto err;
+        }
+    }
 
-static int
-afr_sh_is_wise (int32_t *array, int i, int child_count)
-{
-        return !array[i];  /* wise if does not accuse itself */
+    return xattr;
+err:
+    if (xattr)
+        dict_unref(xattr);
+    return NULL;
 }
 
+int
+afr_selfheal_undo_pending(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                          unsigned char *sources, unsigned char *sinks,
+                          unsigned char *healed_sinks,
+                          unsigned char *undid_pending,
+                          afr_transaction_type type, struct afr_reply *replies,
+                          unsigned char *locked_on)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int i = 0;
+    int j = 0;
+    unsigned char *pending = NULL;
+    int *input_dirty = NULL;
+    int **input_matrix = NULL;
+    int **full_heal_mtx_in = NULL;
+    int **full_heal_mtx_out = NULL;
+    int *output_dirty = NULL;
+    int **output_matrix = NULL;
+    dict_t *xattr = NULL;
+    dict_t *xdata = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    pending = alloca0(priv->child_count);
+
+    input_dirty = alloca0(priv->child_count * sizeof(int));
+    input_matrix = ALLOC_MATRIX(priv->child_count, int);
+    full_heal_mtx_in = ALLOC_MATRIX(priv->child_count, int);
+    full_heal_mtx_out = ALLOC_MATRIX(priv->child_count, int);
+    output_dirty = alloca0(priv->child_count * sizeof(int));
+    output_matrix = ALLOC_MATRIX(priv->child_count, int);
+
+    xdata = dict_new();
+    if (!xdata)
+        return -1;
 
-static int
-afr_sh_all_nodes_innocent (afr_node_character *characters,
-                           int child_count)
-{
-        int i   = 0;
-        int ret = 1;
+    afr_selfheal_extract_xattr(this, replies, type, input_dirty, input_matrix);
 
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].type != AFR_NODE_INNOCENT) {
-                        ret = 0;
-                        break;
-                }
-        }
+    if (local->need_full_crawl)
+        afr_selfheal_extract_xattr(this, replies, AFR_DATA_TRANSACTION, NULL,
+                                   full_heal_mtx_in);
 
-        return ret;
-}
+    for (i = 0; i < priv->child_count; i++)
+        if (sinks[i] && !healed_sinks[i])
+            pending[i] = 1;
 
-
-static int
-afr_sh_wise_nodes_exist (afr_node_character *characters, int child_count)
-{
-        int i   = 0;
-        int ret = 0;
-
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].type == AFR_NODE_WISE) {
-                        ret = 1;
-                        break;
-                }
+    for (i = 0; i < priv->child_count; i++) {
+        for (j = 0; j < priv->child_count; j++) {
+            if (pending[j]) {
+                output_matrix[i][j] = 1;
+                if (type == AFR_ENTRY_TRANSACTION)
+                    full_heal_mtx_out[i][j] = 1;
+            } else if (locked_on[j]) {
+                output_matrix[i][j] = -input_matrix[i][j];
+                if (type == AFR_ENTRY_TRANSACTION)
+                    full_heal_mtx_out[i][j] = -full_heal_mtx_in[i][j];
+            }
         }
+    }
 
-        return ret;
-}
+    for (i = 0; i < priv->child_count; i++) {
+        if (!pending[i])
+            output_dirty[i] = -input_dirty[i];
+    }
 
+    for (i = 0; i < priv->child_count; i++) {
+        if (!locked_on[i])
+            /* perform post-op only on subvols we had locked
+               and inspected on.
+            */
+            continue;
+        if (undid_pending[i])
+            /* We already unset the pending xattrs in
+             * _afr_fav_child_reset_sink_xattrs(). */
+            continue;
 
-/*
- * The 'wisdom' of a wise node is 0 if any other wise node accuses it.
- * It is 1 if no other wise node accuses it.
- * Only wise nodes with wisdom 1 are sources.
- *
- * If no nodes with wisdom 1 exist, a split-brain has occurred.
- */
+        xattr = afr_selfheal_output_xattr(this, local->need_full_crawl, type,
+                                          output_dirty, output_matrix, i,
+                                          full_heal_mtx_out);
+        if (!xattr) {
+            continue;
+        }
 
-static void
-afr_sh_compute_wisdom (int32_t *pending_matrix[],
-                       afr_node_character characters[], int child_count)
-{
-        int i = 0;
-        int j = 0;
+        if ((type == AFR_ENTRY_TRANSACTION) && (priv->esh_granular)) {
+            if (xdata && dict_set_int8(xdata, GF_XATTROP_PURGE_INDEX, 1))
+                gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_DICT_SET_FAILED,
+                       "Failed to set"
+                       " dict value for %s",
+                       GF_XATTROP_PURGE_INDEX);
+        }
 
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].type == AFR_NODE_WISE) {
-                        characters[i].wisdom = 1;
+        afr_selfheal_post_op(frame, this, inode, i, xattr, xdata);
+        dict_unref(xattr);
+    }
 
-                        for (j = 0; j < child_count; j++) {
-                                if ((characters[j].type == AFR_NODE_WISE)
-                                    && pending_matrix[j][i]) {
+    if (xdata)
+        dict_unref(xdata);
 
-                                        characters[i].wisdom = 0;
-                                }
-                        }
-                }
-        }
+    return 0;
 }
 
+void
+afr_reply_copy(struct afr_reply *dst, struct afr_reply *src)
+{
+    dict_t *xdata = NULL;
+
+    dst->valid = src->valid;
+    dst->op_ret = src->op_ret;
+    dst->op_errno = src->op_errno;
+    dst->prestat = src->prestat;
+    dst->poststat = src->poststat;
+    dst->preparent = src->preparent;
+    dst->postparent = src->postparent;
+    dst->preparent2 = src->preparent2;
+    dst->postparent2 = src->postparent2;
+    if (src->xdata)
+        xdata = dict_ref(src->xdata);
+    else
+        xdata = NULL;
+    if (dst->xdata)
+        dict_unref(dst->xdata);
+    dst->xdata = xdata;
+    if (xdata && dict_get_str_boolean(xdata, "fips-mode-rchecksum",
+                                      _gf_false) == _gf_true) {
+        memcpy(dst->checksum, src->checksum, SHA256_DIGEST_LENGTH);
+    } else {
+        memcpy(dst->checksum, src->checksum, MD5_DIGEST_LENGTH);
+    }
+    dst->fips_mode_rchecksum = src->fips_mode_rchecksum;
+}
 
-static int
-afr_sh_wise_nodes_conflict (afr_node_character *characters,
-                            int child_count)
+void
+afr_replies_copy(struct afr_reply *dst, struct afr_reply *src, int count)
 {
-        int i   = 0;
-        int ret = 1;
+    int i = 0;
 
-        for (i = 0; i < child_count; i++) {
-                if ((characters[i].type == AFR_NODE_WISE)
-                    && characters[i].wisdom == 1) {
-
-                        /* There is atleast one bona-fide wise node */
-                        ret = 0;
-                        break;
-                }
-        }
+    if (dst == src)
+        return;
 
-        return ret;
+    for (i = 0; i < count; i++) {
+        afr_reply_copy(&dst[i], &src[i]);
+    }
 }
 
-
-static int
-afr_sh_mark_wisest_as_sources (int sources[],
-                               afr_node_character *characters,
-                               int child_count)
+int
+afr_selfheal_fill_dirty(xlator_t *this, int *dirty, int subvol, int idx,
+                        dict_t *xdata)
 {
-        int nsources = 0;
-        int i        = 0;
+    void *pending_raw = NULL;
+    int pending[3] = {
+        0,
+    };
 
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].wisdom == 1) {
-                        sources[i] = 1;
-                        nsources++;
-                }
-        }
+    if (!dirty)
+        return 0;
 
-        return nsources;
-}
+    if (dict_get_ptr(xdata, AFR_DIRTY, &pending_raw))
+        return -1;
 
-static void
-afr_compute_witness_of_fools (int32_t *witnesses, int32_t **pending_matrix,
-                              afr_node_character *characters,
-                              int32_t child_count)
-{
-        int i       = 0;
-        int j       = 0;
-        int witness = 0;
-
-        GF_ASSERT (witnesses);
-        GF_ASSERT (pending_matrix);
-        GF_ASSERT (characters);
-        GF_ASSERT (child_count > 0);
-
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].type != AFR_NODE_FOOL)
-                        continue;
-
-                witness = 0;
-                for (j = 0; j < child_count; j++) {
-                        if (i == j)
-                                continue;
-                        witness += pending_matrix[i][j];
-                }
-                witnesses[i] = witness;
-        }
+    if (!pending_raw)
+        return -1;
+
+    memcpy(pending, pending_raw, sizeof(pending));
+
+    dirty[subvol] = ntoh32(pending[idx]);
+
+    return 0;
 }
 
-static int32_t
-afr_find_biggest_witness_among_fools (int32_t *witnesses,
-                                      afr_node_character *characters,
-                                      int32_t child_count)
+int
+afr_selfheal_fill_matrix(xlator_t *this, int **matrix, int subvol, int idx,
+                         dict_t *xdata)
 {
-        int i               = 0;
-        int biggest_witness = -1;
-        int biggest_witness_idx = -1;
-        int biggest_witness_cnt = -1;
+    int i = 0;
+    void *pending_raw = NULL;
+    int pending[3] = {
+        0,
+    };
+    afr_private_t *priv = NULL;
 
-        GF_ASSERT (witnesses);
-        GF_ASSERT (characters);
-        GF_ASSERT (child_count > 0);
+    priv = this->private;
 
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].type != AFR_NODE_FOOL)
-                        continue;
+    if (!matrix)
+        return 0;
 
-                if (biggest_witness < witnesses[i]) {
-                        biggest_witness = witnesses[i];
-			biggest_witness_idx = i;
-			biggest_witness_cnt = 1;
-			continue;
-		}
+    for (i = 0; i < priv->child_count; i++) {
+        if (dict_get_ptr(xdata, priv->pending_key[i], &pending_raw))
+            continue;
 
-		if (biggest_witness == witnesses[i])
-			biggest_witness_cnt++;
-        }
+        if (!pending_raw)
+            continue;
 
-	if (biggest_witness_cnt != 1)
-		return -1;
+        memcpy(pending, pending_raw, sizeof(pending));
 
-        return biggest_witness_idx;
+        matrix[subvol][i] = ntoh32(pending[idx]);
+    }
+
+    return 0;
 }
 
 int
-afr_mark_fool_as_source_by_witness (int32_t *sources, int32_t *witnesses,
-                                    afr_node_character *characters,
-                                    int32_t child_count, int32_t witness)
+afr_selfheal_extract_xattr(xlator_t *this, struct afr_reply *replies,
+                           afr_transaction_type type, int *dirty, int **matrix)
 {
-        int i        = 0;
-        int nsources = 0;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    dict_t *xdata = NULL;
+    int idx = -1;
 
-        GF_ASSERT (sources);
-        GF_ASSERT (witnesses);
-        GF_ASSERT (characters);
-        GF_ASSERT (child_count > 0);
+    idx = afr_index_for_transaction_type(type);
 
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].type != AFR_NODE_FOOL)
-                        continue;
+    priv = this->private;
 
-                if (witness == witnesses[i]) {
-                        sources[i] = 1;
-                        nsources++;
-                }
-        }
-        return nsources;
-}
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret != 0)
+            continue;
 
+        if (!replies[i].xdata)
+            continue;
 
-int
-afr_mark_fool_as_source_by_idx (int32_t *sources, int child_count, int idx)
-{
-	if (idx >= 0 && idx < child_count) {
-		sources[idx] = 1;
-		return 1;
-	}
-	return 0;
-}
+        xdata = replies[i].xdata;
 
+        afr_selfheal_fill_dirty(this, dirty, i, idx, xdata);
+        afr_selfheal_fill_matrix(this, matrix, i, idx, xdata);
+    }
 
-static int
-afr_find_largest_file_size (struct iatt *bufs, int32_t *success_children,
-			    int child_count)
-{
-	int idx = -1;
-	int i = -1;
-	int child = -1;
-	uint64_t max_size = 0;
-        uint64_t min_size = 0;
-        int      num_children = 0;
-
-	for (i = 0; i < child_count; i++) {
-		if (success_children[i] == -1)
-			break;
-
-		child = success_children[i];
-		if (bufs[child].ia_size > max_size) {
-			max_size = bufs[child].ia_size;
-			idx = child;
-		}
-
-                if ((num_children == 0) || (bufs[child].ia_size < min_size)) {
-                        min_size = bufs[child].ia_size;
-                }
+    return 0;
+}
 
-                num_children++;
-	}
+/*
+ * If by chance there are multiple sources with differing sizes, select
+ * the largest file as the source.
+ *
+ * This can happen if data was directly modified in the backend or for snapshots
+ */
+void
+afr_mark_largest_file_as_source(xlator_t *this, unsigned char *sources,
+                                struct afr_reply *replies)
+{
+    int i = 0;
+    afr_private_t *priv = NULL;
+    uint64_t size = 0;
 
-        /* If sizes are same for all of them, finding sources will have to
-         * happen with pending changelog. So return -1
-         */
-        if ((num_children > 1) && (min_size == max_size))
-                return -1;
-	return idx;
+    /* Find source with biggest file size */
+    priv = this->private;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i])
+            continue;
+        if (!replies[i].valid || replies[i].op_ret != 0) {
+            sources[i] = 0;
+            continue;
+        }
+        if (size <= replies[i].poststat.ia_size) {
+            size = replies[i].poststat.ia_size;
+        }
+    }
+
+    /* Mark sources with less size as not source */
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i])
+            continue;
+        if (size > replies[i].poststat.ia_size)
+            sources[i] = 0;
+    }
 }
 
+void
+afr_mark_latest_mtime_file_as_source(xlator_t *this, unsigned char *sources,
+                                     struct afr_reply *replies)
+{
+    int i = 0;
+    afr_private_t *priv = NULL;
+    uint32_t mtime = 0;
+    uint32_t mtime_nsec = 0;
+
+    priv = this->private;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i])
+            continue;
+        if (!replies[i].valid || replies[i].op_ret != 0) {
+            sources[i] = 0;
+            continue;
+        }
+        if ((mtime < replies[i].poststat.ia_mtime) ||
+            ((mtime == replies[i].poststat.ia_mtime) &&
+             (mtime_nsec < replies[i].poststat.ia_mtime_nsec))) {
+            mtime = replies[i].poststat.ia_mtime;
+            mtime_nsec = replies[i].poststat.ia_mtime_nsec;
+        }
+    }
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i])
+            continue;
+        if ((mtime > replies[i].poststat.ia_mtime) ||
+            ((mtime == replies[i].poststat.ia_mtime) &&
+             (mtime_nsec > replies[i].poststat.ia_mtime_nsec))) {
+            sources[i] = 0;
+        }
+    }
+}
 
-static int
-afr_find_newest_file (struct iatt *bufs, int32_t *success_children,
-		      int child_count)
+void
+afr_mark_active_sinks(xlator_t *this, unsigned char *sources,
+                      unsigned char *locked_on, unsigned char *sinks)
 {
-	int idx = -1;
-	int i = -1;
-	int child = -1;
-	uint64_t max_ctime = 0;
-
-	for (i = 0; i < child_count; i++) {
-		if (success_children[i] == -1)
-			break;
+    int i = 0;
+    afr_private_t *priv = NULL;
 
-		child = success_children[i];
-		if (bufs[child].ia_ctime > max_ctime) {
-			max_ctime = bufs[child].ia_ctime;
-			idx = child;
-		}
-	}
+    priv = this->private;
 
-	return idx;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i] && locked_on[i])
+            sinks[i] = 1;
+        else
+            sinks[i] = 0;
+    }
 }
 
+gf_boolean_t
+afr_dict_contains_heal_op(call_frame_t *frame)
+{
+    afr_local_t *local = NULL;
+    dict_t *xdata_req = NULL;
+    int ret = 0;
+    int heal_op = -1;
+
+    local = frame->local;
+    xdata_req = local->xdata_req;
+    ret = dict_get_int32_sizen(xdata_req, "heal-op", &heal_op);
+    if (ret)
+        return _gf_false;
+    if (local->xdata_rsp == NULL) {
+        local->xdata_rsp = dict_new();
+        if (!local->xdata_rsp)
+            return _gf_true;
+    }
+    ret = dict_set_sizen_str_sizen(local->xdata_rsp, "sh-fail-msg",
+                                   SFILE_NOT_IN_SPLIT_BRAIN);
 
-static int
-afr_mark_biggest_of_fools_as_source (int32_t *sources, int32_t **pending_matrix,
-                                     afr_node_character *characters,
-				     int32_t *success_children,
-                                     int child_count, struct iatt *bufs)
-{
-        int32_t       biggest_witness = 0;
-        int           nsources        = 0;
-        int32_t       *witnesses      = NULL;
-
-        GF_ASSERT (child_count > 0);
-
-	biggest_witness = afr_find_largest_file_size (bufs, success_children,
-						      child_count);
-	if (biggest_witness != -1)
-		goto found;
-
-        witnesses = GF_CALLOC (child_count, sizeof (*witnesses),
-                               gf_afr_mt_int32_t);
-        if (NULL == witnesses) {
-                nsources = -1;
-                goto out;
-        }
+    return _gf_true;
+}
 
-        afr_compute_witness_of_fools (witnesses, pending_matrix, characters,
-                                      child_count);
-        biggest_witness = afr_find_biggest_witness_among_fools (witnesses,
-                                                                characters,
-                                                                child_count);
-	if (biggest_witness != -1)
-		goto found;
+gf_boolean_t
+afr_can_decide_split_brain_source_sinks(struct afr_reply *replies,
+                                        int child_count)
+{
+    int i = 0;
 
-	biggest_witness = afr_find_newest_file (bufs, success_children,
-						child_count);
+    for (i = 0; i < child_count; i++)
+        if (replies[i].valid != 1 || replies[i].op_ret != 0)
+            return _gf_false;
+
+    return _gf_true;
+}
 
-found:
-	nsources = afr_mark_fool_as_source_by_idx (sources, child_count,
-						   biggest_witness);
+int
+afr_mark_split_brain_source_sinks_by_heal_op(
+    call_frame_t *frame, xlator_t *this, unsigned char *sources,
+    unsigned char *sinks, unsigned char *healed_sinks, unsigned char *locked_on,
+    struct afr_reply *replies, afr_transaction_type type, int heal_op)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    dict_t *xdata_req = NULL;
+    dict_t *xdata_rsp = NULL;
+    int ret = 0;
+    int i = 0;
+    char *name = NULL;
+    int source = -1;
+
+    local = frame->local;
+    priv = this->private;
+    xdata_req = local->xdata_req;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (locked_on[i])
+            if (sources[i] || !sinks[i] || !healed_sinks[i]) {
+                ret = -1;
+                goto out;
+            }
+    }
+    if (local->xdata_rsp == NULL) {
+        local->xdata_rsp = dict_new();
+        if (!local->xdata_rsp) {
+            ret = -1;
+            goto out;
+        }
+    }
+    xdata_rsp = local->xdata_rsp;
+
+    if (!afr_can_decide_split_brain_source_sinks(replies, priv->child_count)) {
+        ret = dict_set_sizen_str_sizen(xdata_rsp, "sh-fail-msg",
+                                       SBRAIN_HEAL_NO_GO_MSG);
+        ret = -1;
+        goto out;
+    }
+
+    for (i = 0; i < priv->child_count; i++)
+        if (locked_on[i])
+            sources[i] = 1;
+    switch (heal_op) {
+        case GF_SHD_OP_SBRAIN_HEAL_FROM_BIGGER_FILE:
+            if (type == AFR_METADATA_TRANSACTION) {
+                ret = dict_set_sizen_str_sizen(xdata_rsp, "sh-fail-msg",
+                                               SUSE_SOURCE_BRICK_TO_HEAL);
+                if (!ret)
+                    ret = -1;
+                goto out;
+            }
+            afr_mark_largest_file_as_source(this, sources, replies);
+            if (AFR_COUNT(sources, priv->child_count) != 1) {
+                ret = dict_set_sizen_str_sizen(xdata_rsp, "sh-fail-msg",
+                                               SNO_BIGGER_FILE);
+                if (!ret)
+                    ret = -1;
+                goto out;
+            }
+            break;
+        case GF_SHD_OP_SBRAIN_HEAL_FROM_LATEST_MTIME:
+            if (type == AFR_METADATA_TRANSACTION) {
+                ret = dict_set_sizen_str_sizen(xdata_rsp, "sh-fail-msg",
+                                               SUSE_SOURCE_BRICK_TO_HEAL);
+                if (!ret)
+                    ret = -1;
+                goto out;
+            }
+            afr_mark_latest_mtime_file_as_source(this, sources, replies);
+            if (AFR_COUNT(sources, priv->child_count) != 1) {
+                ret = dict_set_sizen_str_sizen(xdata_rsp, "sh-fail-msg",
+                                               SNO_DIFF_IN_MTIME);
+                if (!ret)
+                    ret = -1;
+                goto out;
+            }
+            break;
+        case GF_SHD_OP_SBRAIN_HEAL_FROM_BRICK:
+            ret = dict_get_str_sizen(xdata_req, "child-name", &name);
+            if (ret)
+                goto out;
+            source = afr_get_child_index_from_name(this, name);
+            if (source < 0) {
+                ret = dict_set_sizen_str_sizen(xdata_rsp, "sh-fail-msg",
+                                               SINVALID_BRICK_NAME);
+                if (!ret)
+                    ret = -1;
+                goto out;
+            }
+            if (locked_on[source] != 1) {
+                ret = dict_set_sizen_str_sizen(xdata_rsp, "sh-fail-msg",
+                                               SBRICK_IS_NOT_UP);
+                if (!ret)
+                    ret = -1;
+                goto out;
+            }
+            memset(sources, 0, sizeof(*sources) * priv->child_count);
+            sources[source] = 1;
+            break;
+        default:
+            ret = -1;
+            goto out;
+    }
+    for (i = 0; i < priv->child_count; i++) {
+        if (sources[i]) {
+            source = i;
+            break;
+        }
+    }
+    sinks[source] = 0;
+    healed_sinks[source] = 0;
+    ret = source;
 out:
-        GF_FREE (witnesses);
-        return nsources;
+    if (ret < 0)
+        memset(sources, 0, sizeof(*sources) * priv->child_count);
+    return ret;
 }
 
 int
-afr_mark_child_as_source_by_uid (int32_t *sources, struct iatt *bufs,
-                                 int32_t *success_children,
-                                 unsigned int child_count, uint32_t uid)
-{
-        int     i        = 0;
-        int     nsources = 0;
-        int     child    = 0;
-
-        for (i = 0; i < child_count; i++) {
-                if (-1 == success_children[i])
-                        break;
-
-                child = success_children[i];
-                if (uid == bufs[child].ia_uid) {
-                        sources[child] = 1;
-                        nsources++;
+afr_sh_fav_by_majority(xlator_t *this, struct afr_reply *replies,
+                       inode_t *inode)
+{
+    afr_private_t *priv;
+    int vote_count = -1;
+    int fav_child = -1;
+    int i = 0;
+    int k = 0;
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (replies[i].valid == 1) {
+            gf_msg_debug(this->name, 0,
+                         "Child:%s mtime_sec = %" PRId64 ", size = %" PRIu64
+                         " for gfid %s",
+                         priv->children[i]->name, replies[i].poststat.ia_mtime,
+                         replies[i].poststat.ia_size, uuid_utoa(inode->gfid));
+            vote_count = 0;
+            for (k = 0; k < priv->child_count; k++) {
+                if ((replies[k].poststat.ia_mtime ==
+                     replies[i].poststat.ia_mtime) &&
+                    (replies[k].poststat.ia_size ==
+                     replies[i].poststat.ia_size)) {
+                    vote_count++;
                 }
+            }
+            if (vote_count > priv->child_count / 2) {
+                fav_child = i;
+                break;
+            }
         }
-        return nsources;
+    }
+    return fav_child;
 }
 
+/*
+ * afr_sh_fav_by_mtime: Choose favorite child by mtime.
+ */
 int
-afr_get_child_with_lowest_uid (struct iatt *bufs, int32_t *success_children,
-                               unsigned int child_count)
-{
-        int     i        = 0;
-        int     smallest = -1;
-        int     child    = 0;
-
-        for (i = 0; i < child_count; i++) {
-                if (-1 == success_children[i])
-                        break;
-                child = success_children[i];
-                if ((smallest == -1) ||
-                    (bufs[child].ia_uid < bufs[smallest].ia_uid)) {
-                        smallest = child;
-                }
-        }
-        return smallest;
+afr_sh_fav_by_mtime(xlator_t *this, struct afr_reply *replies, inode_t *inode)
+{
+    afr_private_t *priv;
+    int fav_child = -1;
+    int i = 0;
+    uint32_t cmp_mtime = 0;
+    uint32_t cmp_mtime_nsec = 0;
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (replies[i].valid == 1) {
+            gf_msg_debug(this->name, 0,
+                         "Child:%s mtime = %" PRId64
+                         ", mtime_nsec = %d for "
+                         "gfid %s",
+                         priv->children[i]->name, replies[i].poststat.ia_mtime,
+                         replies[i].poststat.ia_mtime_nsec,
+                         uuid_utoa(inode->gfid));
+            if (replies[i].poststat.ia_mtime > cmp_mtime) {
+                cmp_mtime = replies[i].poststat.ia_mtime;
+                cmp_mtime_nsec = replies[i].poststat.ia_mtime_nsec;
+                fav_child = i;
+            } else if ((replies[i].poststat.ia_mtime == cmp_mtime) &&
+                       (replies[i].poststat.ia_mtime_nsec > cmp_mtime_nsec)) {
+                cmp_mtime = replies[i].poststat.ia_mtime;
+                cmp_mtime_nsec = replies[i].poststat.ia_mtime_nsec;
+                fav_child = i;
+            }
+        }
+    }
+    return fav_child;
 }
 
-static int
-afr_sh_mark_lowest_uid_as_source (struct iatt *bufs, int32_t *success_children,
-                                  int child_count, int32_t *sources)
-{
-        int   nsources              = 0;
-        int   smallest              = 0;
+/*
+ * afr_sh_fav_by_ctime: Choose favorite child by ctime.
+ */
+int
+afr_sh_fav_by_ctime(xlator_t *this, struct afr_reply *replies, inode_t *inode)
+{
+    afr_private_t *priv;
+    int fav_child = -1;
+    int i = 0;
+    uint32_t cmp_ctime = 0;
+    uint32_t cmp_ctime_nsec = 0;
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (replies[i].valid == 1) {
+            gf_msg_debug(this->name, 0,
+                         "Child:%s ctime = %" PRId64
+                         ", ctime_nsec = %d for "
+                         "gfid %s",
+                         priv->children[i]->name, replies[i].poststat.ia_ctime,
+                         replies[i].poststat.ia_ctime_nsec,
+                         uuid_utoa(inode->gfid));
+            if (replies[i].poststat.ia_ctime > cmp_ctime) {
+                cmp_ctime = replies[i].poststat.ia_ctime;
+                cmp_ctime_nsec = replies[i].poststat.ia_ctime_nsec;
+                fav_child = i;
+            } else if ((replies[i].poststat.ia_ctime == cmp_ctime) &&
+                       (replies[i].poststat.ia_ctime_nsec > cmp_ctime_nsec)) {
+                cmp_ctime = replies[i].poststat.ia_ctime;
+                cmp_ctime_nsec = replies[i].poststat.ia_ctime_nsec;
+                fav_child = i;
+            }
+        }
+    }
+    return fav_child;
+}
 
-        smallest = afr_get_child_with_lowest_uid (bufs, success_children,
-                                                  child_count);
-        if (smallest < 0) {
-                nsources = -1;
-                goto out;
-        }
-        nsources = afr_mark_child_as_source_by_uid (sources, bufs,
-                                                    success_children, child_count,
-                                                    bufs[smallest].ia_uid);
-out:
-        return nsources;
+/*
+ * afr_sh_fav_by_size: Choose favorite child by size
+ * when not all files are of zero size.
+ */
+int
+afr_sh_fav_by_size(xlator_t *this, struct afr_reply *replies, inode_t *inode)
+{
+    afr_private_t *priv;
+    int fav_child = -1;
+    int i = 0;
+    uint64_t cmp_sz = 0;
+
+    priv = this->private;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid) {
+            continue;
+        }
+        gf_msg_debug(this->name, 0,
+                     "Child:%s file size = %" PRIu64 " for gfid %s",
+                     priv->children[i]->name, replies[i].poststat.ia_size,
+                     uuid_utoa(inode->gfid));
+        if (replies[i].poststat.ia_type == IA_IFDIR) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SBRAIN_FAV_CHILD_POLICY,
+                   "Cannot perform selfheal on %s. "
+                   "Size policy is not applicable to directories.",
+                   uuid_utoa(inode->gfid));
+            break;
+        }
+        if (replies[i].poststat.ia_size > cmp_sz) {
+            cmp_sz = replies[i].poststat.ia_size;
+            fav_child = i;
+        } else if (replies[i].poststat.ia_size == cmp_sz) {
+            fav_child = -1;
+        }
+    }
+    if (fav_child == -1) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN,
+               "No bigger file");
+    }
+    return fav_child;
 }
 
 int
-afr_get_no_xattr_dir_read_child (xlator_t *this, int32_t *success_children,
-                                 struct iatt *bufs)
+afr_sh_get_fav_by_policy(xlator_t *this, struct afr_reply *replies,
+                         inode_t *inode, char **policy_str)
 {
-        afr_private_t *priv = NULL;
-        int            i = 0;
-        int            child = -1;
-        int            read_child = -1;
+    afr_private_t *priv = NULL;
+    int fav_child = -1;
 
-        priv = this->private;
-        for (i = 0; i < priv->child_count; i++) {
-                child = success_children[i];
-                if (child < 0)
-                        break;
-                if (read_child < 0)
-                        read_child = child;
-                else if (bufs[read_child].ia_size < bufs[child].ia_size)
-                        read_child = child;
-        }
-        return read_child;
+    priv = this->private;
+    if (!afr_can_decide_split_brain_source_sinks(replies, priv->child_count)) {
+        return -1;
+    }
+
+    switch (priv->fav_child_policy) {
+        case AFR_FAV_CHILD_BY_SIZE:
+            fav_child = afr_sh_fav_by_size(this, replies, inode);
+            if (policy_str && fav_child >= 0) {
+                *policy_str = "SIZE";
+            }
+            break;
+        case AFR_FAV_CHILD_BY_CTIME:
+            fav_child = afr_sh_fav_by_ctime(this, replies, inode);
+            if (policy_str && fav_child >= 0) {
+                *policy_str = "CTIME";
+            }
+            break;
+        case AFR_FAV_CHILD_BY_MTIME:
+            fav_child = afr_sh_fav_by_mtime(this, replies, inode);
+            if (policy_str && fav_child >= 0) {
+                *policy_str = "MTIME";
+            }
+            break;
+        case AFR_FAV_CHILD_BY_MAJORITY:
+            fav_child = afr_sh_fav_by_majority(this, replies, inode);
+            if (policy_str && fav_child >= 0) {
+                *policy_str = "MAJORITY";
+            }
+            break;
+        case AFR_FAV_CHILD_NONE:
+        default:
+            break;
+    }
+
+    return fav_child;
 }
 
 int
-afr_sh_mark_zero_size_file_as_sink (struct iatt *bufs, int32_t *success_children,
-                                    int child_count, int32_t *sources)
-{
-        int             nsources = 0;
-        int             i = 0;
-        int             child = 0;
-        gf_boolean_t    sink_exists = _gf_false;
-        gf_boolean_t    source_exists = _gf_false;
-        int             source = -1;
-
-        for (i = 0; i < child_count; i++) {
-                child = success_children[i];
-                if (child < 0)
-                        break;
-                if (!bufs[child].ia_size) {
-                        sink_exists = _gf_true;
-                        continue;
-                }
-                if (!source_exists) {
-                        source_exists = _gf_true;
-                        source = child;
-                        continue;
-                }
-                if (bufs[source].ia_size != bufs[child].ia_size) {
-                        nsources = -1;
-                        goto out;
-                }
-        }
-        if (!source_exists && !sink_exists) {
-                nsources = -1;
-                goto out;
-        }
-
-        if (!source_exists || !sink_exists)
-                goto out;
-
-        for (i = 0; i < child_count; i++) {
-                child = success_children[i];
-                if (child < 0)
-                        break;
-                if (bufs[child].ia_size) {
-                        sources[child] = 1;
-                        nsources++;
-                }
-        }
-out:
-        return nsources;
+afr_mark_split_brain_source_sinks_by_policy(
+    call_frame_t *frame, xlator_t *this, inode_t *inode, unsigned char *sources,
+    unsigned char *sinks, unsigned char *healed_sinks, unsigned char *locked_on,
+    struct afr_reply *replies, afr_transaction_type type)
+{
+    afr_private_t *priv = NULL;
+    int fav_child = -1;
+    char mtime_str[256];
+    char ctime_str[256];
+    char *policy_str = NULL;
+    struct tm *tm_ptr;
+    time_t time;
+
+    priv = this->private;
+
+    fav_child = afr_sh_get_fav_by_policy(this, replies, inode, &policy_str);
+    if (fav_child == -1) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SBRAIN_FAV_CHILD_POLICY,
+               "No child selected by favorite-child policy.");
+    } else if (fav_child > priv->child_count - 1) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SBRAIN_FAV_CHILD_POLICY,
+               "Invalid child (%d) "
+               "selected by policy %s.",
+               fav_child, policy_str);
+    } else if (fav_child >= 0) {
+        time = replies[fav_child].poststat.ia_mtime;
+        tm_ptr = localtime(&time);
+        strftime(mtime_str, sizeof(mtime_str), "%Y-%m-%d %H:%M:%S", tm_ptr);
+        time = replies[fav_child].poststat.ia_ctime;
+        tm_ptr = localtime(&time);
+        strftime(ctime_str, sizeof(ctime_str), "%Y-%m-%d %H:%M:%S", tm_ptr);
+
+        gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_SBRAIN_FAV_CHILD_POLICY,
+               "Source %s selected as authentic to resolve conflicting data "
+               "in file (gfid:%s) by %s (%" PRIu64
+               " bytes @ %s mtime, %s "
+               "ctime).",
+               priv->children[fav_child]->name, uuid_utoa(inode->gfid),
+               policy_str, replies[fav_child].poststat.ia_size, mtime_str,
+               ctime_str);
+
+        sources[fav_child] = 1;
+        sinks[fav_child] = 0;
+        healed_sinks[fav_child] = 0;
+    }
+    return fav_child;
 }
 
-char *
-afr_get_character_str (afr_node_type type)
+gf_boolean_t
+afr_is_file_empty_on_all_children(afr_private_t *priv,
+                                  struct afr_reply *replies)
 {
-        char *character = NULL;
+    int i = 0;
 
-        switch (type) {
-        case AFR_NODE_INNOCENT:
-                character = "innocent";
-                break;
-        case AFR_NODE_FOOL:
-                character = "fool";
-                break;
-        case AFR_NODE_WISE:
-                character = "wise";
-                break;
-        default:
-                character = "invalid";
-                break;
-        }
-        return character;
+    for (i = 0; i < priv->child_count; i++) {
+        if ((!replies[i].valid) || (replies[i].op_ret != 0) ||
+            (replies[i].poststat.ia_size != 0))
+            return _gf_false;
+    }
+
+    return _gf_true;
 }
 
-afr_node_type
-afr_find_child_character_type (int32_t *pending_row, int32_t child,
-                               unsigned int child_count)
-{
-        afr_node_type type = AFR_NODE_INVALID;
+int
+afr_mark_source_sinks_if_file_empty(xlator_t *this, unsigned char *sources,
+                                    unsigned char *sinks,
+                                    unsigned char *healed_sinks,
+                                    unsigned char *locked_on,
+                                    struct afr_reply *replies,
+                                    afr_transaction_type type)
+{
+    int source = -1;
+    int i = 0;
+    afr_private_t *priv = this->private;
+    struct iatt stbuf = {
+        0,
+    };
+
+    if ((AFR_COUNT(locked_on, priv->child_count) < priv->child_count) ||
+        (afr_success_count(replies, priv->child_count) < priv->child_count))
+        return -1;
 
-        GF_ASSERT ((child >= 0) && (child < child_count));
+    if (type == AFR_DATA_TRANSACTION) {
+        if (!afr_is_file_empty_on_all_children(priv, replies))
+            return -1;
+        goto mark;
+    }
+
+    /*For AFR_METADATA_TRANSACTION, metadata must be same on all bricks.*/
+    stbuf = replies[0].poststat;
+    for (i = 1; i < priv->child_count; i++) {
+        if ((!IA_EQUAL(stbuf, replies[i].poststat, type)) ||
+            (!IA_EQUAL(stbuf, replies[i].poststat, uid)) ||
+            (!IA_EQUAL(stbuf, replies[i].poststat, gid)) ||
+            (!IA_EQUAL(stbuf, replies[i].poststat, prot)))
+            return -1;
+    }
+    for (i = 1; i < priv->child_count; i++) {
+        if (!afr_xattrs_are_equal(replies[0].xdata, replies[i].xdata))
+            return -1;
+    }
+
+mark:
+    /* data/metadata is same on all bricks. Pick one of them as source. Rest
+     * are sinks.*/
+    for (i = 0; i < priv->child_count; i++) {
+        if (source == -1) {
+            source = i;
+            sources[i] = 1;
+            sinks[i] = 0;
+            healed_sinks[i] = 0;
+            continue;
+        }
+        sources[i] = 0;
+        sinks[i] = 1;
+        healed_sinks[i] = 1;
+    }
+
+    return source;
+}
+
+/* Return a source depending on the type of heal_op, and set sources[source],
+ * sinks[source] and healed_sinks[source] to 1, 0 and 0 respectively. Do so
+ * only if the following condition is met:
+ * ∀i((i ∈ locked_on[] ∧ i=1)==>(sources[i]=0 ∧ sinks[i]=1 ∧ healed_sinks[i]=1))
+ * i.e. for each locked node, sources[node] is 0; healed_sinks[node] and
+ * sinks[node] are 1. This should be the case if the file is in split-brain.
+ */
+int
+afr_mark_split_brain_source_sinks(
+    call_frame_t *frame, xlator_t *this, inode_t *inode, unsigned char *sources,
+    unsigned char *sinks, unsigned char *healed_sinks, unsigned char *locked_on,
+    struct afr_reply *replies, afr_transaction_type type)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    dict_t *xdata_req = NULL;
+    int heal_op = -1;
+    int ret = -1;
+    int source = -1;
+
+    local = frame->local;
+    priv = this->private;
+    xdata_req = local->xdata_req;
+
+    source = afr_mark_source_sinks_if_file_empty(
+        this, sources, sinks, healed_sinks, locked_on, replies, type);
+    if (source >= 0)
+        return source;
+
+    ret = dict_get_int32_sizen(xdata_req, "heal-op", &heal_op);
+    if (ret)
+        goto autoheal;
+
+    source = afr_mark_split_brain_source_sinks_by_heal_op(
+        frame, this, sources, sinks, healed_sinks, locked_on, replies, type,
+        heal_op);
+    return source;
+
+autoheal:
+    /* Automatically heal if fav_child_policy is set. */
+    if (priv->fav_child_policy != AFR_FAV_CHILD_NONE) {
+        source = afr_mark_split_brain_source_sinks_by_policy(
+            frame, this, inode, sources, sinks, healed_sinks, locked_on,
+            replies, type);
+        if (source != -1) {
+            ret = dict_set_int32_sizen(xdata_req, "fav-child-policy", 1);
+            if (ret)
+                return -1;
+        }
+    }
 
-        if (afr_sh_is_innocent (pending_row, child_count))
-                type = AFR_NODE_INNOCENT;
-        else if (afr_sh_is_fool (pending_row, child, child_count))
-                type = AFR_NODE_FOOL;
-        else if (afr_sh_is_wise (pending_row, child, child_count))
-                type = AFR_NODE_WISE;
-        return type;
+    return source;
 }
 
 int
-afr_build_sources (xlator_t *this, dict_t **xattr, struct iatt *bufs,
-                   int32_t **pending_matrix, int32_t *sources,
-                   int32_t *success_children, afr_transaction_type type,
-                   int32_t *subvol_status, gf_boolean_t ignore_ignorant)
-{
-        afr_private_t           *priv = NULL;
-        afr_self_heal_type      sh_type    = AFR_SELF_HEAL_INVALID;
-        int                     nsources   = -1;
-        unsigned char           *ignorant_subvols = NULL;
-        unsigned int            child_count = 0;
+_afr_fav_child_reset_sink_xattrs(call_frame_t *frame, xlator_t *this,
+                                 inode_t *inode, int source,
+                                 unsigned char *healed_sinks,
+                                 unsigned char *undid_pending,
+                                 afr_transaction_type type,
+                                 unsigned char *locked_on,
+                                 struct afr_reply *replies)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int *input_dirty = NULL;
+    int **input_matrix = NULL;
+    int *output_dirty = NULL;
+    int **output_matrix = NULL;
+    dict_t *xattr = NULL;
+    dict_t *xdata = NULL;
+    int i = 0;
+
+    priv = this->private;
+    local = frame->local;
+
+    if (!dict_get_sizen(local->xdata_req, "fav-child-policy"))
+        return 0;
 
-        priv = this->private;
-        child_count = priv->child_count;
+    xdata = dict_new();
+    if (!xdata)
+        return -1;
 
-        if (afr_get_children_count (success_children, priv->child_count) == 0)
-                goto out;
+    input_dirty = alloca0(priv->child_count * sizeof(int));
+    input_matrix = ALLOC_MATRIX(priv->child_count, int);
+    output_dirty = alloca0(priv->child_count * sizeof(int));
+    output_matrix = ALLOC_MATRIX(priv->child_count, int);
 
-        if (!ignore_ignorant) {
-                ignorant_subvols = GF_CALLOC (sizeof (*ignorant_subvols),
-                                              child_count, gf_afr_mt_char);
-                if (NULL == ignorant_subvols)
-                        goto out;
-        }
+    afr_selfheal_extract_xattr(this, replies, type, input_dirty, input_matrix);
 
-        afr_build_pending_matrix (priv->pending_key, pending_matrix,
-                                  ignorant_subvols, xattr, type,
-                                  priv->child_count);
+    for (i = 0; i < priv->child_count; i++) {
+        if (i == source || !healed_sinks[i])
+            continue;
+        output_dirty[i] = -input_dirty[i];
+        output_matrix[i][source] = -input_matrix[i][source];
+    }
 
-        if (!ignore_ignorant)
-                afr_mark_ignorant_subvols_as_pending (pending_matrix,
-                                                      ignorant_subvols,
-                                                      priv->child_count);
-        sh_type = afr_self_heal_type_for_transaction (type);
-        if (AFR_SELF_HEAL_INVALID == sh_type)
-                goto out;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!healed_sinks[i] || !locked_on[i])
+            continue;
+        xattr = afr_selfheal_output_xattr(this, _gf_false, type, output_dirty,
+                                          output_matrix, i, NULL);
 
-        afr_sh_print_pending_matrix (pending_matrix, this);
+        afr_selfheal_post_op(frame, this, inode, i, xattr, xdata);
 
-        nsources = afr_mark_sources (this, sources, pending_matrix, bufs,
-                                     sh_type, success_children, subvol_status);
-out:
-        GF_FREE (ignorant_subvols);
-        return nsources;
+        undid_pending[i] = 1;
+        dict_unref(xattr);
+    }
+
+    if (xdata)
+        dict_unref(xdata);
+
+    return 0;
 }
 
-void
-afr_find_character_types (afr_node_character *characters,
-                          int32_t **pending_matrix, int32_t *success_children,
-                          unsigned int child_count)
-{
-        afr_node_type type  = AFR_NODE_INVALID;
-        int           child = 0;
-        int           i     = 0;
-
-        for (i = 0; i < child_count; i++) {
-                child = success_children[i];
-                if (child == -1)
-                        break;
-                type = afr_find_child_character_type (pending_matrix[child],
-                                                      child, child_count);
-                characters[child].type = type;
-        }
+gf_boolean_t
+afr_does_witness_exist(xlator_t *this, uint64_t *witness)
+{
+    int i = 0;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (witness[i])
+            return _gf_true;
+    }
+    return _gf_false;
+}
+
+unsigned int
+afr_get_quorum_count(afr_private_t *priv)
+{
+    if (priv->quorum_count == AFR_QUORUM_AUTO) {
+        return priv->child_count / 2 + 1;
+    } else {
+        return priv->quorum_count;
+    }
 }
 
 void
-afr_mark_success_children_sources (int32_t *sources, int32_t *success_children,
-                                   unsigned int child_count)
-{
-        int i = 0;
-        for (i = 0; i < child_count; i++) {
-                if (success_children[i] == -1)
-                        break;
-                sources[success_children[i]] = 1;
+afr_selfheal_post_op_failure_accounting(afr_private_t *priv, char *accused,
+                                        unsigned char *sources,
+                                        unsigned char *locked_on)
+{
+    int i = 0;
+    unsigned int quorum_count = 0;
+
+    if (AFR_COUNT(sources, priv->child_count) != 0)
+        return;
+
+    quorum_count = afr_get_quorum_count(priv);
+    for (i = 0; i < priv->child_count; i++) {
+        if ((accused[i] < quorum_count) && locked_on[i]) {
+            sources[i] = 1;
         }
+    }
+    return;
 }
-/**
- * mark_sources: Mark all 'source' nodes and return number of source
- * nodes found
+
+/*
+ * This function determines if a self-heal is required for a given inode,
+ * and if needed, in what direction.
  *
- * A node (a row in the pending matrix) belongs to one of
- * three categories:
+ * locked_on[] is the array representing servers which have been locked and
+ * from which xattrs have been fetched for analysis.
  *
- * M is the pending matrix.
+ * The output of the function is by filling the arrays sources[] and sinks[].
  *
- * 'innocent' - M[i] is all zeroes
- * 'fool'     - M[i] has i'th element = 1 (self-reference)
- * 'wise'     - M[i] has i'th element = 0, others are 1 or 0.
+ * sources[i] is set if i'th server is an eligible source for a selfheal.
  *
- * All 'innocent' nodes are sinks. If all nodes are innocent, no self-heal is
- * needed.
+ * sinks[i] is set if i'th server needs to be healed.
  *
- * A 'wise' node can be a source. If two 'wise' nodes conflict, it is
- * a split-brain. If one wise node refers to the other but the other doesn't
- * refer back, the referrer is a source.
+ * if sources[0..N] are all set, there is no need for a selfheal.
+ *
+ * if sinks[0..N] are all set, the inode is in split brain.
  *
- * All fools are sinks, unless there are no 'wise' nodes. In that case,
- * one of the fools is made a source.
  */
 
 int
-afr_mark_sources (xlator_t *this, int32_t *sources, int32_t **pending_matrix,
-                  struct iatt *bufs, afr_self_heal_type type,
-                  int32_t *success_children, int32_t *subvol_status)
-{
-        /* stores the 'characters' (innocent, fool, wise) of the nodes */
-        afr_node_character *characters =  NULL;
-        int                nsources    = -1;
-        unsigned int       child_count = 0;
-        afr_private_t      *priv       = NULL;
-
-        priv = this->private;
-        child_count = priv->child_count;
-        characters = GF_CALLOC (sizeof (afr_node_character),
-                                child_count, gf_afr_mt_afr_node_character);
-        if (!characters)
-                goto out;
-
-        this = THIS;
-
-        /* start clean */
-        memset (sources, 0, sizeof (*sources) * child_count);
-        nsources = 0;
-        afr_find_character_types (characters, pending_matrix, success_children,
-                                  child_count);
-        if (afr_sh_all_nodes_innocent (characters, child_count)) {
-                switch (type) {
-                case AFR_SELF_HEAL_METADATA:
-                        nsources = afr_sh_mark_lowest_uid_as_source (bufs,
-                                                             success_children,
-                                                             child_count,
-                                                             sources);
-                        break;
-                case AFR_SELF_HEAL_DATA:
-                        nsources = afr_sh_mark_zero_size_file_as_sink (bufs,
-                                                             success_children,
-                                                             child_count,
-                                                             sources);
-                        if ((nsources < 0) && subvol_status)
-                                *subvol_status |= SPLIT_BRAIN;
-                        break;
-                default:
-                        break;
-                }
-                goto out;
+afr_selfheal_find_direction(call_frame_t *frame, xlator_t *this,
+                            struct afr_reply *replies,
+                            afr_transaction_type type, unsigned char *locked_on,
+                            unsigned char *sources, unsigned char *sinks,
+                            uint64_t *witness, unsigned char *pflag)
+{
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int j = 0;
+    int *dirty = NULL;         /* Denotes if dirty xattr is set */
+    int **matrix = NULL;       /* Changelog matrix */
+    char *accused = NULL;      /* Accused others without any self-accusal */
+    char *pending = NULL;      /* Have pending operations on others */
+    char *self_accused = NULL; /* Accused itself */
+
+    priv = this->private;
+
+    dirty = alloca0(priv->child_count * sizeof(int));
+    accused = alloca0(priv->child_count);
+    pending = alloca0(priv->child_count);
+    self_accused = alloca0(priv->child_count);
+    matrix = ALLOC_MATRIX(priv->child_count, int);
+    memset(witness, 0, sizeof(*witness) * priv->child_count);
+
+    /* First construct the pending matrix for further analysis */
+    afr_selfheal_extract_xattr(this, replies, type, dirty, matrix);
+
+    if (pflag) {
+        for (i = 0; i < priv->child_count; i++) {
+            for (j = 0; j < priv->child_count; j++)
+                if (matrix[i][j])
+                    *pflag |= PFLAG_PENDING;
+            if (*pflag)
+                break;
         }
-
-        if (afr_sh_wise_nodes_exist (characters, child_count)) {
-                afr_sh_compute_wisdom (pending_matrix, characters, child_count);
-
-                if (afr_sh_wise_nodes_conflict (characters, child_count)) {
-                        if (subvol_status)
-                                *subvol_status |= SPLIT_BRAIN;
-                        nsources = -1;
-                } else {
-                        nsources = afr_sh_mark_wisest_as_sources (sources,
-                                                                  characters,
-                                                                  child_count);
-                }
-        } else {
-                if (subvol_status)
-                        *subvol_status |= ALL_FOOLS;
-                nsources = afr_mark_biggest_of_fools_as_source (sources,
-                                                                pending_matrix,
-                                                                characters,
-								success_children,
-                                                                child_count, bufs);
+    }
+
+    if (afr_success_count(replies, priv->child_count) < priv->child_count) {
+        /* Treat this just like locks not being acquired */
+        return -ENOTCONN;
+    }
+
+    /* short list all self-accused */
+    for (i = 0; i < priv->child_count; i++) {
+        if (matrix[i][i])
+            self_accused[i] = 1;
+    }
+
+    /* Next short list all accused to exclude them from being sources */
+    /* Self-accused can't accuse others as they are FOOLs */
+    for (i = 0; i < priv->child_count; i++) {
+        for (j = 0; j < priv->child_count; j++) {
+            if (matrix[i][j]) {
+                if (!self_accused[i])
+                    accused[j] += 1;
+                if (i != j)
+                    pending[i] += 1;
+            }
+        }
+    }
+
+    /* Short list all non-accused as sources */
+    for (i = 0; i < priv->child_count; i++) {
+        if (!accused[i] && locked_on[i])
+            sources[i] = 1;
+        else
+            sources[i] = 0;
+    }
+
+    /* Everyone accused by non-self-accused sources are sinks */
+    memset(sinks, 0, priv->child_count);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i])
+            continue;
+        if (self_accused[i])
+            continue;
+        for (j = 0; j < priv->child_count; j++) {
+            if (matrix[i][j])
+                sinks[j] = 1;
+        }
+    }
+
+    /* For breaking ties provide with number of fops they witnessed */
+
+    /*
+     * count the pending fops witnessed from itself to others when it is
+     * self-accused
+     */
+    for (i = 0; i < priv->child_count; i++) {
+        if (!self_accused[i])
+            continue;
+        for (j = 0; j < priv->child_count; j++) {
+            if (i == j)
+                continue;
+            witness[i] += matrix[i][j];
+        }
+    }
+
+    if (type == AFR_DATA_TRANSACTION || type == AFR_METADATA_TRANSACTION)
+        afr_selfheal_post_op_failure_accounting(priv, accused, sources,
+                                                locked_on);
+
+    /* If no sources, all locked nodes are sinks - split brain */
+    if (AFR_COUNT(sources, priv->child_count) == 0) {
+        for (i = 0; i < priv->child_count; i++) {
+            if (locked_on[i])
+                sinks[i] = 1;
+        }
+        if (pflag)
+            *pflag |= PFLAG_SBRAIN;
+    }
+
+    /* One more class of witness similar to dirty in v2 is where no pending
+     * exists but we have self-accusing markers. This can happen in afr-v1
+     * if the brick crashes just after doing xattrop on self but
+     * before xattrop on the other xattrs on the brick in pre-op. */
+    if (AFR_COUNT(pending, priv->child_count) == 0) {
+        for (i = 0; i < priv->child_count; i++) {
+            if (self_accused[i])
+                witness[i] += matrix[i][i];
         }
+    } else {
+        /* In afr-v1 if a file is self-accused and has pending
+         * operations on others then it is similar to 'dirty' in afr-v2.
+         * Consider such cases as witness.
+         */
+        for (i = 0; i < priv->child_count; i++) {
+            if (self_accused[i] && pending[i])
+                witness[i] += matrix[i][i];
+        }
+    }
 
-out:
-        if (nsources == 0)
-                afr_mark_success_children_sources (sources, success_children,
-                                                   child_count);
-        GF_FREE (characters);
+    /* count the number of dirty fops witnessed */
+    for (i = 0; i < priv->child_count; i++)
+        witness[i] += dirty[i];
 
-        gf_log (this->name, GF_LOG_DEBUG, "Number of sources: %d", nsources);
-        return nsources;
+    return 0;
 }
 
 void
-afr_sh_pending_to_delta (afr_private_t *priv, dict_t **xattr,
-                         int32_t *delta_matrix[], unsigned char success[],
-                         int child_count, afr_transaction_type type)
-{
-        int     tgt     = 0;
-        int     src     = 0;
-        int     value   = 0;
-
-        afr_build_pending_matrix (priv->pending_key, delta_matrix, NULL,
-                                  xattr, type, priv->child_count);
-
-        /*
-         * The algorithm here has two parts.  First, for each subvol indexed
-         * as tgt, we try to figure out what count everyone should have for it.
-         * If the self-heal succeeded, that's easy; the value is zero.
-         * Otherwise, the value is the maximum of the succeeding nodes' counts.
-         * Once we know the value, we loop through (possibly for a second time)
-         * setting each count to the difference so that when we're done all
-         * succeeding nodes will have the same count for tgt.
-         */
-        for (tgt = 0; tgt < priv->child_count; ++tgt) {
-                value = 0;
-                if (!success[tgt]) {
-                        /* Find the maximum. */
-                        for (src = 0; src < priv->child_count; ++src) {
-                                if (!success[src]) {
-                                        continue;
-                                }
-                                if (delta_matrix[src][tgt] > value) {
-                                        value = delta_matrix[src][tgt];
-                                }
-                        }
-                }
-                /* Force everyone who succeeded to the chosen value. */
-                for (src = 0; src < priv->child_count; ++src) {
-                        if (success[src]) {
-                                delta_matrix[src][tgt] = value
-                                                       - delta_matrix[src][tgt];
-                        }
-                        else {
-                                delta_matrix[src][tgt] = 0;
-                        }
-                }
-        }
+afr_log_selfheal(uuid_t gfid, xlator_t *this, int ret, char *type, int source,
+                 unsigned char *sources, unsigned char *healed_sinks)
+{
+    char *status = NULL;
+    char *sinks_str = NULL;
+    char *p = NULL;
+    char *sources_str = NULL;
+    char *q = NULL;
+    afr_private_t *priv = NULL;
+    gf_loglevel_t loglevel = GF_LOG_NONE;
+    int i = 0;
+
+    priv = this->private;
+    sinks_str = alloca0(priv->child_count * 8);
+    p = sinks_str;
+    sources_str = alloca0(priv->child_count * 8);
+    q = sources_str;
+    for (i = 0; i < priv->child_count; i++) {
+        if (healed_sinks[i])
+            p += sprintf(p, "%d ", i);
+        if (sources[i]) {
+            if (source == i) {
+                q += sprintf(q, "[%d] ", i);
+            } else {
+                q += sprintf(q, "%d ", i);
+            }
+        }
+    }
+
+    if (ret < 0) {
+        status = "Failed";
+        loglevel = GF_LOG_DEBUG;
+    } else {
+        status = "Completed";
+        loglevel = GF_LOG_INFO;
+    }
+
+    gf_msg(this->name, loglevel, 0, AFR_MSG_SELF_HEAL_INFO,
+           "%s %s selfheal on %s. "
+           "sources=%s sinks=%s",
+           status, type, uuid_utoa(gfid), sources_str, sinks_str);
 }
 
-
 int
-afr_sh_delta_to_xattr (xlator_t *this,
-                       int32_t *delta_matrix[], dict_t *xattr[],
-                       int child_count, afr_transaction_type type)
-{
-        int              i       = 0;
-        int              j       = 0;
-        int              k       = 0;
-        int              ret     = 0;
-        int32_t         *pending = NULL;
-        int32_t         *local_pending = NULL;
-        afr_private_t   *priv = NULL;
-
-        priv = this->private;
-        for (i = 0; i < child_count; i++) {
-                if (!xattr[i])
-                        continue;
-
-                local_pending = NULL;
-                for (j = 0; j < child_count; j++) {
-                        pending = GF_CALLOC (sizeof (int32_t), 3,
-                                             gf_afr_mt_int32_t);
-
-                        if (!pending) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "failed to allocate pending entry "
-                                        "for %s[%d] on %s",
-                                        priv->pending_key[j], type,
-                                        priv->children[i]->name);
-                                continue;
-                        }
-                        /* 3 = data+metadata+entry */
-
-                        k = afr_index_for_transaction_type (type);
-
-                        pending[k] = hton32 (delta_matrix[i][j]);
-
-                        if (j == i) {
-                                local_pending = pending;
-                                continue;
-                        }
-                        ret = dict_set_bin (xattr[i], priv->pending_key[j],
-                                            pending,
-                                        AFR_NUM_CHANGE_LOGS * sizeof (int32_t));
-                        if (ret < 0) {
-                                gf_log (this->name, GF_LOG_WARNING,
-                                        "Unable to set dict value.");
-                                GF_FREE (pending);
-                        }
-                }
-                if (local_pending) {
-                        ret = dict_set_bin (xattr[i], priv->pending_key[i],
-                                            local_pending,
-                                        AFR_NUM_CHANGE_LOGS * sizeof (int32_t));
-                        if (ret < 0) {
-                                gf_log (this->name, GF_LOG_WARNING,
-                                        "Unable to set dict value.");
-                                GF_FREE (local_pending);
-                        }
-                }
-        }
-        return 0;
-}
+afr_selfheal_discover_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int op_ret, int op_errno, inode_t *inode,
+                          struct iatt *buf, dict_t *xdata, struct iatt *parbuf)
+{
+    afr_local_t *local = NULL;
+    int i = -1;
+    GF_UNUSED int ret = -1;
+    int8_t need_heal = 1;
 
+    local = frame->local;
+    i = (long)cookie;
 
-int
-afr_sh_missing_entries_done (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
+    local->replies[i].valid = 1;
+    local->replies[i].op_ret = op_ret;
+    local->replies[i].op_errno = op_errno;
+    if (buf)
+        local->replies[i].poststat = *buf;
+    if (parbuf)
+        local->replies[i].postparent = *parbuf;
+    if (xdata) {
+        local->replies[i].xdata = dict_ref(xdata);
+        ret = dict_get_int8(xdata, "link-count", &need_heal);
+    }
 
-        local = frame->local;
-        sh = &local->self_heal;
+    local->replies[i].need_heal = need_heal;
+    syncbarrier_wake(&local->barrier);
 
-        afr_sh_reset (frame, this);
+    return 0;
+}
 
-        if (local->unhealable) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "split brain found, aborting selfheal of %s",
-                        local->loc.path);
-        }
+inode_t *
+afr_selfheal_unlocked_lookup_on(call_frame_t *frame, inode_t *parent,
+                                const char *name, struct afr_reply *replies,
+                                unsigned char *lookup_on, dict_t *xattr)
+{
+    loc_t loc = {
+        0,
+    };
+    dict_t *xattr_req = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    inode_t *inode = NULL;
 
-        if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
-                sh->completion_cbk (frame, this);
-        } else {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "proceeding to metadata check on %s",
-                        local->loc.path);
-                afr_self_heal_metadata (frame, this);
-        }
+    local = frame->local;
+    priv = frame->this->private;
 
-        return 0;
-}
+    xattr_req = dict_new();
+    if (!xattr_req)
+        return NULL;
 
+    if (xattr)
+        dict_copy(xattr, xattr_req);
 
-static int
-afr_sh_missing_entries_finish (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
+    if (afr_xattr_req_prepare(frame->this, xattr_req) != 0) {
+        dict_unref(xattr_req);
+        return NULL;
+    }
 
-        local = frame->local;
-        int_lock = &local->internal_lock;
+    inode = inode_new(parent->table);
+    if (!inode) {
+        dict_unref(xattr_req);
+        return NULL;
+    }
 
-        int_lock->lock_cbk = afr_sh_missing_entries_done;
-        afr_unlock (frame, this);
+    loc.parent = inode_ref(parent);
+    gf_uuid_copy(loc.pargfid, parent->gfid);
+    loc.name = name;
+    loc.inode = inode_ref(inode);
 
-        return 0;
-}
+    AFR_ONLIST(lookup_on, frame, afr_selfheal_discover_cbk, lookup, &loc,
+               xattr_req);
 
-int
-afr_sh_common_create (afr_self_heal_t *sh, unsigned int child_count)
-{
-        int     ret = -ENOMEM;
-        sh->buf = GF_CALLOC (child_count, sizeof (*sh->buf),
-                             gf_afr_mt_iatt);
-        if (!sh->buf)
-                goto out;
-        sh->parentbufs = GF_CALLOC (child_count, sizeof (*sh->parentbufs),
-                                    gf_afr_mt_iatt);
-        if (!sh->parentbufs)
-                goto out;
-        sh->child_errno = GF_CALLOC (child_count, sizeof (*sh->child_errno),
-                                     gf_afr_mt_int);
-        if (!sh->child_errno)
-                goto out;
-        sh->success_children = afr_children_create (child_count);
-        if (!sh->success_children)
-                goto out;
-        sh->fresh_children = afr_children_create (child_count);
-        if (!sh->fresh_children)
-                goto out;
-        sh->xattr = GF_CALLOC (child_count, sizeof (*sh->xattr),
-                               gf_afr_mt_dict_t);
-        if (!sh->xattr)
-                goto out;
-        ret = 0;
-out:
-        return ret;
-}
+    afr_replies_copy(replies, local->replies, priv->child_count);
 
-void
-afr_sh_common_lookup_resp_handler (call_frame_t *frame, void *cookie,
-                                   xlator_t *this,
-                                   int32_t op_ret, int32_t op_errno,
-                                   inode_t *inode, struct iatt *buf,
-                                   dict_t *xattr, struct iatt *postparent,
-                                   loc_t *loc)
-{
-        int              child_index = 0;
-        afr_local_t     *local = NULL;
-        afr_private_t   *priv = NULL;
-        afr_self_heal_t *sh = NULL;
-
-        local = frame->local;
-        priv = this->private;
-        sh   = &local->self_heal;
-        child_index = (long) cookie;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == 0) {
-                        sh->buf[child_index] = *buf;
-                        sh->parentbufs[child_index] = *postparent;
-                        sh->success_children[sh->success_count] = child_index;
-                        sh->success_count++;
-                        sh->xattr[child_index] = dict_ref (xattr);
-                } else {
-                        gf_log (this->name, GF_LOG_DEBUG, "path %s on subvolume"
-                                " %s => -1 (%s)", loc->path,
-                                priv->children[child_index]->name,
-                                strerror (op_errno));
-                        local->self_heal.child_errno[child_index] = op_errno;
-                }
-        }
-        UNLOCK (&frame->lock);
-        return;
-}
+    loc_wipe(&loc);
+    dict_unref(xattr_req);
 
-gf_boolean_t
-afr_valid_ia_type (ia_type_t ia_type)
-{
-        switch (ia_type) {
-        case IA_IFSOCK:
-        case IA_IFREG:
-        case IA_IFBLK:
-        case IA_IFCHR:
-        case IA_IFIFO:
-        case IA_IFLNK:
-        case IA_IFDIR:
-                return _gf_true;
-        default:
-                return _gf_false;
-        }
-        return _gf_false;
+    return inode;
 }
 
-int
-afr_impunge_frame_create (call_frame_t *frame, xlator_t *this,
-                          int active_source, call_frame_t **impunge_frame)
+static int
+afr_set_multi_dom_lock_count_request(xlator_t *this, dict_t *dict)
 {
-        afr_local_t     *local         = NULL;
-        afr_local_t     *impunge_local = NULL;
-        afr_self_heal_t *impunge_sh    = NULL;
-        int32_t         op_errno       = 0;
-        afr_private_t   *priv          = NULL;
-        int             ret            = 0;
-        call_frame_t    *new_frame     = NULL;
+    int ret = 0;
+    afr_private_t *priv = NULL;
+    char *key1 = NULL;
+    char *key2 = NULL;
 
-        op_errno = ENOMEM;
-        priv = this->private;
-        new_frame = copy_frame (frame);
-        if (!new_frame) {
-                goto out;
-        }
+    priv = this->private;
+    key1 = alloca0(strlen(GLUSTERFS_INODELK_DOM_PREFIX) + 2 +
+                   strlen(this->name));
+    key2 = alloca0(strlen(GLUSTERFS_INODELK_DOM_PREFIX) + 2 +
+                   strlen(priv->sh_domain));
 
-        AFR_LOCAL_ALLOC_OR_GOTO (impunge_local, out);
-
-        local = frame->local;
-        new_frame->local = impunge_local;
-        impunge_sh = &impunge_local->self_heal;
-        impunge_sh->sh_frame = frame;
-        impunge_sh->active_source = active_source;
-        impunge_local->child_up  = memdup (local->child_up,
-                                           sizeof (*local->child_up) *
-                                           priv->child_count);
-        if (!impunge_local->child_up)
-                goto out;
+    ret = dict_set_uint32(dict, GLUSTERFS_MULTIPLE_DOM_LK_CNT_REQUESTS, 1);
+    if (ret)
+        return ret;
 
-        impunge_local->pending = afr_matrix_create (priv->child_count,
-                                                    AFR_NUM_CHANGE_LOGS);
-        if (!impunge_local->pending)
-                goto out;
+    sprintf(key1, "%s:%s", GLUSTERFS_INODELK_DOM_PREFIX, this->name);
+    ret = dict_set_uint32(dict, key1, 1);
+    if (ret)
+        return ret;
 
-        ret = afr_sh_common_create (impunge_sh, priv->child_count);
-        if (ret) {
-                op_errno = -ret;
-                goto out;
-        }
-        op_errno = 0;
-        *impunge_frame = new_frame;
-out:
-        if (op_errno && new_frame)
-                AFR_STACK_DESTROY (new_frame);
-        return -op_errno;
-}
+    sprintf(key2, "%s:%s", GLUSTERFS_INODELK_DOM_PREFIX, priv->sh_domain);
+    ret = dict_set_uint32(dict, key2, 1);
+    if (ret)
+        return ret;
 
-void
-afr_sh_missing_entry_call_impunge_recreate (call_frame_t *frame, xlator_t *this,
-                                            struct iatt *buf,
-                                            struct iatt *postparent,
-                                            afr_impunge_done_cbk_t impunge_done)
-{
-        call_frame_t    *impunge_frame = NULL;
-        afr_local_t     *local = NULL;
-        afr_local_t     *impunge_local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_self_heal_t *impunge_sh = NULL;
-        int             ret = 0;
-        unsigned int    enoent_count = 0;
-        afr_private_t   *priv = NULL;
-        int             i = 0;
-        int32_t         op_errno = 0;
-
-        local = frame->local;
-        sh    = &local->self_heal;
-        priv  = this->private;
-
-        enoent_count = afr_errno_count (NULL, sh->child_errno,
-                                        priv->child_count, ENOENT);
-        if (!enoent_count) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "no missing files - %s. proceeding to metadata check",
-                        local->loc.path);
-                goto out;
-        }
-        sh->impunge_done = impunge_done;
-        ret = afr_impunge_frame_create (frame, this, sh->source, &impunge_frame);
-        if (ret)
-                goto out;
-        impunge_local = impunge_frame->local;
-        impunge_sh    = &impunge_local->self_heal;
-        loc_copy (&impunge_local->loc, &local->loc);
-        ret = afr_build_parent_loc (&impunge_sh->parent_loc,
-                                    &impunge_local->loc, &op_errno);
-        if (ret) {
-                ret = -op_errno;
-                goto out;
-        }
-        impunge_local->call_count = enoent_count;
-        impunge_sh->entrybuf = sh->buf[sh->source];
-        impunge_sh->parentbuf = sh->parentbufs[sh->source];
-        for (i = 0; i < priv->child_count; i++) {
-                if (!impunge_local->child_up[i]) {
-                        impunge_sh->child_errno[i] = ENOTCONN;
-                        continue;
-                }
-                if (sh->child_errno[i] != ENOENT) {
-                        impunge_sh->child_errno[i] = EEXIST;
-                        continue;
-                }
-        }
-        for (i = 0; i < priv->child_count; i++) {
-                if (sh->child_errno[i] != ENOENT)
-                        continue;
-                afr_sh_entry_impunge_create (impunge_frame, this, i);
-                enoent_count--;
-        }
-        GF_ASSERT (!enoent_count);
-        return;
-out:
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "impunge of %s failed, "
-                        "reason: %s", local->loc.path, strerror (-ret));
-                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-        }
-        afr_sh_missing_entries_finish (frame, this);
+    return 0;
 }
 
 int
-afr_sh_create_entry_cbk (call_frame_t *frame, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
+afr_selfheal_unlocked_discover_on(call_frame_t *frame, inode_t *inode,
+                                  uuid_t gfid, struct afr_reply *replies,
+                                  unsigned char *discover_on, dict_t *dict)
+{
+    loc_t loc = {
+        0,
+    };
+    dict_t *xattr_req = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = frame->this->private;
+
+    xattr_req = dict_new();
+    if (!xattr_req)
+        return -ENOMEM;
+    if (dict)
+        dict_copy(dict, xattr_req);
+
+    if (afr_xattr_req_prepare(frame->this, xattr_req) != 0) {
+        dict_unref(xattr_req);
+        return -ENOMEM;
+    }
+
+    if (afr_set_multi_dom_lock_count_request(frame->this, xattr_req)) {
+        dict_unref(xattr_req);
+        return -1;
+    }
 
-        local = frame->local;
-        sh = &local->self_heal;
-        if (op_ret < 0)
-                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-        afr_sh_missing_entries_finish (frame, this);
-        return 0;
-}
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, gfid);
 
-static int
-sh_missing_entries_create (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        int              type = 0;
-        struct iatt     *buf = NULL;
-        struct iatt     *postparent = NULL;
-
-        local = frame->local;
-        sh = &local->self_heal;
-
-        buf = &sh->buf[sh->source];
-        postparent = &sh->parentbufs[sh->source];
-
-        type = buf->ia_type;
-        if (!afr_valid_ia_type (type)) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s: unknown file type: 0%o", local->loc.path, type);
-                afr_set_local_for_unhealable (local);
-                afr_sh_missing_entries_finish (frame, this);
-                goto out;
-        }
+    AFR_ONLIST(discover_on, frame, afr_selfheal_discover_cbk, lookup, &loc,
+               xattr_req);
 
-        afr_sh_missing_entry_call_impunge_recreate (frame, this,
-                                                    buf, postparent,
-                                                    afr_sh_create_entry_cbk);
-out:
-        return 0;
-}
+    afr_replies_copy(replies, local->replies, priv->child_count);
 
-void
-afr_sh_missing_entries_lookup_done (call_frame_t *frame, xlator_t *this,
-                                    int32_t op_ret, int32_t op_errno)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        ia_type_t       ia_type = IA_INVAL;
-        int32_t         nsources = 0;
-        loc_t           *loc = NULL;
-        int32_t         subvol_status = 0;
-        afr_transaction_type txn_type = AFR_DATA_TRANSACTION;
-        gf_boolean_t    split_brain = _gf_false;
-        int             read_child = -1;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-        loc = &local->loc;
-
-        if (op_ret < 0) {
-                if (op_errno == EIO) {
-                        afr_set_local_for_unhealable (local);
-                }
-                // EIO can happen if finding the fresh parent dir failed
-                goto out;
-        }
+    loc_wipe(&loc);
+    dict_unref(xattr_req);
 
-        //now No chance for the ia_type to conflict
-        ia_type = sh->buf[sh->success_children[0]].ia_type;
-        txn_type = afr_transaction_type_get (ia_type);
-        nsources = afr_build_sources (this, sh->xattr, sh->buf,
-                                      sh->pending_matrix, sh->sources,
-                                      sh->success_children, txn_type,
-                                      &subvol_status, _gf_false);
-        if (nsources < 0) {
-                gf_log (this->name, GF_LOG_INFO, "No sources for dir of %s,"
-                        " in missing entry self-heal, continuing with the rest"
-                        " of the self-heals", local->loc.path);
-                if (subvol_status & SPLIT_BRAIN) {
-                        split_brain = _gf_true;
-                        switch (txn_type) {
-                        case AFR_DATA_TRANSACTION:
-                                nsources = 1;
-                                sh->sources[sh->success_children[0]] = 1;
-                                break;
-                        case AFR_ENTRY_TRANSACTION:
-                                read_child = afr_get_no_xattr_dir_read_child
-                                                          (this,
-                                                           sh->success_children,
-                                                           sh->buf);
-                                sh->sources[read_child] = 1;
-                                nsources = 1;
-                                break;
-                        default:
-                                op_errno = EIO;
-                                goto out;
-                        }
-                } else {
-                        op_errno = EIO;
-                        goto out;
-                }
-        }
+    return 0;
+}
 
-        afr_get_fresh_children (sh->success_children, sh->sources,
-                                sh->fresh_children, priv->child_count);
-        sh->source = sh->fresh_children[0];
-        if (sh->source == -1) {
-                gf_log (this->name, GF_LOG_DEBUG, "No active sources found.");
-                op_errno = EIO;
-                goto out;
-        }
+int
+afr_selfheal_unlocked_discover(call_frame_t *frame, inode_t *inode, uuid_t gfid,
+                               struct afr_reply *replies)
+{
+    afr_local_t *local = NULL;
+    dict_t *dict = NULL;
 
-        if (sh->gfid_sh_success_cbk)
-                sh->gfid_sh_success_cbk (frame, this);
-        sh->type = sh->buf[sh->source].ia_type;
-        if (uuid_is_null (loc->inode->gfid))
-                uuid_copy (loc->gfid, sh->buf[sh->source].ia_gfid);
-        if (split_brain) {
-                afr_sh_missing_entries_finish (frame, this);
-        } else {
-                sh_missing_entries_create (frame, this);
-        }
-        return;
-out:
-        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-        afr_sh_set_error (sh, op_errno);
-        afr_sh_missing_entries_finish (frame, this);
-        return;
-}
+    local = frame->local;
 
-static int
-afr_sh_common_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                          int32_t op_ret, int32_t op_errno, inode_t *inode,
-                          struct iatt *buf, dict_t *xattr,
-                          struct iatt *postparent)
-{
-        int                     call_count = 0;
-        afr_local_t             *local = NULL;
-        afr_self_heal_t         *sh    = NULL;
-        afr_private_t           *priv  = NULL;
-
-        local = frame->local;
-        sh    = &local->self_heal;
-        priv  = this->private;
-
-        afr_sh_common_lookup_resp_handler (frame, cookie, this, op_ret,
-                                           op_errno, inode, buf, xattr,
-                                           postparent, &sh->lookup_loc);
-        call_count = afr_frame_return (frame);
-
-        if (call_count)
-                goto out;
-        op_ret = -1;
-        if (!sh->success_count) {
-                op_errno = afr_resultant_errno_get (NULL, sh->child_errno,
-                                                    priv->child_count);
-                gf_log (this->name, GF_LOG_ERROR, "Failed to lookup %s, "
-                        "reason %s", sh->lookup_loc.path,
-                        strerror (op_errno));
-                goto done;
-        }
+    if (local->xattr_req)
+        dict = local->xattr_req;
 
-        if ((sh->lookup_flags & AFR_LOOKUP_FAIL_CONFLICTS) &&
-            (afr_conflicting_iattrs (sh->buf, sh->success_children,
-                                     priv->child_count,
-                                     sh->lookup_loc.path, this->name))) {
-                op_errno = EIO;
-                gf_log (this->name, GF_LOG_ERROR, "Conflicting entries "
-                        "for %s", sh->lookup_loc.path);
-                goto done;
-        }
+    return afr_selfheal_unlocked_discover_on(frame, inode, gfid, replies,
+                                             local->child_up, dict);
+}
 
-        if ((sh->lookup_flags & AFR_LOOKUP_FAIL_MISSING_GFIDS) &&
-            (afr_gfid_missing_count (this->name, sh->success_children,
-                                     sh->buf, priv->child_count,
-                                     sh->lookup_loc.path))) {
-                op_errno = ENODATA;
-                gf_log (this->name, GF_LOG_ERROR, "Missing Gfids "
-                        "for %s", sh->lookup_loc.path);
-                goto done;
-        }
-        op_ret = 0;
+unsigned int
+afr_success_count(struct afr_reply *replies, unsigned int count)
+{
+    int i = 0;
+    unsigned int success = 0;
 
-done:
-        sh->lookup_done (frame, this, op_ret, op_errno);
-out:
-        return 0;
+    for (i = 0; i < count; i++)
+        if (replies[i].valid && replies[i].op_ret == 0)
+            success++;
+    return success;
 }
 
 int
-afr_sh_remove_entry_cbk (call_frame_t *frame, xlator_t *this, int child,
-                         int32_t op_ret, int32_t op_errno)
-{
-        int             call_count = 0;
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-
-        local = frame->local;
-        sh = &local->self_heal;
-
-        GF_ASSERT (sh->post_remove_call);
-        if ((op_ret == -1) && (op_errno != ENOENT)) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "purge entry %s failed, on child %d reason, %s",
-                        local->loc.path, child, strerror (op_errno));
-                LOCK (&frame->lock);
-                {
-                        afr_sh_set_error (sh, EIO);
-                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-                }
-                UNLOCK (&frame->lock);
-        }
-        call_count = afr_frame_return (frame);
-        if (call_count == 0)
-                sh->post_remove_call (frame, this);
-        return 0;
-}
+afr_selfheal_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int i = 0;
 
-void
-afr_sh_call_entry_expunge_remove (call_frame_t *frame, xlator_t *this,
-                                  int child_index, struct iatt *buf,
-                                  struct iatt *parentbuf,
-                                  afr_expunge_done_cbk_t expunge_done)
-{
-        call_frame_t    *expunge_frame = NULL;
-        afr_local_t     *local = NULL;
-        afr_local_t     *expunge_local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_self_heal_t *expunge_sh = NULL;
-        int32_t         op_errno = 0;
-        int             ret = 0;
-
-        expunge_frame = copy_frame (frame);
-        if (!expunge_frame) {
-                goto out;
-        }
+    local = frame->local;
+    i = (long)cookie;
 
-        AFR_LOCAL_ALLOC_OR_GOTO (expunge_local, out);
+    local->replies[i].valid = 1;
+    local->replies[i].op_ret = op_ret;
+    local->replies[i].op_errno = op_errno;
 
-        local = frame->local;
-        sh = &local->self_heal;
-        expunge_frame->local = expunge_local;
-        expunge_sh = &expunge_local->self_heal;
-        expunge_sh->sh_frame = frame;
-        loc_copy (&expunge_local->loc, &local->loc);
-        ret = afr_build_parent_loc (&expunge_sh->parent_loc,
-                                    &expunge_local->loc, &op_errno);
-        if (ret) {
-                ret = -op_errno;
-                goto out;
-        }
-        sh->expunge_done = expunge_done;
-        afr_sh_entry_expunge_remove (expunge_frame, this, child_index, buf,
-                                     parentbuf);
-        return;
-out:
-        gf_log (this->name, GF_LOG_ERROR, "Expunge of %s failed, reason: %s",
-                local->loc.path, strerror (op_errno));
-        expunge_done (frame, this, child_index, -1, op_errno);
-}
+    syncbarrier_wake(&local->barrier);
 
-void
-afr_sh_remove_stale_lookup_info (afr_self_heal_t *sh, int32_t *success_children,
-                                 int32_t *fresh_children,
-                                 unsigned int child_count)
-{
-        int     i = 0;
-
-        for (i = 0; i < child_count; i++) {
-                if (afr_is_child_present (success_children, child_count, i) &&
-                    !afr_is_child_present (fresh_children, child_count, i)) {
-                        sh->child_errno[i] = ENOENT;
-                        GF_ASSERT (sh->xattr[i]);
-                        dict_unref (sh->xattr[i]);
-                        sh->xattr[i] = NULL;
-                }
-        }
+    return 0;
 }
 
 int
-afr_sh_purge_stale_entries_done (call_frame_t *frame, xlator_t *this)
+afr_locked_fill(call_frame_t *frame, xlator_t *this, unsigned char *locked_on)
 {
-        afr_local_t         *local    = NULL;
-        afr_self_heal_t     *sh       = NULL;
-        afr_private_t       *priv     = NULL;
+    int i = 0;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int count = 0;
 
-        local    = frame->local;
-        sh       = &local->self_heal;
-        priv     = this->private;
+    local = frame->local;
+    priv = this->private;
 
-        if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
-                afr_sh_missing_entries_finish (frame, this);
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->replies[i].valid && local->replies[i].op_ret == 0) {
+            locked_on[i] = 1;
+            count++;
         } else {
-                if (afr_gfid_missing_count (this->name, sh->fresh_children,
-                                            sh->buf, priv->child_count,
-                                            local->loc.path)) {
-                        afr_sh_common_lookup (frame, this, &local->loc,
-                                              afr_sh_missing_entries_lookup_done,
-                                              sh->sh_gfid_req,
-                                              AFR_LOOKUP_FAIL_CONFLICTS|
-                                              AFR_LOOKUP_FAIL_MISSING_GFIDS,
-                                              NULL);
-                } else {
-                        //No need to set gfid so goto missing entries lookup done
-                        //Behave as if you have done the lookup
-                        afr_sh_remove_stale_lookup_info (sh,
-                                                         sh->success_children,
-                                                         sh->fresh_children,
-                                                         priv->child_count);
-                        afr_children_copy (sh->success_children,
-                                           sh->fresh_children,
-                                           priv->child_count);
-                        afr_sh_missing_entries_lookup_done (frame, this, 0, 0);
-                }
+            locked_on[i] = 0;
         }
-        return 0;
+    }
+
+    return count;
 }
 
-gf_boolean_t
-afr_sh_purge_entry_condition (afr_local_t *local, afr_private_t *priv,
-                              int child)
+int
+afr_selfheal_tryinodelk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                        char *dom, off_t off, size_t size,
+                        unsigned char *locked_on)
 {
-        afr_self_heal_t *sh = NULL;
+    loc_t loc = {
+        0,
+    };
+    struct gf_flock flock = {
+        0,
+    };
 
-        sh = &local->self_heal;
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-        if (local->child_up[child] &&
-            (!afr_is_child_present (sh->fresh_parent_dirs, priv->child_count,
-                                    child))
-            && (sh->child_errno[child] != ENOENT))
-                return _gf_true;
+    flock.l_type = F_WRLCK;
+    flock.l_start = off;
+    flock.l_len = size;
 
-        return _gf_false;
+    AFR_ONALL(frame, afr_selfheal_lock_cbk, inodelk, dom, &loc, F_SETLK, &flock,
+              NULL);
+
+    loc_wipe(&loc);
+
+    return afr_locked_fill(frame, this, locked_on);
 }
 
-gf_boolean_t
-afr_sh_purge_stale_entry_condition (afr_local_t *local, afr_private_t *priv,
-                                    int child)
+int
+afr_selfheal_inodelk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                     char *dom, off_t off, size_t size,
+                     unsigned char *locked_on)
 {
-        afr_self_heal_t *sh = NULL;
+    loc_t loc = {
+        0,
+    };
+    struct gf_flock flock = {
+        0,
+    };
+    afr_local_t *local = NULL;
+    int i = 0;
+    afr_private_t *priv = NULL;
 
-        sh = &local->self_heal;
+    priv = this->private;
+    local = frame->local;
 
-        if (local->child_up[child] &&
-            (!afr_is_child_present (sh->fresh_children, priv->child_count,
-                                    child))
-             && (sh->child_errno[child] != ENOENT))
-                return _gf_true;
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-        return _gf_false;
-}
+    flock.l_type = F_WRLCK;
+    flock.l_start = off;
+    flock.l_len = size;
 
-void
-afr_sh_purge_entry_common (call_frame_t *frame, xlator_t *this,
-                           gf_boolean_t purge_condition (afr_local_t *local,
-                                                         afr_private_t *priv,
-                                                         int child))
-{
-        afr_local_t     *local = NULL;
-        afr_private_t   *priv = NULL;
-        afr_self_heal_t *sh = NULL;
-        int             i = 0;
-        int             call_count = 0;
+    AFR_ONALL(frame, afr_selfheal_lock_cbk, inodelk, dom, &loc, F_SETLK, &flock,
+              NULL);
 
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->replies[i].op_ret == -1 &&
+            local->replies[i].op_errno == EAGAIN) {
+            afr_locked_fill(frame, this, locked_on);
+            afr_selfheal_uninodelk(frame, this, inode, dom, off, size,
+                                   locked_on);
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (purge_condition (local, priv, i))
-                        call_count++;
+            AFR_SEQ(frame, afr_selfheal_lock_cbk, inodelk, dom, &loc, F_SETLKW,
+                    &flock, NULL);
+            break;
         }
+    }
 
-        if (call_count == 0) {
-                sh->post_remove_call (frame, this);
-                goto out;
-        }
+    loc_wipe(&loc);
 
-        local->call_count = call_count;
-        for (i = 0; i < priv->child_count; i++) {
-                if (!purge_condition (local, priv, i))
-                        continue;
-                gf_log (this->name, GF_LOG_INFO, "purging the stale entry %s "
-                        "on %s", local->loc.path, priv->children[i]->name);
-                afr_sh_call_entry_expunge_remove (frame, this,
-                                                  (long) i, &sh->buf[i],
-                                                  &sh->parentbufs[i],
-                                                  afr_sh_remove_entry_cbk);
-        }
-out:
-        return;
+    return afr_locked_fill(frame, this, locked_on);
 }
 
-void
-afr_sh_purge_entry (call_frame_t *frame, xlator_t *this)
+static void
+afr_get_lock_and_eagain_counts(afr_private_t *priv, struct afr_reply *replies,
+                               int *lock_count, int *eagain_count)
 {
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        sh->post_remove_call = afr_sh_missing_entries_finish;
+    int i = 0;
 
-        afr_sh_purge_entry_common (frame, this, afr_sh_purge_entry_condition);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid)
+            continue;
+        if (replies[i].op_ret == 0) {
+            (*lock_count)++;
+        } else if (replies[i].op_ret == -1 && replies[i].op_errno == EAGAIN) {
+            (*eagain_count)++;
+        }
+    }
 }
 
-void
-afr_sh_purge_stale_entry (call_frame_t *frame, xlator_t *this)
+/*Do blocking locks if number of locks acquired is majority and there were some
+ * EAGAINs. Useful for odd-way replication*/
+int
+afr_selfheal_tie_breaker_inodelk(call_frame_t *frame, xlator_t *this,
+                                 inode_t *inode, char *dom, off_t off,
+                                 size_t size, unsigned char *locked_on)
 {
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int             i = 0;
+    loc_t loc = {
+        0,
+    };
+    struct gf_flock flock = {
+        0,
+    };
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int lock_count = 0;
+    int eagain_count = 0;
 
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
+    priv = this->private;
+    local = frame->local;
 
-        sh->post_remove_call = afr_sh_purge_stale_entries_done;
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (afr_is_child_present (sh->fresh_children,
-                                          priv->child_count, i))
-                        continue;
+    flock.l_type = F_WRLCK;
+    flock.l_start = off;
+    flock.l_len = size;
 
-                if ((!local->child_up[i]) || sh->child_errno[i] != 0)
-                        continue;
+    AFR_ONALL(frame, afr_selfheal_lock_cbk, inodelk, dom, &loc, F_SETLK, &flock,
+              NULL);
 
-                GF_ASSERT (!uuid_is_null (sh->entrybuf.ia_gfid) ||
-                           uuid_is_null (sh->buf[i].ia_gfid));
+    afr_get_lock_and_eagain_counts(priv, local->replies, &lock_count,
+                                   &eagain_count);
 
-                if ((sh->entrybuf.ia_type != sh->buf[i].ia_type) ||
-                    (uuid_compare (sh->buf[i].ia_gfid,
-                                   sh->entrybuf.ia_gfid)))
-                        continue;
+    if (lock_count > priv->child_count / 2 && eagain_count) {
+        afr_locked_fill(frame, this, locked_on);
+        afr_selfheal_uninodelk(frame, this, inode, dom, off, size, locked_on);
 
-                afr_children_add_child (sh->fresh_children, i,
-                                        priv->child_count);
+        AFR_SEQ(frame, afr_selfheal_lock_cbk, inodelk, dom, &loc, F_SETLKW,
+                &flock, NULL);
+    }
 
-        }
-        afr_sh_purge_entry_common (frame, this,
-                                   afr_sh_purge_stale_entry_condition);
-}
+    loc_wipe(&loc);
 
-void
-afr_sh_save_child_iatts_from_policy (int32_t *children, struct iatt *bufs,
-                                     struct iatt *save,
-                                     unsigned int child_count)
-{
-        int             i = 0;
-        int             child = 0;
-        gf_boolean_t    saved = _gf_false;
-
-        GF_ASSERT (save);
-        //if iatt buf with gfid exists sets it
-        for (i = 0; i < child_count; i++) {
-                child = children[i];
-                if (child == -1)
-                        break;
-                *save = bufs[child];
-                saved = _gf_true;
-                if (!uuid_is_null (save->ia_gfid))
-                        break;
-        }
-        GF_ASSERT (saved);
+    return afr_locked_fill(frame, this, locked_on);
 }
 
-void
-afr_get_children_of_fresh_parent_dirs (afr_self_heal_t *sh,
-                                       unsigned int child_count)
+int
+afr_selfheal_uninodelk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                       char *dom, off_t off, size_t size,
+                       const unsigned char *locked_on)
 {
-        afr_children_intersection_get (sh->success_children,
-                                       sh->fresh_parent_dirs,
-                                       sh->sources, child_count);
-        afr_get_fresh_children (sh->success_children, sh->sources,
-                                sh->fresh_children, child_count);
-        memset (sh->sources, 0, sizeof (*sh->sources) * child_count);
-}
+    loc_t loc = {
+        0,
+    };
+    struct gf_flock flock = {
+        0,
+    };
 
-void
-afr_sh_children_lookup_done (call_frame_t *frame, xlator_t *this,
-                             int32_t op_ret, int32_t op_errno)
-{
-        afr_local_t      *local = NULL;
-        afr_self_heal_t  *sh = NULL;
-        afr_private_t    *priv = NULL;
-        int32_t          fresh_child_enoents = 0;
-        int32_t          fresh_parent_count = 0;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        if (op_ret < 0)
-                goto fail;
-        afr_get_children_of_fresh_parent_dirs (sh, priv->child_count);
-        fresh_parent_count = afr_get_children_count (sh->fresh_parent_dirs,
-                                                     priv->child_count);
-        //we need the enoent count of the subvols present in fresh_parent_dirs
-        fresh_child_enoents = afr_errno_count (sh->fresh_parent_dirs,
-                                               sh->child_errno,
-                                               priv->child_count, ENOENT);
-        if (fresh_child_enoents == fresh_parent_count) {
-                afr_sh_set_error (sh, ENOENT);
-                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-                afr_sh_purge_entry (frame, this);
-        } else if (!afr_conflicting_iattrs (sh->buf, sh->fresh_children,
-                                            priv->child_count, local->loc.path,
-                                            this->name)) {
-                afr_sh_save_child_iatts_from_policy (sh->fresh_children,
-                                                     sh->buf, &sh->entrybuf,
-                                                     priv->child_count);
-                afr_update_gfid_from_iatts (sh->sh_gfid_req, sh->buf,
-                                            sh->fresh_children,
-                                            priv->child_count);
-                afr_sh_purge_stale_entry (frame, this);
-        } else {
-                op_errno = EIO;
-                afr_set_local_for_unhealable (local);
-                goto fail;
-        }
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-        return;
+    flock.l_type = F_UNLCK;
+    flock.l_start = off;
+    flock.l_len = size;
 
-fail:
-        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-        afr_sh_set_error (sh, op_errno);
-        afr_sh_missing_entries_finish (frame, this);
-        return;
-}
+    AFR_ONLIST(locked_on, frame, afr_selfheal_lock_cbk, inodelk, dom, &loc,
+               F_SETLK, &flock, NULL);
 
-static void
-afr_sh_find_fresh_parents (call_frame_t *frame, xlator_t *this,
-                           int32_t op_ret, int32_t op_errno)
-{
-        afr_self_heal_t *sh  = NULL;
-        afr_private_t   *priv = NULL;
-        afr_local_t     *local = NULL;
-        int             enoent_count = 0;
-        int             nsources = 0;
-        int             source  = -1;
-        int32_t         subvol_status = 0;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        if (op_ret < 0)
-                goto out;
-        enoent_count = afr_errno_count (NULL, sh->child_errno,
-                                        priv->child_count, ENOENT);
-        if (enoent_count > 0) {
-                gf_log (this->name, GF_LOG_INFO, "Parent dir missing for %s,"
-                        " in missing entry self-heal, aborting missing-entry "
-                        "self-heal",
-                        local->loc.path);
-                afr_sh_missing_entries_finish (frame, this);
-                return;
-        }
-
-        nsources = afr_build_sources (this, sh->xattr, sh->buf,
-                                      sh->pending_matrix, sh->sources,
-                                      sh->success_children,
-                                      AFR_ENTRY_TRANSACTION, &subvol_status,
-                                      _gf_true);
-        if ((subvol_status & ALL_FOOLS) ||
-            (subvol_status & SPLIT_BRAIN)) {
-                gf_log (this->name, GF_LOG_INFO, "%s: Performing conservative "
-                        "merge", sh->parent_loc.path);
-                afr_mark_success_children_sources (sh->sources,
-                                                   sh->success_children,
-                                                   priv->child_count);
-        } else if (nsources < 0) {
-                gf_log (this->name, GF_LOG_ERROR, "No sources for dir "
-                        "of %s, in missing entry self-heal, aborting "
-                        "self-heal", local->loc.path);
-                op_errno = EIO;
-                goto out;
-        }
-
-        source = afr_sh_select_source (sh->sources, priv->child_count);
-        if (source == -1) {
-                GF_ASSERT (0);
-                gf_log (this->name, GF_LOG_DEBUG, "No active sources found.");
-                op_errno = EIO;
-                goto out;
-        }
-        afr_get_fresh_children (sh->success_children, sh->sources,
-                                sh->fresh_parent_dirs, priv->child_count);
-        afr_sh_common_lookup (frame, this, &local->loc,
-                              afr_sh_children_lookup_done, NULL, 0,
-                              NULL);
-        return;
+    loc_wipe(&loc);
 
-out:
-        afr_sh_set_error (sh, op_errno);
-        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-	afr_sh_missing_entries_finish (frame, this);
-        return;
+    return 0;
 }
 
-void
-afr_sh_common_reset (afr_self_heal_t *sh, unsigned int child_count)
+int
+afr_selfheal_tryentrylk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                        char *dom, const char *name, unsigned char *locked_on)
 {
-        int             i = 0;
+    loc_t loc = {
+        0,
+    };
 
-        for (i = 0; i < child_count; i++) {
-                memset (&sh->buf[i], 0, sizeof (sh->buf[i]));
-                memset (&sh->parentbufs[i], 0, sizeof (sh->parentbufs[i]));
-                sh->child_errno[i] = 0;
-        }
-        memset (&sh->parentbuf, 0, sizeof (sh->parentbuf));
-        sh->success_count = 0;
-        afr_reset_children (sh->success_children, child_count);
-        afr_reset_children (sh->fresh_children, child_count);
-        afr_reset_xattr (sh->xattr, child_count);
-        loc_wipe (&sh->lookup_loc);
-}
-
-/* afr self-heal state will be lost if this call is made
- * please check the afr_sh_common_reset that is called in this function
- */
-int
-afr_sh_common_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                      afr_lookup_done_cbk_t lookup_done , uuid_t gfid,
-                      int32_t flags, dict_t *xdata)
-{
-        afr_local_t    *local = NULL;
-        int             i = 0;
-        int             call_count = 0;
-        afr_private_t  *priv = NULL;
-        dict_t         *xattr_req = NULL;
-        afr_self_heal_t *sh = NULL;
-
-        local = frame->local;
-        priv  = this->private;
-        sh    = &local->self_heal;
-
-        call_count = afr_up_children_count (local->child_up, priv->child_count);
-
-        local->call_count = call_count;
-
-        xattr_req = dict_new();
-
-        if (xattr_req) {
-                afr_xattr_req_prepare (this, xattr_req, loc->path);
-                if (gfid) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "looking up %s with gfid: %s",
-                                loc->path, uuid_utoa (gfid));
-                        GF_ASSERT (!uuid_is_null (gfid));
-                        afr_set_dict_gfid (xattr_req, gfid);
-                }
-        }
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-        afr_sh_common_reset (sh, priv->child_count);
-        sh->lookup_done = lookup_done;
-        loc_copy (&sh->lookup_loc, loc);
-        sh->lookup_flags = flags;
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "looking up %s on subvolume %s",
-                                loc->path, priv->children[i]->name);
-
-                        STACK_WIND_COOKIE (frame,
-                                           afr_sh_common_lookup_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->lookup,
-                                           loc, xattr_req);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
+    AFR_ONALL(frame, afr_selfheal_lock_cbk, entrylk, dom, &loc, name,
+              ENTRYLK_LOCK_NB, ENTRYLK_WRLCK, NULL);
 
-        if (xattr_req)
-                dict_unref (xattr_req);
+    loc_wipe(&loc);
 
-        return 0;
+    return afr_locked_fill(frame, this, locked_on);
 }
 
+int
+afr_selfheal_entrylk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                     char *dom, const char *name, unsigned char *locked_on)
+{
+    loc_t loc = {
+        0,
+    };
+    afr_local_t *local = NULL;
+    int i = 0;
+    afr_private_t *priv = NULL;
 
+    priv = this->private;
+    local = frame->local;
 
-int
-afr_sh_post_nb_entrylk_missing_entry_sh_cbk (call_frame_t *frame,
-                                             xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_self_heal_t     *sh       = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        sh       = &local->self_heal;
-
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "Non blocking entrylks failed.");
-                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-                afr_sh_missing_entries_done (frame, this);
-        } else {
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Non blocking entrylks done. Proceeding to FOP");
-                afr_sh_common_lookup (frame, this, &sh->parent_loc,
-                                      afr_sh_find_fresh_parents,
-                                      NULL, AFR_LOOKUP_FAIL_CONFLICTS,
-                                      NULL);
+    AFR_ONALL(frame, afr_selfheal_lock_cbk, entrylk, dom, &loc, name,
+              ENTRYLK_LOCK_NB, ENTRYLK_WRLCK, NULL);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->replies[i].op_ret == -1 &&
+            local->replies[i].op_errno == EAGAIN) {
+            afr_locked_fill(frame, this, locked_on);
+            afr_selfheal_unentrylk(frame, this, inode, dom, name, locked_on,
+                                   NULL);
+
+            AFR_SEQ(frame, afr_selfheal_lock_cbk, entrylk, dom, &loc, name,
+                    ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL);
+            break;
         }
+    }
 
-        return 0;
+    loc_wipe(&loc);
+
+    return afr_locked_fill(frame, this, locked_on);
 }
 
 int
-afr_sh_entrylk (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                char *base_name, afr_lock_cbk_t lock_cbk)
+afr_selfheal_tie_breaker_entrylk(call_frame_t *frame, xlator_t *this,
+                                 inode_t *inode, char *dom, const char *name,
+                                 unsigned char *locked_on)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_private_t       *priv     = NULL;
+    loc_t loc = {
+        0,
+    };
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int lock_count = 0;
+    int eagain_count = 0;
 
-        priv     = this->private;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+    priv = this->private;
+    local = frame->local;
 
-        int_lock->transaction_lk_type = AFR_SELFHEAL_LK;
-        int_lock->selfheal_lk_type    = AFR_ENTRY_SELF_HEAL_LK;
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-        afr_set_lock_number (frame, this);
+    AFR_ONALL(frame, afr_selfheal_lock_cbk, entrylk, dom, &loc, name,
+              ENTRYLK_LOCK_NB, ENTRYLK_WRLCK, NULL);
 
-        int_lock->lk_basename = base_name;
-        int_lock->lk_loc      = loc;
-        int_lock->lock_cbk    = lock_cbk;
-        int_lock->domain      = this->name;
+    afr_get_lock_and_eagain_counts(priv, local->replies, &lock_count,
+                                   &eagain_count);
 
-        int_lock->lockee_count = 0;
-        afr_init_entry_lockee (&int_lock->lockee[0], local, loc,
-                               base_name, priv->child_count);
-        int_lock->lockee_count++;
-        afr_nonblocking_entrylk (frame, this);
+    if (lock_count > priv->child_count / 2 && eagain_count) {
+        afr_locked_fill(frame, this, locked_on);
+        afr_selfheal_unentrylk(frame, this, inode, dom, name, locked_on, NULL);
 
-        return 0;
+        AFR_SEQ(frame, afr_selfheal_lock_cbk, entrylk, dom, &loc, name,
+                ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL);
+    }
+
+    loc_wipe(&loc);
+
+    return afr_locked_fill(frame, this, locked_on);
 }
 
-static int
-afr_self_heal_parent_entrylk (call_frame_t *frame, xlator_t *this,
-                              afr_lock_cbk_t lock_cbk)
+int
+afr_selfheal_unentrylk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                       char *dom, const char *name, unsigned char *locked_on,
+                       dict_t *xdata)
 {
-        afr_local_t         *local    = NULL;
-        afr_self_heal_t     *sh       = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-        int                 ret       = -1;
-        int32_t             op_errno  = 0;
+    loc_t loc = {
+        0,
+    };
 
-        local    = frame->local;
-        sh       = &local->self_heal;
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-        gf_log (this->name, GF_LOG_TRACE,
-                "attempting to recreate missing entries for path=%s",
-                local->loc.path);
+    AFR_ONLIST(locked_on, frame, afr_selfheal_lock_cbk, entrylk, dom, &loc,
+               name, ENTRYLK_UNLOCK, ENTRYLK_WRLCK, xdata);
 
-        ret = afr_build_parent_loc (&sh->parent_loc, &local->loc, &op_errno);
-        if (ret)
-                goto out;
+    loc_wipe(&loc);
 
-        afr_sh_entrylk (frame, this, &sh->parent_loc, NULL,
-                        lock_cbk);
-        return 0;
-out:
-        int_lock = &local->internal_lock;
-        int_lock->lock_op_ret = -1;
-        lock_cbk (frame, this);
-        return 0;
+    return 0;
 }
 
-static int
-afr_self_heal_missing_entries (call_frame_t *frame, xlator_t *this)
+gf_boolean_t
+afr_is_data_set(xlator_t *this, dict_t *xdata)
 {
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-
-        local = frame->local;
-        sh = &local->self_heal;
-
-        sh->sh_type_in_action  = AFR_SELF_HEAL_GFID_OR_MISSING_ENTRY;
-
-        afr_set_self_heal_status (sh, AFR_SELF_HEAL_STARTED);
-
-        afr_self_heal_parent_entrylk (frame, this,
-                                      afr_sh_post_nb_entrylk_missing_entry_sh_cbk);
-        return 0;
+    return afr_is_pending_set(this, xdata, AFR_DATA_TRANSACTION);
 }
 
-afr_local_t*
-afr_self_heal_local_init (afr_local_t *l, xlator_t *this)
+gf_boolean_t
+afr_is_metadata_set(xlator_t *this, dict_t *xdata)
 {
-        afr_private_t   *priv  = NULL;
-        afr_local_t     *lc    = NULL;
-        afr_self_heal_t *sh    = NULL;
-        afr_self_heal_t *shc   = NULL;
-        int             ret    = 0;
+    return afr_is_pending_set(this, xdata, AFR_METADATA_TRANSACTION);
+}
 
-        priv = this->private;
+gf_boolean_t
+afr_is_entry_set(xlator_t *this, dict_t *xdata)
+{
+    return afr_is_pending_set(this, xdata, AFR_ENTRY_TRANSACTION);
+}
 
-        sh = &l->self_heal;
+/*
+ * This function inspects the looked up replies (in an unlocked manner)
+ * and decides whether a locked verification and possible healing is
+ * required or not. It updates the three booleans for each type
+ * of healing. If the boolean flag gets set to FALSE, then we are sure
+ * no healing is required. If the boolean flag gets set to TRUE then
+ * we have to proceed with locked reinspection.
+ */
 
-        lc = mem_get0 (this->local_pool);
-        if (!lc)
-                goto out;
+int
+afr_selfheal_unlocked_inspect(call_frame_t *frame, xlator_t *this, uuid_t gfid,
+                              inode_t **link_inode, gf_boolean_t *data_selfheal,
+                              gf_boolean_t *metadata_selfheal,
+                              gf_boolean_t *entry_selfheal,
+                              struct afr_reply *replies_dst)
+{
+    afr_private_t *priv = NULL;
+    inode_t *inode = NULL;
+    int i = 0;
+    int valid_cnt = 0;
+    struct iatt first = {
+        0,
+    };
+    int first_idx = 0;
+    struct afr_reply *replies = NULL;
+    int ret = -1;
+
+    priv = this->private;
+
+    inode = afr_inode_find(this, gfid);
+    if (!inode)
+        goto out;
+
+    replies = alloca0(sizeof(*replies) * priv->child_count);
+
+    ret = afr_selfheal_unlocked_discover(frame, inode, gfid, replies);
+    if (ret)
+        goto out;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid)
+            continue;
+        if (replies[i].op_ret == -1)
+            continue;
+
+        /* The data segment of the changelog can be non-zero to indicate
+         * the directory needs a full heal. So the check below ensures
+         * it's not a directory before setting the data_selfheal boolean.
+         */
+        if (data_selfheal && !IA_ISDIR(replies[i].poststat.ia_type) &&
+            afr_is_data_set(this, replies[i].xdata))
+            *data_selfheal = _gf_true;
+
+        if (metadata_selfheal && afr_is_metadata_set(this, replies[i].xdata))
+            *metadata_selfheal = _gf_true;
+
+        if (entry_selfheal && afr_is_entry_set(this, replies[i].xdata))
+            *entry_selfheal = _gf_true;
+
+        valid_cnt++;
+        if (valid_cnt == 1) {
+            first = replies[i].poststat;
+            first_idx = i;
+            continue;
+        }
+
+        if (!IA_EQUAL(first, replies[i].poststat, type)) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN,
+                   "TYPE mismatch %d vs %d on %s for gfid:%s",
+                   (int)first.ia_type, (int)replies[i].poststat.ia_type,
+                   priv->children[i]->name,
+                   uuid_utoa(replies[i].poststat.ia_gfid));
+            gf_event(EVENT_AFR_SPLIT_BRAIN,
+                     "client-pid=%d;"
+                     "subvol=%s;"
+                     "type=file;gfid=%s;"
+                     "ia_type-%d=%s;ia_type-%d=%s",
+                     this->ctx->cmd_args.client_pid, this->name,
+                     uuid_utoa(replies[i].poststat.ia_gfid), first_idx,
+                     gf_inode_type_to_str(first.ia_type), i,
+                     gf_inode_type_to_str(replies[i].poststat.ia_type));
+            ret = -EIO;
+            goto out;
+        }
+
+        if (!IA_EQUAL(first, replies[i].poststat, uid)) {
+            gf_msg_debug(this->name, 0,
+                         "UID mismatch "
+                         "%d vs %d on %s for gfid:%s",
+                         (int)first.ia_uid, (int)replies[i].poststat.ia_uid,
+                         priv->children[i]->name,
+                         uuid_utoa(replies[i].poststat.ia_gfid));
+
+            if (metadata_selfheal)
+                *metadata_selfheal = _gf_true;
+        }
+
+        if (!IA_EQUAL(first, replies[i].poststat, gid)) {
+            gf_msg_debug(this->name, 0,
+                         "GID mismatch "
+                         "%d vs %d on %s for gfid:%s",
+                         (int)first.ia_uid, (int)replies[i].poststat.ia_uid,
+                         priv->children[i]->name,
+                         uuid_utoa(replies[i].poststat.ia_gfid));
+
+            if (metadata_selfheal)
+                *metadata_selfheal = _gf_true;
+        }
+
+        if (!IA_EQUAL(first, replies[i].poststat, prot)) {
+            gf_msg_debug(this->name, 0,
+                         "MODE mismatch "
+                         "%d vs %d on %s for gfid:%s",
+                         (int)st_mode_from_ia(first.ia_prot, 0),
+                         (int)st_mode_from_ia(replies[i].poststat.ia_prot, 0),
+                         priv->children[i]->name,
+                         uuid_utoa(replies[i].poststat.ia_gfid));
+
+            if (metadata_selfheal)
+                *metadata_selfheal = _gf_true;
+        }
+
+        if (IA_ISREG(first.ia_type) &&
+            !IA_EQUAL(first, replies[i].poststat, size)) {
+            gf_msg_debug(this->name, 0,
+                         "SIZE mismatch "
+                         "%lld vs %lld on %s for gfid:%s",
+                         (long long)first.ia_size,
+                         (long long)replies[i].poststat.ia_size,
+                         priv->children[i]->name,
+                         uuid_utoa(replies[i].poststat.ia_gfid));
+
+            if (data_selfheal)
+                *data_selfheal = _gf_true;
+        }
+    }
+
+    if (valid_cnt > 0 && link_inode) {
+        *link_inode = inode_link(inode, NULL, NULL, &first);
+        if (!*link_inode) {
+            ret = -EINVAL;
+            goto out;
+        }
+    } else if (valid_cnt < 2) {
+        ret = afr_check_stale_error(replies, priv);
+        goto out;
+    }
+
+    ret = 0;
+out:
+    if (replies && replies_dst)
+        afr_replies_copy(replies_dst, replies, priv->child_count);
+    if (inode)
+        inode_unref(inode);
+    if (replies)
+        afr_replies_wipe(replies, priv->child_count);
 
-        shc = &lc->self_heal;
-
-        shc->unwind = sh->unwind;
-        shc->gfid_sh_success_cbk = sh->gfid_sh_success_cbk;
-        shc->do_missing_entry_self_heal = sh->do_missing_entry_self_heal;
-        shc->do_gfid_self_heal = sh->do_gfid_self_heal;
-        shc->do_data_self_heal = sh->do_data_self_heal;
-        shc->do_metadata_self_heal = sh->do_metadata_self_heal;
-        shc->do_entry_self_heal = sh->do_entry_self_heal;
-        shc->force_confirm_spb = sh->force_confirm_spb;
-        shc->forced_merge = sh->forced_merge;
-        shc->background = sh->background;
-        shc->type = sh->type;
-        shc->data_sh_info = "";
-        shc->metadata_sh_info =  "";
-
-        uuid_copy (shc->sh_gfid_req, sh->sh_gfid_req);
-        if (l->loc.path) {
-                ret = loc_copy (&lc->loc, &l->loc);
-                if (ret < 0)
-                        goto out;
-        }
+    return ret;
+}
 
-        lc->child_up  = memdup (l->child_up,
-                                sizeof (*lc->child_up) * priv->child_count);
-        if (!lc->child_up) {
-                ret = -1;
-                goto out;
-        }
+inode_t *
+afr_inode_find(xlator_t *this, uuid_t gfid)
+{
+    inode_table_t *table = NULL;
+    inode_t *inode = NULL;
 
-        if (l->xattr_req)
-                lc->xattr_req = dict_ref (l->xattr_req);
+    table = this->itable;
+    if (!table)
+        return NULL;
 
-        if (l->cont.lookup.inode)
-                lc->cont.lookup.inode = inode_ref (l->cont.lookup.inode);
-        if (l->cont.lookup.xattr)
-                lc->cont.lookup.xattr = dict_ref (l->cont.lookup.xattr);
+    inode = inode_find(table, gfid);
+    if (inode)
+        return inode;
 
-        lc->internal_lock.locked_nodes =
-                             GF_CALLOC (sizeof (*l->internal_lock.locked_nodes),
-                                        priv->child_count, gf_afr_mt_char);
-        if (!lc->internal_lock.locked_nodes) {
-                ret = -1;
-                goto out;
-        }
+    inode = inode_new(table);
+    if (!inode)
+        return NULL;
 
-        ret = afr_inodelk_init (&lc->internal_lock.inodelk[0],
-                                this->name, priv->child_count);
-        if (ret)
-                goto out;
+    gf_uuid_copy(inode->gfid, gfid);
 
-out:
-        if (ret) {
-                afr_local_cleanup (lc, this);
-                lc = NULL;
-        }
-        return lc;
+    return inode;
 }
 
-int
-afr_self_heal_completion_cbk (call_frame_t *bgsh_frame, xlator_t *this)
-{
-        afr_private_t *   priv  = NULL;
-        afr_local_t *     local = NULL;
-        afr_self_heal_t * sh    = NULL;
-        afr_local_t *     orig_frame_local = NULL;
-        afr_self_heal_t * orig_frame_sh = NULL;
-        char              sh_type_str[256] = {0,};
-        gf_loglevel_t     loglevel = 0;
-
-        priv  = this->private;
-        local = bgsh_frame->local;
-        sh    = &local->self_heal;
-
-        if (local->unhealable) {
-                afr_set_split_brain (this, sh->inode, SPB, SPB);
-        }
-
-        afr_self_heal_type_str_get (sh, sh_type_str,
-                                    sizeof(sh_type_str));
-        if (is_self_heal_failed (sh, AFR_CHECK_ALL) && !priv->shd.iamshd) {
-                loglevel = GF_LOG_ERROR;
-        } else if (!is_self_heal_failed (sh, AFR_CHECK_ALL)) {
-                loglevel = GF_LOG_INFO;
-        } else {
-                loglevel = GF_LOG_DEBUG;
-        }
+call_frame_t *
+afr_frame_create(xlator_t *this, int32_t *op_errno)
+{
+    call_frame_t *frame = NULL;
+    afr_local_t *local = NULL;
+    pid_t pid = GF_CLIENT_PID_SELF_HEALD;
 
-        afr_log_self_heal_completion_status (local, loglevel);
+    frame = create_frame(this, this->ctx->pool);
+    if (!frame) {
+        if (op_errno)
+            *op_errno = ENOMEM;
+        return NULL;
+    }
 
-        FRAME_SU_UNDO (bgsh_frame, afr_local_t);
+    local = AFR_FRAME_INIT(frame, (*op_errno));
+    if (!local) {
+        STACK_DESTROY(frame->root);
+        return NULL;
+    }
 
-        if (!sh->unwound && sh->unwind) {
-                orig_frame_local = sh->orig_frame->local;
-                orig_frame_sh = &orig_frame_local->self_heal;
-                orig_frame_sh->actual_sh_started = _gf_true;
-                sh->unwind (sh->orig_frame, this, sh->op_ret, sh->op_errno,
-                            is_self_heal_failed (sh, AFR_CHECK_ALL));
-        }
+    syncopctx_setfspid(&pid);
 
-        if (sh->background) {
-                LOCK (&priv->lock);
-                {
-                        priv->background_self_heals_started--;
-                }
-                UNLOCK (&priv->lock);
-        }
+    frame->root->pid = pid;
 
-        AFR_STACK_DESTROY (bgsh_frame);
+    afr_set_lk_owner(frame, this, frame->root);
 
-        return 0;
+    return frame;
 }
 
 int
-afr_self_heal (call_frame_t *frame, xlator_t *this, inode_t *inode)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int32_t          op_errno = 0;
-        int              ret = 0;
-        afr_self_heal_t *orig_sh = NULL;
-        call_frame_t    *sh_frame = NULL;
-        afr_local_t     *sh_local = NULL;
-        loc_t           *loc   = NULL;
-
-        local = frame->local;
-        orig_sh = &local->self_heal;
-        priv  = this->private;
-
-        GF_ASSERT (local->loc.path);
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "performing self heal on %s (metadata=%d data=%d entry=%d)",
-                local->loc.path,
-                local->self_heal.do_metadata_self_heal,
-                local->self_heal.do_data_self_heal,
-                local->self_heal.do_entry_self_heal);
-
-        op_errno        = ENOMEM;
-        sh_frame        = copy_frame (frame);
-        if (!sh_frame)
-                goto out;
-        afr_set_lk_owner (sh_frame, this, sh_frame->root);
-        afr_set_low_priority (sh_frame);
-
-        sh_local        = afr_self_heal_local_init (local, this);
-        if (!sh_local)
-                goto out;
-        sh_frame->local = sh_local;
-        sh              = &sh_local->self_heal;
-
-        sh->inode       = inode_ref (inode);
-        sh->orig_frame  = frame;
-
-        sh->completion_cbk = afr_self_heal_completion_cbk;
-
-        sh->success = GF_CALLOC (priv->child_count, sizeof (*sh->success),
-                                 gf_afr_mt_char);
-        if (!sh->success)
-                goto out;
-        sh->sources = GF_CALLOC (sizeof (*sh->sources), priv->child_count,
-                                 gf_afr_mt_int);
-        if (!sh->sources)
-                goto out;
-        sh->locked_nodes = GF_CALLOC (sizeof (*sh->locked_nodes),
-                                      priv->child_count,
-                                      gf_afr_mt_int);
-        if (!sh->locked_nodes)
-                goto out;
-
-        sh->pending_matrix = afr_matrix_create (priv->child_count,
-                                                priv->child_count);
-        if (!sh->pending_matrix)
-                goto out;
-
-        sh->delta_matrix = afr_matrix_create (priv->child_count,
-                                              priv->child_count);
-        if (!sh->delta_matrix)
-                goto out;
-
-        sh->fresh_parent_dirs = afr_children_create (priv->child_count);
-        if (!sh->fresh_parent_dirs)
-                goto out;
-        ret = afr_sh_common_create (sh, priv->child_count);
-        if (ret) {
-                op_errno = -ret;
-                goto out;
-        }
+afr_selfheal_newentry_mark(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                           int source, struct afr_reply *replies,
+                           unsigned char *sources, unsigned char *newentry)
+{
+    int ret = 0;
+    int i = 0;
+    afr_private_t *priv = NULL;
+    dict_t *xattr = NULL;
+    int **changelog = NULL;
 
-        if (local->self_heal.background) {
-                LOCK (&priv->lock);
-                {
-                        if (priv->background_self_heals_started
-                            < priv->background_self_heal_count) {
-                                priv->background_self_heals_started++;
+    priv = this->private;
 
+    gf_uuid_copy(inode->gfid, replies[source].poststat.ia_gfid);
 
-                        } else {
-                                local->self_heal.background = _gf_false;
-                                sh->background = _gf_false;
-                        }
-                }
-                UNLOCK (&priv->lock);
-        }
+    xattr = dict_new();
+    if (!xattr)
+        return -ENOMEM;
 
-        if (!local->loc.parent) {
-                sh->do_missing_entry_self_heal = _gf_false;
-                sh->do_gfid_self_heal = _gf_false;
-        }
+    changelog = afr_mark_pending_changelog(priv, newentry, xattr,
+                                           replies[source].poststat.ia_type);
 
-        sh->sh_type_in_action = AFR_SELF_HEAL_INVALID;
+    if (!changelog) {
+        ret = -ENOMEM;
+        goto out;
+    }
 
-        FRAME_SU_DO (sh_frame, afr_local_t);
-        if (sh->do_missing_entry_self_heal || sh->do_gfid_self_heal) {
-                afr_self_heal_missing_entries (sh_frame, this);
-        } else {
-                loc = &sh_local->loc;
-                if (uuid_is_null (loc->inode->gfid) && uuid_is_null (loc->gfid)) {
-                        if (!uuid_is_null (inode->gfid))
-                                GF_ASSERT (!uuid_compare (inode->gfid,
-                                           sh->sh_gfid_req));
-                        uuid_copy (loc->gfid, sh->sh_gfid_req);
-                }
-                gf_log (this->name, GF_LOG_TRACE,
-                        "proceeding to metadata check on %s",
-                        local->loc.path);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i])
+            continue;
+        ret |= afr_selfheal_post_op(frame, this, inode, i, xattr, NULL);
+    }
+out:
+    if (changelog)
+        afr_matrix_cleanup(changelog, priv->child_count);
+    if (xattr)
+        dict_unref(xattr);
+    return ret;
+}
 
-                afr_sh_missing_entries_done (sh_frame, this);
-        }
-        op_errno = 0;
+int
+afr_selfheal_do(call_frame_t *frame, xlator_t *this, uuid_t gfid)
+{
+    int ret = -1;
+    int entry_ret = 1;
+    int metadata_ret = 1;
+    int data_ret = 1;
+    int or_ret = 0;
+    inode_t *inode = NULL;
+    fd_t *fd = NULL;
+    gf_boolean_t data_selfheal = _gf_false;
+    gf_boolean_t metadata_selfheal = _gf_false;
+    gf_boolean_t entry_selfheal = _gf_false;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+
+    ret = afr_selfheal_unlocked_inspect(frame, this, gfid, &inode,
+                                        &data_selfheal, &metadata_selfheal,
+                                        &entry_selfheal, NULL);
+    if (ret)
+        goto out;
+
+    if (!(data_selfheal || metadata_selfheal || entry_selfheal)) {
+        ret = 2;
+        goto out;
+    }
+
+    if (inode->ia_type == IA_IFREG) {
+        ret = afr_selfheal_data_open(this, inode, &fd);
+        if (!fd) {
+            ret = -EIO;
+            goto out;
+        }
+    }
+
+    if (data_selfheal && priv->data_self_heal)
+        data_ret = afr_selfheal_data(frame, this, fd);
+
+    if (metadata_selfheal && priv->metadata_self_heal)
+        metadata_ret = afr_selfheal_metadata(frame, this, inode);
+
+    if (entry_selfheal && priv->entry_self_heal)
+        entry_ret = afr_selfheal_entry(frame, this, inode);
+
+    or_ret = (data_ret | metadata_ret | entry_ret);
+
+    if (data_ret == -EIO || metadata_ret == -EIO || entry_ret == -EIO)
+        ret = -EIO;
+    else if (data_ret == 1 && metadata_ret == 1 && entry_ret == 1)
+        ret = 1;
+    else if (or_ret < 0)
+        ret = or_ret;
+    else
+        ret = 0;
 
 out:
-        if (op_errno) {
-                orig_sh->unwind (frame, this, -1, op_errno, 1);
-                if (sh_frame)
-                        AFR_STACK_DESTROY (sh_frame);
-        }
-        return 0;
+    if (inode)
+        inode_unref(inode);
+    if (fd)
+        fd_unref(fd);
+    return ret;
 }
+/*
+ * This is the entry point for healing a given GFID. The return values for this
+ * function are as follows:
+ * '0' if the self-heal is successful
+ * '1' if the afr-xattrs are non-zero (due to on-going IO) and no heal is needed
+ * '2' if the afr-xattrs are all-zero and no heal is needed
+ * $errno if the heal on the gfid failed.
+ */
 
-void
-afr_self_heal_type_str_get (afr_self_heal_t *self_heal_p, char *str,
-                            size_t size)
+int
+afr_selfheal(xlator_t *this, uuid_t gfid)
 {
-        GF_ASSERT (str && (size > strlen (" missing-entry gfid "
-                                          "meta-data data entry")));
+    int ret = -1;
+    call_frame_t *frame = NULL;
+    afr_local_t *local = NULL;
 
-        if (self_heal_p->do_metadata_self_heal) {
-                snprintf (str, size, " meta-data");
-        }
+    frame = afr_frame_create(this, NULL);
+    if (!frame)
+        return ret;
 
-        if (self_heal_p->do_data_self_heal) {
-                snprintf (str + strlen(str), size - strlen(str), " data");
-        }
+    local = frame->local;
+    local->xdata_req = dict_new();
 
-        if (self_heal_p->do_entry_self_heal) {
-                snprintf (str + strlen(str), size - strlen(str), " entry");
-        }
+    ret = afr_selfheal_do(frame, this, gfid);
 
-        if (self_heal_p->do_missing_entry_self_heal) {
-                snprintf (str + strlen(str), size - strlen(str),
-                         " missing-entry");
-        }
+    if (frame)
+        AFR_STACK_DESTROY(frame);
 
-        if (self_heal_p->do_gfid_self_heal) {
-                snprintf (str + strlen(str), size - strlen(str), " gfid");
-        }
+    return ret;
 }
 
-afr_self_heal_type
-afr_self_heal_type_for_transaction (afr_transaction_type type)
+afr_local_t *
+__afr_dequeue_heals(afr_private_t *priv)
 {
-        afr_self_heal_type sh_type = AFR_SELF_HEAL_INVALID;
+    afr_local_t *local = NULL;
 
-        switch (type) {
-        case AFR_DATA_TRANSACTION:
-                sh_type = AFR_SELF_HEAL_DATA;
-                break;
-        case AFR_METADATA_TRANSACTION:
-                sh_type = AFR_SELF_HEAL_METADATA;
-                break;
-        case AFR_ENTRY_TRANSACTION:
-                sh_type = AFR_SELF_HEAL_ENTRY;
-                break;
-        case AFR_ENTRY_RENAME_TRANSACTION:
-                GF_ASSERT (0);
-                break;
-        }
-        return sh_type;
+    if (list_empty(&priv->heal_waiting))
+        goto none;
+    if ((priv->background_self_heal_count > 0) &&
+        (priv->healers >= priv->background_self_heal_count))
+        goto none;
+
+    local = list_entry(priv->heal_waiting.next, afr_local_t, healer);
+    priv->heal_waiters--;
+    GF_ASSERT(priv->heal_waiters >= 0);
+    list_del_init(&local->healer);
+    list_add(&local->healer, &priv->healing);
+    priv->healers++;
+    return local;
+none:
+    gf_msg_debug(THIS->name, 0,
+                 "Nothing dequeued. "
+                 "Num healers: %d, Num Waiters: %d",
+                 priv->healers, priv->heal_waiters);
+    return NULL;
 }
 
 int
-afr_build_child_loc (xlator_t *this, loc_t *child, loc_t *parent, char *name)
+afr_refresh_selfheal_wrap(void *opaque)
 {
-        int   ret = -1;
-        uuid_t pargfid = {0};
-
-        if (!child)
-                goto out;
-
-        if (!uuid_is_null (parent->inode->gfid))
-                uuid_copy (pargfid, parent->inode->gfid);
-        else if (!uuid_is_null (parent->gfid))
-                uuid_copy (pargfid, parent->gfid);
-
-        if (uuid_is_null (pargfid))
-                goto out;
-
-        if (strcmp (parent->path, "/") == 0)
-                ret = gf_asprintf ((char **)&child->path, "/%s", name);
-        else
-                ret = gf_asprintf ((char **)&child->path, "%s/%s", parent->path,
-                                   name);
-
-        if (-1 == ret) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "asprintf failed while setting child path");
-        }
-
-        child->name = strrchr (child->path, '/');
-        if (child->name)
-                child->name++;
+    call_frame_t *heal_frame = opaque;
+    afr_local_t *local = heal_frame->local;
+    int ret = 0;
 
-        child->parent = inode_ref (parent->inode);
-        child->inode = inode_new (parent->inode->table);
-        uuid_copy (child->pargfid, pargfid);
-
-        if (!child->inode) {
-                ret = -1;
-                goto out;
-        }
-
-        ret = 0;
-out:
-        if ((ret == -1) && child)
-                loc_wipe (child);
-
-        return ret;
+    ret = afr_selfheal(heal_frame->this, local->refreshinode->gfid);
+    return ret;
 }
 
 int
-afr_sh_erase_pending (call_frame_t *frame, xlator_t *this,
-                      afr_transaction_type type, afr_fxattrop_cbk_t cbk,
-                      int (*finish)(call_frame_t *frame, xlator_t *this))
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int              call_count = 0;
-        int              i = 0;
-        dict_t          **erase_xattr = NULL;
-        int             ret = -1;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        afr_sh_pending_to_delta (priv, sh->xattr, sh->delta_matrix,
-                                 sh->success, priv->child_count, type);
-
-        erase_xattr = GF_CALLOC (sizeof (*erase_xattr), priv->child_count,
-                                 gf_afr_mt_dict_t);
-        if (!erase_xattr)
-                goto out;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (sh->xattr[i]) {
-                        call_count++;
-                        erase_xattr[i] = dict_new ();
-                        if (!erase_xattr[i])
-                                goto out;
-                }
-        }
-
-        afr_sh_delta_to_xattr (this, sh->delta_matrix, erase_xattr,
-                               priv->child_count, type);
-
-        gf_log (this->name, GF_LOG_DEBUG, "Delta matrix for: %s",
-                lkowner_utoa (&frame->root->lk_owner));
-        afr_sh_print_pending_matrix (sh->delta_matrix, this);
-        local->call_count = call_count;
-        if (call_count == 0) {
-                ret = 0;
-                finish (frame, this);
-                goto out;
-        }
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (!erase_xattr[i])
-                        continue;
-
-                if (sh->healing_fd) {//true for ENTRY, reg file DATA transaction
-                        STACK_WIND_COOKIE (frame, cbk, (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->fxattrop,
-                                           sh->healing_fd,
-                                           GF_XATTROP_ADD_ARRAY, erase_xattr[i],
-                                           NULL);
-                } else {
-                        STACK_WIND_COOKIE (frame, cbk, (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->xattrop,
-                                           &local->loc,
-                                           GF_XATTROP_ADD_ARRAY, erase_xattr[i],
-                                           NULL);
-                }
-        }
-
-        ret = 0;
-out:
-        if (erase_xattr) {
-                for (i = 0; i < priv->child_count; i++) {
-                        if (erase_xattr[i]) {
-                                dict_unref (erase_xattr[i]);
-                        }
-                }
-        }
+afr_refresh_heal_done(int ret, call_frame_t *frame, void *opaque)
+{
+    call_frame_t *heal_frame = opaque;
+    xlator_t *this = heal_frame->this;
+    afr_private_t *priv = this->private;
+    afr_local_t *local = heal_frame->local;
 
-        GF_FREE (erase_xattr);
+    LOCK(&priv->lock);
+    {
+        list_del_init(&local->healer);
+        priv->healers--;
+        GF_ASSERT(priv->healers >= 0);
+        local = __afr_dequeue_heals(priv);
+    }
+    UNLOCK(&priv->lock);
 
-        if (ret < 0) {
-                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-                finish (frame, this);
-        }
+    AFR_STACK_DESTROY(heal_frame);
 
-        return 0;
+    if (local)
+        afr_heal_synctask(this, local);
+    return 0;
 }
 
 void
-afr_set_self_heal_status(afr_self_heal_t *sh, afr_self_heal_status status)
+afr_heal_synctask(xlator_t *this, afr_local_t *local)
 {
-        xlator_t                *this = NULL;
-        afr_sh_status_for_all_type *sh_status = &(sh->afr_all_sh_status);
-        afr_self_heal_type  sh_type_in_action = sh->sh_type_in_action;
-        this = THIS;
+    int ret = 0;
+    call_frame_t *heal_frame = NULL;
 
-        if (!sh) {
-                gf_log_callingfn (this->name, GF_LOG_ERROR, "Null self heal"
-                                  "Structure");
-                goto out;
-        }
-
-        switch (sh_type_in_action) {
-                case AFR_SELF_HEAL_GFID_OR_MISSING_ENTRY:
-                       sh_status->gfid_or_missing_entry_self_heal = status;
-                        break;
-                case AFR_SELF_HEAL_METADATA:
-                        sh_status->metadata_self_heal = status;
-                        break;
-                case AFR_SELF_HEAL_DATA:
-                        sh_status->data_self_heal = status;
-                        break;
-                case AFR_SELF_HEAL_ENTRY:
-                        sh_status->entry_self_heal = status;
-                        break;
-                case AFR_SELF_HEAL_INVALID:
-                        gf_log_callingfn (this->name, GF_LOG_ERROR, "Invalid"
-                                          "self heal type in action");
-                        break;
-        }
-out:
-        return;
+    heal_frame = local->heal_frame;
+    ret = synctask_new(this->ctx->env, afr_refresh_selfheal_wrap,
+                       afr_refresh_heal_done, heal_frame, heal_frame);
+    if (ret < 0)
+        /* Heal not launched. Will be queued when the next inode
+         * refresh happens and shd hasn't healed it yet. */
+        afr_refresh_heal_done(ret, heal_frame, heal_frame);
 }
 
-void
-afr_set_local_for_unhealable (afr_local_t *local)
-{
-        afr_self_heal_t  *sh = NULL;
+gf_boolean_t
+afr_throttled_selfheal(call_frame_t *frame, xlator_t *this)
+{
+    gf_boolean_t can_heal = _gf_true;
+    afr_private_t *priv = this->private;
+    afr_local_t *local = frame->local;
+
+    LOCK(&priv->lock);
+    {
+        if ((priv->background_self_heal_count > 0) &&
+            (priv->heal_wait_qlen + priv->background_self_heal_count) >
+                (priv->heal_waiters + priv->healers)) {
+            list_add_tail(&local->healer, &priv->heal_waiting);
+            priv->heal_waiters++;
+            local = __afr_dequeue_heals(priv);
+        } else {
+            can_heal = _gf_false;
+        }
+    }
+    UNLOCK(&priv->lock);
 
-        sh = &local->self_heal;
+    if (can_heal) {
+        if (local)
+            afr_heal_synctask(this, local);
+        else
+            gf_msg_debug(this->name, 0,
+                         "Max number of heals are "
+                         "pending, background self-heal rejected.");
+    }
 
-        local->unhealable = 1;
-        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+    return can_heal;
 }
 
 int
-is_self_heal_failed (afr_self_heal_t *sh, afr_sh_fail_check_type type)
+afr_choose_source_by_policy(afr_private_t *priv, unsigned char *sources,
+                            afr_transaction_type type)
 {
-        afr_sh_status_for_all_type      sh_status = sh->afr_all_sh_status;
-        afr_self_heal_type   sh_type_in_action =  AFR_SELF_HEAL_INVALID;
-        afr_self_heal_status    status = AFR_SELF_HEAL_FAILED;
-        xlator_t                *this = NULL;
-        int                     sh_failed = 0;
+    int source = -1;
+    int i = 0;
 
-        this = THIS;
+    /* Give preference to local child to save on bandwidth */
+    for (i = 0; i < priv->child_count; i++) {
+        if (priv->local[i] && sources[i]) {
+            if ((type == AFR_DATA_TRANSACTION) && AFR_IS_ARBITER_BRICK(priv, i))
+                continue;
 
-        if (!sh) {
-                gf_log_callingfn (this->name, GF_LOG_ERROR, "Null self heal "
-                                  "structure");
-                sh_failed = 1;
-                goto out;
+            source = i;
+            goto out;
         }
+    }
 
-        if (type == AFR_CHECK_ALL) {
-                if ((sh_status.gfid_or_missing_entry_self_heal == AFR_SELF_HEAL_FAILED)
-                    || (sh_status.metadata_self_heal == AFR_SELF_HEAL_FAILED)
-                    || (sh_status.data_self_heal == AFR_SELF_HEAL_FAILED)
-                    || (sh_status.entry_self_heal == AFR_SELF_HEAL_FAILED))
-                sh_failed = 1;
-        } else if (type == AFR_CHECK_SPECIFIC) {
-                sh_type_in_action = sh->sh_type_in_action;
-                switch (sh_type_in_action) {
-                        case AFR_SELF_HEAL_GFID_OR_MISSING_ENTRY:
-                             status = sh_status.gfid_or_missing_entry_self_heal;
-                                break;
-                        case AFR_SELF_HEAL_METADATA:
-                                status = sh_status.metadata_self_heal;
-                                break;
-                        case AFR_SELF_HEAL_ENTRY:
-                                status = sh_status.entry_self_heal;
-                                break;
-                        case AFR_SELF_HEAL_DATA:
-                                status = sh_status.data_self_heal;
-                                break;
-                        case AFR_SELF_HEAL_INVALID:
-                                status = AFR_SELF_HEAL_NOT_ATTEMPTED;
-                                break;
-                }
-                if (status == AFR_SELF_HEAL_FAILED)
-                        sh_failed = 1;
-
+    for (i = 0; i < priv->child_count; i++) {
+        if (sources[i]) {
+            source = i;
+            goto out;
         }
-
+    }
 out:
-        return sh_failed;
-}
-
-char *
-get_sh_completion_status (afr_self_heal_status status)
-{
-
-        char *not_attempted       = " is not attempted";
-        char *failed              = " failed";
-        char *started             = " is started";
-        char *sync_begin          = " is successfully completed";
-        char *result              = " has unknown status";
-
-        switch (status)
-        {
-                case AFR_SELF_HEAL_NOT_ATTEMPTED:
-                        result = not_attempted;
-                        break;
-                case AFR_SELF_HEAL_FAILED:
-                        result = failed;
-                        break;
-                case AFR_SELF_HEAL_STARTED:
-                        result = started;
-                        break;
-                case AFR_SELF_HEAL_SYNC_BEGIN:
-                        result = sync_begin;
-                        break;
+    return source;
+}
+
+static int
+afr_anon_inode_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, inode_t *inode,
+                         struct iatt *buf, struct iatt *preparent,
+                         struct iatt *postparent, dict_t *xdata)
+{
+    afr_local_t *local = frame->local;
+    int i = (long)cookie;
+
+    local->replies[i].valid = 1;
+    local->replies[i].op_ret = op_ret;
+    local->replies[i].op_errno = op_errno;
+    if (op_ret == 0) {
+        local->op_ret = 0;
+        local->replies[i].poststat = *buf;
+        local->replies[i].preparent = *preparent;
+        local->replies[i].postparent = *postparent;
+    }
+    if (xdata) {
+        local->replies[i].xdata = dict_ref(xdata);
+    }
+
+    syncbarrier_wake(&local->barrier);
+    return 0;
+}
+
+int
+afr_anon_inode_create(xlator_t *this, int child, inode_t **linked_inode)
+{
+    call_frame_t *frame = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = this->private;
+    unsigned char *mkdir_on = alloca0(priv->child_count);
+    unsigned char *lookup_on = alloca0(priv->child_count);
+    loc_t loc = {0};
+    int32_t op_errno = 0;
+    int32_t child_op_errno = 0;
+    struct iatt iatt = {0};
+    dict_t *xdata = NULL;
+    uuid_t anon_inode_gfid = {0};
+    int mkdir_count = 0;
+    int i = 0;
+
+    /*Try to mkdir everywhere and return success if the dir exists on 'child'
+     */
+
+    if (!priv->use_anon_inode) {
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    frame = afr_frame_create(this, &op_errno);
+    if (op_errno) {
+        goto out;
+    }
+    local = frame->local;
+    if (!local->child_up[child]) {
+        /*Other bricks may need mkdir so don't error out yet*/
+        child_op_errno = ENOTCONN;
+    }
+    gf_uuid_parse(priv->anon_gfid_str, anon_inode_gfid);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->child_up[i])
+            continue;
+
+        if (priv->anon_inode[i]) {
+            mkdir_on[i] = 0;
+        } else {
+            mkdir_on[i] = 1;
+            mkdir_count++;
         }
+    }
 
-        return result;
+    if (mkdir_count == 0) {
+        *linked_inode = inode_find(this->itable, anon_inode_gfid);
+        if (*linked_inode) {
+            op_errno = 0;
+            goto out;
+        }
+    }
 
-}
+    loc.parent = inode_ref(this->itable->root);
+    loc.name = priv->anon_inode_name;
+    loc.inode = inode_new(this->itable);
+    if (!loc.inode) {
+        op_errno = ENOMEM;
+        goto out;
+    }
 
-void
-afr_log_self_heal_completion_status (afr_local_t *local, gf_loglevel_t loglvl)
-{
-
-        char sh_log[4096]              = {0};
-        afr_self_heal_t *sh            = &local->self_heal;
-        afr_sh_status_for_all_type   all_status = sh->afr_all_sh_status;
-        xlator_t      *this            = NULL;
-        size_t        off              = 0;
-        int           data_sh          = 0;
-        int           metadata_sh      = 0;
-        int           print_log        = 0;
-
-        this = THIS;
-
-        ADD_FMT_STRING (sh_log, off, "gfid or missing entry",
-                        all_status.gfid_or_missing_entry_self_heal, print_log);
-        ADD_FMT_STRING_SYNC (sh_log, off, "metadata",
-                             all_status.metadata_self_heal, print_log);
-        if (sh->background) {
-                ADD_FMT_STRING_SYNC (sh_log, off, "backgroung data",
-                                all_status.data_self_heal, print_log);
+    xdata = dict_new();
+    if (!xdata) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    op_errno = -dict_set_gfuuid(xdata, "gfid-req", anon_inode_gfid, _gf_true);
+    if (op_errno) {
+        goto out;
+    }
+
+    if (mkdir_count == 0) {
+        memcpy(lookup_on, local->child_up, priv->child_count);
+        goto lookup;
+    }
+
+    AFR_ONLIST(mkdir_on, frame, afr_anon_inode_mkdir_cbk, mkdir, &loc, 0755, 0,
+               xdata);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!mkdir_on[i]) {
+            continue;
+        }
+
+        if (local->replies[i].op_ret == 0) {
+            priv->anon_inode[i] = 1;
+            iatt = local->replies[i].poststat;
+        } else if (local->replies[i].op_ret < 0 &&
+                   local->replies[i].op_errno == EEXIST) {
+            lookup_on[i] = 1;
+        } else if (i == child) {
+            child_op_errno = local->replies[i].op_errno;
+        }
+    }
+
+    if (AFR_COUNT(lookup_on, priv->child_count) == 0) {
+        goto link;
+    }
+
+lookup:
+    AFR_ONLIST(lookup_on, frame, afr_selfheal_discover_cbk, lookup, &loc,
+               xdata);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!lookup_on[i]) {
+            continue;
+        }
+
+        if (local->replies[i].op_ret == 0) {
+            if (gf_uuid_compare(anon_inode_gfid,
+                                local->replies[i].poststat.ia_gfid) == 0) {
+                priv->anon_inode[i] = 1;
+                iatt = local->replies[i].poststat;
+            } else {
+                if (i == child)
+                    child_op_errno = EINVAL;
+                gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_INVALID_DATA,
+                       "%s has gfid: %s", priv->anon_inode_name,
+                       uuid_utoa(local->replies[i].poststat.ia_gfid));
+            }
+        } else if (i == child) {
+            child_op_errno = local->replies[i].op_errno;
+        }
+    }
+link:
+    if (!gf_uuid_is_null(iatt.ia_gfid)) {
+        *linked_inode = inode_link(loc.inode, loc.parent, loc.name, &iatt);
+        if (*linked_inode) {
+            op_errno = 0;
+            inode_lookup(*linked_inode);
         } else {
-                ADD_FMT_STRING_SYNC (sh_log, off, "foreground data",
-                                all_status.data_self_heal, print_log);
+            op_errno = ENOMEM;
         }
-        ADD_FMT_STRING_SYNC (sh_log, off, "entry", all_status.entry_self_heal,
-                             print_log);
-
-        if (AFR_SELF_HEAL_SYNC_BEGIN == all_status.data_self_heal &&
-	    strcmp (sh->data_sh_info, "") && sh->data_sh_info )
-                data_sh = 1;
-        if (AFR_SELF_HEAL_SYNC_BEGIN == all_status.metadata_self_heal &&
-	    strcmp (sh->metadata_sh_info, "") && sh->metadata_sh_info)
-                metadata_sh = 1;
-
-        if (!print_log)
-                return;
-
-        gf_log (this->name, loglvl, "%s %s %s on %s", sh_log,
-                ((data_sh == 1) ? sh->data_sh_info : ""),
-                ((metadata_sh == 1) ? sh->metadata_sh_info : ""),
-                local->loc.path);
+        goto out;
+    }
+
+out:
+    if (xdata)
+        dict_unref(xdata);
+    loc_wipe(&loc);
+    /*child_op_errno takes precedence*/
+    if (child_op_errno == 0) {
+        child_op_errno = op_errno;
+    }
+
+    if (child_op_errno && *linked_inode) {
+        inode_unref(*linked_inode);
+        *linked_inode = NULL;
+    }
+    if (frame)
+        AFR_STACK_DESTROY(frame);
+    return -child_op_errno;
 }
diff --git a/xlators/cluster/afr/src/afr-self-heal-common.h b/xlators/cluster/afr/src/afr-self-heal-common.h
deleted file mode 100644
index 4732647760d..00000000000
--- a/xlators/cluster/afr/src/afr-self-heal-common.h
+++ /dev/null
@@ -1,144 +0,0 @@
-/*
-  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
-  This file is part of GlusterFS.
-
-  This file is licensed to you under your choice of the GNU Lesser
-  General Public License, version 3 or any later version (LGPLv3 or
-  later), or the GNU General Public License, version 2 (GPLv2), in all
-  cases as published by the Free Software Foundation.
-*/
-
-#ifndef __AFR_SELF_HEAL_COMMON_H__
-#define __AFR_SELF_HEAL_COMMON_H__
-
-#define FILE_HAS_HOLES(buf) (((buf)->ia_size) > ((buf)->ia_blocks * 512))
-#define AFR_SH_MIN_PARTICIPANTS 2
-
-typedef enum {
-        AFR_LOOKUP_FAIL_CONFLICTS = 1,
-        AFR_LOOKUP_FAIL_MISSING_GFIDS = 2,
-} afr_lookup_flags_t;
-
-int
-afr_sh_select_source (int sources[], int child_count);
-
-int
-afr_sh_source_count (int sources[], int child_count);
-
-void
-afr_sh_print_pending_matrix (int32_t *pending_matrix[], xlator_t *this);
-
-void
-afr_sh_print_split_brain_log (int32_t *pending_matrix[], xlator_t *this,
-                              const char *loc);
-
-int
-afr_build_pending_matrix (char **pending_key, int32_t **pending_matrix,
-                          unsigned char *ignorant_subvols,
-                          dict_t *xattr[], afr_transaction_type type,
-                          size_t child_count);
-
-void
-afr_sh_pending_to_delta (afr_private_t *priv, dict_t **xattr,
-                         int32_t *delta_matrix[], unsigned char success[],
-                         int child_count, afr_transaction_type type);
-
-int
-afr_mark_sources (xlator_t *this, int32_t *sources, int32_t **pending_matrix,
-                  struct iatt *bufs, afr_self_heal_type type,
-                  int32_t *success_children, int32_t *subvol_status);
-
-int
-afr_sh_delta_to_xattr (xlator_t *this,
-                       int32_t *delta_matrix[], dict_t *xattr[],
-		       int child_count, afr_transaction_type type);
-
-void
-afr_self_heal_type_str_get (afr_self_heal_t *self_heal_p, char *str,
-                            size_t size);
-
-afr_self_heal_type
-afr_self_heal_type_for_transaction (afr_transaction_type type);
-
-int
-afr_build_sources (xlator_t *this, dict_t **xattr, struct iatt *bufs,
-                   int32_t **pending_matrix, int32_t *sources,
-                   int32_t *success_children, afr_transaction_type type,
-                   int32_t *subvol_status, gf_boolean_t ignore_ignorant);
-void
-afr_sh_common_reset (afr_self_heal_t *sh, unsigned int child_count);
-
-void
-afr_sh_common_lookup_resp_handler (call_frame_t *frame, void *cookie,
-                                   xlator_t *this,
-                                   int32_t op_ret, int32_t op_errno,
-                                   inode_t *inode, struct iatt *buf,
-                                   dict_t *xattr, struct iatt *postparent,
-                                   loc_t *loc);
-
-int
-afr_sh_common_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                      afr_lookup_done_cbk_t lookup_cbk, uuid_t uuid,
-                      int32_t flags, dict_t *xdata);
-int
-afr_sh_entry_expunge_remove (call_frame_t *expunge_frame, xlator_t *this,
-                             int active_src, struct iatt *buf,
-                             struct iatt *parentbuf);
-int
-afr_sh_entrylk (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                char *base_name, afr_lock_cbk_t lock_cbk);
-int
-afr_sh_entry_impunge_create (call_frame_t *impunge_frame, xlator_t *this,
-                             int child_index);
-int
-afr_sh_data_unlock (call_frame_t *frame, xlator_t *this, char *dom,
-                    afr_lock_cbk_t lock_cbk);
-afr_local_t *
-afr_self_heal_local_init (afr_local_t *l, xlator_t *this);
-int
-afr_sh_data_lock (call_frame_t *frame, xlator_t *this,
-                  off_t start, off_t len, gf_boolean_t block, char *dom,
-                  afr_lock_cbk_t success_handler,
-                  afr_lock_cbk_t failure_handler);
-void
-afr_sh_set_error (afr_self_heal_t *sh, int32_t op_errno);
-void
-afr_sh_mark_source_sinks (call_frame_t *frame, xlator_t *this);
-typedef int
-(*afr_fxattrop_cbk_t) (call_frame_t *frame, void *cookie,
-                       xlator_t *this, int32_t op_ret, int32_t op_errno,
-                       dict_t *xattr, dict_t *xdata);
-int
-afr_build_child_loc (xlator_t *this, loc_t *child, loc_t *parent, char *name);
-int
-afr_impunge_frame_create (call_frame_t *frame, xlator_t *this,
-                          int active_source, call_frame_t **impunge_frame);
-void
-afr_sh_reset (call_frame_t *frame, xlator_t *this);
-
-void
-afr_children_intersection_get (int32_t *set1, int32_t *set2,
-                               int *intersection, unsigned int child_count);
-int
-afr_get_no_xattr_dir_read_child (xlator_t *this, int32_t *success_children,
-                                 struct iatt *bufs);
-int
-afr_sh_erase_pending (call_frame_t *frame, xlator_t *this,
-                      afr_transaction_type type, afr_fxattrop_cbk_t cbk,
-                      int (*finish)(call_frame_t *frame, xlator_t *this));
-
-void
-afr_set_local_for_unhealable (afr_local_t *local);
-
-int
-is_self_heal_failed (afr_self_heal_t *sh, afr_sh_fail_check_type type);
-
-void
-afr_set_self_heal_status (afr_self_heal_t *sh, afr_self_heal_status status);
-
-void
-afr_log_self_heal_completion_status (afr_local_t *local, gf_loglevel_t  logl);
-
-char*
-afr_get_pending_matrix_str (int32_t *pending_matrix[], xlator_t *this);
-#endif /* __AFR_SELF_HEAL_COMMON_H__ */
diff --git a/xlators/cluster/afr/src/afr-self-heal-data.c b/xlators/cluster/afr/src/afr-self-heal-data.c
index 9de26ee569c..37bcc2b3f9e 100644
--- a/xlators/cluster/afr/src/afr-self-heal-data.c
+++ b/xlators/cluster/afr/src/afr-self-heal-data.c
@@ -1,5 +1,5 @@
 /*
-  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
   This file is licensed to you under your choice of the GNU Lesser
@@ -8,1747 +8,884 @@
   cases as published by the Free Software Foundation.
 */
 
-#include <libgen.h>
-#include <unistd.h>
-#include <fnmatch.h>
-#include <sys/time.h>
-#include <stdlib.h>
-#include <signal.h>
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
 #include "afr.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
-#include "byte-order.h"
-
-#include "afr-transaction.h"
 #include "afr-self-heal.h"
-#include "afr-self-heal-common.h"
-#include "afr-self-heal-algorithm.h"
-
-int
-afr_sh_data_fail (call_frame_t *frame, xlator_t *this);
-
-static inline gf_boolean_t
-afr_sh_data_proceed (unsigned int success_count)
-{
-        return (success_count >= AFR_SH_MIN_PARTICIPANTS);
-}
-
-extern int
-sh_loop_finish (call_frame_t *loop_frame, xlator_t *this);
-
-int
-afr_post_sh_big_lock_success (call_frame_t *frame, xlator_t *this);
-
-int
-afr_post_sh_big_lock_failure (call_frame_t *frame, xlator_t *this);
-
-int
-afr_sh_data_finish (call_frame_t *frame, xlator_t *this);
-
-int
-afr_sh_data_done (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-
-        local = frame->local;
-        sh = &local->self_heal;
-
-        sh->completion_cbk (frame, this);
-
-        return 0;
-}
-
-
-int
-afr_sh_data_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-        afr_local_t   *local       = NULL;
-        afr_private_t *priv        = NULL;
-        int            call_count  = 0;
-        int            child_index = (long) cookie;
-
-        local = frame->local;
-        priv = this->private;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "flush failed on %s on subvolume %s: %s",
-                                local->loc.path, priv->children[child_index]->name,
-                                strerror (op_errno));
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                afr_sh_data_done (frame, this);
-        }
-
-        return 0;
-}
-
-int
-afr_sh_data_close (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local      = NULL;
-        afr_private_t   *priv       = NULL;
-        afr_self_heal_t *sh         = NULL;
-        int              i          = 0;
-        int              call_count = 0;
-
-        local = frame->local;
-        sh    = &local->self_heal;
-        priv  = this->private;
-
-        if (!sh->healing_fd) {
-                //This happens when file is non-reg
-                afr_sh_data_done (frame, this);
-                return 0;
-        }
-        call_count        = afr_set_elem_count_get (sh->success,
-                                                    priv->child_count);
-        local->call_count = call_count;
-
-        if (call_count == 0) {
-                afr_sh_data_done (frame, this);
-                return 0;
-        }
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (!sh->success[i])
-                        continue;
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "closing fd of %s on %s",
-                        local->loc.path, priv->children[i]->name);
-
-                STACK_WIND_COOKIE (frame, afr_sh_data_flush_cbk,
-                                   (void *) (long) i,
-                                   priv->children[i],
-                                   priv->children[i]->fops->flush,
-                                   sh->healing_fd, NULL);
-
-                if (!--call_count)
-                        break;
-        }
-
-        return 0;
-}
-
-int
-afr_sh_dom_unlock (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh    = NULL;
-        afr_private_t   *priv  = NULL;
-
-        local = frame->local;
-        sh    = &local->self_heal;
-        priv  = this->private;
-
-        if (sh->sh_dom_lock_held)
-                afr_sh_data_unlock (frame, this, priv->sh_domain,
-                                    afr_sh_data_close);
-        else
-                afr_sh_data_close (frame, this);
-        return 0;
-}
-
-int
-afr_sh_data_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, struct iatt *statpre,
-                         struct iatt *statpost, dict_t *xdata)
-{
-
-        afr_local_t   *local       = NULL;
-        afr_private_t *priv        = NULL;
-        int            call_count  = 0;
-        int            child_index = (long) cookie;
-
-        local = frame->local;
-        priv = this->private;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "setattr failed on %s on subvolume %s: %s",
-                                local->loc.path, priv->children[child_index]->name,
-                                strerror (op_errno));
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                afr_sh_data_finish (frame, this);
-        }
-
-        return 0;
-}
-
-int
-afr_sh_data_setattr (call_frame_t *frame, xlator_t *this, struct iatt* stbuf)
-{
-        afr_local_t     *local      = NULL;
-        afr_private_t   *priv       = NULL;
-        afr_self_heal_t *sh         = NULL;
-        int              i          = 0;
-        int              call_count = 0;
-        int32_t          valid      = 0;
-
-        local = frame->local;
-        sh    = &local->self_heal;
-        priv  = this->private;
-
-        valid = (GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME);
-
-        call_count        = afr_set_elem_count_get (sh->success,
-                                                    priv->child_count);
-        local->call_count = call_count;
-
-        if (call_count == 0) {
-                GF_ASSERT (0);
-                afr_sh_data_finish (frame, this);
-                return 0;
-        }
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (!sh->success[i])
-                        continue;
-
-                STACK_WIND_COOKIE (frame, afr_sh_data_setattr_cbk,
-                                   (void *) (long) i,
-                                   priv->children[i],
-                                   priv->children[i]->fops->setattr,
-                                   &local->loc, stbuf, valid, NULL);
-
-                if (!--call_count)
-                        break;
-        }
-
-        return 0;
-}
-
-int
-afr_sh_data_setattr_fstat_cbk (call_frame_t *frame, void *cookie,
-                               xlator_t *this, int32_t op_ret, int32_t op_errno,
-                               struct iatt *buf, dict_t *xdata)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        int child_index = (long) cookie;
-
-        local = frame->local;
-        sh = &local->self_heal;
-
-        GF_ASSERT (sh->source == child_index);
-        if (op_ret != -1) {
-                sh->buf[child_index] = *buf;
-                afr_sh_data_setattr (frame, this, buf);
-        } else {
-                gf_log (this->name, GF_LOG_ERROR, "%s: Failed to set "
-                        "time-stamps after self-heal", local->loc.path);
-                afr_sh_data_fail (frame, this);
-        }
-
-        return 0;
-}
-
-/*
- * If there are any writes after the self-heal is triggered then the
- * stbuf stored in local->self_heal.buf[] will be invalid so we do one more
- * stat on the source and then set the [am]times
- */
-int
-afr_sh_set_timestamps (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local      = NULL;
-        afr_private_t   *priv       = NULL;
-        afr_self_heal_t *sh         = NULL;
-
-        local = frame->local;
-        sh    = &local->self_heal;
-        priv  = this->private;
-
-        STACK_WIND_COOKIE (frame, afr_sh_data_setattr_fstat_cbk,
-                           (void *) (long) sh->source,
-                           priv->children[sh->source],
-                           priv->children[sh->source]->fops->fstat,
-                           sh->healing_fd, NULL);
-        return 0;
-}
+#include <glusterfs/byte-order.h>
+#include "protocol-common.h"
+#include "afr-messages.h"
+#include <glusterfs/events.h>
 
-//Fun fact, lock_cbk is being used for both lock & unlock
-int
-afr_sh_data_unlock (call_frame_t *frame, xlator_t *this, char *dom,
-                    afr_lock_cbk_t lock_cbk)
-{
-        afr_local_t         *local    = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-        afr_self_heal_t     *sh       = NULL;
-        afr_private_t       *priv     = NULL;
-        int                 ret       = 0;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        sh       = &local->self_heal;
-        priv     = this->private;
-
-        if (strcmp (dom, this->name) == 0) {
-                sh->data_lock_held = _gf_false;
-        } else if (strcmp (dom, priv->sh_domain) == 0) {
-                sh->sh_dom_lock_held = _gf_false;
+#define HAS_HOLES(i) ((i->ia_blocks * 512) < (i->ia_size))
+static int
+__checksum_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, uint32_t weak, uint8_t *strong, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    struct afr_reply *replies = NULL;
+    int i = (long)cookie;
+
+    local = frame->local;
+    replies = local->replies;
+
+    replies[i].valid = 1;
+    replies[i].op_ret = op_ret;
+    replies[i].op_errno = op_errno;
+    if (xdata) {
+        replies[i].buf_has_zeroes = dict_get_str_boolean(
+            xdata, "buf-has-zeroes", _gf_false);
+        replies[i].fips_mode_rchecksum = dict_get_str_boolean(
+            xdata, "fips-mode-rchecksum", _gf_false);
+    }
+    if (strong) {
+        if (replies[i].fips_mode_rchecksum) {
+            memcpy(local->replies[i].checksum, strong, SHA256_DIGEST_LENGTH);
         } else {
-                ret = -1;
-                goto out;
-        }
-        int_lock->lock_cbk = lock_cbk;
-        int_lock->domain = dom;
-        afr_unlock (frame, this);
-
-out:
-        if (ret) {
-                int_lock->lock_op_ret = -1;
-                int_lock->lock_cbk (frame, this);
-        }
-        return 0;
-}
-
-int
-afr_sh_data_finish (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh    = NULL;
-
-        local = frame->local;
-        sh = &local->self_heal;
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "finishing data selfheal of %s", local->loc.path);
+            memcpy(local->replies[i].checksum, strong, MD5_DIGEST_LENGTH);
+        }
+    }
+
+    syncbarrier_wake(&local->barrier);
+    return 0;
+}
+
+static gf_boolean_t
+__afr_can_skip_data_block_heal(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                               int source, unsigned char *healed_sinks,
+                               off_t offset, size_t size, struct iatt *poststat)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    unsigned char *wind_subvols = NULL;
+    gf_boolean_t checksum_match = _gf_true;
+    struct afr_reply *replies = NULL;
+    dict_t *xdata = NULL;
+    int i = 0;
+
+    priv = this->private;
+    local = frame->local;
+    replies = local->replies;
+
+    xdata = dict_new();
+    if (!xdata)
+        goto out;
+    if (dict_set_int32_sizen(xdata, "check-zero-filled", 1)) {
+        dict_unref(xdata);
+        goto out;
+    }
+
+    wind_subvols = alloca0(priv->child_count);
+    for (i = 0; i < priv->child_count; i++) {
+        if (i == source || healed_sinks[i])
+            wind_subvols[i] = 1;
+    }
+
+    AFR_ONLIST(wind_subvols, frame, __checksum_cbk, rchecksum, fd, offset, size,
+               xdata);
+    if (xdata)
+        dict_unref(xdata);
+
+    if (!replies[source].valid || replies[source].op_ret != 0)
+        return _gf_false;
 
-        if (sh->data_lock_held)
-                afr_sh_data_unlock (frame, this, this->name, afr_sh_dom_unlock);
+    for (i = 0; i < priv->child_count; i++) {
+        if (i == source)
+            continue;
+        if (replies[i].valid) {
+            if (memcmp(replies[source].checksum, replies[i].checksum,
+                       replies[source].fips_mode_rchecksum
+                           ? SHA256_DIGEST_LENGTH
+                           : MD5_DIGEST_LENGTH)) {
+                checksum_match = _gf_false;
+                break;
+            }
+        }
+    }
+
+    if (checksum_match) {
+        if (HAS_HOLES(poststat))
+            return _gf_true;
+
+        /* For non-sparse files, we might be better off writing the
+         * zeroes to sinks to avoid mismatch of disk-usage in bricks. */
+        if (local->replies[source].buf_has_zeroes)
+            return _gf_false;
         else
-                afr_sh_dom_unlock (frame, this);
-
-        return 0;
-}
-
-int
-afr_sh_data_fail (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t   *local = NULL;
-        afr_self_heal_t *sh = NULL;
-
-        local = frame->local;
-        sh = &local->self_heal;
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "finishing failed data selfheal of %s", local->loc.path);
-
-        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-        afr_sh_data_finish (frame, this);
-        return 0;
-}
-
-int
-afr_sh_data_erase_pending_cbk (call_frame_t *frame, void *cookie,
-                               xlator_t *this, int32_t op_ret,
-                               int32_t op_errno, dict_t *xattr, dict_t *xdata)
-{
-        int             call_count = 0;
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int32_t         child_index = (long) cookie;
-
-        priv  = this->private;
-        local = frame->local;
-        sh    = &local->self_heal;
-        if (op_ret < 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Erasing of pending change "
-                        "log failed on %s for subvol %s, reason: %s",
-                        local->loc.path, priv->children[child_index]->name,
-                        strerror (op_errno));
-                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-        }
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
-                        if (sh->old_loop_frame)
-                                sh_loop_finish (sh->old_loop_frame, this);
-                        sh->old_loop_frame = NULL;
-                        afr_sh_data_fail (frame, this);
-                        goto out;
-                }
-                if (!IA_ISREG (sh->type)) {
-                        afr_sh_data_finish (frame, this);
-                        goto out;
-                }
-                GF_ASSERT (sh->old_loop_frame);
-                afr_sh_data_lock (frame, this, 0, 0, _gf_true, this->name,
-                                  afr_post_sh_big_lock_success,
-                                  afr_post_sh_big_lock_failure);
-        }
+            return _gf_true;
+    }
 out:
-        return 0;
-}
-
-int
-afr_sh_data_erase_pending (call_frame_t *frame, xlator_t *this)
-{
-        afr_sh_erase_pending (frame, this, AFR_DATA_TRANSACTION,
-                              afr_sh_data_erase_pending_cbk,
-                              afr_sh_data_finish);
-        return 0;
-}
-
-int
-afr_sh_data_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int op_ret, int op_errno, struct iatt *pre,
-                       struct iatt *post, dict_t *xdata)
-{
-        afr_local_t     *local      = NULL;
-        afr_private_t   *priv       = NULL;
-        afr_self_heal_t *sh         = NULL;
-        int             call_count  = 0;
-        int             child_index = (long) cookie;
-
-        local = frame->local;
-        priv = this->private;
-        sh   = &local->self_heal;
-
-        if (op_ret < 0) {
-                gf_log (this->name, GF_LOG_ERROR, "%s: Failed to fsync on "
-                        "%s - %s", local->loc.path,
-                        priv->children[child_index]->name, strerror (op_errno));
-                LOCK (&frame->lock);
-                {
-                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-                }
-                UNLOCK (&frame->lock);
-                if (sh->old_loop_frame)
-                        sh_loop_finish (sh->old_loop_frame, this);
-                sh->old_loop_frame = NULL;
-        }
-
-        call_count = afr_frame_return (frame);
-        if (call_count == 0) {
-                if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC))
-                        afr_sh_data_fail (frame, this);
-                else
-                        afr_sh_data_erase_pending (frame, this);
-        }
-        return 0;
-}
-
-/*
- * Before erasing xattrs, make sure the data is written to disk
- */
-int
-afr_sh_data_fsync (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local     = NULL;
-        afr_private_t   *priv      = NULL;
-        afr_self_heal_t *sh        = NULL;
-        int             i          = 0;
-        int             call_count = 0;
-
-        local = frame->local;
-        priv = this->private;
-        sh   = &local->self_heal;
-
-        call_count        = sh->active_sinks;
-        if (call_count == 0) {
-                afr_sh_data_erase_pending (frame, this);
-                return 0;
-        }
-
-        local->call_count = call_count;
-        for (i = 0; i < priv->child_count; i++) {
-                if (!sh->success[i] || sh->sources[i])
-                        continue;
-
-                STACK_WIND_COOKIE (frame, afr_sh_data_fsync_cbk,
-                                   (void *) (long) i, priv->children[i],
-                                   priv->children[i]->fops->fsync,
-                                   sh->healing_fd, 1, NULL);
-        }
-
-        return 0;
-}
-
-static struct afr_sh_algorithm *
-sh_algo_from_name (xlator_t *this, char *name)
-{
-        int i = 0;
-
-        if (name == NULL)
-                goto out;
-
-        while (afr_self_heal_algorithms[i].name) {
-                if (!strcmp (name, afr_self_heal_algorithms[i].name)) {
-                        return &afr_self_heal_algorithms[i];
-                }
-
-                i++;
-        }
-
+    return _gf_false;
+}
+
+static gf_boolean_t
+__afr_is_sink_zero_filled(xlator_t *this, fd_t *fd, size_t size, off_t offset,
+                          int sink)
+{
+    afr_private_t *priv = NULL;
+    struct iobref *iobref = NULL;
+    struct iovec *iovec = NULL;
+    int count = 0;
+    int ret = 0;
+    gf_boolean_t zero_filled = _gf_false;
+
+    priv = this->private;
+    ret = syncop_readv(priv->children[sink], fd, size, offset, 0, &iovec,
+                       &count, &iobref, NULL, NULL, NULL);
+    if (ret < 0)
+        goto out;
+    ret = iov_0filled(iovec, count);
+    if (!ret)
+        zero_filled = _gf_true;
 out:
-        return NULL;
+    if (iovec)
+        GF_FREE(iovec);
+    if (iobref)
+        iobref_unref(iobref);
+    return zero_filled;
 }
 
-
 static int
-sh_zero_byte_files_exist (afr_local_t *local, int child_count)
-{
-        int             i = 0;
-        int             ret = 0;
-        afr_self_heal_t *sh = NULL;
-
-        sh = &local->self_heal;
-        for (i = 0; i < child_count; i++) {
-                if (!local->child_up[i] || sh->child_errno[i])
-                        continue;
-                if (sh->buf[i].ia_size == 0) {
-                        ret = 1;
-                        break;
-                }
-        }
-
+__afr_selfheal_data_read_write(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                               int source, unsigned char *healed_sinks,
+                               off_t offset, size_t size,
+                               struct afr_reply *replies, int type)
+{
+    struct iovec *iovec = NULL;
+    int count = 0;
+    struct iobref *iobref = NULL;
+    int ret = 0;
+    int i = 0;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+
+    ret = syncop_readv(priv->children[source], fd, size, offset, 0, &iovec,
+                       &count, &iobref, NULL, NULL, NULL);
+    if (ret <= 0)
         return ret;
-}
 
+    for (i = 0; i < priv->child_count; i++) {
+        if (!healed_sinks[i])
+            continue;
+
+            /*
+             * TODO: Use fiemap() and discard() to heal holes
+             * in the future.
+             *
+             * For now,
+             *
+             * - if the source had any holes at all,
+             * AND
+             * - if we are writing past the original file size
+             *   of the sink
+             * AND
+             * - is NOT the last block of the source file. if
+             *   the block contains EOF, it has to be written
+             *   in order to set the file size even if the
+             *   last block is 0-filled.
+             * AND
+             * - if the read buffer is filled with only 0's
+             *
+             * then, skip writing to this source. We don't depend
+             * on the write to happen to update the size as we
+             * have performed an ftruncate() upfront anyways.
+             */
+#define is_last_block(o, b, s) ((s >= o) && (s <= (o + b)))
+        if (HAS_HOLES((&replies[source].poststat)) &&
+            offset >= replies[i].poststat.ia_size &&
+            !is_last_block(offset, size, replies[source].poststat.ia_size) &&
+            (iov_0filled(iovec, count) == 0))
+            continue;
+
+        /* Avoid filling up sparse regions of the sink with 0-filled
+         * writes.*/
+        if (type == AFR_SELFHEAL_DATA_FULL &&
+            HAS_HOLES((&replies[source].poststat)) &&
+            ((offset + size) <= replies[i].poststat.ia_size) &&
+            (iov_0filled(iovec, count) == 0) &&
+            __afr_is_sink_zero_filled(this, fd, size, offset, i)) {
+            continue;
+        }
+
+        ret = syncop_writev(priv->children[i], fd, iovec, count, offset, iobref,
+                            0, NULL, NULL, NULL, NULL);
+        if (ret != iov_length(iovec, count)) {
+            /* write() failed on this sink. unset the corresponding
+               member in sinks[] (which is healed_sinks[] in the
+               caller) so that this server does NOT get considered
+               as successfully healed.
+            */
+            healed_sinks[i] = 0;
+        }
+    }
+    if (iovec)
+        GF_FREE(iovec);
+    if (iobref)
+        iobref_unref(iobref);
+
+    return ret;
+}
+
+static gf_boolean_t
+afr_source_sinks_locked(xlator_t *this, unsigned char *locked_on, int source,
+                        unsigned char *healed_sinks)
+{
+    afr_private_t *priv = this->private;
+    int i = 0;
+
+    if (!locked_on[source])
+        return _gf_false;
 
-struct afr_sh_algorithm *
-afr_sh_data_pick_algo (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t *           priv  = NULL;
-        struct afr_sh_algorithm * algo  = NULL;
-        afr_local_t *             local = NULL;
-        afr_self_heal_t *         sh    = NULL;
-
-        priv  = this->private;
-        local = frame->local;
-        sh    = &local->self_heal;
-        algo  = sh_algo_from_name (this, priv->data_self_heal_algorithm);
-
-        if (algo == NULL) {
-                /* option not set, so fall back on heuristics */
-
-                if (sh_zero_byte_files_exist (local, priv->child_count)
-                    || (sh->file_size <= (priv->data_self_heal_window_size *
-                                          this->ctx->page_size))) {
-
-                        /*
-                         * If the file does not exist on one of the subvolumes,
-                         * or a zero-byte file exists (created by entry self-heal)
-                         * the entire content has to be copied anyway, so there
-                         * is no benefit from using the "diff" algorithm.
-                         *
-                         * If the file size is about the same as page size,
-                         * the entire file can be read and written with a few
-                         * (pipelined) STACK_WINDs, which will be faster
-                         * than "diff" which has to read checksums and then
-                         * read and write.
-                         */
-
-                        algo = sh_algo_from_name (this, "full");
-
-                } else {
-                        algo = sh_algo_from_name (this, "diff");
-                }
-        }
+    for (i = 0; i < priv->child_count; i++) {
+        if (healed_sinks[i] && locked_on[i])
+            return _gf_true;
+    }
 
-        return algo;
+    return _gf_false;
 }
 
-
-int
-afr_sh_data_sync_prepare (call_frame_t *frame, xlator_t *this)
+static int
+afr_selfheal_data_block(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                        int source, unsigned char *healed_sinks, off_t offset,
+                        size_t size, int type, struct afr_reply *replies)
 {
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        struct afr_sh_algorithm *sh_algo = NULL;
+    int ret = -1;
+    afr_private_t *priv = NULL;
+    unsigned char *data_lock = NULL;
 
-        local = frame->local;
-        sh = &local->self_heal;
-
-        sh->algo_completion_cbk = afr_sh_data_fsync;
-        sh->algo_abort_cbk      = afr_sh_data_fail;
-
-        sh_algo = afr_sh_data_pick_algo (frame, this);
-
-        sh->algo = sh_algo;
-        sh_algo->fn (frame, this);
-
-        return 0;
-}
+    priv = this->private;
+    data_lock = alloca0(priv->child_count);
 
-int
-afr_sh_data_trim_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                      struct iatt *postbuf, dict_t *xdata)
-{
-        int              call_count = 0;
-        int              child_index = 0;
-        afr_private_t    *priv = NULL;
-        afr_local_t      *local  = NULL;
-        afr_self_heal_t  *sh = NULL;
-
-        priv  = this->private;
-        local = frame->local;
-        sh    = &local->self_heal;
-
-        child_index = (long) cookie;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "ftruncate of %s on subvolume %s failed (%s)",
-                                local->loc.path,
-                                priv->children[child_index]->name,
-                                strerror (op_errno));
-                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-                } else {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "ftruncate of %s on subvolume %s completed",
-                                local->loc.path,
-                                priv->children[child_index]->name);
-                }
+    ret = afr_selfheal_inodelk(frame, this, fd->inode, this->name, offset, size,
+                               data_lock);
+    {
+        if (!afr_source_sinks_locked(this, data_lock, source, healed_sinks)) {
+            ret = -ENOTCONN;
+            goto unlock;
         }
-        UNLOCK (&frame->lock);
 
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC))
-                        afr_sh_data_fail (frame, this);
-                else
-                        afr_sh_data_sync_prepare (frame, this);
+        if (type == AFR_SELFHEAL_DATA_DIFF &&
+            __afr_can_skip_data_block_heal(frame, this, fd, source,
+                                           healed_sinks, offset, size,
+                                           &replies[source].poststat)) {
+            ret = 0;
+            goto unlock;
         }
 
-        return 0;
+        ret = __afr_selfheal_data_read_write(
+            frame, this, fd, source, healed_sinks, offset, size, replies, type);
+    }
+unlock:
+    afr_selfheal_uninodelk(frame, this, fd->inode, this->name, offset, size,
+                           data_lock);
+    return ret;
 }
 
-
-int
-afr_sh_data_trim_sinks (call_frame_t *frame, xlator_t *this)
+static int
+afr_selfheal_data_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                        unsigned char *healed_sinks)
 {
-        afr_private_t * priv = NULL;
-        afr_local_t * local  = NULL;
-        afr_self_heal_t *sh  = NULL;
-        int             *sources = NULL;
-        int              call_count = 0;
-        int              i = 0;
-
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
 
-        priv = this->private;
-        local = frame->local;
-        sh = &local->self_heal;
+    local = frame->local;
+    priv = this->private;
 
-        sources = sh->sources;
-        call_count = sh->active_sinks;
-
-        local->call_count = call_count;
+    if (!priv->ensure_durability)
+        return 0;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (sources[i] || !local->child_up[i])
-                        continue;
-
-                STACK_WIND_COOKIE (frame, afr_sh_data_trim_cbk,
-                                   (void *) (long) i,
-                                   priv->children[i],
-                                   priv->children[i]->fops->ftruncate,
-                                   sh->healing_fd, sh->file_size,
-                                   NULL);
-
-                if (!--call_count)
-                        break;
-        }
+    AFR_ONLIST(healed_sinks, frame, afr_sh_generic_fop_cbk, fsync, fd, 0, NULL);
 
-        return 0;
+    for (i = 0; i < priv->child_count; i++)
+        if (healed_sinks[i] && local->replies[i].op_ret != 0)
+            /* fsync() failed. Do NOT consider this server
+               as successfully healed. Mark it so.
+            */
+            healed_sinks[i] = 0;
+    return 0;
 }
 
-int
-afr_sh_inode_set_read_ctx (afr_self_heal_t *sh, xlator_t *this)
+static int
+afr_data_self_heal_type_get(afr_private_t *priv, unsigned char *healed_sinks,
+                            int source, struct afr_reply *replies)
 {
-        afr_private_t   *priv = NULL;
-        int             ret = 0;
-        int             i = 0;
-
-        priv = this->private;
-        sh->source = afr_sh_select_source (sh->sources, priv->child_count);
-        if (sh->source < 0) {
-                ret = -1;
-                goto out;
-        }
+    int type = AFR_SELFHEAL_DATA_FULL;
+    int i = 0;
 
-        /* detect changes not visible through pending flags -- JIC */
+    if (priv->data_self_heal_algorithm == AFR_SELFHEAL_DATA_DYNAMIC) {
+        type = AFR_SELFHEAL_DATA_FULL;
         for (i = 0; i < priv->child_count; i++) {
-                if (i == sh->source || sh->child_errno[i])
-                        continue;
-
-                if (SIZE_DIFFERS (&sh->buf[i], &sh->buf[sh->source]))
-                        sh->sources[i] = 0;
+            if (!healed_sinks[i] && i != source)
+                continue;
+            if (replies[i].poststat.ia_size) {
+                type = AFR_SELFHEAL_DATA_DIFF;
+                break;
+            }
         }
-
-        afr_reset_children (sh->fresh_children, priv->child_count);
-        afr_get_fresh_children (sh->success_children, sh->sources,
-                                sh->fresh_children, priv->child_count);
-        afr_inode_set_read_ctx (this, sh->inode, sh->source,
-                                sh->fresh_children);
-out:
-        return ret;
+    } else {
+        type = priv->data_self_heal_algorithm;
+    }
+    return type;
 }
 
-char*
-afr_get_sizes_str (afr_local_t *local, struct iatt *bufs, xlator_t *this)
+static int
+afr_selfheal_data_do(call_frame_t *frame, xlator_t *this, fd_t *fd, int source,
+                     unsigned char *healed_sinks, struct afr_reply *replies)
 {
-        afr_private_t *priv = NULL;
-        int           i     = 0;
-        char          num[1024] = {0};
-        size_t        len = 0;
-        char          *sizes_str = NULL;
-        size_t        off = 0;
-        char          *fmt_str = "%llu bytes on %s, ";
-        char          *child_down =  " %s,";
-        char          *child_unknown = " %s,";
-        int           down_child_present = 0;
-        int           down_count = 0;
-        int           unknown_count = 0;
-        int           unknown_child_present = 0;
-        char          *down_subvol_1 = " down subvolume is ";
-        char          *unknown_subvol_1 = " unknown subvolume is ";
-        char          *down_subvol_2 = " down subvolumes are ";
-        char          *unknown_subvol_2 = " unknown subvolumes are ";
-
-        priv = this->private;
+    afr_private_t *priv = NULL;
+    off_t off = 0;
+    size_t block = 0;
+    int type = AFR_SELFHEAL_DATA_FULL;
+    int ret = -1;
+    call_frame_t *iter_frame = NULL;
+    unsigned char arbiter_sink_status = 0;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i] == 1) {
-                        len += snprintf (num, sizeof (num), fmt_str,
-                                         (unsigned long long) bufs[i].ia_size,
-                                         priv->children[i]->name);
-                } else if (local->child_up[i] == 0) {
-                        len += snprintf (num, sizeof (num), child_down,
-                                         priv->children[i]->name);
-                        if (!down_child_present)
-                                down_child_present = 1;
-                        down_count ++;
-                } else if (local->child_up[i] == -1) {
-                        len += snprintf (num, sizeof (num), child_unknown,
-                                         priv->children[i]->name);
-                        if (!unknown_child_present)
-                                unknown_child_present = 1;
-                        unknown_count++;
-                }
-
-        }
-
-        if (down_child_present) {
-                if (down_count > 1)
-                        len += snprintf (num, sizeof (num), "%s",
-                                         down_subvol_2);
-                else
-                        len += snprintf (num, sizeof (num), "%s",
-                                        down_subvol_1);
-        }
-        if (unknown_child_present) {
-                if (unknown_count > 1)
-                        len += snprintf (num, sizeof (num), "%s",
-                                         unknown_subvol_2);
-                else
-                        len += snprintf (num, sizeof (num), "%s",
-                                         unknown_subvol_1);
-        }
-
-        len++;//for '\0'
-
-        sizes_str = GF_CALLOC (len, sizeof (char), gf_common_mt_char);
-
-        if (!sizes_str)
-                return NULL;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i] == 1) {
-                        off += snprintf (sizes_str + off, len - off, fmt_str,
-                                         (unsigned long long) bufs[i].ia_size,
-                                         priv->children[i]->name);
-                }
-        }
+    gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_SELF_HEAL_INFO,
+           "performing data selfheal on %s", uuid_utoa(fd->inode->gfid));
 
-        if (down_child_present) {
-                if (down_count > 1) {
-                        off += snprintf (sizes_str + off, len - off, "%s",
-                                         down_subvol_2);
-                } else {
-                        off += snprintf (sizes_str + off, len - off, "%s",
-                                         down_subvol_1);
-                }
-        }
+    priv = this->private;
+    if (priv->arbiter_count) {
+        arbiter_sink_status = healed_sinks[ARBITER_BRICK_INDEX];
+        healed_sinks[ARBITER_BRICK_INDEX] = 0;
+    }
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i] == 0) {
-                        off += snprintf (sizes_str + off, len - off, child_down,
-                                         priv->children[i]->name);
-                }
-        }
+    block = 128 * 1024 * priv->data_self_heal_window_size;
 
-        if (unknown_child_present) {
-                if (unknown_count > 1) {
-                        off += snprintf (sizes_str + off, len - off, "%s",
-                                        unknown_subvol_2);
-                } else {
-                        off += snprintf (sizes_str + off, len - off, "%s",
-                                         unknown_subvol_1);
-                }
-        }
+    type = afr_data_self_heal_type_get(priv, healed_sinks, source, replies);
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i] == -1) {
-                        off += snprintf (sizes_str + off, len - off,
-                                         child_unknown,
-                                         priv->children[i]->name);
+    iter_frame = afr_copy_frame(frame);
+    if (!iter_frame) {
+        ret = -ENOMEM;
+        goto out;
+    }
 
-                }
+    for (off = 0; off < replies[source].poststat.ia_size; off += block) {
+        if (AFR_COUNT(healed_sinks, priv->child_count) == 0) {
+            ret = -ENOTCONN;
+            goto out;
         }
 
-        return sizes_str;
-}
+        ret = afr_selfheal_data_block(iter_frame, this, fd, source,
+                                      healed_sinks, off, block, type, replies);
+        if (ret < 0)
+            goto out;
 
-char*
-afr_get_sinks_str (xlator_t *this, afr_local_t *local, afr_self_heal_t *sh)
-{
-        afr_private_t   *priv = NULL;
-        int             i = 0;
-        char            num[1024] = {0};
-        size_t          len = 0;
-        char            *sinks_str = NULL;
-        char            *temp_str = " to sinks ";
-        char            *str_format = " %s,";
-        char            off = 0;
-
-        priv = this->private;
-
-        len += snprintf (num, sizeof (num), "%s", temp_str);
-        for (i = 0; i < priv->child_count; i++) {
-                if ((sh->sources[i] == 0) && (local->child_up[i] == 1)) {
-                        len += snprintf (num, sizeof (num), str_format,
-                                         priv->children[i]->name);
-                }
+        AFR_STACK_RESET(iter_frame);
+        if (iter_frame->local == NULL) {
+            ret = -ENOTCONN;
+            goto out;
         }
+    }
 
-        len ++;
-
-        sinks_str = GF_CALLOC (len, sizeof (char), gf_common_mt_char);
-
-        if (!sinks_str)
-                return NULL;
+    ret = afr_selfheal_data_fsync(frame, this, fd, healed_sinks);
 
-        off += snprintf (sinks_str + off, len - off, "%s", temp_str);
-
-        for (i = 0; i < priv->child_count; i++) {
-                if ((sh->sources[i] == 0) && (local->child_up[i] == 1)) {
-                        off += snprintf (sinks_str + off, len - off,
-                                         str_format,
-                                         priv->children[i]->name);
-                }
-        }
-
-        return sinks_str;
+out:
+    if (arbiter_sink_status)
+        healed_sinks[ARBITER_BRICK_INDEX] = arbiter_sink_status;
 
+    if (iter_frame)
+        AFR_STACK_DESTROY(iter_frame);
+    return ret;
 }
 
-
-void
-afr_set_data_sh_info_str (afr_local_t *local, afr_self_heal_t *sh, xlator_t *this)
+static int
+__afr_selfheal_truncate_sinks(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                              unsigned char *healed_sinks, uint64_t size)
 {
-        char            *pending_matrix_str = NULL;
-        char            *sizes_str = NULL;
-        char            *sinks_str = NULL;
-        afr_private_t   *priv = NULL;
-
-        priv = this->private;
-
-        pending_matrix_str = afr_get_pending_matrix_str (sh->pending_matrix,
-                                                         this);
-        if (!pending_matrix_str)
-                pending_matrix_str = "";
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
 
-        sizes_str = afr_get_sizes_str (local, sh->buf, this);
-        if (!sizes_str)
-                sizes_str = "";
+    local = frame->local;
+    priv = this->private;
 
-        sinks_str = afr_get_sinks_str (this, local, sh);
-        if (!sinks_str)
-                sinks_str = "";
+    /* This will send truncate on the arbiter brick as well if it is marked as
+     * sink. If changelog is enabled on the volume it captures truncate as a
+     * data transactions on the arbiter brick. This will help geo-rep to
+     * properly sync the data from master to slave if arbiter is the ACTIVE
+     * brick during syncing and which had got some entries healed for data as
+     * part of self heal.
+     */
+    AFR_ONLIST(healed_sinks, frame, afr_sh_generic_fop_cbk, ftruncate, fd, size,
+               NULL);
 
-        gf_asprintf (&sh->data_sh_info, " data self heal from %s %s with "
-                     "%s data %s", priv->children[sh->source]->name, sinks_str,
-                     sizes_str, pending_matrix_str);
+    for (i = 0; i < priv->child_count; i++)
+        if (healed_sinks[i] && local->replies[i].op_ret == -1)
+            /* truncate() failed. Do NOT consider this server
+               as successfully healed. Mark it so.
+            */
+            healed_sinks[i] = 0;
 
-        if (pending_matrix_str && strcmp (pending_matrix_str, ""))
-                GF_FREE (pending_matrix_str);
-
-        if (sizes_str && strcmp (sizes_str, ""))
-                GF_FREE (sizes_str);
+    return 0;
 }
 
-void
-afr_sh_data_fix (call_frame_t *frame, xlator_t *this)
+gf_boolean_t
+afr_has_source_witnesses(xlator_t *this, unsigned char *sources,
+                         uint64_t *witness)
 {
-        int              source = 0;
-        afr_local_t     *local      = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        source     = sh->source;
-        sh->block_size = this->ctx->page_size;
-        sh->file_size  = sh->buf[source].ia_size;
-
-        if (FILE_HAS_HOLES (&sh->buf[source]))
-                sh->file_has_holes = 1;
-
-        if (sh->background && sh->unwind && !sh->unwound) {
-                sh->unwind (sh->orig_frame, this, sh->op_ret, sh->op_errno,
-                            is_self_heal_failed (sh, AFR_CHECK_SPECIFIC));
-                sh->unwound = _gf_true;
-        }
-
-        afr_sh_mark_source_sinks (frame, this);
-        if (sh->active_sinks == 0) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "no active sinks for performing self-heal on file %s",
-                        local->loc.path);
-                afr_sh_data_finish (frame, this);
-                return;
-        }
+    int i = 0;
+    afr_private_t *priv = NULL;
 
-        gf_log (this->name, GF_LOG_DEBUG,
-                "self-healing file %s from subvolume %s to %d other",
-                local->loc.path, priv->children[sh->source]->name,
-                sh->active_sinks);
+    priv = this->private;
 
-        sh->actual_sh_started = _gf_true;
-        afr_set_self_heal_status (sh, AFR_SELF_HEAL_SYNC_BEGIN);
-        afr_sh_data_trim_sinks (frame, this);
+    for (i = 0; i < priv->child_count; i++) {
+        if (sources[i] && witness[i])
+            return _gf_true;
+    }
+    return _gf_false;
 }
 
-int
-afr_sh_data_fxattrop_fstat_done (call_frame_t *frame, xlator_t *this)
+static gf_boolean_t
+afr_does_size_mismatch(xlator_t *this, unsigned char *sources,
+                       struct afr_reply *replies)
 {
-        afr_local_t     *local      = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int              nsources = 0;
-        int              ret = 0;
-        int             *old_sources = NULL;
-        int             tstamp_source = 0;
-        int             i = 0;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        gf_log (this->name, GF_LOG_DEBUG, "Pending matrix for: %s",
-                lkowner_utoa (&frame->root->lk_owner));
-        if (sh->sync_done) {
-                //store sources before sync so that mtime can be set using the
-                //iatt buf from one of them.
-                old_sources = alloca (priv->child_count*sizeof (*old_sources));
-                memcpy (old_sources, sh->sources,
-                        priv->child_count * sizeof (*old_sources));
-        }
+    int i = 0;
+    afr_private_t *priv = NULL;
+    struct iatt *min = NULL;
+    struct iatt *max = NULL;
 
-        nsources = afr_build_sources (this, sh->xattr, sh->buf, sh->pending_matrix,
-                                      sh->sources, sh->success_children,
-                                      AFR_DATA_TRANSACTION, NULL, _gf_true);
-        if ((nsources == -1)
-            && (priv->favorite_child != -1)
-            && (sh->child_errno[priv->favorite_child] == 0)) {
+    priv = this->private;
 
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Picking favorite child %s as authentic source to "
-                        "resolve conflicting data of %s",
-                        priv->children[priv->favorite_child]->name,
-                        local->loc.path);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid)
+            continue;
 
-                sh->sources[priv->favorite_child] = 1;
+        if (replies[i].op_ret < 0)
+            continue;
 
-                nsources = afr_sh_source_count (sh->sources,
-                                                priv->child_count);
-        }
+        if (!sources[i])
+            continue;
 
-        if (nsources == -1) {
-                afr_sh_print_split_brain_log (sh->pending_matrix, this,
-                                              local->loc.path);
-                afr_set_split_brain (this, sh->inode, DONT_KNOW, SPB);
+        if (AFR_IS_ARBITER_BRICK(priv, i) && (replies[i].poststat.ia_size == 0))
+            continue;
 
-                afr_sh_data_fail (frame, this);
-                return 0;
-        }
+        if (!min)
+            min = &replies[i].poststat;
 
-        afr_set_split_brain (this, sh->inode, DONT_KNOW, NO_SPB);
+        if (!max)
+            max = &replies[i].poststat;
 
-        ret = afr_sh_inode_set_read_ctx (sh, this);
-        if (ret) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "No active sources found.");
+        if (min->ia_size > replies[i].poststat.ia_size)
+            min = &replies[i].poststat;
 
-                afr_sh_data_fail (frame, this);
-                return 0;
-        }
+        if (max->ia_size < replies[i].poststat.ia_size)
+            max = &replies[i].poststat;
+    }
 
-        if (sh->sync_done) {
-                /* Perform setattr from one of the old_sources if possible
-                 * Because only they have the correct mtime, the new sources
-                 * (i.e. old sinks) have mtime from last writev in sync.
-                 */
-                tstamp_source = sh->source;
-                for (i = 0; i < priv->child_count; i++) {
-                        if (old_sources[i] && sh->sources[i])
-                                tstamp_source = i;
-                }
-                afr_sh_data_setattr (frame, this, &sh->buf[tstamp_source]);
-        } else {
-                afr_set_data_sh_info_str (local, sh, this);
-                if (nsources == 0) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "No self-heal needed for %s",
-                                local->loc.path);
-
-                        afr_sh_data_finish (frame, this);
-                        return 0;
-                }
-
-                if (sh->do_data_self_heal &&
-                    afr_data_self_heal_enabled (priv->data_self_heal))
-                        afr_sh_data_fix (frame, this);
-                else
-                        afr_sh_data_finish (frame, this);
-        }
-        return 0;
-}
+    if (min && max) {
+        if (min->ia_size != max->ia_size)
+            return _gf_true;
+    }
 
-int
-afr_lookup_select_read_child_by_txn_type (xlator_t *this, afr_local_t *local,
-                                          dict_t **xattr,
-                                          afr_transaction_type txn_type,
-                                          uuid_t gfid)
-{
-        afr_private_t            *priv      = NULL;
-        int                      read_child = -1;
-        int32_t                  **pending_matrix = NULL;
-        int32_t                  *sources         = NULL;
-        int32_t                  *success_children   = NULL;
-        struct iatt              *bufs            = NULL;
-        int32_t                  nsources         = 0;
-        int32_t                  prev_read_child  = -1;
-        int32_t                  config_read_child = -1;
-        int32_t                  subvol_status = 0;
-
-        priv = this->private;
-        bufs = local->cont.lookup.bufs;
-        success_children = local->cont.lookup.success_children;
-
-        pending_matrix = local->cont.lookup.pending_matrix;
-        sources = local->cont.lookup.sources;
-        memset (sources, 0, sizeof (*sources) * priv->child_count);
-
-        nsources = afr_build_sources (this, xattr, bufs, pending_matrix,
-                                      sources, success_children, txn_type,
-                                      &subvol_status, _gf_false);
-        if (subvol_status & SPLIT_BRAIN) {
-                gf_log (this->name, GF_LOG_DEBUG, "%s: Possible split-brain",
-                        local->loc.path);
-                switch (txn_type) {
-                case AFR_DATA_TRANSACTION:
-                        local->cont.lookup.possible_spb = _gf_true;
-                        nsources = 1;
-                        sources[success_children[0]] = 1;
-                        break;
-                case AFR_ENTRY_TRANSACTION:
-                        read_child = afr_get_no_xattr_dir_read_child (this,
-                                                             success_children,
-                                                             bufs);
-                        sources[read_child] = 1;
-                        nsources = 1;
-                        break;
-                default:
-                        break;
-                }
-        }
-        if (nsources < 0)
-                goto out;
-
-        prev_read_child = local->read_child_index;
-        config_read_child = priv->read_child;
-        read_child = afr_select_read_child_from_policy (success_children,
-                                                        priv->child_count,
-                                                        prev_read_child,
-                                                        config_read_child,
-                                                        sources,
-                                                        priv->hash_mode, gfid);
-out:
-        gf_log (this->name, GF_LOG_DEBUG, "returning read_child: %d",
-                read_child);
-        return read_child;
+    return _gf_false;
 }
 
-int
-afr_sh_data_fstat_cbk (call_frame_t *frame, void *cookie,
-                       xlator_t *this, int32_t op_ret, int32_t op_errno,
-                       struct iatt *buf, dict_t *xdata)
+static void
+afr_mark_biggest_witness_as_source(xlator_t *this, unsigned char *sources,
+                                   uint64_t *witness)
 {
-        afr_private_t   *priv  = NULL;
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        int call_count  = -1;
-        int child_index = (long) cookie;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret != -1) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "fstat of %s on %s succeeded",
-                                local->loc.path,
-                                priv->children[child_index]->name);
-
-                        sh->buf[child_index] = *buf;
-                        sh->success_children[sh->success_count] = child_index;
-                        sh->success_count++;
-                } else {
-                        gf_log (this->name, GF_LOG_ERROR, "%s: fstat failed "
-                                "on %s, reason %s", local->loc.path,
-                                priv->children[child_index]->name,
-                                strerror (op_errno));
-                        sh->child_errno[child_index] = op_errno;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                /* Previous versions of glusterfs might have set
-                 * the pending data xattrs which need to be erased
-                 */
-                if (!afr_sh_data_proceed (sh->success_count)) {
-                        gf_log (this->name, GF_LOG_ERROR, "inspecting metadata "
-                                "succeeded on < %d children, aborting "
-                                "self-heal for %s", AFR_SH_MIN_PARTICIPANTS,
-                                local->loc.path);
-                        afr_sh_data_fail (frame, this);
-                        goto out;
-                }
-                afr_sh_data_fxattrop_fstat_done (frame, this);
-        }
-out:
-        return 0;
-}
+    int i = 0;
+    afr_private_t *priv = NULL;
+    uint64_t biggest_witness = 0;
 
+    priv = this->private;
+    /* Find source with biggest witness count */
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i])
+            continue;
+        if (biggest_witness < witness[i])
+            biggest_witness = witness[i];
+    }
 
-int
-afr_sh_data_fstat (call_frame_t *frame, xlator_t *this)
-{
-        afr_self_heal_t *sh    = NULL;
-        afr_local_t     *local = NULL;
-        afr_private_t   *priv  = NULL;
-        int             call_count = 0;
-        int             i = 0;
-        int             child = 0;
-        int32_t         *fstat_children = NULL;
-
-        priv  = this->private;
-        local = frame->local;
-        sh    = &local->self_heal;
-
-        fstat_children = memdup (sh->success_children,
-                                 sizeof (*fstat_children) * priv->child_count);
-        if (!fstat_children) {
-                afr_sh_data_fail (frame, this);
-                goto out;
-        }
-        call_count = sh->success_count;
-        local->call_count = call_count;
+    /* Mark files with less witness count as not source */
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i])
+            continue;
+        if (witness[i] < biggest_witness)
+            sources[i] = 0;
+    }
 
-        memset (sh->buf, 0, sizeof (*sh->buf) * priv->child_count);
-        afr_reset_children (sh->success_children, priv->child_count);
-        sh->success_count = 0;
-        for (i = 0; i < priv->child_count; i++) {
-                child = fstat_children[i];
-                if (child == -1)
-                        break;
-                STACK_WIND_COOKIE (frame, afr_sh_data_fstat_cbk,
-                                   (void *) (long) child,
-                                   priv->children[child],
-                                   priv->children[child]->fops->fstat,
-                                   sh->healing_fd, NULL);
-                --call_count;
-        }
-        GF_ASSERT (!call_count);
-out:
-        GF_FREE (fstat_children);
-        return 0;
-}
-
-void
-afr_sh_common_fxattrop_resp_handler (call_frame_t *frame, void *cookie,
-                                     xlator_t *this, int32_t op_ret,
-                                     int32_t op_errno, dict_t *xattr)
-{
-        afr_private_t   *priv  = NULL;
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        int child_index = (long) cookie;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret != -1) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "fxattrop of %s on %s succeeded",
-                                local->loc.path,
-                                priv->children[child_index]->name);
-
-                        sh->xattr[child_index] = dict_ref (xattr);
-                        sh->success_children[sh->success_count] = child_index;
-                        sh->success_count++;
-                } else {
-                        gf_log (this->name, GF_LOG_ERROR, "fxattrop of %s "
-                                "failed on %s, reason %s", local->loc.path,
-                                priv->children[child_index]->name,
-                                strerror (op_errno));
-                        sh->child_errno[child_index] = op_errno;
-                }
-        }
-        UNLOCK (&frame->lock);
+    return;
 }
 
-int
-afr_sh_data_fxattrop_cbk (call_frame_t *frame, void *cookie,
-                          xlator_t *this, int32_t op_ret, int32_t op_errno,
-                          dict_t *xattr, dict_t *xdata)
+/* This is a tie breaker function. Only one source be assigned here */
+static void
+afr_mark_newest_file_as_source(xlator_t *this, unsigned char *sources,
+                               struct afr_reply *replies)
 {
-        int             call_count  = -1;
-        afr_local_t     *local  = NULL;
-        afr_self_heal_t *sh     = NULL;
-
-        local = frame->local;
-        sh    = &local->self_heal;
-
-        afr_sh_common_fxattrop_resp_handler (frame, cookie, this, op_ret,
-                                             op_errno, xattr);
-
-        call_count = afr_frame_return (frame);
-        if (call_count == 0) {
-                if (!afr_sh_data_proceed (sh->success_count)) {
-                        gf_log (this->name, GF_LOG_ERROR, "%s, inspecting "
-                                "change log succeeded on < %d children",
-                                local->loc.path, AFR_SH_MIN_PARTICIPANTS);
-                        afr_sh_data_fail (frame, this);
-                        goto out;
-                }
-                afr_sh_data_fstat (frame, this);
-        }
-out:
-        return 0;
-}
+    int i = 0;
+    afr_private_t *priv = NULL;
+    int source = -1;
+    uint32_t max_ctime = 0;
 
+    priv = this->private;
+    /* Find source with latest ctime */
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i])
+            continue;
 
-int
-afr_sh_data_fxattrop (call_frame_t *frame, xlator_t *this)
-{
-        afr_self_heal_t *sh    = NULL;
-        afr_local_t     *local = NULL;
-        afr_private_t   *priv  = NULL;
-        dict_t          **xattr_req;
-        int32_t         *zero_pending = NULL;
-        int call_count = 0;
-        int i = 0;
-        int ret = 0;
-	int j;
-
-        priv  = this->private;
-        local = frame->local;
-        sh    = &local->self_heal;
-
-        call_count = afr_up_children_count (local->child_up,
-                                            priv->child_count);
-
-        local->call_count = call_count;
-
-	xattr_req = GF_CALLOC(priv->child_count, sizeof(struct dict_t *),
-			      gf_afr_mt_dict_t);
-	if (!xattr_req)
-		goto out;
-
-	for (i = 0; i < priv->child_count; i++) {
-		xattr_req[i] = dict_new();
-		if (!xattr_req[i]) {
-			ret = -1;
-			goto out;
-		}
-	}
-
-	for (i = 0; i < priv->child_count; i++) {
-		for (j = 0; j < priv->child_count; j++) {
-			zero_pending = GF_CALLOC (3, sizeof (*zero_pending),
-						  gf_afr_mt_int32_t);
-			if (!zero_pending) {
-				ret = -1;
-				goto out;
-			}
-			ret = dict_set_dynptr (xattr_req[i], priv->pending_key[j],
-					       zero_pending,
-					       3 * sizeof (*zero_pending));
-			if (ret < 0) {
-				gf_log (this->name, GF_LOG_WARNING,
-					"Unable to set dict value");
-				goto out;
-			} else {
-				zero_pending = NULL;
-			}
-		}
-	}
-
-        afr_reset_xattr (sh->xattr, priv->child_count);
-        afr_reset_children (sh->success_children, priv->child_count);
-        memset (sh->child_errno, 0,
-                sizeof (*sh->child_errno) * priv->child_count);
-        sh->success_count = 0;
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND_COOKIE (frame, afr_sh_data_fxattrop_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->fxattrop,
-                                           sh->healing_fd, GF_XATTROP_ADD_ARRAY,
-                                           xattr_req[i], NULL);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-out:
-	if (xattr_req) {
-		for (i = 0; i < priv->child_count; i++)
-			if (xattr_req[i])
-				dict_unref(xattr_req[i]);
-		GF_FREE(xattr_req);
-	}
-
-        if (ret) {
-                GF_FREE (zero_pending);
-                afr_sh_data_fail (frame, this);
+        if (max_ctime <= replies[i].poststat.ia_ctime) {
+            source = i;
+            max_ctime = replies[i].poststat.ia_ctime;
         }
+    }
 
-        return 0;
+    /* Only mark one of the files as source to break ties */
+    memset(sources, 0, sizeof(*sources) * priv->child_count);
+    sources[source] = 1;
 }
 
-int
-afr_sh_data_big_lock_success (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t   *local = NULL;
-        afr_self_heal_t *sh = NULL;
+static int
+__afr_selfheal_data_finalize_source(
+    call_frame_t *frame, xlator_t *this, inode_t *inode, unsigned char *sources,
+    unsigned char *sinks, unsigned char *healed_sinks, unsigned char *locked_on,
+    unsigned char *undid_pending, struct afr_reply *replies, uint64_t *witness)
+{
+    afr_private_t *priv = NULL;
+    int source = -1;
+    int sources_count = 0;
+    priv = this->private;
+
+    sources_count = AFR_COUNT(sources, priv->child_count);
+
+    if ((AFR_CMP(locked_on, healed_sinks, priv->child_count) == 0) ||
+        !sources_count) {
+        /* split brain */
+        source = afr_mark_split_brain_source_sinks(
+            frame, this, inode, sources, sinks, healed_sinks, locked_on,
+            replies, AFR_DATA_TRANSACTION);
+        if (source < 0) {
+            gf_event(EVENT_AFR_SPLIT_BRAIN,
+                     "client-pid=%d;"
+                     "subvol=%s;type=data;"
+                     "file=%s",
+                     this->ctx->cmd_args.client_pid, this->name,
+                     uuid_utoa(inode->gfid));
+            return -EIO;
+        }
+
+        _afr_fav_child_reset_sink_xattrs(
+            frame, this, inode, source, healed_sinks, undid_pending,
+            AFR_DATA_TRANSACTION, locked_on, replies);
+        goto out;
+    }
+
+    /* No split brain at this point. If we were called from
+     * afr_heal_splitbrain_file(), abort.*/
+    if (afr_dict_contains_heal_op(frame))
+        return -EIO;
+
+    /* If there are no witnesses/size-mismatches on sources we are done*/
+    if (!afr_does_size_mismatch(this, sources, replies) &&
+        !afr_has_source_witnesses(this, sources, witness))
+        goto out;
+
+    afr_mark_largest_file_as_source(this, sources, replies);
+    afr_mark_biggest_witness_as_source(this, sources, witness);
+    afr_mark_newest_file_as_source(this, sources, replies);
+    if (priv->arbiter_count)
+        /* Choose non-arbiter brick as source for empty files. */
+        afr_mark_source_sinks_if_file_empty(this, sources, sinks, healed_sinks,
+                                            locked_on, replies,
+                                            AFR_DATA_TRANSACTION);
 
-        local = frame->local;
-        sh = &local->self_heal;
+out:
+    afr_mark_active_sinks(this, sources, locked_on, healed_sinks);
+    source = afr_choose_source_by_policy(priv, sources, AFR_DATA_TRANSACTION);
 
-        sh->data_lock_held = _gf_true;
-        afr_sh_data_fxattrop (frame, this);
-        return 0;
+    return source;
 }
 
+/*
+ * __afr_selfheal_data_prepare:
+ *
+ * This function inspects the on-disk xattrs and determines which subvols
+ * are sources and sinks.
+ *
+ * The return value is the index of the subvolume to be used as the source
+ * for self-healing, or -1 if no healing is necessary/split brain.
+ */
 int
-afr_sh_dom_lock_success (call_frame_t *frame, xlator_t *this)
+__afr_selfheal_data_prepare(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                            unsigned char *locked_on, unsigned char *sources,
+                            unsigned char *sinks, unsigned char *healed_sinks,
+                            unsigned char *undid_pending,
+                            struct afr_reply *replies, unsigned char *pflag)
 {
-        afr_local_t   *local = NULL;
-        afr_self_heal_t *sh = NULL;
-
-        local = frame->local;
-        sh = &local->self_heal;
-
-        sh->sh_dom_lock_held = _gf_true;
-        afr_sh_data_lock (frame, this, 0, 0, _gf_true, this->name,
-                          afr_sh_data_big_lock_success,
-                          afr_sh_data_fail);
-        return 0;
-}
+    int ret = -1;
+    int source = -1;
+    afr_private_t *priv = NULL;
+    uint64_t *witness = NULL;
 
-int
-afr_sh_data_post_blocking_inodelk_cbk (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_self_heal_t     *sh       = NULL;
+    priv = this->private;
 
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        sh       = &local->self_heal;
+    ret = afr_selfheal_unlocked_discover(frame, inode, inode->gfid, replies);
 
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Blocking data inodelks "
-                        "failed for %s. by %s",
-                        local->loc.path, lkowner_utoa (&frame->root->lk_owner));
+    if (ret)
+        return ret;
 
-                sh->data_lock_failure_handler (frame, this);
-        } else {
+    witness = alloca0(priv->child_count * sizeof(*witness));
+    ret = afr_selfheal_find_direction(frame, this, replies,
+                                      AFR_DATA_TRANSACTION, locked_on, sources,
+                                      sinks, witness, pflag);
+    if (ret)
+        return ret;
 
-                gf_log (this->name, GF_LOG_DEBUG, "Blocking data inodelks "
-                        "done for %s by %s. Proceding to self-heal",
-                        local->loc.path, lkowner_utoa (&frame->root->lk_owner));
+    /* Initialize the healed_sinks[] array optimistically to
+       the intersection of to-be-healed (i.e sinks[]) and
+       the list of servers which are up (i.e locked_on[]).
+       As we encounter failures in the healing process, we
+       will unmark the respective servers in the healed_sinks[]
+       array.
+    */
+    AFR_INTERSECT(healed_sinks, sinks, locked_on, priv->child_count);
 
-                sh->data_lock_success_handler (frame, this);
-        }
+    source = __afr_selfheal_data_finalize_source(
+        frame, this, inode, sources, sinks, healed_sinks, locked_on,
+        undid_pending, replies, witness);
+    if (source < 0)
+        return -EIO;
 
-        return 0;
+    return source;
 }
 
-int
-afr_sh_data_post_nonblocking_inodelk_cbk (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_self_heal_t     *sh       = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        sh       = &local->self_heal;
-
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG, "Non Blocking data inodelks "
-                        "failed for %s. by %s",
-                        local->loc.path, lkowner_utoa (&frame->root->lk_owner));
-
-		if (!sh->data_lock_block) {
-			sh->data_lock_failure_handler(frame, this);
-		} else {
-			int_lock->lock_cbk =
-				afr_sh_data_post_blocking_inodelk_cbk;
-			afr_blocking_lock (frame, this);
-		}
-        } else {
+static int
+__afr_selfheal_data(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                    unsigned char *locked_on)
+{
+    afr_private_t *priv = NULL;
+    int ret = -1;
+    unsigned char *sources = NULL;
+    unsigned char *sinks = NULL;
+    unsigned char *data_lock = NULL;
+    unsigned char *healed_sinks = NULL;
+    unsigned char *undid_pending = NULL;
+    struct afr_reply *locked_replies = NULL;
+    int source = -1;
+    gf_boolean_t did_sh = _gf_true;
+    gf_boolean_t is_arbiter_the_only_sink = _gf_false;
+    gf_boolean_t empty_file = _gf_false;
+
+    priv = this->private;
+
+    sources = alloca0(priv->child_count);
+    sinks = alloca0(priv->child_count);
+    healed_sinks = alloca0(priv->child_count);
+    data_lock = alloca0(priv->child_count);
+    undid_pending = alloca0(priv->child_count);
+
+    locked_replies = alloca0(sizeof(*locked_replies) * priv->child_count);
+
+    ret = afr_selfheal_inodelk(frame, this, fd->inode, this->name, 0, 0,
+                               data_lock);
+    {
+        if (ret < priv->child_count) {
+            gf_msg_debug(this->name, 0,
+                         "%s: Skipping "
+                         "self-heal as only %d number "
+                         "of subvolumes "
+                         "could be locked",
+                         uuid_utoa(fd->inode->gfid), ret);
+            ret = -ENOTCONN;
+            goto unlock;
+        }
+
+        ret = __afr_selfheal_data_prepare(frame, this, fd->inode, data_lock,
+                                          sources, sinks, healed_sinks,
+                                          undid_pending, locked_replies, NULL);
+        if (ret < 0)
+            goto unlock;
+
+        if (AFR_COUNT(healed_sinks, priv->child_count) == 0) {
+            did_sh = _gf_false;
+            goto unlock;
+        }
+
+        source = ret;
+
+        if (AFR_IS_ARBITER_BRICK(priv, source)) {
+            empty_file = afr_is_file_empty_on_all_children(priv,
+                                                           locked_replies);
+            if (empty_file)
+                goto restore_time;
+
+            did_sh = _gf_false;
+            goto unlock;
+        }
+
+        ret = __afr_selfheal_truncate_sinks(
+            frame, this, fd, healed_sinks,
+            locked_replies[source].poststat.ia_size);
+        if (ret < 0)
+            goto unlock;
+
+        if (priv->arbiter_count &&
+            AFR_COUNT(healed_sinks, priv->child_count) == 1 &&
+            healed_sinks[ARBITER_BRICK_INDEX]) {
+            is_arbiter_the_only_sink = _gf_true;
+            goto restore_time;
+        }
+        ret = 0;
+    }
+unlock:
+    afr_selfheal_uninodelk(frame, this, fd->inode, this->name, 0, 0, data_lock);
+    if (ret < 0)
+        goto out;
+
+    if (!did_sh)
+        goto out;
+
+    ret = afr_selfheal_data_do(frame, this, fd, source, healed_sinks,
+                               locked_replies);
+    if (ret)
+        goto out;
+restore_time:
+    afr_selfheal_restore_time(frame, this, fd->inode, source, healed_sinks,
+                              locked_replies);
+
+    if (!is_arbiter_the_only_sink && !empty_file) {
+        ret = afr_selfheal_inodelk(frame, this, fd->inode, this->name, 0, 0,
+                                   data_lock);
+        if (ret < priv->child_count) {
+            ret = -ENOTCONN;
+            did_sh = _gf_false;
+            goto skip_undo_pending;
+        }
+    }
+    ret = afr_selfheal_undo_pending(
+        frame, this, fd->inode, sources, sinks, healed_sinks, undid_pending,
+        AFR_DATA_TRANSACTION, locked_replies, data_lock);
+skip_undo_pending:
+    afr_selfheal_uninodelk(frame, this, fd->inode, this->name, 0, 0, data_lock);
+out:
 
-                gf_log (this->name, GF_LOG_DEBUG, "Non Blocking data inodelks "
-                        "done for %s by %s. Proceeding to self-heal",
-                        local->loc.path, lkowner_utoa (&frame->root->lk_owner));
-                sh->data_lock_success_handler (frame, this);
-        }
+    if (did_sh)
+        afr_log_selfheal(fd->inode->gfid, this, ret, "data", source, sources,
+                         healed_sinks);
+    else
+        ret = 1;
 
-        return 0;
+    if (locked_replies)
+        afr_replies_wipe(locked_replies, priv->child_count);
+
+    return ret;
 }
 
 int
-afr_sh_data_lock_rec (call_frame_t *frame, xlator_t *this, char *dom,
-                      off_t start, off_t len)
+afr_selfheal_data_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int32_t op_ret, int32_t op_errno, fd_t *fd,
+                           dict_t *xdata)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_inodelk_t       *inodelk  = NULL;
-        afr_local_t         *local    = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+    afr_local_t *local = NULL;
+    int i = (long)cookie;
 
-        int_lock->transaction_lk_type = AFR_SELFHEAL_LK;
-        int_lock->selfheal_lk_type    = AFR_DATA_SELF_HEAL_LK;
+    local = frame->local;
 
-        afr_set_lock_number (frame, this);
+    local->replies[i].valid = 1;
+    local->replies[i].op_ret = op_ret;
+    local->replies[i].op_errno = op_errno;
 
-        int_lock->lock_cbk         = afr_sh_data_post_nonblocking_inodelk_cbk;
-
-        int_lock->domain = dom;
-        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
-        inodelk->flock.l_start = start;
-        inodelk->flock.l_len   = len;
-        inodelk->flock.l_type  = F_WRLCK;
-
-        afr_nonblocking_inodelk (frame, this);
-
-        return 0;
-}
+    syncbarrier_wake(&local->barrier);
 
-int
-afr_post_sh_big_lock_success (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local     = NULL;
-        afr_self_heal_t *sh        = NULL;
-
-        local = frame->local;
-        sh = &local->self_heal;
-
-        GF_ASSERT (sh->old_loop_frame);
-        sh_loop_finish (sh->old_loop_frame, this);
-        sh->old_loop_frame = NULL;
-        sh->data_lock_held = _gf_true;
-        sh->sync_done = _gf_true;
-        afr_sh_data_fxattrop (frame, this);
-        return 0;
+    return 0;
 }
 
 int
-afr_post_sh_big_lock_failure (call_frame_t *frame, xlator_t *this)
+afr_selfheal_data_open(xlator_t *this, inode_t *inode, fd_t **fd)
 {
-        afr_local_t     *local     = NULL;
-        afr_self_heal_t *sh        = NULL;
+    int ret = 0;
+    fd_t *fd_tmp = NULL;
+    loc_t loc = {
+        0,
+    };
+    call_frame_t *frame = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
 
-        local = frame->local;
-        sh = &local->self_heal;
+    priv = this->private;
 
-        GF_ASSERT (sh->old_loop_frame);
-        sh_loop_finish (sh->old_loop_frame, this);
-        sh->old_loop_frame = NULL;
-        afr_sh_set_timestamps (frame, this);
-        return 0;
-}
+    fd_tmp = fd_create(inode, 0);
+    if (!fd_tmp)
+        return -ENOMEM;
 
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-int
-afr_sh_data_lock (call_frame_t *frame, xlator_t *this,
-                  off_t start, off_t len, gf_boolean_t block,
-                  char *dom, afr_lock_cbk_t success_handler,
-                  afr_lock_cbk_t failure_handler)
-{
-        afr_local_t *   local = NULL;
-        afr_self_heal_t * sh  = NULL;
+    frame = afr_frame_create(this, &ret);
+    if (!frame) {
+        ret = -ret;
+        fd_unref(fd_tmp);
+        goto out;
+    }
+    local = frame->local;
 
-        local = frame->local;
-        sh    = &local->self_heal;
+    AFR_ONLIST(local->child_up, frame, afr_selfheal_data_open_cbk, open, &loc,
+               O_RDWR | O_LARGEFILE, fd_tmp, NULL);
 
-        sh->data_lock_success_handler = success_handler;
-        sh->data_lock_failure_handler = failure_handler;
-	sh->data_lock_block = block;
-        return afr_sh_data_lock_rec (frame, this, dom, start, len);
-}
+    ret = -ENOTCONN;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
 
-int
-afr_sh_data_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int              call_count = 0;
-        int              child_index = 0;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        child_index = (long) cookie;
-
-        /* TODO: some of the open's might fail.
-           In that case, modify cleanup fn to send flush on those
-           fd's which are already open */
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "open of %s failed on child %s (%s)",
-                                local->loc.path,
-                                priv->children[child_index]->name,
-                                strerror (op_errno));
-                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-                } else {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "open of %s succeeded on child %s",
-                                local->loc.path,
-                                priv->children[child_index]->name);
-                }
+        if (local->replies[i].op_ret < 0) {
+            ret = -local->replies[i].op_errno;
+            continue;
         }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
 
-        if (call_count == 0) {
-                if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
-                        afr_sh_data_fail (frame, this);
-                        return 0;
-                }
+        ret = 0;
+        break;
+    }
 
-                gf_log (this->name, GF_LOG_TRACE,
-                        "fd for %s opened, commencing sync",
-                        local->loc.path);
+    if (ret < 0) {
+        fd_unref(fd_tmp);
+        goto out;
+    } else {
+        fd_bind(fd_tmp);
+    }
 
-                afr_sh_data_lock (frame, this, 0, 0, _gf_true, priv->sh_domain,
-                                  afr_sh_dom_lock_success, afr_sh_data_fail);
-        }
-
-        return 0;
+    *fd = fd_tmp;
+out:
+    loc_wipe(&loc);
+    if (frame)
+        AFR_STACK_DESTROY(frame);
+    return ret;
 }
 
-
 int
-afr_sh_data_open (call_frame_t *frame, xlator_t *this)
+afr_selfheal_data(call_frame_t *frame, xlator_t *this, fd_t *fd)
 {
-        int i = 0;
-        int call_count = 0;
-        fd_t *fd = NULL;
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
-        afr_self_heal_t *sh = NULL;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        call_count = afr_up_children_count (local->child_up, priv->child_count);
-        local->call_count = call_count;
-
-        fd = fd_create (local->loc.inode, frame->root->pid);
-        sh->healing_fd = fd;
-
-        /* open sinks */
-        for (i = 0; i < priv->child_count; i++) {
-                if(!local->child_up[i])
-                        continue;
-
-                STACK_WIND_COOKIE (frame, afr_sh_data_open_cbk,
-                                   (void *) (long) i,
-                                   priv->children[i],
-                                   priv->children[i]->fops->open,
-                                   &local->loc,
-                                   O_RDWR|O_LARGEFILE, fd, NULL);
-
-                if (!--call_count)
-                        break;
-        }
+    afr_private_t *priv = NULL;
+    unsigned char *locked_on = NULL;
+    int ret = 0;
+    inode_t *inode = fd->inode;
 
-        return 0;
-}
-
-void
-afr_sh_non_reg_fix (call_frame_t *frame, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno)
-{
-        afr_private_t   *priv = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_local_t     *local = NULL;
-        int             i = 0;
-
-        if (op_ret < 0) {
-                afr_sh_data_fail (frame, this);
-                return;
-        }
+    priv = this->private;
 
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
+    locked_on = alloca0(priv->child_count);
 
-        for (i = 0; i < priv->child_count ; i++) {
-                if (1 == local->child_up[i])
-                        sh->success[i] = 1;
+    ret = afr_selfheal_tie_breaker_inodelk(frame, this, inode, priv->sh_domain,
+                                           0, 0, locked_on);
+    {
+        if (ret < priv->child_count) {
+            gf_msg_debug(this->name, 0,
+                         "%s: Skipping "
+                         "self-heal as only %d number of "
+                         "subvolumes could be locked",
+                         uuid_utoa(fd->inode->gfid), ret);
+            /* Either less than two subvols available, or another
+               selfheal (from another server) is in progress. Skip
+               for now in any case there isn't anything to do.
+            */
+            ret = -ENOTCONN;
+            goto unlock;
         }
 
-        afr_sh_erase_pending (frame, this, AFR_DATA_TRANSACTION,
-                              afr_sh_data_erase_pending_cbk,
-                              afr_sh_data_finish);
-}
-
-int
-afr_sh_non_reg_lock_success (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t   *local = NULL;
-        afr_self_heal_t *sh = NULL;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        sh->data_lock_held = _gf_true;
-        afr_sh_common_lookup (frame, this, &local->loc,
-                              afr_sh_non_reg_fix, NULL,
-                              AFR_LOOKUP_FAIL_CONFLICTS |
-                              AFR_LOOKUP_FAIL_MISSING_GFIDS,
-                              NULL);
-        return 0;
-}
-
-gf_boolean_t
-afr_can_start_data_self_heal (afr_self_heal_t *sh, afr_private_t *priv)
-{
-        if (sh->force_confirm_spb)
-                return _gf_true;
-        if (sh->do_data_self_heal &&
-            afr_data_self_heal_enabled (priv->data_self_heal))
-                return _gf_true;
-        return _gf_false;
-}
+        ret = __afr_selfheal_data(frame, this, fd, locked_on);
+    }
+unlock:
+    afr_selfheal_uninodelk(frame, this, inode, priv->sh_domain, 0, 0,
+                           locked_on);
 
-int
-afr_self_heal_data (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh    = NULL;
-        afr_private_t   *priv  = this->private;
-        int             ret    = -1;
-
-        local = frame->local;
-        sh = &local->self_heal;
-
-        sh->sh_type_in_action = AFR_SELF_HEAL_DATA;
-
-        if (afr_can_start_data_self_heal (sh, priv)) {
-                afr_set_self_heal_status (sh, AFR_SELF_HEAL_STARTED);
-                ret = afr_inodelk_init (&local->internal_lock.inodelk[1],
-                                        priv->sh_domain, priv->child_count);
-                if (ret < 0) {
-                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-                        afr_sh_data_done (frame, this);
-                        return 0;
-                }
-
-                if (IA_ISREG (sh->type)) {
-                        afr_sh_data_open (frame, this);
-                } else {
-                        afr_sh_data_lock (frame, this, 0, 0, _gf_true,
-                                          this->name,
-                                          afr_sh_non_reg_lock_success,
-                                          afr_sh_data_fail);
-                }
-        } else {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "not doing data self heal on %s",
-                        local->loc.path);
-                afr_sh_data_done (frame, this);
-        }
-
-        return 0;
+    return ret;
 }
diff --git a/xlators/cluster/afr/src/afr-self-heal-entry.c b/xlators/cluster/afr/src/afr-self-heal-entry.c
index 53491a1d74c..64893f441e3 100644
--- a/xlators/cluster/afr/src/afr-self-heal-entry.c
+++ b/xlators/cluster/afr/src/afr-self-heal-entry.c
@@ -1,5 +1,5 @@
 /*
-  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
   This file is licensed to you under your choice of the GNU Lesser
@@ -8,2363 +8,1269 @@
   cases as published by the Free Software Foundation.
 */
 
-#include <libgen.h>
-#include <unistd.h>
-#include <fnmatch.h>
-#include <sys/time.h>
-#include <stdlib.h>
-#include <signal.h>
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "inode.h"
 #include "afr.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
-#include "byte-order.h"
-
-#include "afr-transaction.h"
 #include "afr-self-heal.h"
-#include "afr-self-heal-common.h"
-
-#define AFR_INIT_SH_FRAME_VALS(_frame, _local, _sh, _sh_frame, _sh_local, _sh_sh)\
-        do {\
-                _local = _frame->local;\
-                _sh = &_local->self_heal;\
-                _sh_frame = _sh->sh_frame;\
-                _sh_local = _sh_frame->local;\
-                _sh_sh    = &_sh_local->self_heal;\
-        } while (0);
-
-int
-afr_sh_entry_impunge_create_file (call_frame_t *impunge_frame, xlator_t *this,
-                                  int child_index);
-int
-afr_sh_entry_done (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-
-        local = frame->local;
-        sh = &local->self_heal;
-
-        sh->completion_cbk (frame, this);
-
-        return 0;
-}
-
-
-int
-afr_sh_entry_unlock (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t         *local    = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        int_lock->lock_cbk = afr_sh_entry_done;
-        afr_unlock (frame, this);
-
-        return 0;
-}
-
-
-int
-afr_sh_entry_finish (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t   *local = NULL;
-
-        local = frame->local;
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "finishing entry selfheal of %s", local->loc.path);
-
-        afr_sh_entry_unlock (frame, this);
-
-        return 0;
-}
-
-
-int
-afr_sh_entry_erase_pending_cbk (call_frame_t *frame, void *cookie,
-                                xlator_t *this, int32_t op_ret,
-                                int32_t op_errno, dict_t *xattr, dict_t *xdata)
-{
-        long                 i          = 0;
-        int                  call_count = 0;
-        afr_local_t         *local      = NULL;
-        afr_self_heal_t     *sh         = NULL;
-        afr_local_t         *orig_local = NULL;
-        call_frame_t        *orig_frame = NULL;
-        afr_private_t       *priv       = NULL;
-        int32_t             read_child  = -1;
-
-        local = frame->local;
-        priv  = this->private;
-        sh = &local->self_heal;
-        i = (long)cookie;
-
-
-        afr_children_add_child (sh->fresh_children, i, priv->child_count);
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "%s: failed to erase pending xattrs on %s (%s)",
-                        local->loc.path, priv->children[i]->name,
-                        strerror (op_errno));
-        }
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                if (sh->source == -1) {
-                        //this happens if the forced merge option is set
-                        read_child = sh->fresh_children[0];
-                } else {
-                        read_child = sh->source;
-                }
-                afr_inode_set_read_ctx (this, sh->inode, read_child,
-                                        sh->fresh_children);
-                orig_frame = sh->orig_frame;
-                orig_local = orig_frame->local;
-
-                if (sh->source != -1) {
-                        orig_local->cont.lookup.buf.ia_nlink = sh->buf[sh->source].ia_nlink;
-                }
-
-                afr_sh_entry_finish (frame, this);
-        }
-
-        return 0;
-}
-
+#include <glusterfs/byte-order.h>
+#include "afr-transaction.h"
+#include "afr-messages.h"
+#include <glusterfs/syncop-utils.h>
+#include <glusterfs/events.h>
 
 int
-afr_sh_entry_erase_pending (call_frame_t *frame, xlator_t *this)
+afr_selfheal_entry_anon_inode(xlator_t *this, inode_t *dir, const char *name,
+                              inode_t *inode, int child,
+                              struct afr_reply *replies,
+                              gf_boolean_t *anon_inode)
 {
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-
-        local = frame->local;
-        sh = &local->self_heal;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    xlator_t *subvol = NULL;
+    int ret = 0;
+    int i = 0;
+    char g[64] = {0};
+    unsigned char *lookup_success = NULL;
+    call_frame_t *frame = NULL;
+    loc_t loc2 = {
+        0,
+    };
+    loc_t loc = {
+        0,
+    };
+
+    priv = this->private;
+    subvol = priv->children[child];
+    lookup_success = alloca0(priv->child_count);
+    uuid_utoa_r(replies[child].poststat.ia_gfid, g);
+    loc.inode = inode_new(inode->table);
+    if (!loc.inode) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    if (replies[child].poststat.ia_type == IA_IFDIR) {
+        /* This directory may have sub-directory hierarchy which may need to
+         * be preserved for subsequent heals. So unconditionally move the
+         * directory to anonymous-inode directory*/
+        *anon_inode = _gf_true;
+        goto anon_inode;
+    }
+
+    frame = afr_frame_create(this, &ret);
+    if (!frame) {
+        ret = -ret;
+        goto out;
+    }
+    local = frame->local;
+    gf_uuid_copy(loc.gfid, replies[child].poststat.ia_gfid);
+    AFR_ONLIST(local->child_up, frame, afr_selfheal_discover_cbk, lookup, &loc,
+               NULL);
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->replies[i].op_ret == 0) {
+            lookup_success[i] = 1;
+        } else if (local->replies[i].op_errno != ENOENT &&
+                   local->replies[i].op_errno != ESTALE) {
+            ret = -local->replies[i].op_errno;
+        }
+    }
+
+    if (priv->quorum_count) {
+        if (afr_has_quorum(lookup_success, this, NULL)) {
+            *anon_inode = _gf_true;
+        }
+    } else if (AFR_COUNT(lookup_success, priv->child_count) > 1) {
+        *anon_inode = _gf_true;
+    } else if (ret) {
+        goto out;
+    }
+
+anon_inode:
+    if (!*anon_inode) {
+        ret = 0;
+        goto out;
+    }
+
+    loc.parent = inode_ref(dir);
+    gf_uuid_copy(loc.pargfid, dir->gfid);
+    loc.name = name;
+
+    ret = afr_anon_inode_create(this, child, &loc2.parent);
+    if (ret < 0)
+        goto out;
+
+    loc2.name = g;
+    ret = syncop_rename(subvol, &loc, &loc2, NULL, NULL);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, AFR_MSG_EXPUNGING_FILE_OR_DIR,
+               "Rename to %s dir %s/%s (%s) on %s failed",
+               priv->anon_inode_name, uuid_utoa(dir->gfid), name, g,
+               subvol->name);
+    } else {
+        gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_EXPUNGING_FILE_OR_DIR,
+               "Rename to %s dir %s/%s (%s) on %s successful",
+               priv->anon_inode_name, uuid_utoa(dir->gfid), name, g,
+               subvol->name);
+    }
 
-        if (sh->entries_skipped) {
-                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-                goto out;
-        }
-        afr_sh_erase_pending (frame, this, AFR_ENTRY_TRANSACTION,
-                              afr_sh_entry_erase_pending_cbk,
-                              afr_sh_entry_finish);
-        return 0;
 out:
-        afr_sh_entry_finish (frame, this);
-        return 0;
-}
-
-
-
-static int
-next_active_source (call_frame_t *frame, xlator_t *this,
-                    int current_active_source)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *local  = NULL;
-        afr_self_heal_t *sh  = NULL;
-        int              source = -1;
-        int              next_active_source = -1;
-        int              i = 0;
-
-        priv = this->private;
-        local = frame->local;
-        sh = &local->self_heal;
-
-        source = sh->source;
-
-        if (source != -1) {
-                if (current_active_source != source)
-                        next_active_source = source;
-                goto out;
-        }
-
-        /*
-          the next active sink becomes the source for the
-          'conservative decision' of merging all entries
-        */
-
-        for (i = 0; i < priv->child_count; i++) {
-                if ((sh->sources[i] == 0)
-                    && (local->child_up[i] == 1)
-                    && (i > current_active_source)) {
+    loc_wipe(&loc);
+    loc_wipe(&loc2);
+    if (frame) {
+        AFR_STACK_DESTROY(frame);
+    }
 
-                        next_active_source = i;
-                        break;
-                }
-        }
-out:
-        return next_active_source;
+    return ret;
 }
 
-
-
-static int
-next_active_sink (call_frame_t *frame, xlator_t *this,
-                  int current_active_sink)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *local  = NULL;
-        afr_self_heal_t *sh  = NULL;
-        int              next_active_sink = -1;
-        int              i = 0;
-
-        priv = this->private;
-        local = frame->local;
-        sh = &local->self_heal;
-
-        /*
-          the next active sink becomes the source for the
-          'conservative decision' of merging all entries
-        */
-
-        for (i = 0; i < priv->child_count; i++) {
-                if ((sh->sources[i] == 0)
-                    && (local->child_up[i] == 1)
-                    && (i > current_active_sink)) {
-
-                        next_active_sink = i;
-                        break;
-                }
-        }
-
-        return next_active_sink;
-}
-
-int
-afr_sh_entry_impunge_all (call_frame_t *frame, xlator_t *this);
-
-int
-afr_sh_entry_impunge_subvol (call_frame_t *frame, xlator_t *this);
-
-int
-afr_sh_entry_expunge_all (call_frame_t *frame, xlator_t *this);
-
-int
-afr_sh_entry_expunge_subvol (call_frame_t *frame, xlator_t *this,
-                             int active_src);
-
-int
-afr_sh_entry_expunge_entry_done (call_frame_t *frame, xlator_t *this,
-                                 int active_src, int32_t op_ret,
-                                 int32_t op_errno)
-{
-        int              call_count = 0;
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0)
-                afr_sh_entry_expunge_subvol (frame, this, active_src);
-
-        return 0;
-}
-
-int
-afr_sh_entry_expunge_parent_setattr_cbk (call_frame_t *expunge_frame,
-                                         void *cookie, xlator_t *this,
-                                         int32_t op_ret, int32_t op_errno,
-                                         struct iatt *preop, struct iatt *postop,
-                                         dict_t *xdata)
-{
-        afr_private_t   *priv          = NULL;
-        afr_local_t     *expunge_local = NULL;
-        afr_self_heal_t *expunge_sh    = NULL;
-        call_frame_t    *frame         = NULL;
-        int              active_src    = (long) cookie;
-        afr_self_heal_t *sh            = NULL;
-        afr_local_t     *local         = NULL;
-
-        priv          = this->private;
-        expunge_local = expunge_frame->local;
-        expunge_sh    = &expunge_local->self_heal;
-        frame         = expunge_sh->sh_frame;
-        local         = frame->local;
-        sh            = &local->self_heal;
-
-        if (op_ret != 0) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "setattr on parent directory of %s on subvolume %s failed: %s",
-                        expunge_local->loc.path,
-                        priv->children[active_src]->name, strerror (op_errno));
-        }
-
-        AFR_STACK_DESTROY (expunge_frame);
-        sh->expunge_done (frame, this, active_src, op_ret, op_errno);
-
-        return 0;
-}
-
-
 int
-afr_sh_entry_expunge_remove_cbk (call_frame_t *expunge_frame, void *cookie,
-                                 xlator_t *this,
-                                 int32_t op_ret, int32_t op_errno,
-                                 struct iatt *preparent,
-                                 struct iatt *postparent, dict_t *xdata)
+afr_selfheal_entry_delete(xlator_t *this, inode_t *dir, const char *name,
+                          inode_t *inode, int child, struct afr_reply *replies)
 {
-        afr_private_t   *priv = NULL;
-        afr_local_t     *expunge_local = NULL;
-        afr_self_heal_t *expunge_sh = NULL;
-        int              active_src = 0;
-        int32_t          valid = 0;
-
-        priv = this->private;
-        expunge_local = expunge_frame->local;
-        expunge_sh = &expunge_local->self_heal;
-
-        active_src = (long) cookie;
-
-        if (op_ret == 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "removed %s on %s",
-                        expunge_local->loc.path,
-                        priv->children[active_src]->name);
-        } else {
-                gf_log (this->name, GF_LOG_INFO,
-                        "removing %s on %s failed (%s)",
-                        expunge_local->loc.path,
-                        priv->children[active_src]->name,
-                        strerror (op_errno));
-        }
-
-        valid = GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME;
-
-        STACK_WIND_COOKIE (expunge_frame, afr_sh_entry_expunge_parent_setattr_cbk,
-                           (void *) (long) active_src,
-                           priv->children[active_src],
-                           priv->children[active_src]->fops->setattr,
-                           &expunge_sh->parent_loc,
-                           &expunge_sh->parentbuf,
-                           valid, NULL);
-
-        return 0;
-}
-
-
-int
-afr_sh_entry_expunge_unlink (call_frame_t *expunge_frame, xlator_t *this,
-                             int active_src)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *expunge_local = NULL;
-
-        priv          = this->private;
-        expunge_local = expunge_frame->local;
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "expunging file %s on %s",
-                expunge_local->loc.path, priv->children[active_src]->name);
-
-        STACK_WIND_COOKIE (expunge_frame, afr_sh_entry_expunge_remove_cbk,
-                           (void *) (long) active_src,
-                           priv->children[active_src],
-                           priv->children[active_src]->fops->unlink,
-                           &expunge_local->loc, 0, NULL);
-
-        return 0;
-}
-
-
-
-int
-afr_sh_entry_expunge_rmdir (call_frame_t *expunge_frame, xlator_t *this,
-                            int active_src)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *expunge_local = NULL;
-
-        priv          = this->private;
-        expunge_local = expunge_frame->local;
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "expunging directory %s on %s",
-                expunge_local->loc.path, priv->children[active_src]->name);
-
-        STACK_WIND_COOKIE (expunge_frame, afr_sh_entry_expunge_remove_cbk,
-                           (void *) (long) active_src,
-                           priv->children[active_src],
-                           priv->children[active_src]->fops->rmdir,
-                           &expunge_local->loc, 1, NULL);
-
-        return 0;
-}
-
-
-int
-afr_sh_entry_expunge_remove (call_frame_t *expunge_frame, xlator_t *this,
-                             int active_src, struct iatt *buf,
-                             struct iatt *parentbuf)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *expunge_local = NULL;
-        afr_self_heal_t *expunge_sh = NULL;
-        call_frame_t    *frame = NULL;
-        int              type = 0;
-        afr_self_heal_t *sh            = NULL;
-        afr_local_t     *local         = NULL;
-        loc_t           *loc           = NULL;
-
-        priv = this->private;
-        expunge_local = expunge_frame->local;
-        expunge_sh = &expunge_local->self_heal;
-        frame = expunge_sh->sh_frame;
-        local         = frame->local;
-        sh            = &local->self_heal;
-        loc           = &expunge_local->loc;
-
-        type = buf->ia_type;
-        if (loc->parent && uuid_is_null (loc->parent->gfid))
-                uuid_copy (loc->pargfid, parentbuf->ia_gfid);
-
-        switch (type) {
-        case IA_IFSOCK:
-        case IA_IFREG:
-        case IA_IFBLK:
-        case IA_IFCHR:
-        case IA_IFIFO:
-        case IA_IFLNK:
-                afr_sh_entry_expunge_unlink (expunge_frame, this, active_src);
-                break;
+    char g[64] = {0};
+    afr_private_t *priv = NULL;
+    xlator_t *subvol = NULL;
+    int ret = 0;
+    loc_t loc = {
+        0,
+    };
+    gf_boolean_t anon_inode = _gf_false;
+
+    priv = this->private;
+    subvol = priv->children[child];
+
+    if ((!replies[child].valid) || (replies[child].op_ret < 0)) {
+        /*Nothing to do*/
+        ret = 0;
+        goto out;
+    }
+
+    if (priv->use_anon_inode) {
+        ret = afr_selfheal_entry_anon_inode(this, dir, name, inode, child,
+                                            replies, &anon_inode);
+        if (ret < 0 || anon_inode)
+            goto out;
+    }
+
+    loc.parent = inode_ref(dir);
+    loc.inode = inode_new(inode->table);
+    if (!loc.inode) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    loc.name = name;
+    switch (replies[child].poststat.ia_type) {
         case IA_IFDIR:
-                afr_sh_entry_expunge_rmdir (expunge_frame, this, active_src);
-                break;
+            gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_EXPUNGING_FILE_OR_DIR,
+                   "expunging dir %s/%s (%s) on %s", uuid_utoa(dir->gfid), name,
+                   uuid_utoa_r(replies[child].poststat.ia_gfid, g),
+                   subvol->name);
+            ret = syncop_rmdir(subvol, &loc, 1, NULL, NULL);
+            break;
         default:
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s has unknown file type on %s: 0%o",
-                        expunge_local->loc.path,
-                        priv->children[active_src]->name, type);
-                goto out;
-                break;
-        }
+            gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_EXPUNGING_FILE_OR_DIR,
+                   "expunging file %s/%s (%s) on %s", uuid_utoa(dir->gfid),
+                   name, uuid_utoa_r(replies[child].poststat.ia_gfid, g),
+                   subvol->name);
+            ret = syncop_unlink(subvol, &loc, NULL, NULL);
+            break;
+    }
 
-        return 0;
 out:
-        AFR_STACK_DESTROY (expunge_frame);
-        sh->expunge_done (frame, this, active_src, -1, EINVAL);
-
-        return 0;
+    loc_wipe(&loc);
+    return ret;
 }
 
-
 int
-afr_sh_entry_expunge_lookup_cbk (call_frame_t *expunge_frame, void *cookie,
-                                 xlator_t *this,
-                                 int32_t op_ret, int32_t op_errno,
-                                 inode_t *inode, struct iatt *buf, dict_t *x,
-                                 struct iatt *postparent)
+afr_selfheal_recreate_entry(call_frame_t *frame, int dst, int source,
+                            unsigned char *sources, inode_t *dir,
+                            const char *name, inode_t *inode,
+                            struct afr_reply *replies)
 {
-        afr_private_t   *priv = NULL;
-        afr_local_t     *expunge_local = NULL;
-        afr_self_heal_t *expunge_sh = NULL;
-        call_frame_t    *frame = NULL;
-        int              active_src = 0;
-        afr_self_heal_t *sh            = NULL;
-        afr_local_t     *local         = NULL;
-
-        priv = this->private;
-        expunge_local = expunge_frame->local;
-        expunge_sh = &expunge_local->self_heal;
-        frame = expunge_sh->sh_frame;
-        active_src = (long) cookie;
-        local         = frame->local;
-        sh            = &local->self_heal;
-
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "lookup of %s on %s failed (%s)",
-                        expunge_local->loc.path,
-                        priv->children[active_src]->name,
-                        strerror (op_errno));
+    int ret = 0;
+    loc_t loc = {
+        0,
+    };
+    loc_t srcloc = {
+        0,
+    };
+    loc_t anonloc = {
+        0,
+    };
+    xlator_t *this = frame->this;
+    afr_private_t *priv = NULL;
+    dict_t *xdata = NULL;
+    struct iatt *iatt = NULL;
+    char *linkname = NULL;
+    mode_t mode = 0;
+    struct iatt newent = {
+        0,
+    };
+    unsigned char *newentry = NULL;
+    char iatt_uuid_str[64] = {0};
+    char dir_uuid_str[64] = {0};
+
+    priv = this->private;
+    iatt = &replies[source].poststat;
+    uuid_utoa_r(iatt->ia_gfid, iatt_uuid_str);
+    if (iatt->ia_type == IA_INVAL || gf_uuid_is_null(iatt->ia_gfid)) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SELF_HEAL_FAILED,
+               "Invalid ia_type (%d) or gfid(%s). source brick=%d, "
+               "pargfid=%s, name=%s",
+               iatt->ia_type, iatt_uuid_str, source,
+               uuid_utoa_r(dir->gfid, dir_uuid_str), name);
+        ret = -EINVAL;
+        goto out;
+    }
+
+    xdata = dict_new();
+    if (!xdata)
+        return -ENOMEM;
+    newentry = alloca0(priv->child_count);
+    loc.parent = inode_ref(dir);
+    gf_uuid_copy(loc.pargfid, dir->gfid);
+    loc.name = name;
+    loc.inode = inode_ref(inode);
+
+    ret = afr_selfheal_entry_delete(this, dir, name, inode, dst, replies);
+    if (ret)
+        goto out;
+
+    ret = dict_set_gfuuid(xdata, "gfid-req", replies[source].poststat.ia_gfid,
+                          true);
+    if (ret)
+        goto out;
+
+    srcloc.inode = inode_ref(inode);
+    gf_uuid_copy(srcloc.gfid, iatt->ia_gfid);
+    ret = syncop_lookup(priv->children[dst], &srcloc, 0, 0, 0, 0);
+    if (ret == -ENOENT || ret == -ESTALE) {
+        newentry[dst] = 1;
+        ret = afr_selfheal_newentry_mark(frame, this, inode, source, replies,
+                                         sources, newentry);
+        if (ret)
+            goto out;
+    } else if (ret == 0 && iatt->ia_type == IA_IFDIR && priv->use_anon_inode) {
+        // Try rename from hidden directory
+        ret = afr_anon_inode_create(this, dst, &anonloc.parent);
+        if (ret < 0)
+            goto out;
+        anonloc.inode = inode_ref(inode);
+        anonloc.name = iatt_uuid_str;
+        ret = syncop_rename(priv->children[dst], &anonloc, &loc, NULL, NULL);
+        if (ret == -ENOENT || ret == -ESTALE)
+            ret = -1; /*This sets 'mismatch' to true*/
+        goto out;
+    }
+
+    mode = st_mode_from_ia(iatt->ia_prot, iatt->ia_type);
+
+    switch (iatt->ia_type) {
+        case IA_IFDIR:
+            ret = syncop_mkdir(priv->children[dst], &loc, mode, 0, xdata, NULL);
+            break;
+        case IA_IFLNK:
+            if (!newentry[dst]) {
+                ret = syncop_link(priv->children[dst], &srcloc, &loc, &newent,
+                                  NULL, NULL);
+            } else {
+                ret = syncop_readlink(priv->children[source], &srcloc,
+                                      &linkname, 4096, NULL, NULL);
+                if (ret <= 0)
+                    goto out;
+                ret = syncop_symlink(priv->children[dst], &loc, linkname, NULL,
+                                     xdata, NULL);
+            }
+            break;
+        default:
+            ret = dict_set_int32_sizen(xdata, GLUSTERFS_INTERNAL_FOP_KEY, 1);
+            if (ret)
                 goto out;
-        }
-
-        afr_sh_entry_expunge_remove (expunge_frame, this, active_src, buf,
-                                     postparent);
+            ret = syncop_mknod(
+                priv->children[dst], &loc, mode,
+                makedev(ia_major(iatt->ia_rdev), ia_minor(iatt->ia_rdev)),
+                &newent, xdata, NULL);
+            break;
+    }
 
-        return 0;
 out:
-        AFR_STACK_DESTROY (expunge_frame);
-        sh->expunge_done (frame, this, active_src, op_ret, op_errno);
-
-        return 0;
+    if (xdata)
+        dict_unref(xdata);
+    GF_FREE(linkname);
+    loc_wipe(&loc);
+    loc_wipe(&srcloc);
+    loc_wipe(&anonloc);
+    return ret;
 }
 
-
-int
-afr_sh_entry_expunge_purge (call_frame_t *expunge_frame, xlator_t *this,
-                            int active_src)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *expunge_local = NULL;
-
-        priv = this->private;
-        expunge_local = expunge_frame->local;
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "looking up %s on %s",
-                expunge_local->loc.path, priv->children[active_src]->name);
-
-        STACK_WIND_COOKIE (expunge_frame, afr_sh_entry_expunge_lookup_cbk,
-                           (void *) (long) active_src,
-                           priv->children[active_src],
-                           priv->children[active_src]->fops->lookup,
-                           &expunge_local->loc, NULL);
-
-        return 0;
-}
-
-int
-afr_sh_entry_expunge_entry_cbk (call_frame_t *expunge_frame, void *cookie,
-                                xlator_t *this,
-                                int32_t op_ret, int32_t op_errno,
-                                inode_t *inode, struct iatt *buf, dict_t *x,
-                                struct iatt *postparent)
+static int
+__afr_selfheal_heal_dirent(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                           char *name, inode_t *inode, int source,
+                           unsigned char *sources, unsigned char *healed_sinks,
+                           unsigned char *locked_on, struct afr_reply *replies)
 {
-        afr_private_t   *priv = NULL;
-        afr_local_t     *expunge_local = NULL;
-        afr_self_heal_t *expunge_sh = NULL;
-        int              source = 0;
-        call_frame_t    *frame = NULL;
-        int              active_src = 0;
-        int              need_expunge = 0;
-        afr_self_heal_t *sh            = NULL;
-        afr_local_t     *local         = NULL;
-
-        priv = this->private;
-        expunge_local = expunge_frame->local;
-        expunge_sh = &expunge_local->self_heal;
-        frame = expunge_sh->sh_frame;
-        active_src = expunge_sh->active_source;
-        source = (long) cookie;
-        local         = frame->local;
-        sh            = &local->self_heal;
-
-        if (op_ret == -1 && op_errno == ENOENT)
-                need_expunge = 1;
-        else if (op_ret == -1)
-                goto out;
-
-        if (!uuid_is_null (expunge_sh->entrybuf.ia_gfid) &&
-            !uuid_is_null (buf->ia_gfid) &&
-            (uuid_compare (expunge_sh->entrybuf.ia_gfid, buf->ia_gfid) != 0)) {
-                char uuidbuf1[64];
-                char uuidbuf2[64];
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "entry %s found on %s with mismatching gfid (%s/%s)",
-                        expunge_local->loc.path,
-                        priv->children[source]->name,
-                        uuid_utoa_r (expunge_sh->entrybuf.ia_gfid, uuidbuf1),
-                        uuid_utoa_r (buf->ia_gfid, uuidbuf2));
-                need_expunge = 1;
-        }
-
-        if (need_expunge) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "Entry %s is missing on %s and deleting from "
-                        "replica's other bricks",
-                        expunge_local->loc.path,
-                        priv->children[source]->name);
+    int ret = 0;
+    afr_private_t *priv = NULL;
+    int i = 0;
 
-                if (postparent)
-                        expunge_sh->parentbuf = *postparent;
+    priv = this->private;
 
-                afr_sh_entry_expunge_purge (expunge_frame, this, active_src);
+    if (!replies[source].valid)
+        return -EIO;
 
-                return 0;
-        }
+    /* Skip healing this entry if the last lookup on it failed for reasons
+     * other than ENOENT.
+     */
+    if ((replies[source].op_ret < 0) && (replies[source].op_errno != ENOENT))
+        return -replies[source].op_errno;
 
-out:
-        if (op_ret == 0) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "%s exists under %s",
-                        expunge_local->loc.path,
-                        priv->children[source]->name);
+    if (replies[source].op_ret == 0) {
+        ret = afr_lookup_and_heal_gfid(this, fd->inode, name, inode, replies,
+                                       source, sources,
+                                       &replies[source].poststat.ia_gfid, NULL);
+        if (ret)
+            return ret;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!healed_sinks[i])
+            continue;
+        if (replies[source].op_ret == -1 &&
+            replies[source].op_errno == ENOENT) {
+            ret = afr_selfheal_entry_delete(this, fd->inode, name, inode, i,
+                                            replies);
         } else {
-                gf_log (this->name, GF_LOG_INFO,
-                        "looking up %s under %s failed (%s)",
-                        expunge_local->loc.path,
-                        priv->children[source]->name,
-                        strerror (op_errno));
-        }
-
-        AFR_STACK_DESTROY (expunge_frame);
-        sh->expunge_done (frame, this, active_src, op_ret, op_errno);
-
-        return 0;
-}
-
-static gf_boolean_t
-can_skip_entry_self_heal (char *name, loc_t *parent_loc)
-{
-        if (strcmp (name, ".") == 0) {
-                return _gf_true;
-        } else if (strcmp (name, "..") == 0) {
-                return _gf_true;
-        } else if (loc_is_root (parent_loc) &&
-                   (strcmp (name, GF_REPLICATE_TRASH_DIR) == 0)) {
-                return _gf_true;
-        }
-        return _gf_false;
-}
-
-int
-afr_sh_entry_expunge_entry (call_frame_t *frame, xlator_t *this,
-                            gf_dirent_t *entry)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *local  = NULL;
-        afr_self_heal_t *sh  = NULL;
-        int              ret = -1;
-        call_frame_t    *expunge_frame = NULL;
-        afr_local_t     *expunge_local = NULL;
-        afr_self_heal_t *expunge_sh = NULL;
-        int              active_src = 0;
-        int              source = 0;
-        int              op_errno = 0;
-        char            *name = NULL;
-        int             op_ret = -1;
-
-        priv = this->private;
-        local = frame->local;
-        sh = &local->self_heal;
-
-        active_src = sh->active_source;
-        source = sh->source;
-        sh->expunge_done = afr_sh_entry_expunge_entry_done;
-
-        name = entry->d_name;
-        if (can_skip_entry_self_heal (name, &local->loc)) {
-                op_ret = 0;
-                goto out;
-        }
+            if (!gf_uuid_compare(replies[i].poststat.ia_gfid,
+                                 replies[source].poststat.ia_gfid))
+                continue;
 
-        gf_log (this->name, GF_LOG_TRACE,
-                "inspecting existence of %s under %s",
-                name, local->loc.path);
-
-        expunge_frame = copy_frame (frame);
-        if (!expunge_frame) {
-                op_errno = ENOMEM;
-                goto out;
-        }
-
-        AFR_LOCAL_ALLOC_OR_GOTO (expunge_local, out);
-
-        expunge_frame->local = expunge_local;
-        expunge_sh = &expunge_local->self_heal;
-        expunge_sh->sh_frame = frame;
-        expunge_sh->active_source = active_src;
-        expunge_sh->entrybuf = entry->d_stat;
-        loc_copy (&expunge_sh->parent_loc, &local->loc);
-
-        ret = afr_build_child_loc (this, &expunge_local->loc, &local->loc,
-                                   name);
-        if (ret != 0) {
-                op_errno = EINVAL;
-                goto out;
-        }
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "looking up %s on %s", expunge_local->loc.path,
-                priv->children[source]->name);
-
-        STACK_WIND_COOKIE (expunge_frame,
-                           afr_sh_entry_expunge_entry_cbk,
-                           (void *) (long) source,
-                           priv->children[source],
-                           priv->children[source]->fops->lookup,
-                           &expunge_local->loc, NULL);
-
-        ret = 0;
-out:
-        if (ret == -1)
-                sh->expunge_done (frame, this, active_src, op_ret, op_errno);
-
-        return 0;
-}
-
-
-int
-afr_sh_entry_expunge_readdir_cbk (call_frame_t *frame, void *cookie,
-                                  xlator_t *this,
-                                  int32_t op_ret, int32_t op_errno,
-                                  gf_dirent_t *entries, dict_t *xdata)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *local  = NULL;
-        afr_self_heal_t *sh  = NULL;
-        gf_dirent_t     *entry = NULL;
-        off_t            last_offset = 0;
-        int              active_src = 0;
-        int              entry_count = 0;
-
-        priv = this->private;
-        local = frame->local;
-        sh = &local->self_heal;
-
-        active_src = sh->active_source;
-
-        if (op_ret <= 0) {
-                if (op_ret < 0) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "readdir of %s on subvolume %s failed (%s)",
-                                local->loc.path,
-                                priv->children[active_src]->name,
-                                strerror (op_errno));
-                } else {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "readdir of %s on subvolume %s complete",
-                                local->loc.path,
-                                priv->children[active_src]->name);
-                }
-
-                afr_sh_entry_expunge_all (frame, this);
-                return 0;
-        }
-
-        list_for_each_entry (entry, &entries->list, list) {
-                last_offset = entry->d_off;
-                entry_count++;
-        }
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "readdir'ed %d entries from %s",
-                entry_count, priv->children[active_src]->name);
-
-        sh->offset = last_offset;
-        local->call_count = entry_count;
-
-        list_for_each_entry (entry, &entries->list, list) {
-                afr_sh_entry_expunge_entry (frame, this, entry);
+            ret = afr_selfheal_recreate_entry(frame, i, source, sources,
+                                              fd->inode, name, inode, replies);
         }
+        if (ret < 0)
+            break;
+    }
 
-        return 0;
+    return ret;
 }
 
-int
-afr_sh_entry_expunge_subvol (call_frame_t *frame, xlator_t *this,
-                             int active_src)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *local  = NULL;
-        afr_self_heal_t *sh  = NULL;
-
-        priv = this->private;
-        local = frame->local;
-        sh = &local->self_heal;
-
-        STACK_WIND (frame, afr_sh_entry_expunge_readdir_cbk,
-                    priv->children[active_src],
-                    priv->children[active_src]->fops->readdirp,
-                    sh->healing_fd, sh->block_size, sh->offset, NULL);
-
-        return 0;
-}
-
-
-int
-afr_sh_entry_expunge_all (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *local  = NULL;
-        afr_self_heal_t *sh  = NULL;
-        int              active_src = -1;
-
-        priv = this->private;
-        local = frame->local;
-        sh = &local->self_heal;
-
-        sh->offset = 0;
-
-        if (sh->source == -1) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no active sources for %s to expunge entries",
-                        local->loc.path);
-                goto out;
-        }
-
-        active_src = next_active_sink (frame, this, sh->active_source);
-        sh->active_source = active_src;
-
-        if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
-                goto out;
-        }
-
-        if (active_src == -1) {
-                /* completed creating missing files on all subvolumes */
-                goto out;
-        }
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "expunging entries of %s on %s to other sinks",
-                local->loc.path, priv->children[active_src]->name);
-
-        afr_sh_entry_expunge_subvol (frame, this, active_src);
-
-        return 0;
-out:
-        afr_sh_entry_impunge_all (frame, this);
-        return 0;
-
-}
-
-
-int
-afr_sh_entry_impunge_entry_done (call_frame_t *frame, xlator_t *this,
-                                 int32_t op_ret, int32_t op_errno)
+static int
+afr_selfheal_detect_gfid_and_type_mismatch(xlator_t *this,
+                                           struct afr_reply *replies,
+                                           inode_t *inode, uuid_t pargfid,
+                                           char *bname, int src_idx,
+                                           unsigned char *locked_on, int *src)
 {
-        int              call_count = 0;
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        if (op_ret < 0)
-                sh->entries_skipped = _gf_true;
-        call_count = afr_frame_return (frame);
-        if (call_count == 0)
-                afr_sh_entry_impunge_subvol (frame, this);
-
-        return 0;
+    int i = 0;
+    int ret = -1;
+    afr_private_t *priv = NULL;
+    void *gfid = NULL;
+    ia_type_t ia_type = IA_INVAL;
+
+    priv = this->private;
+    gfid = &replies[src_idx].poststat.ia_gfid;
+    ia_type = replies[src_idx].poststat.ia_type;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (i == src_idx)
+            continue;
+
+        if (!replies[i].valid)
+            continue;
+
+        if (replies[i].op_ret != 0)
+            continue;
+
+        if (gf_uuid_is_null(replies[i].poststat.ia_gfid))
+            continue;
+
+        if (replies[i].poststat.ia_type == IA_INVAL)
+            continue;
+
+        if (ia_type == IA_INVAL || gf_uuid_is_null(gfid)) {
+            src_idx = i;
+            ia_type = replies[src_idx].poststat.ia_type;
+            gfid = &replies[src_idx].poststat.ia_gfid;
+            continue;
+        }
+
+        if (gf_uuid_compare(gfid, replies[i].poststat.ia_gfid) &&
+            (ia_type == replies[i].poststat.ia_type)) {
+            ret = afr_gfid_split_brain_source(this, replies, inode, pargfid,
+                                              bname, src_idx, i, locked_on, src,
+                                              NULL);
+            if (ret)
+                gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN,
+                       "Skipping conservative merge on the "
+                       "file.");
+            return ret;
+        }
+
+        if (ia_type != replies[i].poststat.ia_type) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN,
+                   "Type mismatch detected "
+                   "for <gfid:%s>/%s>, %s on %s and %s on %s. "
+                   "Skipping conservative merge on the file.",
+                   uuid_utoa(pargfid), bname,
+                   gf_inode_type_to_str(replies[i].poststat.ia_type),
+                   priv->children[i]->name,
+                   gf_inode_type_to_str(replies[src_idx].poststat.ia_type),
+                   priv->children[src_idx]->name);
+            gf_event(EVENT_AFR_SPLIT_BRAIN,
+                     "client-pid=%d;"
+                     "subvol=%s;type=file;"
+                     "file=<gfid:%s>/%s>;count=2;child-%d=%s;type-"
+                     "%d=%s;child-%d=%s;type-%d=%s",
+                     this->ctx->cmd_args.client_pid, this->name,
+                     uuid_utoa(pargfid), bname, i, priv->children[i]->name, i,
+                     gf_inode_type_to_str(replies[i].poststat.ia_type), src_idx,
+                     priv->children[src_idx]->name, src_idx,
+                     gf_inode_type_to_str(replies[src_idx].poststat.ia_type));
+            return -1;
+        }
+    }
+
+    return 0;
 }
 
-void
-afr_sh_entry_call_impunge_done (call_frame_t *impunge_frame, xlator_t *this,
-                                int32_t op_ret, int32_t op_errno)
+static int
+__afr_selfheal_merge_dirent(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                            char *name, inode_t *inode, unsigned char *sources,
+                            unsigned char *healed_sinks,
+                            unsigned char *locked_on, struct afr_reply *replies)
 {
-        afr_local_t     *impunge_local = NULL;
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_self_heal_t *impunge_sh = NULL;
-        call_frame_t    *frame = NULL;
+    int ret = 0;
+    int i = 0;
+    int source = -1;
+    int src = -1;
+    afr_private_t *priv = NULL;
 
-        AFR_INIT_SH_FRAME_VALS (impunge_frame, impunge_local, impunge_sh,
-                                frame, local, sh);
-
-        AFR_STACK_DESTROY (impunge_frame);
-        sh->impunge_done (frame, this, op_ret, op_errno);
-}
+    priv = this->private;
 
-int
-afr_sh_entry_impunge_setattr_cbk (call_frame_t *impunge_frame, void *cookie,
-                                  xlator_t *this,
-                                  int32_t op_ret, int32_t op_errno,
-                                  struct iatt *preop, struct iatt *postop,
-                                  dict_t *xdata)
-{
-        int              call_count = 0;
-        afr_private_t   *priv = NULL;
-        afr_local_t     *impunge_local = NULL;
-        int              child_index = 0;
-
-        priv = this->private;
-        impunge_local = impunge_frame->local;
-        child_index = (long) cookie;
-
-        if (op_ret == 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "setattr done for %s on %s",
-                        impunge_local->loc.path,
-                        priv->children[child_index]->name);
-        } else {
-                gf_log (this->name, GF_LOG_INFO,
-                        "setattr (%s) on %s failed (%s)",
-                        impunge_local->loc.path,
-                        priv->children[child_index]->name,
-                        strerror (op_errno));
-        }
-
-        call_count = afr_frame_return (impunge_frame);
-        if (call_count == 0) {
-                afr_sh_entry_call_impunge_done (impunge_frame, this,
-                                                0, op_errno);
+    for (i = 0; i < priv->child_count; i++) {
+        if (replies[i].valid && replies[i].op_ret == 0) {
+            source = i;
+            break;
         }
+    }
 
+    if (source == -1) {
+        /* entry got deleted in the mean time? */
         return 0;
-}
-
-int
-afr_sh_entry_impunge_parent_setattr_cbk (call_frame_t *setattr_frame,
-                                         void *cookie, xlator_t *this,
-                                         int32_t op_ret, int32_t op_errno,
-                                         struct iatt *preop, struct iatt *postop,
-                                         dict_t *xdata)
-{
-        int             call_count = 0;
-        afr_local_t     *setattr_local = NULL;
-
-        setattr_local = setattr_frame->local;
-        if (op_ret != 0) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "setattr on parent directory (%s) failed: %s",
-                        setattr_local->loc.path, strerror (op_errno));
-        }
-
-        call_count = afr_frame_return (setattr_frame);
-        if (call_count == 0)
-                AFR_STACK_DESTROY (setattr_frame);
-        return 0;
-}
-
-int
-afr_sh_entry_impunge_setattr (call_frame_t *impunge_frame, xlator_t *this)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *impunge_local = NULL;
-        afr_local_t     *setattr_local = NULL;
-        afr_self_heal_t *impunge_sh = NULL;
-        call_frame_t    *setattr_frame = NULL;
-        int32_t          valid = 0;
-        int32_t          op_errno = 0;
-        int              child_index = 0;
-        int              call_count = 0;
-        int              i = 0;
-
-        priv          = this->private;
-        impunge_local = impunge_frame->local;
-        impunge_sh    = &impunge_local->self_heal;
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "setting ownership of %s on %s to %d/%d",
-                impunge_local->loc.path,
-                priv->children[child_index]->name,
-                impunge_sh->entrybuf.ia_uid,
-                impunge_sh->entrybuf.ia_gid);
-
-        setattr_frame = copy_frame (impunge_frame);
-        if (!setattr_frame) {
-                op_errno = ENOMEM;
-                goto out;
-        }
-        AFR_LOCAL_ALLOC_OR_GOTO (setattr_frame->local, out);
-        setattr_local = setattr_frame->local;
-        call_count = afr_errno_count (NULL, impunge_sh->child_errno,
-                                      priv->child_count, 0);
-        loc_copy (&setattr_local->loc, &impunge_sh->parent_loc);
-        impunge_local->call_count = call_count;
-        setattr_local->call_count = call_count;
+    }
+
+    /* Set all the sources as 1, otheriwse newentry_mark won't be set */
+    for (i = 0; i < priv->child_count; i++) {
+        if (replies[i].valid && replies[i].op_ret == 0) {
+            sources[i] = 1;
+        }
+    }
+
+    ret = afr_lookup_and_heal_gfid(this, fd->inode, name, inode, replies,
+                                   source, sources,
+                                   &replies[source].poststat.ia_gfid, NULL);
+    if (ret)
+        return ret;
+
+    /* In case of type mismatch / unable to resolve gfid mismatch on the
+     * entry, return -1.*/
+    ret = afr_selfheal_detect_gfid_and_type_mismatch(
+        this, replies, inode, fd->inode->gfid, name, source, locked_on, &src);
+
+    if (ret < 0)
+        return ret;
+    if (src != -1) {
+        source = src;
         for (i = 0; i < priv->child_count; i++) {
-                if (impunge_sh->child_errno[i])
-                        continue;
-                valid         = GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME;
-                STACK_WIND_COOKIE (setattr_frame,
-                                   afr_sh_entry_impunge_parent_setattr_cbk,
-                                   (void *) (long) i, priv->children[i],
-                                   priv->children[i]->fops->setattr,
-                                   &setattr_local->loc,
-                                   &impunge_sh->parentbuf, valid, NULL);
-
-                valid = GF_SET_ATTR_UID   | GF_SET_ATTR_GID |
-                        GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME;
-                STACK_WIND_COOKIE (impunge_frame,
-                                   afr_sh_entry_impunge_setattr_cbk,
-                                   (void *) (long) i, priv->children[i],
-                                   priv->children[i]->fops->setattr,
-                                   &impunge_local->loc,
-                                   &impunge_sh->entrybuf, valid, NULL);
-                call_count--;
+            if (i != src && replies[i].valid &&
+                gf_uuid_compare(replies[src].poststat.ia_gfid,
+                                replies[i].poststat.ia_gfid)) {
+                sources[i] = 0;
+            }
         }
-        GF_ASSERT (!call_count);
-        return 0;
-out:
-        if (setattr_frame)
-                AFR_STACK_DESTROY (setattr_frame);
-        afr_sh_entry_call_impunge_done (impunge_frame, this, 0, op_errno);
-        return 0;
-}
+    }
 
-int
-afr_sh_entry_impunge_xattrop_cbk (call_frame_t *impunge_frame, void *cookie,
-                                  xlator_t *this,
-                                  int32_t op_ret, int32_t op_errno,
-                                  dict_t *xattr, dict_t *xdata)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *impunge_local = NULL;
-        int              child_index = 0;
+    for (i = 0; i < priv->child_count; i++) {
+        if (i == source || !healed_sinks[i])
+            continue;
 
-        priv          = this->private;
-        impunge_local = impunge_frame->local;
-
-        child_index = (long) cookie;
-
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "%s: failed to perform xattrop on %s (%s)",
-                        impunge_local->loc.path,
-                        priv->children[child_index]->name,
-                        strerror (op_errno));
-                goto out;
+        if (src != -1) {
+            if (!gf_uuid_compare(replies[src].poststat.ia_gfid,
+                                 replies[i].poststat.ia_gfid))
+                continue;
+        } else if (replies[i].op_errno != ENOENT) {
+            continue;
         }
 
-        afr_sh_entry_impunge_setattr (impunge_frame, this);
-        return 0;
-out:
-        afr_sh_entry_call_impunge_done (impunge_frame, this,
-                                        -1, op_errno);
-        return 0;
-}
-
-int
-afr_sh_entry_impunge_perform_xattrop (call_frame_t *impunge_frame,
-                                      xlator_t *this)
-{
-        int              active_src       = 0;
-        dict_t          *xattr            = NULL;
-        afr_private_t   *priv             = NULL;
-        afr_local_t     *impunge_local    = NULL;
-        afr_self_heal_t *impunge_sh       = NULL;
-        int32_t         op_errno          = 0;
-
-        priv = this->private;
-        impunge_local = impunge_frame->local;
-        impunge_sh = &impunge_local->self_heal;
-        active_src = impunge_sh->active_source;
-
-        afr_prepare_new_entry_pending_matrix (impunge_local->pending,
-                                              afr_is_errno_unset,
-                                              impunge_sh->child_errno,
-                                              &impunge_sh->entrybuf,
-                                              priv->child_count);
-        xattr = dict_new ();
-        if (!xattr) {
-                op_errno = ENOMEM;
-                goto out;
-        }
-
-        afr_set_pending_dict (priv, xattr, impunge_local->pending, active_src,
-                              LOCAL_LAST);
+        ret |= afr_selfheal_recreate_entry(frame, i, source, sources, fd->inode,
+                                           name, inode, replies);
+    }
 
-        STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_xattrop_cbk,
-                           (void *) (long) active_src,
-                           priv->children[active_src],
-                           priv->children[active_src]->fops->xattrop,
-                           &impunge_local->loc, GF_XATTROP_ADD_ARRAY, xattr, NULL);
-
-        if (xattr)
-                dict_unref (xattr);
-        return 0;
-out:
-        afr_sh_entry_call_impunge_done (impunge_frame, this,
-                                        -1, op_errno);
-        return 0;
+    return ret;
 }
 
-int
-afr_sh_entry_impunge_newfile_cbk (call_frame_t *impunge_frame, void *cookie,
-                                  xlator_t *this,
-                                  int32_t op_ret, int32_t op_errno,
-                                  inode_t *inode, struct iatt *stbuf,
-                                  struct iatt *preparent,
-                                  struct iatt *postparent, dict_t *xdata)
+static int
+__afr_selfheal_entry_dirent(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                            char *name, inode_t *inode, int source,
+                            unsigned char *sources, unsigned char *healed_sinks,
+                            unsigned char *locked_on, struct afr_reply *replies)
 {
-        int              call_count       = 0;
-        afr_private_t   *priv             = NULL;
-        afr_local_t     *impunge_local    = NULL;
-        afr_self_heal_t *impunge_sh       = NULL;
-        int              child_index      = 0;
-
-        priv = this->private;
-        impunge_local = impunge_frame->local;
-        impunge_sh = &impunge_local->self_heal;
-
-        child_index = (long) cookie;
-
-        if (op_ret == -1) {
-                impunge_sh->child_errno[child_index] = op_errno;
-                gf_log (this->name, GF_LOG_ERROR,
-                        "creation of %s on %s failed (%s)",
-                        impunge_local->loc.path,
-                        priv->children[child_index]->name,
-                        strerror (op_errno));
-        } else {
-                impunge_sh->child_errno[child_index] = 0;
-        }
-
-        call_count = afr_frame_return (impunge_frame);
-        if (call_count == 0) {
-                if (!afr_errno_count (NULL, impunge_sh->child_errno,
-                                      priv->child_count, 0)) {
-                        // new_file creation failed every where
-                        afr_sh_entry_call_impunge_done (impunge_frame, this,
-                                                        -1, op_errno);
-                        goto out;
-                }
-                afr_sh_entry_impunge_perform_xattrop (impunge_frame, this);
-        }
-out:
-        return 0;
+    int ret = -1;
+
+    if (source < 0)
+        ret = __afr_selfheal_merge_dirent(frame, this, fd, name, inode, sources,
+                                          healed_sinks, locked_on, replies);
+    else
+        ret = __afr_selfheal_heal_dirent(frame, this, fd, name, inode, source,
+                                         sources, healed_sinks, locked_on,
+                                         replies);
+    return ret;
 }
 
-int
-afr_sh_entry_impunge_hardlink_cbk (call_frame_t *impunge_frame, void *cookie,
-                                   xlator_t *this, int32_t op_ret,
-                                   int32_t op_errno, inode_t *inode,
-                                   struct iatt *buf, struct iatt *preparent,
-                                   struct iatt *postparent, dict_t *xdata)
+static gf_boolean_t
+is_full_heal_marker_present(xlator_t *this, dict_t *xdata, int idx)
 {
-        int              call_count        = 0;
-        afr_local_t     *impunge_local = NULL;
-        afr_self_heal_t *impunge_sh  = NULL;
-
-        impunge_local = impunge_frame->local;
-        impunge_sh = &impunge_local->self_heal;
-
-        if (IA_IFLNK == impunge_sh->entrybuf.ia_type) {
-                //For symlinks impunge is attempted un-conditionally
-                //So the file can already exist.
-                if ((op_ret < 0) && (op_errno == EEXIST))
-                        op_ret = 0;
-        }
+    int i = 0;
+    int pending[3] = {
+        0,
+    };
+    void *pending_raw = NULL;
+    afr_private_t *priv = NULL;
 
-        call_count = afr_frame_return (impunge_frame);
-        if (call_count == 0)
-                afr_sh_entry_call_impunge_done (impunge_frame, this,
-                                                op_ret, op_errno);
+    priv = this->private;
 
-        return 0;
-}
+    if (!xdata)
+        return _gf_false;
 
-int
-afr_sh_entry_impunge_hardlink (call_frame_t *impunge_frame, xlator_t *this,
-                               int child_index)
-{
-        afr_private_t   *priv          = NULL;
-        afr_local_t     *impunge_local = NULL;
-        afr_self_heal_t *impunge_sh  = NULL;
-        loc_t           *loc           = NULL;
-        struct iatt     *buf           = NULL;
-        loc_t            oldloc        = {0};
-
-        priv = this->private;
-        impunge_local = impunge_frame->local;
-        impunge_sh = &impunge_local->self_heal;
-        loc = &impunge_local->loc;
-        buf = &impunge_sh->entrybuf;
-
-        oldloc.inode = inode_ref (loc->inode);
-        uuid_copy (oldloc.gfid, buf->ia_gfid);
-        gf_log (this->name, GF_LOG_DEBUG, "linking missing file %s on %s",
-                loc->path, priv->children[child_index]->name);
-
-        STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_hardlink_cbk,
-                           (void *) (long) child_index,
-                           priv->children[child_index],
-                           priv->children[child_index]->fops->link,
-                           &oldloc, loc, NULL);
-        loc_wipe (&oldloc);
+    /* Iterate over each of the priv->pending_keys[] elements and then
+     * see if any of them have data segment non-zero. If they do, return
+     * true. Else return false.
+     */
+    for (i = 0; i < priv->child_count; i++) {
+        if (dict_get_ptr(xdata, priv->pending_key[i], &pending_raw))
+            continue;
 
-        return 0;
-}
+        if (!pending_raw)
+            continue;
 
-int
-afr_sh_nameless_lookup_cbk (call_frame_t *impunge_frame, void *cookie,
-                            xlator_t *this,
-                            int32_t op_ret, int32_t op_errno, inode_t *inode,
-                            struct iatt *buf, dict_t *xattr,
-                            struct iatt *postparent)
-{
-        if (op_ret < 0) {
-                 afr_sh_entry_impunge_create_file (impunge_frame, this,
-                                                   (long)cookie);
-        } else {
-                afr_sh_entry_impunge_hardlink (impunge_frame, this,
-                                               (long)cookie);
-        }
-        return 0;
-}
+        memcpy(pending, pending_raw, sizeof(pending));
+        if (ntoh32(pending[idx]))
+            return _gf_true;
+    }
 
-int
-afr_sh_entry_impunge_check_hardlink (call_frame_t *impunge_frame,
-                                     xlator_t *this,
-                                     int child_index, struct iatt *stbuf)
-{
-        afr_private_t   *priv          = NULL;
-        call_frame_t    *frame             = NULL;
-        afr_local_t     *impunge_local     = NULL;
-        afr_local_t     *local             = NULL;
-        afr_self_heal_t *impunge_sh        = NULL;
-        afr_self_heal_t *sh                = NULL;
-        loc_t           *loc           = NULL;
-        dict_t          *xattr_req     = NULL;
-        loc_t            oldloc        = {0};
-        int              ret           = -1;
-
-        priv = this->private;
-        AFR_INIT_SH_FRAME_VALS (impunge_frame, impunge_local, impunge_sh,
-                                frame, local, sh);
-        loc = &impunge_local->loc;
-
-        xattr_req = dict_new ();
-        if (!xattr_req)
-                goto out;
-        oldloc.inode = inode_ref (loc->inode);
-        uuid_copy (oldloc.gfid, stbuf->ia_gfid);
-
-        STACK_WIND_COOKIE (impunge_frame, afr_sh_nameless_lookup_cbk,
-                           (void *) (long) child_index,
-                           priv->children[child_index],
-                           priv->children[child_index]->fops->lookup,
-                           &oldloc, xattr_req);
-        ret = 0;
-out:
-        if (xattr_req)
-                dict_unref (xattr_req);
-        loc_wipe (&oldloc);
-        if (ret)
-                sh->impunge_done (frame, this, -1, ENOMEM);
-        return 0;
+    return _gf_false;
 }
 
-int
-afr_sh_entry_impunge_mknod (call_frame_t *impunge_frame, xlator_t *this,
-                            int child_index, struct iatt *stbuf)
+static gf_boolean_t
+afr_need_full_heal(xlator_t *this, struct afr_reply *replies, int source,
+                   unsigned char *healed_sinks, afr_transaction_type type)
 {
-        afr_private_t *priv          = NULL;
-        afr_local_t   *impunge_local = NULL;
-        dict_t        *dict          = NULL;
-        int            ret           = 0;
-
-        priv = this->private;
-        impunge_local = impunge_frame->local;
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "creating missing file %s on %s",
-                impunge_local->loc.path,
-                priv->children[child_index]->name);
-
-        dict = dict_new ();
-        if (!dict)
-                gf_log (this->name, GF_LOG_ERROR, "Out of memory");
-
-        GF_ASSERT (!uuid_is_null (stbuf->ia_gfid));
-        ret = afr_set_dict_gfid (dict, stbuf->ia_gfid);
-        if (ret)
-                gf_log (this->name, GF_LOG_INFO, "%s: gfid set failed",
-                        impunge_local->loc.path);
-
-        /*
-         * Reason for adding GLUSTERFS_INTERNAL_FOP_KEY :
-         *
-         * Problem:
-         * While a brick is down in a replica pair, lets say the user creates
-         * one file(file-A) and a hard link to that file(h-file-A). After the
-         * brick comes back up, entry self-heal is attempted on parent dir of
-         * these two files. As part of readdir in self-heal it reads both the
-         * entries file-A and h-file-A for both of them it does name less lookup
-         * to check if there are any hardlinks already present in the
-         * destination brick. It finds that there are no hard links already
-         * present for files file-A, h-file-A. Self-heal does mknods for both
-         * file-A and h-file-A. This leads to file-A and h-file-A not being
-         * hardlinks anymore.
-         *
-         * Fix: (More like shrinking of race-window, the race itself is still
-         * present in posix-mknod).
-         * If mknod comes with the presence of GLUSTERFS_INTERNAL_FOP_KEY then
-         * posix_mknod checks if there are already any gfid-links and does
-         * link() instead of mknod. There still can be a race where two
-         * posix_mknods same gfid see that
-         * gfid-link file is not present and proceeds with mknods and result in
-         * two different files with same gfid.
-         */
-        ret = dict_set_str (dict, GLUSTERFS_INTERNAL_FOP_KEY, "yes");
-        if (ret)
-                gf_log (this->name, GF_LOG_INFO, "%s: %s set failed",
-                        impunge_local->loc.path, GLUSTERFS_INTERNAL_FOP_KEY);
-
-        STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_newfile_cbk,
-                           (void *) (long) child_index,
-                           priv->children[child_index],
-                           priv->children[child_index]->fops->mknod,
-                           &impunge_local->loc,
-                           st_mode_from_ia (stbuf->ia_prot, stbuf->ia_type),
-                           makedev (ia_major (stbuf->ia_rdev),
-                                    ia_minor (stbuf->ia_rdev)), 0, dict);
-
-        if (dict)
-                dict_unref (dict);
-
-        return 0;
-}
-
+    int i = 0;
+    int idx = 0;
+    afr_private_t *priv = NULL;
 
+    priv = this->private;
 
-int
-afr_sh_entry_impunge_mkdir (call_frame_t *impunge_frame, xlator_t *this,
-                            int child_index, struct iatt *stbuf)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *impunge_local = NULL;
-        dict_t          *dict = NULL;
+    if (!priv->esh_granular)
+        return _gf_true;
 
-        int ret = 0;
+    if (type != AFR_ENTRY_TRANSACTION)
+        return _gf_true;
 
-        priv = this->private;
-        impunge_local = impunge_frame->local;
+    priv = this->private;
+    idx = afr_index_for_transaction_type(AFR_DATA_TRANSACTION);
 
-        dict = dict_new ();
-        if (!dict) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                return 0;
-        }
+    /* If there is a clear source, check whether the full-heal-indicator
+     * is present in its xdata. Otherwise, we need to examine all the
+     * participating bricks and then figure if *even* one of them has a
+     * full-heal-indicator.
+     */
 
-        GF_ASSERT (!uuid_is_null (stbuf->ia_gfid));
-        ret = afr_set_dict_gfid (dict, stbuf->ia_gfid);
-        if (ret)
-                gf_log (this->name, GF_LOG_INFO, "%s: gfid set failed",
-                        impunge_local->loc.path);
+    if (source != -1) {
+        if (is_full_heal_marker_present(this, replies[source].xdata, idx))
+            return _gf_true;
+    }
 
-        gf_log (this->name, GF_LOG_DEBUG,
-                "creating missing directory %s on %s",
-                impunge_local->loc.path,
-                priv->children[child_index]->name);
+    /* else ..*/
 
-        STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_newfile_cbk,
-                           (void *) (long) child_index,
-                           priv->children[child_index],
-                           priv->children[child_index]->fops->mkdir,
-                           &impunge_local->loc,
-                           st_mode_from_ia (stbuf->ia_prot, stbuf->ia_type),
-                           0, dict);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!healed_sinks[i])
+            continue;
 
-        if (dict)
-                dict_unref (dict);
+        if (is_full_heal_marker_present(this, replies[i].xdata, idx))
+            return _gf_true;
+    }
 
-        return 0;
+    return _gf_false;
 }
 
-
-int
-afr_sh_entry_impunge_symlink (call_frame_t *impunge_frame, xlator_t *this,
-                              int child_index, const char *linkname)
+static int
+__afr_selfheal_entry_finalize_source(xlator_t *this, unsigned char *sources,
+                                     unsigned char *healed_sinks,
+                                     unsigned char *locked_on,
+                                     struct afr_reply *replies,
+                                     uint64_t *witness)
 {
-        afr_private_t   *priv          = NULL;
-        afr_local_t     *impunge_local = NULL;
-        dict_t          *dict          = NULL;
-        struct iatt     *buf           = NULL;
-        int              ret           = 0;
-
-        priv = this->private;
-        impunge_local = impunge_frame->local;
-
-        buf = &impunge_local->cont.dir_fop.buf;
+    afr_private_t *priv = NULL;
+    int source = -1;
+    int sources_count = 0;
+    int i = 0;
 
-        dict = dict_new ();
-        if (!dict) {
-                afr_sh_entry_call_impunge_done (impunge_frame, this,
-                                                -1, ENOMEM);
-                goto out;
-        }
-
-        GF_ASSERT (!uuid_is_null (buf->ia_gfid));
-        ret = afr_set_dict_gfid (dict, buf->ia_gfid);
-        if (ret)
-                gf_log (this->name, GF_LOG_INFO,
-                        "%s: dict set gfid failed",
-                        impunge_local->loc.path);
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "creating missing symlink %s -> %s on %s",
-                impunge_local->loc.path, linkname,
-                priv->children[child_index]->name);
-
-        STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_newfile_cbk,
-                           (void *) (long) child_index,
-                           priv->children[child_index],
-                           priv->children[child_index]->fops->symlink,
-                           linkname, &impunge_local->loc, 0, dict);
-
-        if (dict)
-                dict_unref (dict);
-out:
-        return 0;
-}
+    priv = this->private;
 
+    sources_count = AFR_COUNT(sources, priv->child_count);
 
-int
-afr_sh_entry_impunge_symlink_unlink_cbk (call_frame_t *impunge_frame,
-                                         void *cookie, xlator_t *this,
-                                         int32_t op_ret, int32_t op_errno,
-                                         struct iatt *preparent,
-                                         struct iatt *postparent, dict_t *xdata)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *impunge_local = NULL;
-        afr_self_heal_t *impunge_sh = NULL;
-        int              child_index = -1;
-        int              call_count = -1;
-
-        priv          = this->private;
-        impunge_local = impunge_frame->local;
-        impunge_sh    = &impunge_local->self_heal;
-
-        child_index = (long) cookie;
-
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "unlink of %s on %s failed (%s)",
-                        impunge_local->loc.path,
-                        priv->children[child_index]->name,
-                        strerror (op_errno));
-                goto out;
-        }
+    if ((AFR_CMP(locked_on, healed_sinks, priv->child_count) == 0) ||
+        !sources_count || afr_does_witness_exist(this, witness)) {
+        memset(sources, 0, sizeof(*sources) * priv->child_count);
+        afr_mark_active_sinks(this, sources, locked_on, healed_sinks);
+        return -1;
+    }
 
-        afr_sh_entry_impunge_symlink (impunge_frame, this, child_index,
-                                      impunge_sh->linkname);
+    source = afr_choose_source_by_policy(priv, sources, AFR_ENTRY_TRANSACTION);
 
-        return 0;
-out:
-        LOCK (&impunge_frame->lock);
-        {
-                call_count = --impunge_local->call_count;
+    /*If the selected source does not blame any other brick, then mark
+     * everything as sink to trigger conservative merge.
+     */
+    if (source != -1 && !AFR_COUNT(healed_sinks, priv->child_count)) {
+        for (i = 0; i < priv->child_count; i++) {
+            if (locked_on[i]) {
+                sources[i] = 0;
+                healed_sinks[i] = 1;
+            }
         }
-        UNLOCK (&impunge_frame->lock);
-
-        if (call_count == 0)
-                afr_sh_entry_call_impunge_done (impunge_frame, this,
-                                                op_ret, op_errno);
+        return -1;
+    }
 
-        return 0;
+    return source;
 }
 
-
 int
-afr_sh_entry_impunge_symlink_unlink (call_frame_t *impunge_frame, xlator_t *this,
-                                     int child_index)
+__afr_selfheal_entry_prepare(call_frame_t *frame, xlator_t *this,
+                             inode_t *inode, unsigned char *locked_on,
+                             unsigned char *sources, unsigned char *sinks,
+                             unsigned char *healed_sinks,
+                             struct afr_reply *replies, int *source_p,
+                             unsigned char *pflag)
 {
-        afr_private_t   *priv          = NULL;
-        afr_local_t     *impunge_local = NULL;
-
-        priv          = this->private;
-        impunge_local = impunge_frame->local;
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "unlinking symlink %s with wrong target on %s",
-                impunge_local->loc.path,
-                priv->children[child_index]->name);
-
-        STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_symlink_unlink_cbk,
-                           (void *) (long) child_index,
-                           priv->children[child_index],
-                           priv->children[child_index]->fops->unlink,
-                           &impunge_local->loc, 0, NULL);
-
-        return 0;
+    int ret = -1;
+    int source = -1;
+    afr_private_t *priv = NULL;
+    uint64_t *witness = NULL;
+
+    priv = this->private;
+
+    ret = afr_selfheal_unlocked_discover(frame, inode, inode->gfid, replies);
+    if (ret)
+        return ret;
+
+    witness = alloca0(sizeof(*witness) * priv->child_count);
+    ret = afr_selfheal_find_direction(frame, this, replies,
+                                      AFR_ENTRY_TRANSACTION, locked_on, sources,
+                                      sinks, witness, pflag);
+    if (ret)
+        return ret;
+
+    /* Initialize the healed_sinks[] array optimistically to
+       the intersection of to-be-healed (i.e sinks[]) and
+       the list of servers which are up (i.e locked_on[]).
+
+       As we encounter failures in the healing process, we
+       will unmark the respective servers in the healed_sinks[]
+       array.
+    */
+    AFR_INTERSECT(healed_sinks, sinks, locked_on, priv->child_count);
+
+    source = __afr_selfheal_entry_finalize_source(this, sources, healed_sinks,
+                                                  locked_on, replies, witness);
+
+    if (source < 0) {
+        /* If source is < 0 (typically split-brain), we perform a
+           conservative merge of entries rather than erroring out */
+    }
+    *source_p = source;
+
+    return ret;
 }
 
-
-int
-afr_sh_entry_impunge_readlink_sink_cbk (call_frame_t *impunge_frame, void *cookie,
-                                        xlator_t *this,
-                                        int32_t op_ret, int32_t op_errno,
-                                        const char *linkname, struct iatt *sbuf, dict_t *xdata)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *impunge_local = NULL;
-        afr_self_heal_t *impunge_sh = NULL;
-        int              child_index = -1;
-        int              call_count = -1;
-        int              active_src = -1;
-
-        priv          = this->private;
-        impunge_local = impunge_frame->local;
-        impunge_sh    = &impunge_local->self_heal;
-        active_src    = impunge_sh->active_source;
-
-        child_index = (long) cookie;
-
-        if ((op_ret == -1) && (op_errno != ENOENT)) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "readlink of %s on %s failed (%s)",
-                        impunge_local->loc.path,
-                        priv->children[active_src]->name,
-                        strerror (op_errno));
-                goto out;
-        }
-
-        /* symlink doesn't exist on the sink */
-
-        if ((op_ret == -1) && (op_errno == ENOENT)) {
-                afr_sh_entry_impunge_symlink (impunge_frame, this,
-                                              child_index, impunge_sh->linkname);
-                return 0;
-        }
-
-
-        /* symlink exists on the sink, so check if targets match */
-
-        if (strcmp (linkname, impunge_sh->linkname) == 0) {
-                /* targets match, nothing to do */
-
-                goto out;
-        } else {
-                /*
-                 * Hah! Sneaky wolf in sheep's clothing!
-                 */
-                afr_sh_entry_impunge_symlink_unlink (impunge_frame, this,
-                                                     child_index);
-                return 0;
-        }
-
-out:
-        LOCK (&impunge_frame->lock);
-        {
-                call_count = --impunge_local->call_count;
-        }
-        UNLOCK (&impunge_frame->lock);
-
-        if (call_count == 0)
-                afr_sh_entry_call_impunge_done (impunge_frame, this,
-                                                op_ret, op_errno);
-
-        return 0;
-}
-
-
-int
-afr_sh_entry_impunge_readlink_sink (call_frame_t *impunge_frame, xlator_t *this,
-                                    int child_index)
+static int
+afr_selfheal_entry_dirent(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                          char *name, inode_t *parent_idx_inode,
+                          xlator_t *subvol, gf_boolean_t full_crawl)
 {
-        afr_private_t   *priv = NULL;
-        afr_local_t     *impunge_local = NULL;
-
-        priv = this->private;
-        impunge_local = impunge_frame->local;
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "checking symlink target of %s on %s",
-                impunge_local->loc.path, priv->children[child_index]->name);
-
-        STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_readlink_sink_cbk,
-                           (void *) (long) child_index,
-                           priv->children[child_index],
-                           priv->children[child_index]->fops->readlink,
-                           &impunge_local->loc, 4096, NULL);
-
+    int ret = 0;
+    int source = -1;
+    unsigned char *locked_on = NULL;
+    unsigned char *sources = NULL;
+    unsigned char *sinks = NULL;
+    unsigned char *healed_sinks = NULL;
+    inode_t *inode = NULL;
+    struct afr_reply *replies = NULL;
+    struct afr_reply *par_replies = NULL;
+    afr_private_t *priv = NULL;
+    dict_t *xattr = NULL;
+
+    priv = this->private;
+
+    if (afr_is_private_directory(priv, fd->inode->gfid, name,
+                                 GF_CLIENT_PID_SELF_HEALD)) {
         return 0;
+    }
+
+    xattr = dict_new();
+    if (!xattr)
+        return -ENOMEM;
+    ret = dict_set_int32_sizen(xattr, GF_GFIDLESS_LOOKUP, 1);
+    if (ret) {
+        dict_unref(xattr);
+        return -1;
+    }
+
+    sources = alloca0(priv->child_count);
+    sinks = alloca0(priv->child_count);
+    healed_sinks = alloca0(priv->child_count);
+    locked_on = alloca0(priv->child_count);
+
+    replies = alloca0(priv->child_count * sizeof(*replies));
+    par_replies = alloca0(priv->child_count * sizeof(*par_replies));
+
+    ret = afr_selfheal_entrylk(frame, this, fd->inode, this->name, NULL,
+                               locked_on);
+    {
+        if (ret < priv->child_count) {
+            gf_msg_debug(this->name, 0,
+                         "%s: Skipping "
+                         "entry self-heal as only %d sub-volumes "
+                         " could be locked in %s domain",
+                         uuid_utoa(fd->inode->gfid), ret, this->name);
+            ret = -ENOTCONN;
+            goto unlock;
+        }
+
+        ret = __afr_selfheal_entry_prepare(frame, this, fd->inode, locked_on,
+                                           sources, sinks, healed_sinks,
+                                           par_replies, &source, NULL);
+        if (ret < 0)
+            goto unlock;
+
+        inode = afr_selfheal_unlocked_lookup_on(frame, fd->inode, name, replies,
+                                                locked_on, xattr);
+        if (!inode) {
+            ret = -ENOMEM;
+            goto unlock;
+        }
+
+        ret = __afr_selfheal_entry_dirent(frame, this, fd, name, inode, source,
+                                          sources, healed_sinks, locked_on,
+                                          replies);
+
+        if ((ret == 0) && (priv->esh_granular) && parent_idx_inode) {
+            ret = afr_shd_entry_purge(subvol, parent_idx_inode, name,
+                                      inode->ia_type);
+            /* Why is ret force-set to 0? We do not care about
+             * index purge failing for full heal as it is quite
+             * possible during replace-brick that not all files
+             * and directories have their name indices present in
+             * entry-changes/.
+             */
+            ret = 0;
+        }
+    }
+
+unlock:
+    afr_selfheal_unentrylk(frame, this, fd->inode, this->name, NULL, locked_on,
+                           NULL);
+    if (inode)
+        inode_unref(inode);
+    if (replies)
+        afr_replies_wipe(replies, priv->child_count);
+    if (par_replies)
+        afr_replies_wipe(par_replies, priv->child_count);
+    if (xattr)
+        dict_unref(xattr);
+
+    return ret;
 }
 
-
-int
-afr_sh_entry_impunge_readlink_cbk (call_frame_t *impunge_frame, void *cookie,
-                                   xlator_t *this,
-                                   int32_t op_ret, int32_t op_errno,
-                                   const char *linkname, struct iatt *sbuf, dict_t *xdata)
+static inode_t *
+afr_shd_entry_changes_index_inode(xlator_t *this, xlator_t *subvol,
+                                  uuid_t pargfid)
 {
-        afr_private_t   *priv = NULL;
-        afr_local_t     *impunge_local = NULL;
-        afr_self_heal_t *impunge_sh = NULL;
-        int              child_index = -1;
-        int              call_count = -1;
-        int              active_src = -1;
-
-        priv = this->private;
-        impunge_local = impunge_frame->local;
-        impunge_sh = &impunge_local->self_heal;
-        active_src = impunge_sh->active_source;
-
-        child_index = (long) cookie;
-
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "readlink of %s on %s failed (%s)",
-                        impunge_local->loc.path,
-                        priv->children[active_src]->name,
-                        strerror (op_errno));
-                goto out;
-        }
-
-        impunge_sh->linkname = gf_strdup (linkname);
-        afr_sh_entry_impunge_readlink_sink (impunge_frame, this, child_index);
-
-        return 0;
+    int ret = -1;
+    void *index_gfid = NULL;
+    loc_t rootloc = {
+        0,
+    };
+    loc_t loc = {
+        0,
+    };
+    dict_t *xattr = NULL;
+    inode_t *inode = NULL;
+    struct iatt iatt = {
+        0,
+    };
+
+    rootloc.inode = inode_ref(this->itable->root);
+    gf_uuid_copy(rootloc.gfid, rootloc.inode->gfid);
+
+    ret = syncop_getxattr(subvol, &rootloc, &xattr,
+                          GF_XATTROP_ENTRY_CHANGES_GFID, NULL, NULL);
+    if (ret || !xattr) {
+        errno = -ret;
+        goto out;
+    }
+
+    ret = dict_get_ptr(xattr, GF_XATTROP_ENTRY_CHANGES_GFID, &index_gfid);
+    if (ret) {
+        errno = EINVAL;
+        goto out;
+    }
+
+    loc.inode = inode_new(this->itable);
+    if (!loc.inode) {
+        errno = ENOMEM;
+        goto out;
+    }
+
+    gf_uuid_copy(loc.pargfid, index_gfid);
+    loc.name = gf_strdup(uuid_utoa(pargfid));
+
+    ret = syncop_lookup(subvol, &loc, &iatt, NULL, NULL, NULL);
+    if (ret < 0) {
+        errno = -ret;
+        goto out;
+    }
+
+    inode = inode_link(loc.inode, NULL, NULL, &iatt);
 
 out:
-        LOCK (&impunge_frame->lock);
-        {
-                call_count = --impunge_local->call_count;
-        }
-        UNLOCK (&impunge_frame->lock);
+    if (xattr)
+        dict_unref(xattr);
+    loc_wipe(&rootloc);
+    GF_FREE((char *)loc.name);
+    loc_wipe(&loc);
 
-        if (call_count == 0)
-                afr_sh_entry_call_impunge_done (impunge_frame, this,
-                                                op_ret, op_errno);
-
-        return 0;
+    return inode;
 }
 
-
-int
-afr_sh_entry_impunge_readlink (call_frame_t *impunge_frame, xlator_t *this,
-                               int child_index, struct iatt *stbuf)
+static int
+afr_selfheal_entry_do_subvol(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                             int child)
 {
-        afr_private_t   *priv = NULL;
-        afr_local_t     *impunge_local = NULL;
-        afr_self_heal_t *impunge_sh = NULL;
-        int              active_src = -1;
-
-        priv = this->private;
-        impunge_local = impunge_frame->local;
-        impunge_sh = &impunge_local->self_heal;
-        active_src = impunge_sh->active_source;
-        impunge_local->cont.dir_fop.buf = *stbuf;
-
-        STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_readlink_cbk,
-                           (void *) (long) child_index,
-                           priv->children[active_src],
-                           priv->children[active_src]->fops->readlink,
-                           &impunge_local->loc, 4096, NULL);
+    int ret = 0;
+    gf_dirent_t entries;
+    gf_dirent_t *entry = NULL;
+    off_t offset = 0;
+    call_frame_t *iter_frame = NULL;
+    xlator_t *subvol = NULL;
+    afr_private_t *priv = NULL;
+    gf_boolean_t mismatch = _gf_false;
+    afr_local_t *local = NULL;
+    loc_t loc = {
+        0,
+    };
+
+    priv = this->private;
+    subvol = priv->children[child];
+
+    INIT_LIST_HEAD(&entries.list);
+
+    local = frame->local;
+
+    iter_frame = afr_copy_frame(frame);
+    if (!iter_frame)
+        return -ENOMEM;
+
+    loc.inode = afr_shd_entry_changes_index_inode(this, subvol,
+                                                  fd->inode->gfid);
+
+    while ((ret = syncop_readdir(subvol, fd, 131072, offset, &entries, NULL,
+                                 NULL))) {
+        if (ret > 0)
+            ret = 0;
+        list_for_each_entry(entry, &entries.list, list)
+        {
+            offset = entry->d_off;
 
-        return 0;
-}
+            if (!strcmp(entry->d_name, ".") || !strcmp(entry->d_name, ".."))
+                continue;
 
-int
-afr_sh_entry_impunge_create (call_frame_t *impunge_frame, xlator_t *this,
-                             int child_index)
-{
-        call_frame_t    *frame             = NULL;
-        afr_local_t     *impunge_local     = NULL;
-        afr_local_t     *local             = NULL;
-        afr_self_heal_t *impunge_sh        = NULL;
-        afr_self_heal_t *sh                = NULL;
-        afr_private_t   *priv = NULL;
-        ia_type_t       type = IA_INVAL;
-        int             active_src = 0;
-        struct iatt     *buf = NULL;
-
-        AFR_INIT_SH_FRAME_VALS (impunge_frame, impunge_local, impunge_sh,
-                                frame, local, sh);
-        active_src = impunge_sh->active_source;
-        afr_update_loc_gfids (&impunge_local->loc, &impunge_sh->entrybuf,
-                              &impunge_sh->parentbuf);
-
-        buf = &impunge_sh->entrybuf;
-        type = buf->ia_type;
-
-        switch (type) {
-        case IA_IFSOCK:
-        case IA_IFREG:
-        case IA_IFBLK:
-        case IA_IFCHR:
-        case IA_IFIFO:
-        case IA_IFLNK:
-                afr_sh_entry_impunge_check_hardlink (impunge_frame, this,
-                                                     child_index, buf);
+            ret = afr_selfheal_entry_dirent(iter_frame, this, fd, entry->d_name,
+                                            loc.inode, subvol,
+                                            local->need_full_crawl);
+            AFR_STACK_RESET(iter_frame);
+            if (iter_frame->local == NULL) {
+                ret = -ENOTCONN;
                 break;
-        case IA_IFDIR:
-                afr_sh_entry_impunge_mkdir (impunge_frame, this,
-                                            child_index, buf);
-                break;
-        default:
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s has unknown file type on %s: 0%o",
-                        impunge_local->loc.path,
-                        priv->children[active_src]->name, type);
-                sh->impunge_done (frame, this, -1, EINVAL);
+            }
+
+            if (ret == -1) {
+                /* gfid or type mismatch. */
+                mismatch = _gf_true;
+                ret = 0;
+            }
+            if (ret)
                 break;
         }
 
-        return 0;
-}
-
-int
-afr_sh_entry_impunge_create_file (call_frame_t *impunge_frame, xlator_t *this,
-                                  int child_index)
-{
-        call_frame_t    *frame             = NULL;
-        afr_local_t     *impunge_local     = NULL;
-        afr_local_t     *local             = NULL;
-        afr_self_heal_t *impunge_sh        = NULL;
-        afr_self_heal_t *sh                = NULL;
-        afr_private_t   *priv = NULL;
-        ia_type_t       type = IA_INVAL;
-        int             active_src = 0;
-        struct iatt     *buf = NULL;
-
-        AFR_INIT_SH_FRAME_VALS (impunge_frame, impunge_local, impunge_sh,
-                                frame, local, sh);
-        active_src = impunge_sh->active_source;
-        buf = &impunge_sh->entrybuf;
-        type = buf->ia_type;
-
-        switch (type) {
-        case IA_IFSOCK:
-        case IA_IFREG:
-        case IA_IFBLK:
-        case IA_IFCHR:
-        case IA_IFIFO:
-                afr_sh_entry_impunge_mknod (impunge_frame, this,
-                                            child_index, buf);
-                break;
-        case IA_IFLNK:
-                afr_sh_entry_impunge_readlink (impunge_frame, this,
-                                               child_index, buf);
-                break;
-        default:
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s has unknown file type on %s: 0%o",
-                        impunge_local->loc.path,
-                        priv->children[active_src]->name, type);
-                sh->impunge_done (frame, this, -1, EINVAL);
-                break;
-        }
-
-        return 0;
-}
-
-gf_boolean_t
-afr_sh_need_recreate (afr_self_heal_t *impunge_sh, unsigned int child,
-                      unsigned int child_count)
-{
-        gf_boolean_t    recreate = _gf_false;
-
-        GF_ASSERT (impunge_sh->child_errno);
-
-        if (child == impunge_sh->active_source)
-                goto out;
+        gf_dirent_free(&entries);
+        if (ret)
+            break;
+    }
 
-        if (IA_IFLNK == impunge_sh->entrybuf.ia_type) {
-                recreate = _gf_true;
-                goto out;
-        }
+    loc_wipe(&loc);
 
-        if (impunge_sh->child_errno[child] == ENOENT)
-                recreate = _gf_true;
-out:
-        return recreate;
+    AFR_STACK_DESTROY(iter_frame);
+    if (mismatch == _gf_true)
+        /* undo pending will be skipped */
+        ret = -1;
+    return ret;
 }
 
-unsigned int
-afr_sh_recreate_count (afr_self_heal_t *impunge_sh, int *sources,
-                       unsigned int child_count)
+static int
+afr_selfheal_entry_granular_dirent(xlator_t *subvol, gf_dirent_t *entry,
+                                   loc_t *parent, void *data)
 {
-        int             count = 0;
-        int             i = 0;
+    int ret = 0;
+    loc_t loc = {
+        0,
+    };
+    struct iatt iatt = {
+        0,
+    };
+    afr_granular_esh_args_t *args = data;
+
+    /* Look up the actual inode associated with entry. If the lookup returns
+     * ESTALE or ENOENT, then it means we have a stale index. Remove it.
+     * This is analogous to the check in afr_shd_index_heal() except that
+     * here it is achieved through LOOKUP and in afr_shd_index_heal() through
+     * a GETXATTR.
+     */
+
+    loc.inode = inode_new(args->xl->itable);
+    loc.parent = inode_ref(args->heal_fd->inode);
+    gf_uuid_copy(loc.pargfid, loc.parent->gfid);
+    loc.name = entry->d_name;
+
+    ret = syncop_lookup(args->xl, &loc, &iatt, NULL, NULL, NULL);
+    if ((ret == -ENOENT) || (ret == -ESTALE)) {
+        /* The name indices under the pgfid index dir are guaranteed
+         * to be regular files. Hence the hardcoding.
+         */
+        afr_shd_entry_purge(subvol, parent->inode, entry->d_name, IA_IFREG);
+        ret = 0;
+        goto out;
+    }
+    /* TBD: afr_shd_zero_xattrop? */
 
-        for (i = 0; i < child_count; i++) {
-                if (afr_sh_need_recreate (impunge_sh, i, child_count))
-                        count++;
-        }
+    ret = afr_selfheal_entry_dirent(args->frame, args->xl, args->heal_fd,
+                                    entry->d_name, parent->inode, subvol,
+                                    _gf_false);
+    AFR_STACK_RESET(args->frame);
+    if (args->frame->local == NULL)
+        ret = -ENOTCONN;
 
-        return count;
-}
+    if (ret == -1)
+        args->mismatch = _gf_true;
 
-int
-afr_sh_entry_call_impunge_recreate (call_frame_t *impunge_frame,
-                                    xlator_t *this)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *impunge_local = NULL;
-        afr_self_heal_t *impunge_sh = NULL;
-        call_frame_t    *frame = NULL;
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        unsigned int     recreate_count = 0;
-        int              i = 0;
-        int              active_src = 0;
-
-        priv          = this->private;
-        AFR_INIT_SH_FRAME_VALS (impunge_frame, impunge_local, impunge_sh,
-                                frame, local, sh);
-        active_src = impunge_sh->active_source;
-        impunge_sh->entrybuf = impunge_sh->buf[active_src];
-        impunge_sh->parentbuf = impunge_sh->parentbufs[active_src];
-        recreate_count = afr_sh_recreate_count (impunge_sh, sh->sources,
-                                                priv->child_count);
-        if (!recreate_count) {
-                afr_sh_entry_call_impunge_done (impunge_frame, this, 0, 0);
-                goto out;
-        }
-        impunge_local->call_count = recreate_count;
-        for (i = 0; i < priv->child_count; i++) {
-                if (!impunge_local->child_up[i]) {
-                        impunge_sh->child_errno[i] = ENOTCONN;
-                        continue;
-                }
-                if (!afr_sh_need_recreate (impunge_sh, i, priv->child_count)) {
-                        impunge_sh->child_errno[i] = EEXIST;
-                        continue;
-                }
-        }
-        for (i = 0; i < priv->child_count; i++) {
-                if (!afr_sh_need_recreate (impunge_sh, i, priv->child_count))
-                        continue;
-                (void)afr_sh_entry_impunge_create (impunge_frame, this, i);
-                recreate_count--;
-        }
-        GF_ASSERT (!recreate_count);
 out:
-        return 0;
-}
-
-void
-afr_sh_entry_common_lookup_done (call_frame_t *impunge_frame, xlator_t *this,
-                                 int32_t op_ret, int32_t op_errno)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *impunge_local = NULL;
-        afr_self_heal_t *impunge_sh = NULL;
-        call_frame_t    *frame = NULL;
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        unsigned int     gfid_miss_count = 0;
-        unsigned int     children_up_count = 0;
-        uuid_t           gfid = {0};
-        int              active_src = 0;
-
-        priv          = this->private;
-        AFR_INIT_SH_FRAME_VALS (impunge_frame, impunge_local, impunge_sh,
-                                frame, local, sh);
-        active_src    = impunge_sh->active_source;
-
-        if (op_ret < 0)
-                goto done;
-        if (impunge_sh->child_errno[active_src]) {
-                op_ret = -1;
-                op_errno = impunge_sh->child_errno[active_src];
-                goto done;
-        }
-
-        gfid_miss_count = afr_gfid_missing_count (this->name,
-                                                  impunge_sh->success_children,
-                                                  impunge_sh->buf, priv->child_count,
-                                                  impunge_local->loc.path);
-        children_up_count = afr_up_children_count (impunge_local->child_up,
-                                                   priv->child_count);
-        if ((gfid_miss_count == children_up_count) &&
-            (children_up_count < priv->child_count)) {
-                op_ret = -1;
-                op_errno = ENODATA;
-                gf_log (this->name, GF_LOG_ERROR, "Not all children are up, "
-                        "gfid should not be assigned in this state for %s",
-                        impunge_local->loc.path);
-                goto done;
-        }
-
-        if (gfid_miss_count) {
-                afr_update_gfid_from_iatts (gfid, impunge_sh->buf,
-                                            impunge_sh->success_children,
-                                            priv->child_count);
-                if (uuid_is_null (gfid)) {
-                        sh->entries_skipped = _gf_true;
-                        gf_log (this->name, GF_LOG_INFO, "%s: Skipping entry "
-                                "self-heal because of gfid absence",
-                                impunge_local->loc.path);
-                        goto done;
-                }
-                afr_sh_common_lookup (impunge_frame, this, &impunge_local->loc,
-                                      afr_sh_entry_common_lookup_done, gfid,
-                                      AFR_LOOKUP_FAIL_CONFLICTS |
-                                      AFR_LOOKUP_FAIL_MISSING_GFIDS,
-                                      NULL);
-        } else {
-                afr_sh_entry_call_impunge_recreate (impunge_frame, this);
-        }
-        return;
-done:
-        afr_sh_entry_call_impunge_done (impunge_frame, this,
-                                        op_ret, op_errno);
-        return;
+    loc_wipe(&loc);
+    return 0;
 }
 
-int
-afr_sh_entry_impunge_entry (call_frame_t *frame, xlator_t *this,
-                            gf_dirent_t *entry)
+static int
+afr_selfheal_entry_granular(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                            int subvol_idx, gf_boolean_t is_src)
 {
-        afr_local_t     *local  = NULL;
-        afr_self_heal_t *sh  = NULL;
-        afr_self_heal_t *impunge_sh  = NULL;
-        int              ret = -1;
-        call_frame_t    *impunge_frame = NULL;
-        afr_local_t     *impunge_local = NULL;
-        int              active_src = 0;
-        int              op_errno = 0;
-        int              op_ret = -1;
-
-        local = frame->local;
-        sh = &local->self_heal;
-
-        active_src = sh->active_source;
-        sh->impunge_done = afr_sh_entry_impunge_entry_done;
-
-        if (can_skip_entry_self_heal (entry->d_name, &local->loc)) {
-                op_ret = 0;
-                goto out;
-        }
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "inspecting existence of %s under %s",
-                entry->d_name, local->loc.path);
-
-        ret = afr_impunge_frame_create (frame, this, active_src,
-                                        &impunge_frame);
-        if (ret) {
-                op_errno = -ret;
-                goto out;
-        }
+    int ret = 0;
+    loc_t loc = {
+        0,
+    };
+    xlator_t *subvol = NULL;
+    afr_private_t *priv = NULL;
+    afr_granular_esh_args_t args = {
+        0,
+    };
+
+    priv = this->private;
+    subvol = priv->children[subvol_idx];
+
+    args.frame = afr_copy_frame(frame);
+    if (!args.frame)
+        goto out;
+    args.xl = this;
+    /* args.heal_fd represents the fd associated with the original directory
+     * on which entry heal is being attempted.
+     */
+    args.heal_fd = fd;
+
+    /* @subvol here represents the subvolume of AFR where
+     * indices/entry-changes/<pargfid> will be processed
+     */
+    loc.inode = afr_shd_entry_changes_index_inode(this, subvol,
+                                                  fd->inode->gfid);
+    if (!loc.inode) {
+        /* If granular heal failed on the sink (as it might sometimes
+         * because it is the src that would mostly contain the granular
+         * changelogs and the sink's entry-changes would be empty),
+         * do not treat heal as failure.
+         */
+        if (is_src)
+            ret = -errno;
+        else
+            ret = 0;
+        goto out;
+    }
 
-        impunge_local = impunge_frame->local;
-        impunge_sh = &impunge_local->self_heal;
-        ret = afr_build_child_loc (this, &impunge_local->loc, &local->loc,
-                                   entry->d_name);
-        loc_copy (&impunge_sh->parent_loc, &local->loc);
-        if (ret != 0) {
-                op_errno = ENOMEM;
-                goto out;
-        }
+    ret = syncop_dir_scan(subvol, &loc, GF_CLIENT_PID_SELF_HEALD, &args,
+                          afr_selfheal_entry_granular_dirent);
 
-        afr_sh_common_lookup (impunge_frame, this, &impunge_local->loc,
-                              afr_sh_entry_common_lookup_done, NULL,
-                              AFR_LOOKUP_FAIL_CONFLICTS, NULL);
+    loc_wipe(&loc);
 
-        op_ret = 0;
+    if (args.mismatch == _gf_true)
+        ret = -1;
 out:
-        if (ret) {
-                if (impunge_frame)
-                        AFR_STACK_DESTROY (impunge_frame);
-                sh->impunge_done (frame, this, op_ret, op_errno);
-        }
-
-        return 0;
-}
-
-
-int
-afr_sh_entry_impunge_readdir_cbk (call_frame_t *frame, void *cookie,
-                                  xlator_t *this,
-                                  int32_t op_ret, int32_t op_errno,
-                                  gf_dirent_t *entries, dict_t *xdata)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *local  = NULL;
-        afr_self_heal_t *sh  = NULL;
-        gf_dirent_t     *entry = NULL;
-        off_t            last_offset = 0;
-        int              active_src = 0;
-        int              entry_count = 0;
-
-        priv = this->private;
-        local = frame->local;
-        sh = &local->self_heal;
-
-        active_src = sh->active_source;
-
-        if (op_ret <= 0) {
-                if (op_ret < 0) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "readdir of %s on subvolume %s failed (%s)",
-                                local->loc.path,
-                                priv->children[active_src]->name,
-                                strerror (op_errno));
-                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-                } else {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "readdir of %s on subvolume %s complete",
-                                local->loc.path,
-                                priv->children[active_src]->name);
-                }
-
-                afr_sh_entry_impunge_all (frame, this);
-                return 0;
-        }
-
-        list_for_each_entry (entry, &entries->list, list) {
-                last_offset = entry->d_off;
-                entry_count++;
-        }
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "readdir'ed %d entries from %s",
-                entry_count, priv->children[active_src]->name);
-
-        sh->offset = last_offset;
-        local->call_count = entry_count;
-
-        list_for_each_entry (entry, &entries->list, list) {
-                afr_sh_entry_impunge_entry (frame, this, entry);
-        }
-
-        return 0;
-}
-
-
-int
-afr_sh_entry_impunge_subvol (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *local  = NULL;
-        afr_self_heal_t *sh  = NULL;
-        int32_t         active_src = 0;
-
-        priv = this->private;
-        local = frame->local;
-        sh = &local->self_heal;
-        active_src = sh->active_source;
-        gf_log (this->name, GF_LOG_DEBUG, "%s: readdir from offset %zd",
-                local->loc.path, sh->offset);
-
-        STACK_WIND (frame, afr_sh_entry_impunge_readdir_cbk,
-                    priv->children[active_src],
-                    priv->children[active_src]->fops->readdirp,
-                    sh->healing_fd, sh->block_size, sh->offset, NULL);
-
-        return 0;
-}
-
-
-int
-afr_sh_entry_impunge_all (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t   *priv = NULL;
-        afr_local_t     *local  = NULL;
-        afr_self_heal_t *sh  = NULL;
-        int              active_src = -1;
-
-        priv = this->private;
-        local = frame->local;
-        sh = &local->self_heal;
-
-        sh->offset = 0;
-
-        active_src = next_active_source (frame, this, sh->active_source);
-        sh->active_source = active_src;
-
-        if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
-                afr_sh_entry_finish (frame, this);
-                return 0;
-        }
-
-        if (active_src == -1) {
-                /* completed creating missing files on all subvolumes */
-                afr_sh_entry_erase_pending (frame, this);
-                return 0;
-        }
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "impunging entries of %s on %s to other sinks",
-                local->loc.path, priv->children[active_src]->name);
-
-        afr_sh_entry_impunge_subvol (frame, this);
-
-        return 0;
+    if (args.frame)
+        AFR_STACK_DESTROY(args.frame);
+    return ret;
 }
 
-
-int
-afr_sh_entry_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                          int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int              call_count = 0;
-        int              child_index = 0;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        child_index = (long) cookie;
-
-        /* TODO: some of the open's might fail.
-           In that case, modify cleanup fn to send flush on those
-           fd's which are already open */
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "opendir of %s failed on child %s (%s)",
-                                local->loc.path,
-                                priv->children[child_index]->name,
-                                strerror (op_errno));
-                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
-                        afr_sh_entry_finish (frame, this);
-                        return 0;
-                }
-                gf_log (this->name, GF_LOG_TRACE,
-                        "fd for %s opened, commencing sync",
-                        local->loc.path);
-
-                sh->active_source = -1;
-                afr_sh_entry_expunge_all (frame, this);
-        }
-
-        return 0;
-}
-
-
-int
-afr_sh_entry_open (call_frame_t *frame, xlator_t *this)
+static int
+afr_selfheal_entry_do(call_frame_t *frame, xlator_t *this, fd_t *fd, int source,
+                      unsigned char *sources, unsigned char *healed_sinks)
 {
-        int i = 0;
-        int call_count = 0;
-
-        int source = -1;
-        int *sources = NULL;
-
-        fd_t *fd = NULL;
-
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
-        afr_self_heal_t *sh = NULL;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        source  = local->self_heal.source;
-        sources = local->self_heal.sources;
-
-        sh->block_size = priv->sh_readdir_size;
-        sh->offset = 0;
-
-        call_count = sh->active_sinks;
-        if (source != -1)
-                call_count++;
-
-        local->call_count = call_count;
-
-        fd = fd_create (local->loc.inode, frame->root->pid);
-        sh->healing_fd = fd;
-
-        if (source != -1) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "opening directory %s on subvolume %s (source)",
-                        local->loc.path, priv->children[source]->name);
-
-                /* open source */
-                STACK_WIND_COOKIE (frame, afr_sh_entry_opendir_cbk,
-                                   (void *) (long) source,
-                                   priv->children[source],
-                                   priv->children[source]->fops->opendir,
-                                   &local->loc, fd, NULL);
-                call_count--;
-        }
-
-        /* open sinks */
-        for (i = 0; i < priv->child_count; i++) {
-                if (sources[i] || !local->child_up[i])
-                        continue;
-
-                gf_log (this->name, GF_LOG_TRACE,
-                        "opening directory %s on subvolume %s (sink)",
-                        local->loc.path, priv->children[i]->name);
-
-                STACK_WIND_COOKIE (frame, afr_sh_entry_opendir_cbk,
-                                   (void *) (long) i,
-                                   priv->children[i],
-                                   priv->children[i]->fops->opendir,
-                                   &local->loc, fd, NULL);
-
-                if (!--call_count)
-                        break;
-        }
-
-        return 0;
-}
-
+    int i = 0;
+    int ret = 0;
+    gf_boolean_t mismatch = _gf_false;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_SELF_HEAL_INFO,
+           "performing entry selfheal on %s", uuid_utoa(fd->inode->gfid));
+
+    for (i = 0; i < priv->child_count; i++) {
+        /* Expunge */
+        if (!healed_sinks[i])
+            continue;
+
+        if (!local->need_full_crawl)
+            /* Why call afr_selfheal_entry_granular() on a "healed sink",
+             * given that it is the source that contains the granular
+             * indices?
+             * If the index for this directory is non-existent or empty on
+             * this subvol (=> clear sink), then it will return early
+             * without failure status.
+             * If the index is non-empty and it is yet a 'healed sink', then
+             * it is due to a split-brain in which case we anyway need to
+             * crawl the indices/entry-changes/pargfid directory.
+             */
+            ret = afr_selfheal_entry_granular(frame, this, fd, i, _gf_false);
+        else
+            ret = afr_selfheal_entry_do_subvol(frame, this, fd, i);
 
-int
-afr_sh_entry_sync_prepare (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int              source = 0;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        source = sh->source;
-
-        afr_sh_mark_source_sinks (frame, this);
-        if (source != -1)
-                sh->success[source] = 1;
-
-        if (sh->active_sinks == 0) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "no active sinks for self-heal on dir %s",
-                        local->loc.path);
-                afr_sh_entry_finish (frame, this);
-                return 0;
-        }
-        if (source == -1 && sh->active_sinks < 2) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "cannot sync with 0 sources and 1 sink on dir %s",
-                        local->loc.path);
-                afr_sh_entry_finish (frame, this);
-                return 0;
+        if (ret == -1) {
+            /* gfid or type mismatch. */
+            mismatch = _gf_true;
+            ret = 0;
         }
+        if (ret)
+            break;
+    }
 
-        if (source != -1)
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "self-healing directory %s from subvolume %s to "
-                        "%d other",
-                        local->loc.path, priv->children[source]->name,
-                        sh->active_sinks);
+    if (!ret && source != -1) {
+        /* Impunge */
+        if (local->need_full_crawl)
+            ret = afr_selfheal_entry_do_subvol(frame, this, fd, source);
         else
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no active sources for %s found. "
-                        "merging all entries as a conservative decision",
-                        local->loc.path);
-
-        sh->actual_sh_started = _gf_true;
-        afr_set_self_heal_status (sh, AFR_SELF_HEAL_SYNC_BEGIN);
-        afr_sh_entry_open (frame, this);
-
-        return 0;
+            ret = afr_selfheal_entry_granular(frame, this, fd, source,
+                                              _gf_true);
+    }
+
+    if (mismatch == _gf_true)
+        /* undo pending will be skipped */
+        ret = -1;
+    return ret;
 }
 
-
-void
-afr_sh_entry_fix (call_frame_t *frame, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno)
+static int
+__afr_selfheal_entry(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                     unsigned char *locked_on)
 {
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int              source = 0;
-        int              nsources = 0;
-        int32_t          subvol_status = 0;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        if (op_ret < 0) {
-                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-                afr_sh_set_error (sh, op_errno);
-                afr_sh_entry_finish (frame, this);
-                goto out;
-        }
-
-        if (sh->forced_merge) {
-                sh->source = -1;
-                goto heal;
-        }
-
-        nsources = afr_build_sources (this, sh->xattr, sh->buf,
-                                      sh->pending_matrix, sh->sources,
-                                      sh->success_children,
-                                      AFR_ENTRY_TRANSACTION, &subvol_status,
-                                      _gf_true);
-        if ((subvol_status & ALL_FOOLS) ||
-            (subvol_status & SPLIT_BRAIN)) {
-                gf_log (this->name, GF_LOG_INFO, "%s: Performing conservative "
-                        "merge", local->loc.path);
-                source = -1;
-                memset (sh->sources, 0,
-                        sizeof (*sh->sources) * priv->child_count);
-        } else if (nsources == 0) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "No self-heal needed for %s",
-                        local->loc.path);
-
-                afr_sh_entry_finish (frame, this);
-                return;
-        } else {
-                source = afr_sh_select_source (sh->sources, priv->child_count);
-        }
-
-        sh->source = source;
-
-        afr_reset_children (sh->fresh_children, priv->child_count);
-        afr_get_fresh_children (sh->success_children, sh->sources,
-                                sh->fresh_children, priv->child_count);
-        if (sh->source >= 0)
-                afr_inode_set_read_ctx (this, sh->inode, sh->source,
-                                        sh->fresh_children);
-
-heal:
-        afr_sh_entry_sync_prepare (frame, this);
+    int ret = -1;
+    int source = -1;
+    unsigned char *sources = NULL;
+    unsigned char *sinks = NULL;
+    unsigned char *data_lock = NULL;
+    unsigned char *postop_lock = NULL;
+    unsigned char *healed_sinks = NULL;
+    unsigned char *undid_pending = NULL;
+    struct afr_reply *locked_replies = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    gf_boolean_t did_sh = _gf_true;
+
+    priv = this->private;
+    local = frame->local;
+
+    sources = alloca0(priv->child_count);
+    sinks = alloca0(priv->child_count);
+    healed_sinks = alloca0(priv->child_count);
+    undid_pending = alloca0(priv->child_count);
+    data_lock = alloca0(priv->child_count);
+    postop_lock = alloca0(priv->child_count);
+
+    locked_replies = alloca0(sizeof(*locked_replies) * priv->child_count);
+
+    ret = afr_selfheal_entrylk(frame, this, fd->inode, this->name, NULL,
+                               data_lock);
+    {
+        if (ret < priv->child_count) {
+            gf_msg_debug(this->name, 0,
+                         "%s: Skipping "
+                         "entry self-heal as only %d sub-volumes could "
+                         "be locked in %s domain",
+                         uuid_utoa(fd->inode->gfid), ret, this->name);
+            ret = -ENOTCONN;
+            goto unlock;
+        }
+
+        ret = __afr_selfheal_entry_prepare(frame, this, fd->inode, data_lock,
+                                           sources, sinks, healed_sinks,
+                                           locked_replies, &source, NULL);
+        if (AFR_COUNT(healed_sinks, priv->child_count) == 0) {
+            did_sh = _gf_false;
+            goto unlock;
+        }
+
+        local->need_full_crawl = afr_need_full_heal(
+            this, locked_replies, source, healed_sinks, AFR_ENTRY_TRANSACTION);
+    }
+unlock:
+    afr_selfheal_unentrylk(frame, this, fd->inode, this->name, NULL, data_lock,
+                           NULL);
+    if (ret < 0)
+        goto out;
+
+    if (!did_sh)
+        goto out;
+
+    ret = afr_selfheal_entry_do(frame, this, fd, source, sources, healed_sinks);
+    if (ret)
+        goto out;
+
+    /* Take entrylks in xlator domain before doing post-op (undo-pending) in
+     * entry self-heal. This is to prevent a parallel name self-heal on
+     * an entry under @fd->inode from reading pending xattrs while it is
+     * being modified by SHD after entry sh below, given that
+     * name self-heal takes locks ONLY in xlator domain and is free to read
+     * pending changelog in the absence of the following locking.
+     */
+    ret = afr_selfheal_entrylk(frame, this, fd->inode, this->name, NULL,
+                               postop_lock);
+    {
+        if (AFR_CMP(data_lock, postop_lock, priv->child_count) != 0) {
+            gf_msg_debug(this->name, 0,
+                         "%s: Skipping "
+                         "post-op after entry self-heal as %d "
+                         "sub-volumes, as opposed to %d, "
+                         "could be locked in %s domain",
+                         uuid_utoa(fd->inode->gfid), ret,
+                         AFR_COUNT(data_lock, priv->child_count), this->name);
+            ret = -ENOTCONN;
+            goto postop_unlock;
+        }
+
+        afr_selfheal_restore_time(frame, this, fd->inode, source, healed_sinks,
+                                  locked_replies);
+        ret = afr_selfheal_undo_pending(
+            frame, this, fd->inode, sources, sinks, healed_sinks, undid_pending,
+            AFR_ENTRY_TRANSACTION, locked_replies, postop_lock);
+    }
+postop_unlock:
+    afr_selfheal_unentrylk(frame, this, fd->inode, this->name, NULL,
+                           postop_lock, NULL);
 out:
-        return;
+    if (did_sh)
+        afr_log_selfheal(fd->inode->gfid, this, ret, "entry", source, sources,
+                         healed_sinks);
+    else
+        ret = 1;
+
+    if (locked_replies)
+        afr_replies_wipe(locked_replies, priv->child_count);
+    return ret;
 }
 
-int
-afr_sh_post_nonblocking_entry_cbk (call_frame_t *frame, xlator_t *this)
+static fd_t *
+afr_selfheal_data_opendir(xlator_t *this, inode_t *inode)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_self_heal_t     *sh       = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        sh       = &local->self_heal;
-
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Non Blocking entrylks "
-                        "failed for %s.", local->loc.path);
-                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-                afr_sh_entry_done (frame, this);
-        } else {
-
-                gf_log (this->name, GF_LOG_DEBUG, "Non Blocking entrylks done "
-                        "for %s. Proceeding to FOP", local->loc.path);
-                afr_sh_common_lookup (frame, this, &local->loc,
-                                      afr_sh_entry_fix, NULL,
-                                      AFR_LOOKUP_FAIL_CONFLICTS |
-                                      AFR_LOOKUP_FAIL_MISSING_GFIDS,
-                                      NULL);
-        }
-
-        return 0;
+    loc_t loc = {
+        0,
+    };
+    int ret = 0;
+    fd_t *fd = NULL;
+
+    fd = fd_create(inode, 0);
+    if (!fd)
+        return NULL;
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+
+    ret = syncop_opendir(this, &loc, fd, NULL, NULL);
+    if (ret) {
+        fd_unref(fd);
+        fd = NULL;
+    } else {
+        fd_bind(fd);
+    }
+
+    loc_wipe(&loc);
+    return fd;
 }
 
 int
-afr_self_heal_entry (call_frame_t *frame, xlator_t *this)
+afr_selfheal_entry(call_frame_t *frame, xlator_t *this, inode_t *inode)
 {
-        afr_local_t     *local = NULL;
-        afr_private_t   *priv = NULL;
-        afr_self_heal_t *sh = NULL;
-
-        priv = this->private;
-        local = frame->local;
-        sh = &local->self_heal;
-
-        sh->sh_type_in_action = AFR_SELF_HEAL_ENTRY;
-
-        if (local->self_heal.do_entry_self_heal && priv->entry_self_heal) {
-                afr_set_self_heal_status (sh, AFR_SELF_HEAL_STARTED);
-                afr_sh_entrylk (frame, this, &local->loc, NULL,
-                                afr_sh_post_nonblocking_entry_cbk);
-        } else {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "proceeding to completion on %s",
-                        local->loc.path);
-                afr_sh_entry_done (frame, this);
-        }
-
-        return 0;
+    afr_private_t *priv = NULL;
+    unsigned char *locked_on = NULL;
+    fd_t *fd = NULL;
+    int ret = 0;
+
+    priv = this->private;
+
+    fd = afr_selfheal_data_opendir(this, inode);
+    if (!fd)
+        return -EIO;
+
+    locked_on = alloca0(priv->child_count);
+
+    ret = afr_selfheal_tie_breaker_entrylk(frame, this, inode, priv->sh_domain,
+                                           NULL, locked_on);
+    {
+        if (ret < priv->child_count) {
+            gf_msg_debug(this->name, 0,
+                         "%s: Skipping "
+                         "entry self-heal as only %d sub-volumes could "
+                         "be locked in %s domain",
+                         uuid_utoa(fd->inode->gfid), ret, priv->sh_domain);
+            /* Either less than two subvols available, or another
+               selfheal (from another server) is in progress. Skip
+               for now in any case there isn't anything to do.
+            */
+            ret = -ENOTCONN;
+            goto unlock;
+        }
+
+        ret = __afr_selfheal_entry(frame, this, fd, locked_on);
+    }
+unlock:
+    afr_selfheal_unentrylk(frame, this, inode, priv->sh_domain, NULL, locked_on,
+                           NULL);
+
+    if (fd)
+        fd_unref(fd);
+
+    return ret;
 }
diff --git a/xlators/cluster/afr/src/afr-self-heal-metadata.c b/xlators/cluster/afr/src/afr-self-heal-metadata.c
index fd5da6cfd33..03f43bad16e 100644
--- a/xlators/cluster/afr/src/afr-self-heal-metadata.c
+++ b/xlators/cluster/afr/src/afr-self-heal-metadata.c
@@ -1,5 +1,5 @@
 /*
-  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
   This file is licensed to you under your choice of the GNU Lesser
@@ -8,763 +8,539 @@
   cases as published by the Free Software Foundation.
 */
 
-#include <libgen.h>
-#include <unistd.h>
-#include <fnmatch.h>
-#include <sys/time.h>
-#include <stdlib.h>
-#include <signal.h>
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
 #include "afr.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
-#include "byte-order.h"
-
-#include "afr-transaction.h"
 #include "afr-self-heal.h"
-#include "afr-self-heal-common.h"
-
-
-int
-afr_sh_metadata_done (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-
-        local = frame->local;
-        sh = &local->self_heal;
-
-        afr_sh_reset (frame, this);
-        if (IA_ISDIR (sh->type)) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "proceeding to entry check on %s",
-                        local->loc.path);
-                afr_self_heal_entry (frame, this);
-        } else {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "proceeding to data check on %s",
-                        local->loc.path);
-                afr_self_heal_data (frame, this);
-        }
-
-        return 0;
-}
-
-int
-afr_sh_inode_unlock (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        int_lock->lock_cbk = afr_sh_metadata_done;
-        afr_unlock (frame, this);
-
-        return 0;
-}
-
-int
-afr_sh_metadata_finish (call_frame_t *frame, xlator_t *this)
-{
-        afr_sh_inode_unlock (frame, this);
+#include <glusterfs/byte-order.h>
+#include "protocol-common.h"
+#include <glusterfs/events.h>
 
-        return 0;
-}
+#define AFR_HEAL_ATTR (GF_SET_ATTR_UID | GF_SET_ATTR_GID | GF_SET_ATTR_MODE)
 
-int
-afr_sh_metadata_fail (call_frame_t *frame, xlator_t *this)
+static gf_boolean_t
+_afr_ignorable_key_match(dict_t *d, char *k, data_t *val, void *mdata)
 {
-        afr_local_t         *local    = NULL;
-        afr_self_heal_t     *sh       = NULL;
-
-        local    = frame->local;
-        sh       = &local->self_heal;
-
-        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-        afr_sh_metadata_finish (frame, this);
-        return 0;
+    return afr_is_xattr_ignorable(k);
 }
 
-int
-afr_sh_metadata_erase_pending_cbk (call_frame_t *frame, void *cookie,
-                                   xlator_t *this, int32_t op_ret,
-                                   int32_t op_errno, dict_t *xattr, dict_t *xdata)
-{
-        afr_local_t     *local     = NULL;
-        int             call_count = 0;
-        long            i          = 0;
-        afr_self_heal_t *sh        = NULL;
-        afr_private_t   *priv      = NULL;
-
-        local = frame->local;
-        priv  = this->private;
-        sh = &local->self_heal;
-        i = (long)cookie;
-
-        if ((!IA_ISREG (sh->buf[sh->source].ia_type)) &&
-            (!IA_ISDIR (sh->buf[sh->source].ia_type))) {
-                afr_children_add_child (sh->fresh_children, i,
-                                        priv->child_count);
-        }
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                if ((!IA_ISREG (sh->buf[sh->source].ia_type)) &&
-                    (!IA_ISDIR (sh->buf[sh->source].ia_type))) {
-                        afr_inode_set_read_ctx (this, sh->inode, sh->source,
-                                                sh->fresh_children);
-                }
-                afr_sh_metadata_finish (frame, this);
-        }
-
-        return 0;
-}
-
-int
-afr_sh_metadata_erase_pending (call_frame_t *frame, xlator_t *this)
+void
+afr_delete_ignorable_xattrs(dict_t *xattr)
 {
-         afr_sh_erase_pending (frame, this, AFR_METADATA_TRANSACTION,
-                               afr_sh_metadata_erase_pending_cbk,
-                               afr_sh_metadata_finish);
-         return 0;
+    dict_foreach_match(xattr, _afr_ignorable_key_match, NULL,
+                       dict_remove_foreach_fn, NULL);
 }
 
-
 int
-afr_sh_metadata_sync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                          int32_t op_ret, int32_t op_errno, dict_t *xdata)
+__afr_selfheal_metadata_do(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                           int source, unsigned char *healed_sinks,
+                           struct afr_reply *locked_replies)
 {
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int              call_count = 0;
-        int              child_index = 0;
-
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        child_index = (long) cookie;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "setting attributes failed for %s on %s (%s)",
-                                local->loc.path,
-                                priv->children[child_index]->name,
-                                strerror (op_errno));
-
-                        sh->success[child_index] = 0;
-                }
+    int ret = -1;
+    loc_t loc = {
+        0,
+    };
+    dict_t *xattr = NULL;
+    dict_t *old_xattr = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+
+    priv = this->private;
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+
+    gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_SELF_HEAL_INFO,
+           "performing metadata selfheal on %s", uuid_utoa(inode->gfid));
+
+    ret = syncop_getxattr(priv->children[source], &loc, &xattr, NULL, NULL,
+                          NULL);
+    if (ret < 0) {
+        ret = -EIO;
+        goto out;
+    }
+
+    afr_delete_ignorable_xattrs(xattr);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (old_xattr) {
+            dict_unref(old_xattr);
+            old_xattr = NULL;
         }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
 
-        if (call_count == 0) {
-                if (local->xattr_req) {
-                        dict_unref (local->xattr_req);
-                        local->xattr_req = NULL;
-                }
-                afr_sh_metadata_erase_pending (frame, this);
+        if (!healed_sinks[i])
+            continue;
+
+        ret = syncop_setattr(priv->children[i], &loc,
+                             &locked_replies[source].poststat, AFR_HEAL_ATTR,
+                             NULL, NULL, NULL, NULL);
+        if (ret)
+            healed_sinks[i] = 0;
+
+        ret = syncop_getxattr(priv->children[i], &loc, &old_xattr, 0, NULL,
+                              NULL);
+        if (old_xattr) {
+            afr_delete_ignorable_xattrs(old_xattr);
+            ret = syncop_removexattr(priv->children[i], &loc, "", old_xattr,
+                                     NULL);
+            if (ret)
+                healed_sinks[i] = 0;
         }
 
-        return 0;
-}
-
-
-int
-afr_sh_metadata_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                             int32_t op_ret, int32_t op_errno,
-                             struct iatt *preop, struct iatt *postop, dict_t *xdata)
-{
-        afr_sh_metadata_sync_cbk (frame, cookie, this, op_ret, op_errno, xdata);
+        ret = syncop_setxattr(priv->children[i], &loc, xattr, 0, NULL, NULL);
+        if (ret)
+            healed_sinks[i] = 0;
+    }
+    ret = 0;
 
-        return 0;
-}
-
-
-int
-afr_sh_metadata_xattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                           int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-        afr_sh_metadata_sync_cbk (frame, cookie, this, op_ret, op_errno, xdata);
+out:
+    loc_wipe(&loc);
+    if (xattr)
+        dict_unref(xattr);
+    if (old_xattr)
+        dict_unref(old_xattr);
 
-        return 0;
+    return ret;
 }
 
-int
-afr_sh_removexattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno,
-                        dict_t *xdata)
+static uint64_t
+mtime_ns(struct iatt *ia)
 {
-        int            i     = 0;
-        afr_private_t *priv  = NULL;
-        afr_local_t   *local = NULL;
-
-        priv = this->private;
-        local = frame->local;
+    uint64_t ret;
 
-        if (op_ret < 0) {
-                afr_sh_metadata_sync_cbk (frame, cookie,
-                                          this, -1, op_errno, xdata);
-                goto out;
-        }
-
-        i = (long) cookie;
+    ret = (((uint64_t)(ia->ia_mtime)) * 1000000000) +
+          (uint64_t)(ia->ia_mtime_nsec);
 
-        STACK_WIND_COOKIE (frame, afr_sh_metadata_xattr_cbk,
-                           (void *) (long) i,
-                           priv->children[i],
-                           priv->children[i]->fops->setxattr,
-                           &local->loc, local->xattr_req, 0, NULL);
-
- out:
-        return 0;
+    return ret;
 }
 
-inline void
-afr_prune_special_keys (dict_t *xattr_dict)
+/*
+ * When directory content is modified, [mc]time is updated. On
+ * Linux, the filesystem does it, while at least on NetBSD, the
+ * kernel file-system independent code does it. This means that
+ * when entries are added while bricks are down, the kernel sends
+ * a SETATTR [mc]time which will cause metadata split brain for
+ * the directory. In this case, clear the split brain by finding
+ * the source with the most recent modification date.
+ */
+static int
+afr_dirtime_splitbrain_source(call_frame_t *frame, xlator_t *this,
+                              struct afr_reply *replies,
+                              unsigned char *locked_on)
 {
-        dict_del (xattr_dict, GF_SELINUX_XATTR_KEY);
+    afr_private_t *priv = NULL;
+    int source = -1;
+    struct iatt source_ia;
+    struct iatt child_ia;
+    uint64_t mtime = 0;
+    int i;
+    int ret = -1;
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!locked_on[i])
+            continue;
+
+        if (!replies[i].valid)
+            continue;
+
+        if (replies[i].op_ret != 0)
+            continue;
+
+        if (mtime_ns(&replies[i].poststat) <= mtime)
+            continue;
+
+        mtime = mtime_ns(&replies[i].poststat);
+        source = i;
+    }
+
+    if (source == -1)
+        goto out;
+
+    source_ia = replies[source].poststat;
+    if (source_ia.ia_type != IA_IFDIR)
+        goto out;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (i == source)
+            continue;
+
+        if (!replies[i].valid)
+            continue;
+
+        if (replies[i].op_ret != 0)
+            continue;
+
+        child_ia = replies[i].poststat;
+
+        if (!IA_EQUAL(source_ia, child_ia, gfid) ||
+            !IA_EQUAL(source_ia, child_ia, type) ||
+            !IA_EQUAL(source_ia, child_ia, prot) ||
+            !IA_EQUAL(source_ia, child_ia, uid) ||
+            !IA_EQUAL(source_ia, child_ia, gid) ||
+            !afr_xattrs_are_equal(replies[source].xdata, replies[i].xdata))
+            goto out;
+    }
+
+    /*
+     * Metadata split brain is just about [amc]time
+     * We return our source.
+     */
+    ret = source;
+out:
+    return ret;
 }
 
-inline void
-afr_prune_pending_keys (dict_t *xattr_dict, afr_private_t *priv)
+static int
+__afr_selfheal_metadata_mark_pending_xattrs(call_frame_t *frame, xlator_t *this,
+                                            inode_t *inode,
+                                            struct afr_reply *replies,
+                                            unsigned char *sources)
 {
-        int i = 0;
-
-        for (; i < priv->child_count; i++) {
-                dict_del (xattr_dict, priv->pending_key[i]);
+    int ret = 0;
+    int i = 0;
+    int m_idx = 0;
+    afr_private_t *priv = NULL;
+    int raw[AFR_NUM_CHANGE_LOGS] = {0};
+    dict_t *xattr = NULL;
+
+    priv = this->private;
+    m_idx = afr_index_for_transaction_type(AFR_METADATA_TRANSACTION);
+    raw[m_idx] = 1;
+
+    xattr = dict_new();
+    if (!xattr)
+        return -ENOMEM;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (sources[i])
+            continue;
+        ret = dict_set_static_bin(xattr, priv->pending_key[i], raw,
+                                  sizeof(int) * AFR_NUM_CHANGE_LOGS);
+        if (ret) {
+            ret = -1;
+            goto out;
         }
-}
-
-int
-afr_sh_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, dict_t *xattr,
-                     dict_t *xdata)
-{
-        int            i     = 0;
-        afr_private_t *priv  = NULL;
-        afr_local_t   *local = NULL;
-
-        priv = this->private;
-        local = frame->local;
-
-        if (op_ret < 0) {
-                afr_sh_metadata_sync_cbk (frame, cookie,
-                                          this, -1, op_errno, xdata);
-                goto out;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i])
+            continue;
+        ret = afr_selfheal_post_op(frame, this, inode, i, xattr, NULL);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_INFO, -ret, AFR_MSG_SELF_HEAL_INFO,
+                   "Failed to set pending metadata xattr on child %d for %s", i,
+                   uuid_utoa(inode->gfid));
+            goto out;
         }
+    }
 
-        afr_prune_pending_keys (xattr, priv);
-
-        afr_prune_special_keys (xattr);
+    afr_replies_wipe(replies, priv->child_count);
+    ret = afr_selfheal_unlocked_discover(frame, inode, inode->gfid, replies);
 
-        i = (long) cookie;
-
-        /* send removexattr in bulk via xdata */
-        STACK_WIND_COOKIE (frame, afr_sh_removexattr_cbk,
-                           cookie,
-                           priv->children[i],
-                           priv->children[i]->fops->removexattr,
-                           &local->loc, "", xattr);
-
- out:
-        return 0;
-}
-
-int
-afr_sh_metadata_sync (call_frame_t *frame, xlator_t *this, dict_t *xattr)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int              source = 0;
-        int              active_sinks = 0;
-        int              call_count = 0;
-        int              i = 0;
-
-        struct iatt      stbuf = {0,};
-        int32_t          valid = 0;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        source = sh->source;
-        active_sinks = sh->active_sinks;
-
-        /*
-         * 2 calls per sink - setattr, setxattr
-         */
-        if (xattr) {
-                call_count = active_sinks * 2;
-                local->xattr_req = dict_ref (xattr);
-        } else
-                call_count = active_sinks;
-
-        local->call_count = call_count;
-
-        stbuf.ia_atime = sh->buf[source].ia_atime;
-        stbuf.ia_atime_nsec = sh->buf[source].ia_atime_nsec;
-        stbuf.ia_mtime = sh->buf[source].ia_mtime;
-        stbuf.ia_mtime_nsec = sh->buf[source].ia_mtime_nsec;
-
-        stbuf.ia_uid = sh->buf[source].ia_uid;
-        stbuf.ia_gid = sh->buf[source].ia_gid;
-
-        stbuf.ia_type = sh->buf[source].ia_type;
-        stbuf.ia_prot = sh->buf[source].ia_prot;
-
-        valid = GF_SET_ATTR_MODE  |
-                GF_SET_ATTR_UID   | GF_SET_ATTR_GID |
-                GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (call_count == 0) {
-                        break;
-                }
-                if (sh->sources[i] || !local->child_up[i])
-                        continue;
-
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "self-healing metadata of %s from %s to %s",
-                        local->loc.path, priv->children[source]->name,
-                        priv->children[i]->name);
-
-                STACK_WIND_COOKIE (frame, afr_sh_metadata_setattr_cbk,
-                                   (void *) (long) i,
-                                   priv->children[i],
-                                   priv->children[i]->fops->setattr,
-                                   &local->loc, &stbuf, valid, NULL);
-
-                call_count--;
-
-                if (!xattr)
-                        continue;
-
-                STACK_WIND_COOKIE (frame, afr_sh_getxattr_cbk,
-                                   (void *) (long) i,
-                                   priv->children[i],
-                                   priv->children[i]->fops->getxattr,
-                                   &local->loc, NULL, NULL);
-                call_count--;
-        }
-
-        return 0;
-}
-
-
-int
-afr_sh_metadata_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                              int32_t op_ret, int32_t op_errno, dict_t *xattr,
-                              dict_t *xdata)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int              source = 0;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        source = sh->source;
-
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "getxattr of %s failed on subvolume %s (%s). proceeding without xattr",
-                        local->loc.path, priv->children[source]->name,
-                        strerror (op_errno));
-
-                afr_sh_metadata_sync (frame, this, NULL);
-        } else {
-                afr_prune_pending_keys (xattr, priv);
-                afr_sh_metadata_sync (frame, this, xattr);
-        }
-
-        return 0;
+out:
+    if (xattr)
+        dict_unref(xattr);
+    return ret;
 }
 
-static void
-afr_set_metadata_sh_info_str (afr_local_t *local, afr_self_heal_t *sh,
-                              xlator_t *this)
+/*
+ * Look for mismatching uid/gid or mode or user xattrs even if
+ * AFR xattrs don't say so, and pick one arbitrarily as winner. */
+
+static int
+__afr_selfheal_metadata_finalize_source(call_frame_t *frame, xlator_t *this,
+                                        inode_t *inode, unsigned char *sources,
+                                        unsigned char *sinks,
+                                        unsigned char *healed_sinks,
+                                        unsigned char *undid_pending,
+                                        unsigned char *locked_on,
+                                        struct afr_reply *replies)
 {
-        afr_private_t    *priv = NULL;
-        int              i = 0;
-        char             num[1024] = {0};
-        size_t           len = 0;
-        char             *string = NULL;
-        size_t           off = 0;
-        char             *source_child =  " from source %s to";
-        char             *format = " %s, ";
-        char             *string_msg = " metadata self heal";
-        char             *pending_matrix_str = NULL;
-        int              down_child_present = 0;
-        int              unknown_child_present = 0;
-        char             *down_subvol_1 = " down subvolume is ";
-        char             *unknown_subvol_1 = " unknown subvolume is";
-        char             *down_subvol_2 = " down subvolumes are ";
-        char             *unknown_subvol_2 = " unknown subvolumes are ";
-        int              down_count = 0;
-        int              unknown_count = 0;
-
-        priv = this->private;
-
-        pending_matrix_str = afr_get_pending_matrix_str (sh->pending_matrix,
-                                                         this);
-
-        if (!pending_matrix_str)
-                pending_matrix_str = "";
-
-        len += snprintf (num, sizeof (num), "%s", string_msg);
-
-        for (i = 0; i < priv->child_count; i++) {
-                if ((sh->source == i) && (local->child_up[i] == 1)) {
-                        len += snprintf (num, sizeof (num), source_child,
-                                         priv->children[i]->name);
-                } else if ((local->child_up[i] == 1) && (sh->sources[i] == 0)) {
-                        len += snprintf (num, sizeof (num), format,
-                                         priv->children[i]->name);
-                } else if (local->child_up[i] == 0) {
-                        len += snprintf (num, sizeof (num), format,
-                                         priv->children[i]->name);
-                        if (!down_child_present)
-                                down_child_present = 1;
-                        down_count++;
-                } else if (local->child_up[i] == -1) {
-                        len += snprintf (num, sizeof (num), format,
-                                         priv->children[i]->name);
-                        if (!unknown_child_present)
-                                unknown_child_present = 1;
-                        unknown_count++;
-                }
-        }
-
-        if (down_child_present) {
-                if (down_count > 1) {
-                        len += snprintf (num, sizeof (num), "%s",
-                                         down_subvol_2);
-                } else {
-                        len += snprintf (num, sizeof (num), "%s",
-                                         down_subvol_1);
-                }
-        }
-        if (unknown_child_present) {
-                if (unknown_count > 1) {
-                        len += snprintf (num, sizeof (num), "%s",
-                                         unknown_subvol_2);
-                } else {
-                        len += snprintf (num, sizeof (num), "%s",
-                                         unknown_subvol_1);
-                }
-        }
-
-        len ++;
-
-        string = GF_CALLOC (len, sizeof (char), gf_common_mt_char);
-        if (!string)
-                return;
-
-        off += snprintf (string + off, len - off, "%s", string_msg);
-        for (i=0; i < priv->child_count; i++) {
-                if ((sh->source == i) && (local->child_up[i] == 1))
-                        off += snprintf (string + off, len - off, source_child,
-                                         priv->children[i]->name);
+    int i = 0;
+    afr_private_t *priv = NULL;
+    struct iatt srcstat = {
+        0,
+    };
+    int source = -1;
+    int sources_count = 0;
+    int ret = 0;
+
+    priv = this->private;
+
+    sources_count = AFR_COUNT(sources, priv->child_count);
+
+    if ((AFR_CMP(locked_on, healed_sinks, priv->child_count) == 0) ||
+        !sources_count) {
+        source = afr_mark_split_brain_source_sinks(
+            frame, this, inode, sources, sinks, healed_sinks, locked_on,
+            replies, AFR_METADATA_TRANSACTION);
+        if (source >= 0) {
+            _afr_fav_child_reset_sink_xattrs(
+                frame, this, inode, source, healed_sinks, undid_pending,
+                AFR_METADATA_TRANSACTION, locked_on, replies);
+            goto out;
         }
 
-        for (i = 0; i < priv->child_count; i++) {
-                if ((local->child_up[i] == 1)&& (sh->sources[i] == 0))
-                        off += snprintf (string + off, len - off, format,
-                                         priv->children[i]->name);
+        /* If this is a directory mtime/ctime only split brain
+           use the most recent */
+        source = afr_dirtime_splitbrain_source(frame, this, replies, locked_on);
+        if (source != -1) {
+            gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_SPLIT_BRAIN,
+                   "clear time "
+                   "split brain on %s",
+                   uuid_utoa(replies[source].poststat.ia_gfid));
+            sources[source] = 1;
+            healed_sinks[source] = 0;
+            goto out;
         }
 
-        if (down_child_present) {
-                if (down_count > 1) {
-                        off += snprintf (string + off, len - off, "%s",
-                                         down_subvol_2);
-                } else {
-                        off += snprintf (string + off, len - off, "%s",
-                                         down_subvol_1);
-                }
+        if (!priv->metadata_splitbrain_forced_heal) {
+            gf_event(EVENT_AFR_SPLIT_BRAIN,
+                     "client-pid=%d;"
+                     "subvol=%s;"
+                     "type=metadata;file=%s",
+                     this->ctx->cmd_args.client_pid, this->name,
+                     uuid_utoa(inode->gfid));
+            return -EIO;
         }
 
+        /* Metadata split brain, select one subvol
+           arbitrarily */
         for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i] == 0)
-                        off += snprintf (string + off, len - off, format,
-                                         priv->children[i]->name);
+            if (locked_on[i] && healed_sinks[i]) {
+                sources[i] = 1;
+                healed_sinks[i] = 0;
+                break;
+            }
         }
-
-        if (unknown_child_present) {
-                if (unknown_count > 1) {
-                        off += snprintf (string + off, len - off, "%s",
-                                 unknown_subvol_2);
-                } else {
-                        off += snprintf (string + off, len - off, "%s",
-                                         unknown_subvol_1);
-                }
+    }
+
+    /* No split brain at this point. If we were called from
+     * afr_heal_splitbrain_file(), abort.*/
+    if (afr_dict_contains_heal_op(frame))
+        return -EIO;
+
+    source = afr_choose_source_by_policy(priv, sources,
+                                         AFR_METADATA_TRANSACTION);
+    srcstat = replies[source].poststat;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i] || i == source)
+            continue;
+        if (!IA_EQUAL(srcstat, replies[i].poststat, type) ||
+            !IA_EQUAL(srcstat, replies[i].poststat, uid) ||
+            !IA_EQUAL(srcstat, replies[i].poststat, gid) ||
+            !IA_EQUAL(srcstat, replies[i].poststat, prot)) {
+            gf_msg_debug(this->name, 0,
+                         "%s: iatt mismatch "
+                         "for source(%d) vs (%d)",
+                         uuid_utoa(replies[source].poststat.ia_gfid), source,
+                         i);
+            sources[i] = 0;
+            healed_sinks[i] = 1;
         }
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i] == -1)
-                        off += snprintf (string + off, len - off, format,
-                                         priv->children[i]->name);
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i] || i == source)
+            continue;
+        if (!afr_xattrs_are_equal(replies[source].xdata, replies[i].xdata)) {
+            gf_msg_debug(this->name, 0,
+                         "%s: xattr mismatch "
+                         "for source(%d) vs (%d)",
+                         uuid_utoa(replies[source].poststat.ia_gfid), source,
+                         i);
+            sources[i] = 0;
+            healed_sinks[i] = 1;
         }
-
-        gf_asprintf (&sh->metadata_sh_info, "%s metadata %s,", string,
-                     pending_matrix_str);
-
-        if (pending_matrix_str && strcmp (pending_matrix_str, ""))
-                GF_FREE (pending_matrix_str);
-
-        if (string && strcmp (string, ""))
-                GF_FREE (string);
+    }
+    if ((sources_count == priv->child_count) && (source > -1) &&
+        (AFR_COUNT(healed_sinks, priv->child_count) != 0)) {
+        ret = __afr_selfheal_metadata_mark_pending_xattrs(frame, this, inode,
+                                                          replies, sources);
+        if (ret < 0)
+            return ret;
+    }
+out:
+    afr_mark_active_sinks(this, sources, locked_on, healed_sinks);
+    return source;
 }
 
 int
-afr_sh_metadata_sync_prepare (call_frame_t *frame, xlator_t *this)
+__afr_selfheal_metadata_prepare(call_frame_t *frame, xlator_t *this,
+                                inode_t *inode, unsigned char *locked_on,
+                                unsigned char *sources, unsigned char *sinks,
+                                unsigned char *healed_sinks,
+                                unsigned char *undid_pending,
+                                struct afr_reply *replies, unsigned char *pflag)
 {
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int              source = 0;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        source = sh->source;
-
-        afr_sh_mark_source_sinks (frame, this);
-        if (sh->active_sinks == 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no active sinks for performing self-heal on file %s",
-                        local->loc.path);
-                afr_sh_metadata_finish (frame, this);
-                return 0;
+    int ret = -1;
+    int source = -1;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    uint64_t *witness = NULL;
+
+    priv = this->private;
+
+    ret = afr_selfheal_unlocked_discover(frame, inode, inode->gfid, replies);
+    if (ret)
+        return ret;
+
+    witness = alloca0(sizeof(*witness) * priv->child_count);
+    ret = afr_selfheal_find_direction(frame, this, replies,
+                                      AFR_METADATA_TRANSACTION, locked_on,
+                                      sources, sinks, witness, pflag);
+    if (ret)
+        return ret;
+
+    /* Initialize the healed_sinks[] array optimistically to
+       the intersection of to-be-healed (i.e sinks[]) and
+       the list of servers which are up (i.e locked_on[]).
+
+       As we encounter failures in the healing process, we
+       will unmark the respective servers in the healed_sinks[]
+       array.
+    */
+    AFR_INTERSECT(healed_sinks, sinks, locked_on, priv->child_count);
+
+    /* If any source has witness, pick first
+     * witness source and make everybody else sinks */
+    for (i = 0; i < priv->child_count; i++) {
+        if (sources[i] && witness[i]) {
+            source = i;
+            break;
         }
+    }
 
-        gf_log (this->name, GF_LOG_TRACE,
-                "syncing metadata of %s from subvolume %s to %d active sinks",
-                local->loc.path, priv->children[source]->name,
-                sh->active_sinks);
-
-        sh->actual_sh_started = _gf_true;
-        afr_set_self_heal_status (sh, AFR_SELF_HEAL_SYNC_BEGIN);
-        afr_set_metadata_sh_info_str (local, sh, this);
-        STACK_WIND (frame, afr_sh_metadata_getxattr_cbk,
-                    priv->children[source],
-                    priv->children[source]->fops->getxattr,
-                    &local->loc, NULL, NULL);
-
-        return 0;
-}
-
-
-void
-afr_sh_metadata_fix (call_frame_t *frame, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno)
-{
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int              nsources = 0;
-        int              source = 0;
-        int              i = 0;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        priv = this->private;
-
-        if (op_ret < 0) {
-                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
-                afr_sh_set_error (sh, op_errno);
-                afr_sh_metadata_finish (frame, this);
-                goto out;
-        }
-        nsources = afr_build_sources (this, sh->xattr, sh->buf,
-                                      sh->pending_matrix, sh->sources,
-                                      sh->success_children,
-                                      AFR_METADATA_TRANSACTION, NULL, _gf_false);
-        if ((nsources == -1)
-            && (priv->favorite_child != -1)
-            && (sh->child_errno[priv->favorite_child] == 0)) {
-
-                gf_log (this->name, GF_LOG_WARNING,
-                        "Picking favorite child %s as authentic source to resolve conflicting metadata of %s",
-                        priv->children[priv->favorite_child]->name,
-                        local->loc.path);
-
-                sh->sources[priv->favorite_child] = 1;
-
-                nsources = afr_sh_source_count (sh->sources,
-                                                priv->child_count);
-        }
-
-        if (nsources == -1) {
-                afr_sh_print_split_brain_log (sh->pending_matrix, this,
-                                              local->loc.path);
-                afr_set_split_brain (this, sh->inode, SPB, DONT_KNOW);
-                afr_sh_metadata_fail (frame, this);
-                goto out;
-        }
-
-        afr_set_split_brain (this, sh->inode, NO_SPB, DONT_KNOW);
-        if (nsources == 0) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "No self-heal needed for %s",
-                        local->loc.path);
-
-                afr_sh_metadata_finish (frame, this);
-                goto out;
-        }
-
-        source = afr_sh_select_source (sh->sources, priv->child_count);
-
-        if (source == -1) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "No active sources found.");
-
-                afr_sh_metadata_finish (frame, this);
-                goto out;
-        }
-
-        sh->source = source;
-
-        /* detect changes not visible through pending flags -- JIC */
+    if (source != -1) {
         for (i = 0; i < priv->child_count; i++) {
-                if (i == source || sh->child_errno[i])
-                        continue;
-
-                if (PERMISSION_DIFFERS (&sh->buf[i], &sh->buf[source]))
-                        sh->sources[i] = 0;
-
-                if (OWNERSHIP_DIFFERS (&sh->buf[i], &sh->buf[source]))
-                        sh->sources[i] = 0;
+            if (i != source && sources[i]) {
+                sources[i] = 0;
+                healed_sinks[i] = 1;
+            }
         }
+    }
 
-        if ((!IA_ISREG (sh->buf[source].ia_type)) &&
-            (!IA_ISDIR (sh->buf[source].ia_type))) {
-                afr_reset_children (sh->fresh_children, priv->child_count);
-                afr_get_fresh_children (sh->success_children, sh->sources,
-                                        sh->fresh_children, priv->child_count);
-                afr_inode_set_read_ctx (this, sh->inode, sh->source,
-                                        sh->fresh_children);
-        }
+    source = __afr_selfheal_metadata_finalize_source(
+        frame, this, inode, sources, sinks, healed_sinks, undid_pending,
+        locked_on, replies);
 
-        if (sh->do_metadata_self_heal && priv->metadata_self_heal)
-                afr_sh_metadata_sync_prepare (frame, this);
-        else
-                afr_sh_metadata_finish (frame, this);
-out:
-        return;
-}
-
-int
-afr_sh_metadata_post_nonblocking_inodelk_cbk (call_frame_t *frame,
-                                              xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG, "Non Blocking metadata "
-                        "inodelks failed for %s.", local->loc.path);
-                gf_log (this->name, GF_LOG_DEBUG, "Metadata self-heal "
-                        "failed for %s.", local->loc.path);
-                afr_sh_metadata_done (frame, this);
-        } else {
-
-                gf_log (this->name, GF_LOG_DEBUG, "Non Blocking metadata "
-                        "inodelks done for %s. Proceeding to FOP",
-                        local->loc.path);
-                afr_sh_common_lookup (frame, this, &local->loc,
-                                      afr_sh_metadata_fix, NULL,
-                                      AFR_LOOKUP_FAIL_CONFLICTS |
-                                      AFR_LOOKUP_FAIL_MISSING_GFIDS,
-                                      NULL);
-        }
+    if (source < 0)
+        return -EIO;
 
-        return 0;
+    return source;
 }
 
 int
-afr_sh_metadata_lock (call_frame_t *frame, xlator_t *this)
+afr_selfheal_metadata(call_frame_t *frame, xlator_t *this, inode_t *inode)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_inodelk_t       *inodelk  = NULL;
-        afr_local_t         *local    = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        int_lock->domain           = this->name;
-        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
-        int_lock->transaction_lk_type = AFR_SELFHEAL_LK;
-        int_lock->selfheal_lk_type    = AFR_METADATA_SELF_HEAL_LK;
-
-        afr_set_lock_number (frame, this);
+    afr_private_t *priv = NULL;
+    int ret = -1;
+    unsigned char *sources = NULL;
+    unsigned char *sinks = NULL;
+    unsigned char *data_lock = NULL;
+    unsigned char *healed_sinks = NULL;
+    unsigned char *undid_pending = NULL;
+    struct afr_reply *locked_replies = NULL;
+    gf_boolean_t did_sh = _gf_true;
+    int source = -1;
+
+    priv = this->private;
+
+    sources = alloca0(priv->child_count);
+    sinks = alloca0(priv->child_count);
+    healed_sinks = alloca0(priv->child_count);
+    undid_pending = alloca0(priv->child_count);
+    data_lock = alloca0(priv->child_count);
+
+    locked_replies = alloca0(sizeof(*locked_replies) * priv->child_count);
+
+    ret = afr_selfheal_inodelk(frame, this, inode, this->name, LLONG_MAX - 1, 0,
+                               data_lock);
+    {
+        if (ret < priv->child_count) {
+            ret = -ENOTCONN;
+            goto unlock;
+        }
 
-        inodelk->flock.l_start = LLONG_MAX - 1;
-        inodelk->flock.l_len   = 0;
-        inodelk->flock.l_type  = F_WRLCK;
-        int_lock->lock_cbk         = afr_sh_metadata_post_nonblocking_inodelk_cbk;
+        ret = __afr_selfheal_metadata_prepare(
+            frame, this, inode, data_lock, sources, sinks, healed_sinks,
+            undid_pending, locked_replies, NULL);
+        if (ret < 0)
+            goto unlock;
 
-        afr_nonblocking_inodelk (frame, this);
+        source = ret;
 
-        return 0;
-}
+        if (AFR_COUNT(healed_sinks, priv->child_count) == 0) {
+            did_sh = _gf_false;
+            goto unlock;
+        }
 
-gf_boolean_t
-afr_can_start_metadata_self_heal (afr_self_heal_t *sh, afr_private_t *priv)
-{
-        if (sh->force_confirm_spb)
-                return _gf_true;
-        if (sh->do_metadata_self_heal && priv->metadata_self_heal)
-                return _gf_true;
-        return _gf_false;
+        ret = __afr_selfheal_metadata_do(frame, this, inode, source,
+                                         healed_sinks, locked_replies);
+        if (ret)
+            goto unlock;
+
+        afr_selfheal_restore_time(frame, this, inode, source, healed_sinks,
+                                  locked_replies);
+
+        ret = afr_selfheal_undo_pending(
+            frame, this, inode, sources, sinks, healed_sinks, undid_pending,
+            AFR_METADATA_TRANSACTION, locked_replies, data_lock);
+    }
+unlock:
+    afr_selfheal_uninodelk(frame, this, inode, this->name, LLONG_MAX - 1, 0,
+                           data_lock);
+
+    if (did_sh)
+        afr_log_selfheal(inode->gfid, this, ret, "metadata", source, sources,
+                         healed_sinks);
+    else
+        ret = 1;
+
+    if (locked_replies)
+        afr_replies_wipe(locked_replies, priv->child_count);
+    return ret;
 }
 
 int
-afr_self_heal_metadata (call_frame_t *frame, xlator_t *this)
+afr_selfheal_metadata_by_stbuf(xlator_t *this, struct iatt *stbuf)
 {
-        afr_local_t   *local = NULL;
-        afr_private_t *priv = this->private;
-        afr_self_heal_t *sh = &local->self_heal;
-
-        local = frame->local;
-        sh = &local->self_heal;
-        sh->sh_type_in_action = AFR_SELF_HEAL_METADATA;
-
-        if (afr_can_start_metadata_self_heal (sh, priv)) {
-                afr_set_self_heal_status (sh, AFR_SELF_HEAL_STARTED);
-                afr_sh_metadata_lock (frame, this);
-        } else {
-                afr_sh_metadata_done (frame, this);
-        }
-
-        return 0;
+    inode_t *inode = NULL;
+    inode_t *link_inode = NULL;
+    call_frame_t *frame = NULL;
+    int ret = 0;
+
+    if (gf_uuid_is_null(stbuf->ia_gfid)) {
+        ret = -EINVAL;
+        goto out;
+    }
+
+    inode = inode_new(this->itable);
+    if (!inode) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    link_inode = inode_link(inode, NULL, NULL, stbuf);
+    if (!link_inode) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    frame = afr_frame_create(this, &ret);
+    if (!frame) {
+        ret = -ret;
+        goto out;
+    }
+
+    ret = afr_selfheal_metadata(frame, this, link_inode);
+out:
+    if (inode)
+        inode_unref(inode);
+    if (link_inode)
+        inode_unref(link_inode);
+    if (frame)
+        AFR_STACK_DESTROY(frame);
+    return ret;
 }
diff --git a/xlators/cluster/afr/src/afr-self-heal-name.c b/xlators/cluster/afr/src/afr-self-heal-name.c
new file mode 100644
index 00000000000..834aac86d48
--- /dev/null
+++ b/xlators/cluster/afr/src/afr-self-heal-name.c
@@ -0,0 +1,616 @@
+/*
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/events.h>
+#include "afr.h"
+#include "afr-self-heal.h"
+#include "afr-messages.h"
+
+int
+__afr_selfheal_assign_gfid(xlator_t *this, inode_t *parent, uuid_t pargfid,
+                           const char *bname, inode_t *inode,
+                           struct afr_reply *replies, void *gfid,
+                           unsigned char *locked_on, int source,
+                           unsigned char *sources, gf_boolean_t is_gfid_absent,
+                           int *gfid_idx)
+{
+    int ret = 0;
+    int up_count = 0;
+    int locked_count = 0;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+
+    gf_uuid_copy(parent->gfid, pargfid);
+
+    if (is_gfid_absent) {
+        /* Ensure all children of AFR are up before performing gfid heal, to
+         * guard against the possibility of gfid split brain. */
+
+        up_count = AFR_COUNT(priv->child_up, priv->child_count);
+        if (up_count != priv->child_count) {
+            ret = -EIO;
+            goto out;
+        }
+
+        locked_count = AFR_COUNT(locked_on, priv->child_count);
+        if (locked_count != priv->child_count) {
+            ret = -EIO;
+            goto out;
+        }
+    }
+
+    ret = afr_lookup_and_heal_gfid(this, parent, bname, inode, replies, source,
+                                   sources, gfid, gfid_idx);
+
+out:
+    return ret;
+}
+
+int
+__afr_selfheal_name_impunge(call_frame_t *frame, xlator_t *this,
+                            inode_t *parent, uuid_t pargfid, const char *bname,
+                            inode_t *inode, struct afr_reply *replies,
+                            int gfid_idx)
+{
+    int i = 0;
+    afr_private_t *priv = NULL;
+    int ret = 0;
+    unsigned char *sources = NULL;
+
+    priv = this->private;
+
+    sources = alloca0(priv->child_count);
+
+    gf_uuid_copy(parent->gfid, pargfid);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret != 0)
+            continue;
+
+        if (gf_uuid_compare(replies[i].poststat.ia_gfid,
+                            replies[gfid_idx].poststat.ia_gfid) == 0) {
+            sources[i] = 1;
+            continue;
+        }
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (sources[i])
+            continue;
+
+        ret |= afr_selfheal_recreate_entry(frame, i, gfid_idx, sources, parent,
+                                           bname, inode, replies);
+    }
+
+    return ret;
+}
+
+int
+__afr_selfheal_name_expunge(xlator_t *this, inode_t *parent, uuid_t pargfid,
+                            const char *bname, inode_t *inode,
+                            struct afr_reply *replies)
+{
+    int i = 0;
+    afr_private_t *priv = NULL;
+    int ret = 0;
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid)
+            continue;
+
+        if (replies[i].op_ret)
+            continue;
+
+        ret |= afr_selfheal_entry_delete(this, parent, bname, inode, i,
+                                         replies);
+    }
+
+    return ret;
+}
+
+static gf_boolean_t
+afr_selfheal_name_need_heal_check(xlator_t *this, struct afr_reply *replies)
+{
+    int i = 0;
+    int first_idx = -1;
+    gf_boolean_t need_heal = _gf_false;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid)
+            continue;
+
+        if ((replies[i].op_ret == -1) && (replies[i].op_errno == ENODATA))
+            need_heal = _gf_true;
+
+        if (first_idx == -1) {
+            first_idx = i;
+            continue;
+        }
+
+        if (replies[i].op_ret != replies[first_idx].op_ret)
+            need_heal = _gf_true;
+
+        if (gf_uuid_compare(replies[i].poststat.ia_gfid,
+                            replies[first_idx].poststat.ia_gfid))
+            need_heal = _gf_true;
+
+        if ((replies[i].op_ret == 0) &&
+            (gf_uuid_is_null(replies[i].poststat.ia_gfid)))
+            need_heal = _gf_true;
+    }
+
+    return need_heal;
+}
+
+static int
+afr_selfheal_name_type_mismatch_check(xlator_t *this, struct afr_reply *replies,
+                                      int source, unsigned char *sources,
+                                      uuid_t pargfid, const char *bname)
+{
+    int i = 0;
+    int type_idx = -1;
+    ia_type_t inode_type = IA_INVAL;
+    ia_type_t inode_type1 = IA_INVAL;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret != 0)
+            continue;
+
+        if (replies[i].poststat.ia_type == IA_INVAL)
+            continue;
+
+        if (inode_type == IA_INVAL) {
+            inode_type = replies[i].poststat.ia_type;
+            type_idx = i;
+            continue;
+        }
+        inode_type1 = replies[i].poststat.ia_type;
+        if (sources[i] || source == -1) {
+            if ((sources[type_idx] || source == -1) &&
+                (inode_type != inode_type1)) {
+                gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_SPLIT_BRAIN,
+                       "Type mismatch for <gfid:%s>/%s: "
+                       "%s on %s and %s on %s",
+                       uuid_utoa(pargfid), bname,
+                       gf_inode_type_to_str(inode_type1),
+                       priv->children[i]->name,
+                       gf_inode_type_to_str(inode_type),
+                       priv->children[type_idx]->name);
+                gf_event(EVENT_AFR_SPLIT_BRAIN,
+                         "client-pid=%d;"
+                         "subvol=%s;type=file;"
+                         "file=<gfid:%s>/%s;count=2;"
+                         "child-%d=%s;type-%d=%s;child-%d=%s;"
+                         "type-%d=%s",
+                         this->ctx->cmd_args.client_pid, this->name,
+                         uuid_utoa(pargfid), bname, i, priv->children[i]->name,
+                         i, gf_inode_type_to_str(inode_type1), type_idx,
+                         priv->children[type_idx]->name, type_idx,
+                         gf_inode_type_to_str(inode_type));
+                return -EIO;
+            }
+            inode_type = replies[i].poststat.ia_type;
+            type_idx = i;
+        }
+    }
+    return 0;
+}
+
+static int
+afr_selfheal_name_gfid_mismatch_check(xlator_t *this, struct afr_reply *replies,
+                                      int source, unsigned char *sources,
+                                      int *gfid_idx, uuid_t pargfid,
+                                      const char *bname, inode_t *inode,
+                                      unsigned char *locked_on, dict_t *xdata)
+{
+    int i = 0;
+    int gfid_idx_iter = -1;
+    int ret = -1;
+    void *gfid = NULL;
+    void *gfid1 = NULL;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret != 0)
+            continue;
+
+        if (gf_uuid_is_null(replies[i].poststat.ia_gfid))
+            continue;
+
+        if (!gfid) {
+            gfid = &replies[i].poststat.ia_gfid;
+            gfid_idx_iter = i;
+            continue;
+        }
+
+        gfid1 = &replies[i].poststat.ia_gfid;
+        if (sources[i] || source == -1) {
+            if ((sources[gfid_idx_iter] || source == -1) &&
+                gf_uuid_compare(gfid, gfid1)) {
+                ret = afr_gfid_split_brain_source(this, replies, inode, pargfid,
+                                                  bname, gfid_idx_iter, i,
+                                                  locked_on, gfid_idx, xdata);
+                if (!ret && *gfid_idx >= 0) {
+                    ret = dict_set_sizen_str_sizen(xdata, "gfid-heal-msg",
+                                                   "GFID split-brain resolved");
+                    if (ret)
+                        gf_msg(this->name, GF_LOG_ERROR, 0,
+                               AFR_MSG_DICT_SET_FAILED,
+                               "Error setting gfid-"
+                               "heal-msg dict");
+                }
+                return ret;
+            }
+            gfid = &replies[i].poststat.ia_gfid;
+            gfid_idx_iter = i;
+        }
+    }
+
+    *gfid_idx = gfid_idx_iter;
+    return 0;
+}
+
+static gf_boolean_t
+afr_selfheal_name_source_empty_check(xlator_t *this, struct afr_reply *replies,
+                                     unsigned char *sources, int source)
+{
+    int i = 0;
+    afr_private_t *priv = NULL;
+    gf_boolean_t source_is_empty = _gf_true;
+
+    priv = this->private;
+
+    if (source == -1) {
+        source_is_empty = _gf_false;
+        goto out;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i])
+            continue;
+
+        if (replies[i].op_ret == -1 && replies[i].op_errno == ENOENT)
+            continue;
+
+        source_is_empty = _gf_false;
+        break;
+    }
+out:
+    return source_is_empty;
+}
+
+int
+__afr_selfheal_name_do(call_frame_t *frame, xlator_t *this, inode_t *parent,
+                       uuid_t pargfid, const char *bname, inode_t *inode,
+                       unsigned char *sources, unsigned char *sinks,
+                       unsigned char *healed_sinks, int source,
+                       unsigned char *locked_on, struct afr_reply *replies,
+                       void *gfid_req, dict_t *xdata)
+{
+    int gfid_idx = -1;
+    int ret = -1;
+    void *gfid = NULL;
+    gf_boolean_t source_is_empty = _gf_true;
+    gf_boolean_t need_heal = _gf_false;
+    gf_boolean_t is_gfid_absent = _gf_false;
+
+    need_heal = afr_selfheal_name_need_heal_check(this, replies);
+    if (!need_heal)
+        return 0;
+
+    source_is_empty = afr_selfheal_name_source_empty_check(this, replies,
+                                                           sources, source);
+    if (source_is_empty) {
+        ret = __afr_selfheal_name_expunge(this, parent, pargfid, bname, inode,
+                                          replies);
+        if (ret == -EIO)
+            ret = -1;
+        return ret;
+    }
+
+    ret = afr_selfheal_name_type_mismatch_check(this, replies, source, sources,
+                                                pargfid, bname);
+    if (ret)
+        return ret;
+
+    ret = afr_selfheal_name_gfid_mismatch_check(this, replies, source, sources,
+                                                &gfid_idx, pargfid, bname,
+                                                inode, locked_on, xdata);
+    if (ret)
+        return ret;
+
+    if (gfid_idx == -1) {
+        if (!gfid_req || gf_uuid_is_null(gfid_req))
+            return -1;
+        gfid = gfid_req;
+    } else {
+        gfid = &replies[gfid_idx].poststat.ia_gfid;
+        if (source == -1)
+            /* Either entry split-brain or dirty xattrs are present on parent.*/
+            source = gfid_idx;
+    }
+
+    is_gfid_absent = (gfid_idx == -1) ? _gf_true : _gf_false;
+    ret = __afr_selfheal_assign_gfid(this, parent, pargfid, bname, inode,
+                                     replies, gfid, locked_on, source, sources,
+                                     is_gfid_absent, &gfid_idx);
+    if (ret || (gfid_idx < 0))
+        return ret;
+
+    ret = __afr_selfheal_name_impunge(frame, this, parent, pargfid, bname,
+                                      inode, replies, gfid_idx);
+    if (ret == -EIO)
+        ret = -1;
+
+    return ret;
+}
+
+int
+__afr_selfheal_name_finalize_source(xlator_t *this, unsigned char *sources,
+                                    unsigned char *healed_sinks,
+                                    unsigned char *locked_on, uint64_t *witness)
+{
+    int i = 0;
+    afr_private_t *priv = NULL;
+    int source = -1;
+    int sources_count = 0;
+
+    priv = this->private;
+
+    sources_count = AFR_COUNT(sources, priv->child_count);
+
+    if ((AFR_CMP(locked_on, healed_sinks, priv->child_count) == 0) ||
+        !sources_count || afr_does_witness_exist(this, witness)) {
+        memset(sources, 0, sizeof(*sources) * priv->child_count);
+        afr_mark_active_sinks(this, sources, locked_on, healed_sinks);
+        return -1;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (sources[i]) {
+            source = i;
+            break;
+        }
+    }
+
+    return source;
+}
+
+int
+__afr_selfheal_name_prepare(call_frame_t *frame, xlator_t *this,
+                            inode_t *parent, uuid_t pargfid,
+                            unsigned char *locked_on, unsigned char *sources,
+                            unsigned char *sinks, unsigned char *healed_sinks,
+                            int *source_p)
+{
+    int ret = -1;
+    int source = -1;
+    afr_private_t *priv = NULL;
+    struct afr_reply *replies = NULL;
+    uint64_t *witness = NULL;
+
+    priv = this->private;
+
+    replies = alloca0(priv->child_count * sizeof(*replies));
+
+    ret = afr_selfheal_unlocked_discover(frame, parent, pargfid, replies);
+    if (ret)
+        goto out;
+
+    witness = alloca0(sizeof(*witness) * priv->child_count);
+    ret = afr_selfheal_find_direction(frame, this, replies,
+                                      AFR_ENTRY_TRANSACTION, locked_on, sources,
+                                      sinks, witness, NULL);
+    if (ret)
+        goto out;
+
+    /* Initialize the healed_sinks[] array optimistically to
+       the intersection of to-be-healed (i.e sinks[]) and
+       the list of servers which are up (i.e locked_on[]).
+
+       As we encounter failures in the healing process, we
+       will unmark the respective servers in the healed_sinks[]
+       array.
+    */
+    AFR_INTERSECT(healed_sinks, sinks, locked_on, priv->child_count);
+
+    source = __afr_selfheal_name_finalize_source(this, sources, healed_sinks,
+                                                 locked_on, witness);
+    if (source < 0) {
+        /* If source is < 0 (typically split-brain), we perform a
+           conservative merge of entries rather than erroring out */
+    }
+    *source_p = source;
+
+out:
+    if (replies)
+        afr_replies_wipe(replies, priv->child_count);
+
+    return ret;
+}
+
+int
+afr_selfheal_name_do(call_frame_t *frame, xlator_t *this, inode_t *parent,
+                     uuid_t pargfid, const char *bname, void *gfid_req,
+                     dict_t *xdata)
+{
+    afr_private_t *priv = NULL;
+    unsigned char *sources = NULL;
+    unsigned char *sinks = NULL;
+    unsigned char *healed_sinks = NULL;
+    unsigned char *locked_on = NULL;
+    int source = -1;
+    struct afr_reply *replies = NULL;
+    int ret = -1;
+    inode_t *inode = NULL;
+    dict_t *xattr = NULL;
+
+    xattr = dict_new();
+    if (!xattr)
+        return -ENOMEM;
+
+    ret = dict_set_int32_sizen(xattr, GF_GFIDLESS_LOOKUP, 1);
+    if (ret) {
+        dict_unref(xattr);
+        return -1;
+    }
+
+    priv = this->private;
+
+    locked_on = alloca0(priv->child_count);
+    sources = alloca0(priv->child_count);
+    sinks = alloca0(priv->child_count);
+    healed_sinks = alloca0(priv->child_count);
+
+    replies = alloca0(priv->child_count * sizeof(*replies));
+
+    ret = afr_selfheal_entrylk(frame, this, parent, this->name, bname,
+                               locked_on);
+    {
+        if (ret < priv->child_count) {
+            ret = -ENOTCONN;
+            goto unlock;
+        }
+
+        ret = __afr_selfheal_name_prepare(frame, this, parent, pargfid,
+                                          locked_on, sources, sinks,
+                                          healed_sinks, &source);
+        if (ret)
+            goto unlock;
+
+        inode = afr_selfheal_unlocked_lookup_on(frame, parent, bname, replies,
+                                                locked_on, xattr);
+        if (!inode) {
+            ret = -ENOMEM;
+            goto unlock;
+        }
+
+        ret = __afr_selfheal_name_do(frame, this, parent, pargfid, bname, inode,
+                                     sources, sinks, healed_sinks, source,
+                                     locked_on, replies, gfid_req, xdata);
+    }
+unlock:
+    afr_selfheal_unentrylk(frame, this, parent, this->name, bname, locked_on,
+                           NULL);
+    if (inode)
+        inode_unref(inode);
+
+    if (replies)
+        afr_replies_wipe(replies, priv->child_count);
+    if (xattr)
+        dict_unref(xattr);
+
+    return ret;
+}
+
+int
+afr_selfheal_name_unlocked_inspect(call_frame_t *frame, xlator_t *this,
+                                   inode_t *parent, uuid_t pargfid,
+                                   const char *bname, gf_boolean_t *need_heal)
+{
+    afr_private_t *priv = NULL;
+    int i = 0;
+    struct afr_reply *replies = NULL;
+    inode_t *inode = NULL;
+    int first_idx = -1;
+    afr_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    replies = alloca0(sizeof(*replies) * priv->child_count);
+
+    inode = afr_selfheal_unlocked_lookup_on(frame, parent, bname, replies,
+                                            local->child_up, NULL);
+    if (!inode)
+        return -ENOMEM;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid)
+            continue;
+
+        if ((replies[i].op_ret == -1) && (replies[i].op_errno == ENODATA)) {
+            *need_heal = _gf_true;
+            break;
+        }
+
+        if (first_idx == -1) {
+            first_idx = i;
+            continue;
+        }
+
+        if (replies[i].op_ret != replies[first_idx].op_ret) {
+            *need_heal = _gf_true;
+            break;
+        }
+
+        if (gf_uuid_compare(replies[i].poststat.ia_gfid,
+                            replies[first_idx].poststat.ia_gfid)) {
+            *need_heal = _gf_true;
+            break;
+        }
+    }
+
+    if (inode)
+        inode_unref(inode);
+    if (replies)
+        afr_replies_wipe(replies, priv->child_count);
+    return 0;
+}
+
+int
+afr_selfheal_name(xlator_t *this, uuid_t pargfid, const char *bname,
+                  void *gfid_req, dict_t *xdata)
+{
+    inode_t *parent = NULL;
+    call_frame_t *frame = NULL;
+    int ret = -1;
+    gf_boolean_t need_heal = _gf_false;
+
+    parent = afr_inode_find(this, pargfid);
+    if (!parent)
+        goto out;
+
+    frame = afr_frame_create(this, NULL);
+    if (!frame)
+        goto out;
+
+    ret = afr_selfheal_name_unlocked_inspect(frame, this, parent, pargfid,
+                                             bname, &need_heal);
+    if (ret)
+        goto out;
+
+    if (need_heal) {
+        ret = afr_selfheal_name_do(frame, this, parent, pargfid, bname,
+                                   gfid_req, xdata);
+        if (ret)
+            goto out;
+    }
+
+    ret = 0;
+out:
+    if (parent)
+        inode_unref(parent);
+    if (frame)
+        AFR_STACK_DESTROY(frame);
+
+    return ret;
+}
diff --git a/xlators/cluster/afr/src/afr-self-heal.h b/xlators/cluster/afr/src/afr-self-heal.h
index 7c9bc81119c..48e6dbcfb18 100644
--- a/xlators/cluster/afr/src/afr-self-heal.h
+++ b/xlators/cluster/afr/src/afr-self-heal.h
@@ -1,5 +1,5 @@
 /*
-  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
   This file is licensed to you under your choice of the GNU Lesser
@@ -8,36 +8,370 @@
   cases as published by the Free Software Foundation.
 */
 
-#ifndef __AFR_SELF_HEAL_H__
-#define __AFR_SELF_HEAL_H__
+#ifndef _AFR_SELFHEAL_H
+#define _AFR_SELFHEAL_H
 
-#include <sys/stat.h>
+/* Perform fop on all UP subvolumes and wait for all callbacks to return */
 
-#define FILETYPE_DIFFERS(buf1,buf2) ((buf1)->ia_type != (buf2)->ia_type)
-#define PERMISSION_DIFFERS(buf1,buf2) (st_mode_from_ia ((buf1)->ia_prot, (buf1)->ia_type) != st_mode_from_ia ((buf2)->ia_prot, (buf2)->ia_type))
-#define OWNERSHIP_DIFFERS(buf1,buf2) (((buf1)->ia_uid != (buf2)->ia_uid) || ((buf1)->ia_gid != (buf2)->ia_gid))
-#define SIZE_DIFFERS(buf1,buf2) ((buf1)->ia_size != (buf2)->ia_size)
+#define AFR_ONALL(frame, rfn, fop, args...)                                    \
+    do {                                                                       \
+        afr_local_t *__local = frame->local;                                   \
+        afr_private_t *__priv = frame->this->private;                          \
+        int __i = 0, __count = 0;                                              \
+        unsigned char *__child_up = alloca(__priv->child_count);               \
+                                                                               \
+        memcpy(__child_up, __priv->child_up,                                   \
+               sizeof(*__child_up) * __priv->child_count);                     \
+        __count = AFR_COUNT(__child_up, __priv->child_count);                  \
+                                                                               \
+        __local->barrier.waitfor = __count;                                    \
+        afr_local_replies_wipe(__local, __priv);                               \
+                                                                               \
+        for (__i = 0; __i < __priv->child_count; __i++) {                      \
+            if (!__child_up[__i])                                              \
+                continue;                                                      \
+            STACK_WIND_COOKIE(frame, rfn, (void *)(long)__i,                   \
+                              __priv->children[__i],                           \
+                              __priv->children[__i]->fops->fop, args);         \
+        }                                                                      \
+        syncbarrier_wait(&__local->barrier, __count);                          \
+    } while (0)
 
-#define SIZE_GREATER(buf1,buf2) ((buf1)->ia_size > (buf2)->ia_size)
+/* Perform fop on all subvolumes represented by list[] array and wait
+   for all callbacks to return */
+
+#define AFR_ONLIST(list, frame, rfn, fop, args...)                             \
+    do {                                                                       \
+        afr_local_t *__local = frame->local;                                   \
+        afr_private_t *__priv = frame->this->private;                          \
+        int __i = 0;                                                           \
+        int __count = 0;                                                       \
+        unsigned char *__list = alloca(__priv->child_count);                   \
+                                                                               \
+        memcpy(__list, list, sizeof(*__list) * __priv->child_count);           \
+        __count = AFR_COUNT(__list, __priv->child_count);                      \
+        __local->barrier.waitfor = __count;                                    \
+        afr_local_replies_wipe(__local, __priv);                               \
+                                                                               \
+        for (__i = 0; __i < __priv->child_count; __i++) {                      \
+            if (!__list[__i])                                                  \
+                continue;                                                      \
+            STACK_WIND_COOKIE(frame, rfn, (void *)(long)__i,                   \
+                              __priv->children[__i],                           \
+                              __priv->children[__i]->fops->fop, args);         \
+        }                                                                      \
+        syncbarrier_wait(&__local->barrier, __count);                          \
+    } while (0)
+
+#define AFR_SEQ(frame, rfn, fop, args...)                                      \
+    do {                                                                       \
+        afr_local_t *__local = frame->local;                                   \
+        afr_private_t *__priv = frame->this->private;                          \
+        int __i = 0;                                                           \
+                                                                               \
+        afr_local_replies_wipe(__local, __priv);                               \
+                                                                               \
+        for (__i = 0; __i < __priv->child_count; __i++) {                      \
+            if (!__priv->child_up[__i])                                        \
+                continue;                                                      \
+            STACK_WIND_COOKIE(frame, rfn, (void *)(long)__i,                   \
+                              __priv->children[__i],                           \
+                              __priv->children[__i]->fops->fop, args);         \
+            syncbarrier_wait(&__local->barrier, 1);                            \
+        }                                                                      \
+    } while (0)
+
+#define ALLOC_MATRIX(n, type)                                                  \
+    ({                                                                         \
+        int __i;                                                               \
+        type **__ptr = alloca(n * sizeof(type *));                             \
+                                                                               \
+        for (__i = 0; __i < n; __i++)                                          \
+            __ptr[__i] = alloca0(n * sizeof(type));                            \
+        __ptr;                                                                 \
+    })
+
+#define IA_EQUAL(f, s, field)                                                  \
+    (memcmp(&(f.ia_##field), &(s.ia_##field), sizeof(s.ia_##field)) == 0)
+
+#define SBRAIN_HEAL_NO_GO_MSG                                                  \
+    "Failed to obtain replies from all bricks of "                             \
+    "the replica (are they up?). Cannot resolve split-brain."
+#define SFILE_NOT_IN_SPLIT_BRAIN "File not in split-brain"
+#define SNO_BIGGER_FILE "No bigger file"
+#define SNO_DIFF_IN_MTIME "No difference in mtime"
+#define SUSE_SOURCE_BRICK_TO_HEAL                                              \
+    "Use source-brick option to heal metadata"                                 \
+    " split-brain"
+#define SINVALID_BRICK_NAME "Invalid brick name"
+#define SBRICK_IS_NOT_UP "Brick is not up"
+#define SBRICK_NOT_CONNECTED "Brick is not connected"
+#define SLESS_THAN2_BRICKS_in_REP "< 2 bricks in replica are up"
+#define SBRICK_IS_REMOTE "Brick is remote"
+#define SSTARTED_SELF_HEAL "Started self-heal"
+#define SOP_NOT_SUPPORTED "Operation Not Supported"
+#define SFILE_NOT_UNDER_DATA                                                   \
+    "The file is not under data or metadata "                                  \
+    "split-brain"
+#define SFILE_NOT_IN_SPLIT_BRAIN "File not in split-brain"
+#define SALL_BRICKS_UP_TO_RESOLVE                                              \
+    "All the bricks should be up to resolve the"                               \
+    " gfid split brain"
+#define SERROR_GETTING_SRC_BRICK "Error getting the source brick"
+int
+afr_selfheal(xlator_t *this, uuid_t gfid);
+
+gf_boolean_t
+afr_throttled_selfheal(call_frame_t *frame, xlator_t *this);
+
+int
+afr_selfheal_name(xlator_t *this, uuid_t gfid, const char *name, void *gfid_req,
+                  dict_t *xdata);
+
+int
+afr_selfheal_data(call_frame_t *frame, xlator_t *this, fd_t *fd);
+
+int
+afr_selfheal_metadata(call_frame_t *frame, xlator_t *this, inode_t *inode);
+
+int
+afr_selfheal_entry(call_frame_t *frame, xlator_t *this, inode_t *inode);
+
+int
+afr_lookup_and_heal_gfid(xlator_t *this, inode_t *parent, const char *name,
+                         inode_t *inode, struct afr_reply *replies, int source,
+                         unsigned char *sources, void *gfid, int *gfid_idx);
+
+int
+afr_selfheal_inodelk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                     char *dom, off_t off, size_t size,
+                     unsigned char *locked_on);
+
+int
+afr_selfheal_tryinodelk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                        char *dom, off_t off, size_t size,
+                        unsigned char *locked_on);
+
+int
+afr_selfheal_tie_breaker_inodelk(call_frame_t *frame, xlator_t *this,
+                                 inode_t *inode, char *dom, off_t off,
+                                 size_t size, unsigned char *locked_on);
+
+int
+afr_selfheal_uninodelk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                       char *dom, off_t off, size_t size,
+                       const unsigned char *locked_on);
+
+int
+afr_selfheal_entrylk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                     char *dom, const char *name, unsigned char *locked_on);
+
+int
+afr_selfheal_tryentrylk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                        char *dom, const char *name, unsigned char *locked_on);
+
+int
+afr_selfheal_tie_breaker_entrylk(call_frame_t *frame, xlator_t *this,
+                                 inode_t *inode, char *dom, const char *name,
+                                 unsigned char *locked_on);
+
+int
+afr_selfheal_unentrylk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                       char *dom, const char *name, unsigned char *locked_on,
+                       dict_t *xdata);
+
+int
+afr_selfheal_unlocked_discover(call_frame_t *frame, inode_t *inode, uuid_t gfid,
+                               struct afr_reply *replies);
+
+int
+afr_selfheal_unlocked_discover_on(call_frame_t *frame, inode_t *inode,
+                                  uuid_t gfid, struct afr_reply *replies,
+                                  unsigned char *discover_on, dict_t *dict);
+inode_t *
+afr_selfheal_unlocked_lookup_on(call_frame_t *frame, inode_t *parent,
+                                const char *name, struct afr_reply *replies,
+                                unsigned char *lookup_on, dict_t *xattr);
+
+int
+afr_selfheal_find_direction(call_frame_t *frame, xlator_t *this,
+                            struct afr_reply *replies,
+                            afr_transaction_type type, unsigned char *locked_on,
+                            unsigned char *sources, unsigned char *sinks,
+                            uint64_t *witness, unsigned char *flag);
+int
+afr_selfheal_fill_matrix(xlator_t *this, int **matrix, int subvol, int idx,
+                         dict_t *xdata);
+
+int
+afr_selfheal_extract_xattr(xlator_t *this, struct afr_reply *replies,
+                           afr_transaction_type type, int *dirty, int **matrix);
+
+int
+afr_sh_generic_fop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno, struct iatt *pre,
+                       struct iatt *post, dict_t *xdata);
+
+int
+afr_selfheal_restore_time(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                          int source, unsigned char *healed_sinks,
+                          struct afr_reply *replies);
+int
+afr_selfheal_undo_pending(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                          unsigned char *sources, unsigned char *sinks,
+                          unsigned char *healed_sinks,
+                          unsigned char *undid_pending,
+                          afr_transaction_type type, struct afr_reply *replies,
+                          unsigned char *locked_on);
+
+int
+afr_selfheal_recreate_entry(call_frame_t *frame, int dst, int source,
+                            unsigned char *sources, inode_t *dir,
+                            const char *name, inode_t *inode,
+                            struct afr_reply *replies);
+
+int
+afr_selfheal_post_op(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                     int subvol, dict_t *xattr, dict_t *xdata);
+
+call_frame_t *
+afr_frame_create(xlator_t *this, int32_t *op_errno);
+
+inode_t *
+afr_inode_find(xlator_t *this, uuid_t gfid);
+
+int
+afr_selfheal_discover_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int op_ret, int op_errno, inode_t *inode,
+                          struct iatt *buf, dict_t *xdata, struct iatt *parbuf);
+void
+afr_reply_copy(struct afr_reply *dst, struct afr_reply *src);
+
+void
+afr_replies_copy(struct afr_reply *dst, struct afr_reply *src, int count);
 
 int
-afr_self_heal_entry (call_frame_t *frame, xlator_t *this);
+afr_selfheal_newentry_mark(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                           int source, struct afr_reply *replies,
+                           unsigned char *sources, unsigned char *newentry);
+
+unsigned int
+afr_success_count(struct afr_reply *replies, unsigned int count);
+
+void
+afr_log_selfheal(uuid_t gfid, xlator_t *this, int ret, char *type, int source,
+                 unsigned char *sources, unsigned char *healed_sinks);
+
+void
+afr_mark_largest_file_as_source(xlator_t *this, unsigned char *sources,
+                                struct afr_reply *replies);
+void
+afr_mark_active_sinks(xlator_t *this, unsigned char *sources,
+                      unsigned char *locked_on, unsigned char *sinks);
+
+gf_boolean_t
+afr_dict_contains_heal_op(call_frame_t *frame);
 
+gf_boolean_t
+afr_can_decide_split_brain_source_sinks(struct afr_reply *replies,
+                                        int child_count);
 int
-afr_self_heal_data (call_frame_t *frame, xlator_t *this);
+afr_mark_split_brain_source_sinks(
+    call_frame_t *frame, xlator_t *this, inode_t *inode, unsigned char *sources,
+    unsigned char *sinks, unsigned char *healed_sinks, unsigned char *locked_on,
+    struct afr_reply *replies, afr_transaction_type type);
 
 int
-afr_self_heal_metadata (call_frame_t *frame, xlator_t *this);
+afr_sh_get_fav_by_policy(xlator_t *this, struct afr_reply *replies,
+                         inode_t *inode, char **policy_str);
 
 int
-afr_self_heal_get_source (xlator_t *this, afr_local_t *local, dict_t **xattr);
+_afr_fav_child_reset_sink_xattrs(call_frame_t *frame, xlator_t *this,
+                                 inode_t *inode, int source,
+                                 unsigned char *healed_sinks,
+                                 unsigned char *undid_pending,
+                                 afr_transaction_type type,
+                                 unsigned char *locked_on,
+                                 struct afr_reply *replies);
 
 int
-afr_self_heal (call_frame_t *frame, xlator_t *this, inode_t *inode);
+afr_get_child_index_from_name(xlator_t *this, char *name);
 
+gf_boolean_t
+afr_does_witness_exist(xlator_t *this, uint64_t *witness);
+
+int
+__afr_selfheal_data_prepare(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                            unsigned char *locked_on, unsigned char *sources,
+                            unsigned char *sinks, unsigned char *healed_sinks,
+                            unsigned char *undid_pending,
+                            struct afr_reply *replies, unsigned char *flag);
+
+int
+__afr_selfheal_metadata_prepare(call_frame_t *frame, xlator_t *this,
+                                inode_t *inode, unsigned char *locked_on,
+                                unsigned char *sources, unsigned char *sinks,
+                                unsigned char *healed_sinks,
+                                unsigned char *undid_pending,
+                                struct afr_reply *replies, unsigned char *flag);
+int
+__afr_selfheal_entry_prepare(call_frame_t *frame, xlator_t *this,
+                             inode_t *inode, unsigned char *locked_on,
+                             unsigned char *sources, unsigned char *sinks,
+                             unsigned char *healed_sinks,
+                             struct afr_reply *replies, int *source_p,
+                             unsigned char *flag);
+
+int
+afr_selfheal_unlocked_inspect(call_frame_t *frame, xlator_t *this, uuid_t gfid,
+                              inode_t **link_inode, gf_boolean_t *data_selfheal,
+                              gf_boolean_t *metadata_selfheal,
+                              gf_boolean_t *entry_selfheal,
+                              struct afr_reply *replies);
+
+int
+afr_selfheal_do(call_frame_t *frame, xlator_t *this, uuid_t gfid);
+
+int
+afr_selfheal_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, dict_t *xdata);
+
+int
+afr_locked_fill(call_frame_t *frame, xlator_t *this, unsigned char *locked_on);
+int
+afr_choose_source_by_policy(afr_private_t *priv, unsigned char *sources,
+                            afr_transaction_type type);
+
+int
+afr_selfheal_metadata_by_stbuf(xlator_t *this, struct iatt *stbuf);
+
+int
+afr_sh_fav_by_size(xlator_t *this, struct afr_reply *replies, inode_t *inode);
+int
+afr_sh_fav_by_mtime(xlator_t *this, struct afr_reply *replies, inode_t *inode);
+int
+afr_sh_fav_by_ctime(xlator_t *this, struct afr_reply *replies, inode_t *inode);
+
+int
+afr_gfid_split_brain_source(xlator_t *this, struct afr_reply *replies,
+                            inode_t *inode, uuid_t pargfid, const char *bname,
+                            int src_idx, int child_idx,
+                            unsigned char *locked_on, int *src, dict_t *xdata);
+int
+afr_mark_source_sinks_if_file_empty(xlator_t *this, unsigned char *sources,
+                                    unsigned char *sinks,
+                                    unsigned char *healed_sinks,
+                                    unsigned char *locked_on,
+                                    struct afr_reply *replies,
+                                    afr_transaction_type type);
+
+gf_boolean_t
+afr_is_file_empty_on_all_children(afr_private_t *priv,
+                                  struct afr_reply *replies);
+
+int
+afr_selfheal_entry_delete(xlator_t *this, inode_t *dir, const char *name,
+                          inode_t *inode, int child, struct afr_reply *replies);
 int
-afr_lookup_select_read_child_by_txn_type (xlator_t *this, afr_local_t *local,
-                                          dict_t **xattr,
-                                          afr_transaction_type txn_type,
-                                          uuid_t gfid);
-#endif /* __AFR_SELF_HEAL_H__ */
+afr_anon_inode_create(xlator_t *this, int child, inode_t **linked_inode);
+#endif /* !_AFR_SELFHEAL_H */
diff --git a/xlators/cluster/afr/src/afr-self-heald.c b/xlators/cluster/afr/src/afr-self-heald.c
index f33b04eed06..109fd4b7421 100644
--- a/xlators/cluster/afr/src/afr-self-heald.c
+++ b/xlators/cluster/afr/src/afr-self-heald.c
@@ -1,5 +1,5 @@
 /*
-  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
   This file is licensed to you under your choice of the GNU Lesser
@@ -8,1266 +8,1709 @@
   cases as published by the Free Software Foundation.
 */
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
 #include "afr.h"
-#include "syncop.h"
+#include "afr-self-heal.h"
 #include "afr-self-heald.h"
-#include "afr-self-heal-common.h"
 #include "protocol-common.h"
-#include "event-history.h"
-
-typedef enum {
-        STOP_CRAWL_ON_SINGLE_SUBVOL = 1
-} afr_crawl_flags_t;
-
-typedef enum {
-        HEAL = 1,
-        INFO
-} shd_crawl_op;
-
-typedef struct shd_dump {
-        dict_t   *dict;
-        xlator_t *this;
-        int      child;
-} shd_dump_t;
-
-typedef struct shd_event_ {
-        int     child;
-        char    *path;
-} shd_event_t;
+#include <glusterfs/syncop-utils.h>
+#include "afr-messages.h"
+#include <glusterfs/byte-order.h>
+
+#define AFR_EH_SPLIT_BRAIN_LIMIT 1024
+#define AFR_STATISTICS_HISTORY_SIZE 50
+
+#define ASSERT_LOCAL(this, healer)                                             \
+    if (!afr_shd_is_subvol_local(this, healer->subvol)) {                      \
+        healer->local = _gf_false;                                             \
+        if (safe_break(healer)) {                                              \
+            break;                                                             \
+        } else {                                                               \
+            continue;                                                          \
+        }                                                                      \
+    } else {                                                                   \
+        healer->local = _gf_true;                                              \
+    }
+
+#define NTH_INDEX_HEALER(this, n)                                              \
+    &((((afr_private_t *)this->private))->shd.index_healers[n])
+#define NTH_FULL_HEALER(this, n)                                               \
+    &((((afr_private_t *)this->private))->shd.full_healers[n])
+
+char *
+afr_subvol_name(xlator_t *this, int subvol)
+{
+    afr_private_t *priv = NULL;
 
-typedef struct shd_pos_ {
-        int     child;
-        xlator_t *this;
-        afr_child_pos_t pos;
-} shd_pos_t;
+    priv = this->private;
+    if (subvol < 0 || subvol > priv->child_count)
+        return NULL;
 
-typedef int
-(*afr_crawl_done_cbk_t)  (int ret, call_frame_t *sync_frame, void *crawl_data);
+    return priv->children[subvol]->name;
+}
 
 void
-afr_start_crawl (xlator_t *this, int idx, afr_crawl_type_t crawl,
-                 process_entry_cbk_t process_entry, void *op_data,
-                 gf_boolean_t exclusive, int crawl_flags,
-                 afr_crawl_done_cbk_t crawl_done);
-
-static int
-_crawl_directory (fd_t *fd, loc_t *loc, afr_crawl_data_t *crawl_data);
-
-/* For calling straight through (e.g. already in a synctask). */
-int
-afr_find_child_position (xlator_t *this, int child, afr_child_pos_t *pos);
-
-/* For deferring through a new synctask. */
-int
-afr_syncop_find_child_position (void *data);
-
-static int
-_loc_assign_gfid_path (loc_t *loc)
+afr_destroy_crawl_event_data(void *data)
 {
-        int  ret = -1;
-        char gfid_path[64] = {0};
-
-        if (loc->inode && !uuid_is_null (loc->inode->gfid)) {
-                ret = inode_path (loc->inode, NULL, (char**)&loc->path);
-        } else if (!uuid_is_null (loc->gfid)) {
-                snprintf (gfid_path, sizeof (gfid_path), "<gfid:%s>",
-                          uuid_utoa (loc->gfid));
-                loc->path = gf_strdup (gfid_path);
-                if (loc->path)
-                        ret = 0;
-        }
-        return ret;
+    return;
 }
 
 void
-shd_cleanup_event (void *event)
+afr_destroy_shd_event_data(void *data)
 {
-        shd_event_t *shd_event = event;
+    shd_event_t *shd_event = data;
 
-        if (!shd_event)
-                goto out;
-        GF_FREE (shd_event->path);
-        GF_FREE (shd_event);
-out:
+    if (!shd_event)
         return;
+    GF_FREE(shd_event->path);
+
+    return;
 }
 
-int
-afr_get_local_child (afr_self_heald_t *shd, unsigned int child_count)
+gf_boolean_t
+afr_shd_is_subvol_local(xlator_t *this, int subvol)
 {
-        int i = 0;
-        int ret = -1;
-        for (i = 0; i < child_count; i++) {
-                if (shd->pos[i] == AFR_POS_LOCAL) {
-                        ret = i;
-                        break;
-                }
-        }
-        return ret;
+    afr_private_t *priv = NULL;
+    gf_boolean_t is_local = _gf_false;
+    loc_t loc = {
+        0,
+    };
+
+    loc.inode = this->itable->root;
+    gf_uuid_copy(loc.gfid, loc.inode->gfid);
+    priv = this->private;
+    syncop_is_subvol_local(priv->children[subvol], &loc, &is_local);
+    return is_local;
 }
 
-static int
-_build_index_loc (xlator_t *this, loc_t *loc, char *name, loc_t *parent)
+int
+__afr_shd_healer_wait(struct subvol_healer *healer)
 {
-        int             ret = 0;
-
-        uuid_copy (loc->pargfid, parent->inode->gfid);
-        loc->path = "";
-        loc->name = name;
-        loc->parent = inode_ref (parent->inode);
-        if (!loc->parent) {
-                loc->path = NULL;
-                loc_wipe (loc);
-                ret = -1;
-        }
-        return ret;
+    afr_private_t *priv = NULL;
+    struct timespec wait_till = {
+        0,
+    };
+    int ret = 0;
+
+    priv = healer->this->private;
+
+disabled_loop:
+    wait_till.tv_sec = gf_time() + priv->shd.timeout;
+
+    while (!healer->rerun) {
+        ret = pthread_cond_timedwait(&healer->cond, &healer->mutex, &wait_till);
+        if (ret == ETIMEDOUT)
+            break;
+    }
+
+    ret = healer->rerun;
+    healer->rerun = 0;
+
+    if (!priv->shd.enabled)
+        goto disabled_loop;
+
+    return ret;
 }
 
 int
-_add_path_to_dict (xlator_t *this, dict_t *output, int child, char *path,
-                   struct timeval *tv, gf_boolean_t dyn)
+afr_shd_healer_wait(struct subvol_healer *healer)
 {
-        //subkey not used for now
-        int             ret = -1;
-        uint64_t        count = 0;
-        char            key[256] = {0};
-        int             xl_id = 0;
+    int ret = 0;
 
-        ret = dict_get_int32 (output, this->name, &xl_id);
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "xl does not have id");
-                goto out;
-        }
+    pthread_mutex_lock(&healer->mutex);
+    {
+        ret = __afr_shd_healer_wait(healer);
+    }
+    pthread_mutex_unlock(&healer->mutex);
 
-        snprintf (key, sizeof (key), "%d-%d-count", xl_id, child);
-        ret = dict_get_uint64 (output, key, &count);
+    return ret;
+}
 
-        snprintf (key, sizeof (key), "%d-%d-%"PRIu64, xl_id, child, count);
-        if (dyn)
-                ret = dict_set_dynstr (output, key, path);
-        else
-                ret = dict_set_str (output, key, path);
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "%s: Could not add to output",
-                        path);
-                goto out;
-        }
+gf_boolean_t
+safe_break(struct subvol_healer *healer)
+{
+    gf_boolean_t ret = _gf_false;
 
-        if (!tv)
-                goto inc_count;
-        snprintf (key, sizeof (key), "%d-%d-%"PRIu64"-time", xl_id,
-                  child, count);
-        ret = dict_set_uint32 (output, key, tv->tv_sec);
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "%s: Could not set time",
-                        path);
-                goto out;
-        }
+    pthread_mutex_lock(&healer->mutex);
+    {
+        if (healer->rerun)
+            goto unlock;
 
-inc_count:
-        snprintf (key, sizeof (key), "%d-%d-count", xl_id, child);
-        ret = dict_set_uint64 (output, key, count + 1);
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "Could not increment count");
-                goto out;
-        }
-        ret = 0;
-out:
-        return ret;
+        healer->running = _gf_false;
+        ret = _gf_true;
+    }
+unlock:
+    pthread_mutex_unlock(&healer->mutex);
+
+    return ret;
 }
 
-int
-_get_path_from_gfid_loc (xlator_t *this, xlator_t *readdir_xl, loc_t *child,
-                         char **fpath, gf_boolean_t *missing)
+inode_t *
+afr_shd_inode_find(xlator_t *this, xlator_t *subvol, uuid_t gfid)
 {
-        dict_t          *xattr = NULL;
-        char            *path = NULL;
-        int             ret = -1;
-
-        ret = syncop_getxattr (readdir_xl, child, &xattr, GFID_TO_PATH_KEY);
-        if (ret < 0) {
-                if ((errno == ENOENT) && missing)
-                        *missing = _gf_true;
-                goto out;
-        }
-        ret = dict_get_str (xattr, GFID_TO_PATH_KEY, &path);
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "Failed to get path for "
-                        "gfid %s", uuid_utoa (child->gfid));
-                goto out;
-        }
-        path = gf_strdup (path);
-        if (!path) {
-                ret = -1;
-                goto out;
-        }
-        ret = 0;
+    int ret = 0;
+    uint64_t val = IA_INVAL;
+    dict_t *xdata = NULL;
+    dict_t *rsp_dict = NULL;
+    inode_t *inode = NULL;
+
+    xdata = dict_new();
+    if (!xdata)
+        goto out;
+
+    ret = dict_set_int8(xdata, GF_INDEX_IA_TYPE_GET_REQ, 1);
+    if (ret)
+        goto out;
+
+    ret = syncop_inode_find(this, subvol, gfid, &inode, xdata, &rsp_dict);
+    if (ret < 0)
+        goto out;
+
+    if (rsp_dict) {
+        ret = dict_get_uint64(rsp_dict, GF_INDEX_IA_TYPE_GET_RSP, &val);
+        if (ret)
+            goto out;
+    }
+    ret = inode_ctx_set2(inode, subvol, 0, &val);
 out:
-        if (!ret)
-                *fpath = path;
-        if (xattr)
-                dict_unref (xattr);
-        return ret;
+    if (ret && inode) {
+        inode_unref(inode);
+        inode = NULL;
+    }
+    if (xdata)
+        dict_unref(xdata);
+    if (rsp_dict)
+        dict_unref(rsp_dict);
+    return inode;
 }
 
-int
-_add_event_to_dict (circular_buffer_t *cb, void *data)
+inode_t *
+afr_shd_index_inode(xlator_t *this, xlator_t *subvol, char *vgfid)
 {
-        int               ret = 0;
-        shd_dump_t        *dump_data = NULL;
-        shd_event_t       *shd_event = NULL;
+    loc_t rootloc = {
+        0,
+    };
+    inode_t *inode = NULL;
+    int ret = 0;
+    dict_t *xattr = NULL;
+    void *index_gfid = NULL;
+
+    rootloc.inode = inode_ref(this->itable->root);
+    gf_uuid_copy(rootloc.gfid, rootloc.inode->gfid);
+
+    ret = syncop_getxattr(subvol, &rootloc, &xattr, vgfid, NULL, NULL);
+    if (ret || !xattr) {
+        errno = -ret;
+        goto out;
+    }
+
+    ret = dict_get_ptr(xattr, vgfid, &index_gfid);
+    if (ret)
+        goto out;
+
+    gf_msg_debug(this->name, 0, "%s dir gfid for %s: %s", vgfid, subvol->name,
+                 uuid_utoa(index_gfid));
+
+    inode = afr_shd_inode_find(this, subvol, index_gfid);
 
-        dump_data = data;
-        shd_event = cb->data;
-        if (shd_event->child != dump_data->child)
-                goto out;
-        ret = _add_path_to_dict (dump_data->this, dump_data->dict,
-                                 dump_data->child, shd_event->path, &cb->tv,
-                                 _gf_false);
 out:
-        return ret;
+    loc_wipe(&rootloc);
+
+    if (xattr)
+        dict_unref(xattr);
+
+    return inode;
 }
 
 int
-_add_eh_to_dict (xlator_t *this, eh_t *eh, dict_t *dict, int child)
+afr_shd_entry_purge(xlator_t *subvol, inode_t *inode, char *name,
+                    ia_type_t type)
 {
-        shd_dump_t dump_data = {0};
+    int ret = 0;
+    loc_t loc = {
+        0,
+    };
 
-        dump_data.this = this;
-        dump_data.dict = dict;
-        dump_data.child = child;
-        eh_dump (eh, &dump_data, _add_event_to_dict);
-        return 0;
+    loc.parent = inode_ref(inode);
+    loc.name = name;
+
+    if (IA_ISDIR(type))
+        ret = syncop_rmdir(subvol, &loc, 1, NULL, NULL);
+    else
+        ret = syncop_unlink(subvol, &loc, NULL, NULL);
+
+    loc_wipe(&loc);
+    return ret;
 }
 
 void
-_remove_stale_index (xlator_t *this, xlator_t *readdir_xl,
-                     loc_t *parent, char *fname)
+afr_shd_zero_xattrop(xlator_t *this, uuid_t gfid)
 {
-        int              ret = 0;
-        loc_t            index_loc = {0};
-
-        ret = _build_index_loc (this, &index_loc, fname, parent);
+    call_frame_t *frame = NULL;
+    inode_t *inode = NULL;
+    afr_private_t *priv = NULL;
+    dict_t *xattr = NULL;
+    int ret = 0;
+    int i = 0;
+    int raw[AFR_NUM_CHANGE_LOGS] = {0};
+
+    priv = this->private;
+    frame = afr_frame_create(this, NULL);
+    if (!frame)
+        goto out;
+    inode = afr_inode_find(this, gfid);
+    if (!inode)
+        goto out;
+    xattr = dict_new();
+    if (!xattr)
+        goto out;
+    ret = dict_set_static_bin(xattr, AFR_DIRTY, raw,
+                              sizeof(int) * AFR_NUM_CHANGE_LOGS);
+    if (ret)
+        goto out;
+    for (i = 0; i < priv->child_count; i++) {
+        ret = dict_set_static_bin(xattr, priv->pending_key[i], raw,
+                                  sizeof(int) * AFR_NUM_CHANGE_LOGS);
         if (ret)
-                goto out;
-        gf_log (this->name, GF_LOG_DEBUG, "Removing stale index "
-                "for %s on %s", index_loc.name, readdir_xl->name);
-        ret = syncop_unlink (readdir_xl, &index_loc);
-        if(ret && (errno != ENOENT)) {
-                gf_log(this->name, GF_LOG_ERROR, "%s: Failed to remove index "
-                       "on %s - %s",index_loc.name, readdir_xl->name,
-                       strerror (errno));
-        }
-        index_loc.path = NULL;
-        loc_wipe (&index_loc);
+            goto out;
+    }
+
+    /*Send xattrop to all bricks. Doing a lookup to see if bricks are up or
+     * has valid repies for this gfid seems a bit of an overkill.*/
+    for (i = 0; i < priv->child_count; i++)
+        afr_selfheal_post_op(frame, this, inode, i, xattr, NULL);
+
 out:
-        return;
+    if (frame)
+        AFR_STACK_DESTROY(frame);
+    if (inode)
+        inode_unref(inode);
+    if (xattr)
+        dict_unref(xattr);
+    return;
 }
 
 int
-_add_summary_to_dict (xlator_t *this, afr_crawl_data_t *crawl_data,
-                      gf_dirent_t *entry,
-                      loc_t *childloc, loc_t *parentloc, struct iatt *iattr)
+afr_shd_selfheal_name(struct subvol_healer *healer, int child, uuid_t parent,
+                      const char *bname)
 {
-        dict_t          *output = NULL;
-        xlator_t        *readdir_xl = NULL;
-        int             ret = -1;
-        char            *path = NULL;
-        gf_boolean_t    missing = _gf_false;
-        char            gfid_str[64] = {0};
-
-        if (uuid_is_null (childloc->gfid))
-                goto out;
+    int ret = -1;
 
-        output = crawl_data->op_data;
-        readdir_xl = crawl_data->readdir_xl;
+    ret = afr_selfheal_name(THIS, parent, bname, NULL, NULL);
 
-        ret = _get_path_from_gfid_loc (this, readdir_xl, childloc, &path,
-                                       &missing);
-        if (ret == 0) {
-                ret = _add_path_to_dict (this, output, crawl_data->child, path,
-                                         NULL, _gf_true);
-        } else if (missing) {
-                _remove_stale_index (this, readdir_xl, parentloc,
-                                     uuid_utoa_r (childloc->gfid, gfid_str));
+    return ret;
+}
+
+int
+afr_shd_selfheal(struct subvol_healer *healer, int child, uuid_t gfid)
+{
+    int ret = 0;
+    eh_t *eh = NULL;
+    afr_private_t *priv = NULL;
+    afr_self_heald_t *shd = NULL;
+    shd_event_t *shd_event = NULL;
+    char *path = NULL;
+    xlator_t *subvol = NULL;
+    xlator_t *this = NULL;
+    crawl_event_t *crawl_event = NULL;
+
+    this = healer->this;
+    priv = this->private;
+    shd = &priv->shd;
+    crawl_event = &healer->crawl_event;
+
+    subvol = priv->children[child];
+
+    // If this fails with ENOENT/ESTALE index is stale
+    ret = syncop_gfid_to_path(this->itable, subvol, gfid, &path);
+    if (ret < 0)
+        return ret;
+
+    ret = afr_selfheal(this, gfid);
+
+    LOCK(&priv->lock);
+    {
+        if (ret == -EIO) {
+            eh = shd->split_brain;
+            crawl_event->split_brain_count++;
+        } else if (ret < 0) {
+            crawl_event->heal_failed_count++;
+        } else if (ret == 0) {
+            crawl_event->healed_count++;
         }
+    }
+    UNLOCK(&priv->lock);
+
+    if (eh) {
+        shd_event = GF_CALLOC(1, sizeof(*shd_event), gf_afr_mt_shd_event_t);
+        if (!shd_event)
+            goto out;
 
+        shd_event->child = child;
+        shd_event->path = path;
+
+        if (eh_save_history(eh, shd_event) < 0)
+            goto out;
+
+        shd_event = NULL;
+        path = NULL;
+    }
 out:
-        if (ret && path)
-                GF_FREE (path);
-        return ret;
+    GF_FREE(shd_event);
+    GF_FREE(path);
+    return ret;
 }
 
 void
-_crawl_post_sh_action (xlator_t *this, loc_t *parent, loc_t *child,
-                       int32_t op_ret, int32_t op_errno, dict_t *xattr_rsp,
-                       afr_crawl_data_t *crawl_data)
+afr_shd_sweep_prepare(struct subvol_healer *healer)
 {
-        int              ret = 0;
-        afr_private_t    *priv = NULL;
-        afr_self_heald_t *shd = NULL;
-        eh_t             *eh = NULL;
-        char             *path = NULL;
-        char             gfid_str[64] = {0};
-        shd_event_t      *event = NULL;
-        int32_t          sh_failed = 0;
-        gf_boolean_t     split_brain = 0;
-        int32_t          actual_sh_done = 0;
-        priv = this->private;
-        shd  = &priv->shd;
-        if (crawl_data->crawl == INDEX) {
-                if ((op_ret < 0) && (op_errno == ENOENT)) {
-                        _remove_stale_index (this, crawl_data->readdir_xl,
-                                             parent, uuid_utoa_r (child->gfid,
-                                                                  gfid_str));
-                        goto out;
-                }
-                ret = _get_path_from_gfid_loc (this, crawl_data->readdir_xl,
-                                               child, &path, NULL);
-                if (ret)
-                        goto out;
-        } else {
-                path = gf_strdup (child->path);
-                if (!path) {
-                        ret = -1;
-                        goto out;
-                }
-        }
-
-        if (xattr_rsp) {
-                ret = dict_get_int32 (xattr_rsp, "sh-failed", &sh_failed);
-                ret = dict_get_int32 (xattr_rsp, "actual-sh-done", &actual_sh_done);
-        }
+    crawl_event_t *event = NULL;
 
-        split_brain = afr_is_split_brain (this, child->inode);
+    event = &healer->crawl_event;
 
-        if ((op_ret < 0 && op_errno == EIO) || split_brain) {
-                eh = shd->split_brain;
-        } else if ((op_ret < 0) || sh_failed) {
-                eh = shd->heal_failed;
-        } else if (actual_sh_done == 1) {
-                      eh = shd->healed;
-        }
+    event->healed_count = 0;
+    event->split_brain_count = 0;
+    event->heal_failed_count = 0;
 
-        ret = -1;
+    event->start_time = gf_time();
+    event->end_time = 0;
+    _mask_cancellation();
+}
 
-        if (eh != NULL) {
-                event = GF_CALLOC (1, sizeof (*event), gf_afr_mt_shd_event_t);
-                if (!event)
-                        goto out;
-                event->child = crawl_data->child;
-                event->path = path;
+void
+afr_shd_sweep_done(struct subvol_healer *healer)
+{
+    crawl_event_t *event = NULL;
+    crawl_event_t *history = NULL;
+    afr_self_heald_t *shd = NULL;
 
-                ret = eh_save_history (eh, event);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_ERROR, "%s:Failed to save "
-                                "to event history, (%d, %s)", path, op_ret,
-                                strerror (op_errno));
+    event = &healer->crawl_event;
+    shd = &(((afr_private_t *)healer->this->private)->shd);
 
-                        goto out;
-                }
-        } else {
-                gf_log (this->name, GF_LOG_DEBUG, "%s:Self heal already done ",
-                        path);
+    event->end_time = gf_time();
+    history = gf_memdup(event, sizeof(*event));
+    event->start_time = 0;
 
-        }
-        ret = 0;
-out:
-        if (ret && path)
-                GF_FREE (path);
+    if (!history)
         return;
+
+    if (eh_save_history(shd->statistics[healer->subvol], history) < 0)
+        GF_FREE(history);
+    _unmask_cancellation();
 }
 
 int
-_link_inode_update_loc (xlator_t *this, loc_t *loc, struct iatt *iattr)
+afr_shd_index_heal(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent,
+                   void *data)
 {
-        inode_t       *link_inode = NULL;
-        int           ret = -1;
+    struct subvol_healer *healer = data;
+    afr_private_t *priv = NULL;
+    uuid_t gfid = {0};
+    int ret = 0;
+    uint64_t val = IA_INVAL;
 
-        link_inode = inode_link (loc->inode, NULL, NULL, iattr);
-        if (link_inode == NULL) {
-                gf_log (this->name, GF_LOG_ERROR, "inode link failed "
-                        "on the inode (%s)", uuid_utoa (iattr->ia_gfid));
-                goto out;
-        }
-        inode_unref (loc->inode);
-        loc->inode = link_inode;
-        ret = 0;
-out:
-        return ret;
-}
+    priv = healer->this->private;
+    if (!priv->shd.enabled)
+        return -EBUSY;
 
-int
-_self_heal_entry (xlator_t *this, afr_crawl_data_t *crawl_data, gf_dirent_t *entry,
-                  loc_t *child, loc_t *parent, struct iatt *iattr)
-{
-        struct iatt      parentbuf = {0};
-        int              ret = 0;
-        dict_t           *xattr_rsp = NULL;
+    gf_msg_debug(healer->this->name, 0, "got entry: %s from %s", entry->d_name,
+                 priv->children[healer->subvol]->name);
 
-        gf_log (this->name, GF_LOG_DEBUG, "lookup %s", child->path);
+    ret = gf_uuid_parse(entry->d_name, gfid);
+    if (ret)
+        return 0;
 
-        ret = syncop_lookup (this, child, NULL,
-                             iattr, &xattr_rsp, &parentbuf);
-        _crawl_post_sh_action (this, parent, child, ret, errno, xattr_rsp,
-                               crawl_data);
-        if (xattr_rsp)
-                dict_unref (xattr_rsp);
-        if (ret == 0)
-                ret = _link_inode_update_loc (this, child, iattr);
+    inode_ctx_get2(parent->inode, subvol, NULL, &val);
 
-        return ret;
-}
+    ret = afr_shd_selfheal(healer, healer->subvol, gfid);
 
-static int
-afr_crawl_done  (int ret, call_frame_t *sync_frame, void *data)
-{
-        GF_FREE (data);
-        STACK_DESTROY (sync_frame->root);
-        return 0;
-}
+    if (ret == -ENOENT || ret == -ESTALE)
+        afr_shd_entry_purge(subvol, parent->inode, entry->d_name, val);
 
-void
-_do_self_heal_on_subvol (xlator_t *this, int child, afr_crawl_type_t crawl)
-{
-        afr_start_crawl (this, child, crawl, _self_heal_entry,
-                         NULL, _gf_true, STOP_CRAWL_ON_SINGLE_SUBVOL,
-                         afr_crawl_done);
+    if (ret == 2)
+        /* If bricks crashed in pre-op after creating indices/xattrop
+         * link but before setting afr changelogs, we end up with stale
+         * xattrop links but zero changelogs. Remove such entries by
+         * sending a post-op with zero changelogs.
+         */
+        afr_shd_zero_xattrop(healer->this, gfid);
+
+    return 0;
 }
 
-gf_boolean_t
-_crawl_proceed (xlator_t *this, int child, int crawl_flags, char **reason)
+int
+afr_shd_index_sweep(struct subvol_healer *healer, char *vgfid)
 {
-        afr_private_t           *priv = NULL;
-        afr_self_heald_t        *shd = NULL;
-        gf_boolean_t            proceed = _gf_false;
-        char                    *msg = NULL;
-
-        priv = this->private;
-        shd  = &priv->shd;
-        if (!shd->enabled) {
-                msg = "Self-heal daemon is not enabled";
-                gf_log (this->name, GF_LOG_DEBUG, "%s", msg);
-                goto out;
-        }
-        if (!priv->child_up[child]) {
-                gf_log (this->name, GF_LOG_DEBUG, "Stopping crawl for %s , "
-                        "subvol went down", priv->children[child]->name);
-                msg = "Brick is Not connected";
-                goto out;
-        }
+    loc_t loc = {0};
+    afr_private_t *priv = NULL;
+    int ret = 0;
+    xlator_t *subvol = NULL;
+    dict_t *xdata = NULL;
+    call_frame_t *frame = NULL;
+
+    priv = healer->this->private;
+    subvol = priv->children[healer->subvol];
+
+    frame = afr_frame_create(healer->this, &ret);
+    if (!frame) {
+        ret = -ret;
+        goto out;
+    }
+
+    loc.inode = afr_shd_index_inode(healer->this, subvol, vgfid);
+    if (!loc.inode) {
+        gf_msg(healer->this->name, GF_LOG_WARNING, 0,
+               AFR_MSG_INDEX_DIR_GET_FAILED, "unable to get index-dir on %s",
+               subvol->name);
+        ret = -errno;
+        goto out;
+    }
+
+    xdata = dict_new();
+    if (!xdata || dict_set_int32_sizen(xdata, "get-gfid-type", 1)) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    ret = syncop_mt_dir_scan(frame, subvol, &loc, GF_CLIENT_PID_SELF_HEALD,
+                             healer, afr_shd_index_heal, xdata,
+                             priv->shd.max_threads, priv->shd.wait_qlength);
+
+    if (ret == 0)
+        ret = healer->crawl_event.healed_count;
 
-        if (crawl_flags & STOP_CRAWL_ON_SINGLE_SUBVOL) {
-                if (afr_up_children_count (priv->child_up,
-                                           priv->child_count) < 2) {
-                        gf_log (this->name, GF_LOG_DEBUG, "Stopping crawl as "
-                                "< 2 children are up");
-                        msg = "< 2 bricks in replica are running";
-                        goto out;
-                }
-        }
-        proceed = _gf_true;
 out:
-        if (reason)
-                *reason = msg;
-        return proceed;
+    loc_wipe(&loc);
+
+    if (xdata)
+        dict_unref(xdata);
+    if (frame)
+        AFR_STACK_DESTROY(frame);
+    return ret;
 }
 
 int
-_do_crawl_op_on_local_subvols (xlator_t *this, afr_crawl_type_t crawl,
-                               shd_crawl_op op, dict_t *output)
+afr_shd_index_sweep_all(struct subvol_healer *healer)
 {
-        afr_private_t       *priv = NULL;
-        char                *status = NULL;
-        char                *subkey = NULL;
-        char                key[256] = {0};
-        shd_pos_t           pos_data = {0};
-        int                 op_ret = -1;
-        int                 xl_id = -1;
-        int                 i = 0;
-        int                 ret = 0;
-        int                 crawl_flags = 0;
-
-        priv = this->private;
-        if (op == HEAL)
-                crawl_flags |= STOP_CRAWL_ON_SINGLE_SUBVOL;
-
-        if (output) {
-                ret = dict_get_int32 (output, this->name, &xl_id);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_ERROR, "Invalid input, "
-                                "translator-id is not available");
-                        goto out;
-                }
-        }
-        pos_data.this = this;
-        subkey = "status";
-        for (i = 0; i < priv->child_count; i++) {
-                if (_crawl_proceed (this, i, crawl_flags, &status)) {
-                        pos_data.child = i;
-                        /*
-                         * We're already in a synctask in this case, so we
-                         * don't need to defer through a second (and in fact
-                         * that can cause deadlock).  Just call straight
-                         * through instead.
-                         */
-                        ret = afr_find_child_position(pos_data.this,
-                                                      pos_data.child,
-                                                      &pos_data.pos);
-                        if (ret) {
-                                status = "Not able to find brick location";
-                        } else if (pos_data.pos == AFR_POS_REMOTE) {
-                                status = "brick is remote";
-                        } else {
-                                op_ret = 0;
-                                if (op == HEAL) {
-                                        status = "Started self-heal";
-                                        _do_self_heal_on_subvol (this, i,
-                                                                 crawl);
-                                } else if (output) {
-                                        status = "";
-                                        afr_start_crawl (this, i, INDEX,
-                                                         _add_summary_to_dict,
-                                                         output, _gf_false, 0,
-                                                         NULL);
-                                }
-                        }
-                        if (output) {
-                                snprintf (key, sizeof (key), "%d-%d-%s", xl_id,
-                                          i, subkey);
-                                ret = dict_set_str (output, key, status);
-                        }
-                        if (!op_ret && (crawl == FULL))
-                                break;
-                }
-                if (output) {
-                        snprintf (key, sizeof (key), "%d-%d-%s", xl_id, i,
-                                  subkey);
-                        ret = dict_set_str (output, key, status);
-                }
-        }
+    int ret = 0;
+    int count = 0;
+
+    ret = afr_shd_index_sweep(healer, GF_XATTROP_INDEX_GFID);
+    if (ret < 0)
+        goto out;
+    count = ret;
+
+    ret = afr_shd_index_sweep(healer, GF_XATTROP_DIRTY_GFID);
+    if (ret < 0)
+        goto out;
+    count += ret;
+
+    ret = afr_shd_index_sweep(healer, GF_XATTROP_ENTRY_CHANGES_GFID);
+    if (ret < 0)
+        goto out;
+    count += ret;
 out:
-        return op_ret;
+    if (ret < 0)
+        return ret;
+    else
+        return count;
 }
 
 int
-_do_self_heal_on_local_subvols (xlator_t *this, afr_crawl_type_t crawl,
-                                dict_t *output)
+afr_shd_full_heal(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent,
+                  void *data)
 {
-        return _do_crawl_op_on_local_subvols (this, crawl, HEAL, output);
+    struct subvol_healer *healer = data;
+    xlator_t *this = healer->this;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+
+    if (this->cleanup_starting) {
+        return -ENOTCONN;
+    }
+
+    if (!priv->shd.enabled)
+        return -EBUSY;
+
+    afr_shd_selfheal_name(healer, healer->subvol, parent->inode->gfid,
+                          entry->d_name);
+
+    afr_shd_selfheal(healer, healer->subvol, entry->d_stat.ia_gfid);
+
+    return 0;
 }
 
 int
-_get_index_summary_on_local_subvols (xlator_t *this, dict_t *output)
+afr_shd_full_sweep(struct subvol_healer *healer, inode_t *inode)
 {
-        return _do_crawl_op_on_local_subvols (this, INDEX, INFO, output);
+    afr_private_t *priv = NULL;
+    loc_t loc = {0};
+
+    priv = healer->this->private;
+    loc.inode = inode;
+    return syncop_ftw(priv->children[healer->subvol], &loc,
+                      GF_CLIENT_PID_SELF_HEALD, healer, afr_shd_full_heal);
 }
 
 int
-_add_all_subvols_eh_to_dict (xlator_t *this, eh_t *eh, dict_t *dict)
+afr_shd_fill_ta_loc(xlator_t *this, loc_t *loc)
 {
-        afr_private_t           *priv = NULL;
-        afr_self_heald_t        *shd = NULL;
-        int                     i = 0;
+    afr_private_t *priv = NULL;
+    struct iatt stbuf = {
+        0,
+    };
+    int ret = -1;
+
+    priv = this->private;
+    loc->parent = inode_ref(this->itable->root);
+    gf_uuid_copy(loc->pargfid, loc->parent->gfid);
+    loc->name = priv->pending_key[THIN_ARBITER_BRICK_INDEX];
+    loc->inode = inode_new(loc->parent->table);
+    GF_CHECK_ALLOC(loc->inode, ret, out);
+
+    if (!gf_uuid_is_null(priv->ta_gfid))
+        goto assign_gfid;
+
+    ret = syncop_lookup(priv->children[THIN_ARBITER_BRICK_INDEX], loc, &stbuf,
+                        0, 0, 0);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed lookup on file %s.", loc->name);
+        goto out;
+    }
+
+    gf_uuid_copy(priv->ta_gfid, stbuf.ia_gfid);
+
+assign_gfid:
+    gf_uuid_copy(loc->gfid, priv->ta_gfid);
+    ret = 0;
 
-        priv = this->private;
-        shd = &priv->shd;
+out:
+    if (ret)
+        loc_wipe(loc);
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (shd->pos[i] != AFR_POS_LOCAL)
-                        continue;
-                _add_eh_to_dict (this, eh, dict, i);
-        }
-        return 0;
+    return ret;
 }
 
 int
-afr_xl_op (xlator_t *this, dict_t *input, dict_t *output)
+_afr_shd_ta_get_xattrs(xlator_t *this, loc_t *loc, dict_t **xdata)
 {
-        gf_xl_afr_op_t   op = GF_AFR_OP_INVALID;
-        int              ret = 0;
-        afr_private_t    *priv = NULL;
-        afr_self_heald_t *shd = NULL;
-        int              xl_id = 0;
+    afr_private_t *priv = NULL;
+    dict_t *xattr = NULL;
+    int raw[AFR_NUM_CHANGE_LOGS] = {
+        0,
+    };
+    int ret = -1;
+    int i = 0;
+
+    priv = this->private;
+
+    xattr = dict_new();
+    if (!xattr) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_DICT_GET_FAILED,
+               "Failed to create dict.");
+        goto out;
+    }
+    for (i = 0; i < priv->child_count; i++) {
+        ret = dict_set_static_bin(xattr, priv->pending_key[i], &raw,
+                                  AFR_NUM_CHANGE_LOGS * sizeof(int));
+        if (ret)
+            goto out;
+    }
 
-        priv = this->private;
-        shd = &priv->shd;
+    ret = syncop_xattrop(priv->children[THIN_ARBITER_BRICK_INDEX], loc,
+                         GF_XATTROP_ADD_ARRAY, xattr, NULL, xdata, NULL);
+    if (ret || !(*xdata)) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Xattrop failed on %s.", loc->name);
+    }
 
-        ret = dict_get_int32 (input, "xl-op", (int32_t*)&op);
-        if (ret)
-                goto out;
-        ret = dict_get_int32 (input, this->name, &xl_id);
-        if (ret)
-                goto out;
-        ret = dict_set_int32 (output, this->name, xl_id);
-        if (ret)
-                goto out;
-        switch (op) {
-        case GF_AFR_OP_HEAL_INDEX:
-                ret = _do_self_heal_on_local_subvols (this, INDEX, output);
-                break;
-        case GF_AFR_OP_HEAL_FULL:
-                ret = _do_self_heal_on_local_subvols (this, FULL, output);
-                break;
-        case GF_AFR_OP_INDEX_SUMMARY:
-                (void)_get_index_summary_on_local_subvols (this, output);
-                ret = 0;
-                break;
-        case GF_AFR_OP_HEALED_FILES:
-                ret = _add_all_subvols_eh_to_dict (this, shd->healed, output);
-                break;
-        case GF_AFR_OP_HEAL_FAILED_FILES:
-                ret = _add_all_subvols_eh_to_dict (this, shd->heal_failed,
-                                                   output);
-                break;
-        case GF_AFR_OP_SPLIT_BRAIN_FILES:
-                ret = _add_all_subvols_eh_to_dict (this, shd->split_brain,
-                                                   output);
-                break;
-        default:
-                gf_log (this->name, GF_LOG_ERROR, "Unknown set op %d", op);
-                break;
-        }
 out:
-        dict_del (output, this->name);
-        return ret;
+    if (xattr)
+        dict_unref(xattr);
+
+    return ret;
 }
 
 void
-afr_poll_self_heal (void *data)
+afr_shd_ta_get_xattrs(xlator_t *this, loc_t *loc, struct subvol_healer *healer,
+                      dict_t **xdata)
 {
-        afr_private_t    *priv = NULL;
-        afr_self_heald_t *shd = NULL;
-        struct timeval   timeout = {0};
-        xlator_t         *this = NULL;
-        long             child = (long)data;
-        gf_timer_t       *old_timer = NULL;
-        gf_timer_t       *new_timer = NULL;
-        shd_pos_t        pos_data = {0};
-        int              ret = 0;
-
-        this = THIS;
-        priv = this->private;
-        shd = &priv->shd;
-
-        if (shd->pos[child] == AFR_POS_UNKNOWN) {
-                pos_data.this = this;
-                pos_data.child = child;
-                ret = synctask_new (this->ctx->env,
-                                    afr_syncop_find_child_position,
-                                    NULL, NULL, &pos_data);
-                if (!ret)
-                        shd->pos[child] = pos_data.pos;
-        }
-        if (shd->enabled && (shd->pos[child] == AFR_POS_LOCAL))
-                _do_self_heal_on_subvol (this, child, INDEX);
-        timeout.tv_sec = shd->timeout;
-        timeout.tv_usec = 0;
-        //notify and previous timer should be synchronized.
-        LOCK (&priv->lock);
-        {
-                old_timer = shd->timer[child];
-                if (shd->pos[child] == AFR_POS_REMOTE)
-                        goto unlock;
-                shd->timer[child] = gf_timer_call_after (this->ctx, timeout,
-                                                         afr_poll_self_heal,
-                                                         data);
-                new_timer = shd->timer[child];
-        }
-unlock:
-        UNLOCK (&priv->lock);
-
-        if (old_timer)
-                gf_timer_call_cancel (this->ctx, old_timer);
-        if (!new_timer && (shd->pos[child] != AFR_POS_REMOTE)) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "Could not create self-heal polling timer for %s",
-                        priv->children[child]->name);
+    int ret = 0;
+
+    loc_wipe(loc);
+    if (afr_shd_fill_ta_loc(this, loc)) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed to populate thin-arbiter loc for: %s.", loc->name);
+        ret = -1;
+        goto out;
+    }
+
+    ret = afr_ta_post_op_lock(this, loc);
+    if (ret)
+        goto out;
+
+    ret = _afr_shd_ta_get_xattrs(this, loc, xdata);
+    if (ret) {
+        if (*xdata) {
+            dict_unref(*xdata);
+            *xdata = NULL;
         }
-        return;
+    }
+
+    afr_ta_post_op_unlock(this, loc);
+
+out:
+    if (ret)
+        healer->rerun = 1;
 }
 
-static int
-afr_handle_child_up  (int ret, call_frame_t *sync_frame, void *data)
+int
+afr_shd_ta_unset_xattrs(xlator_t *this, loc_t *loc, dict_t **xdata, int healer)
 {
-        afr_self_heald_t *shd = NULL;
-        shd_pos_t        *pos_data = data;
-        afr_private_t    *priv = NULL;
+    afr_private_t *priv = NULL;
+    dict_t *xattr = NULL;
+    gf_boolean_t need_xattrop = _gf_false;
+    void *pending_raw = NULL;
+    int *raw = NULL;
+    int pending[AFR_NUM_CHANGE_LOGS] = {
+        0,
+    };
+    int i = 0;
+    int j = 0;
+    int val = 0;
+    int ret = -1;
+
+    priv = this->private;
+
+    xattr = dict_new();
+    if (!xattr) {
+        goto out;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        raw = GF_CALLOC(AFR_NUM_CHANGE_LOGS, sizeof(int), gf_afr_mt_int32_t);
+        if (!raw) {
+            goto out;
+        }
 
-        if (ret)
-                goto out;
+        ret = dict_get_ptr(*xdata, priv->pending_key[i], &pending_raw);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_GET_FAILED,
+                   "Error getting value "
+                   "of pending key %s",
+                   priv->pending_key[i]);
+            GF_FREE(raw);
+            goto out;
+        }
+
+        memcpy(pending, pending_raw, sizeof(pending));
+        for (j = 0; j < AFR_NUM_CHANGE_LOGS; j++) {
+            val = ntoh32(pending[j]);
+            if (val) {
+                if (i == healer) {
+                    gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_THIN_ARB,
+                           "I am "
+                           "not the good shd. Skipping. "
+                           "SHD = %d.",
+                           healer);
+                    ret = 0;
+                    GF_FREE(raw);
+                    goto out;
+                }
+                need_xattrop = _gf_true;
+                raw[j] = hton32(-val);
+            }
+        }
+
+        ret = dict_set_bin(xattr, priv->pending_key[i], raw,
+                           AFR_NUM_CHANGE_LOGS * sizeof(int));
+        if (ret) {
+            GF_FREE(raw);
+            goto out;
+        }
+
+        if (need_xattrop)
+            break;
+    }
+
+    if (!need_xattrop) {
+        ret = 0;
+        goto out;
+    }
+
+    ret = syncop_xattrop(priv->children[THIN_ARBITER_BRICK_INDEX], loc,
+                         GF_XATTROP_ADD_ARRAY, xattr, NULL, NULL, NULL);
+    if (ret)
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Xattrop failed.");
 
-        priv = pos_data->this->private;
-        shd = &priv->shd;
-        shd->pos[pos_data->child] = pos_data->pos;
-        if (pos_data->pos != AFR_POS_REMOTE)
-                afr_poll_self_heal ((void*)(long)pos_data->child);
-        _do_self_heal_on_local_subvols (THIS, INDEX, NULL);
 out:
-        GF_FREE (data);
-        return 0;
+    if (xattr)
+        dict_unref(xattr);
+
+    return ret;
 }
 
 void
-afr_proactive_self_heal (void *data)
+afr_shd_ta_check_and_unset_xattrs(xlator_t *this, loc_t *loc,
+                                  struct subvol_healer *healer,
+                                  dict_t *pre_crawl_xdata)
 {
-        xlator_t         *this = NULL;
-        long             child = (long)data;
-        shd_pos_t        *pos_data = NULL;
-        int              ret = 0;
+    int ret_lock = 0;
+    int ret = 0;
+    dict_t *post_crawl_xdata = NULL;
 
-        this = THIS;
+    ret_lock = afr_ta_post_op_lock(this, loc);
+    if (ret_lock)
+        goto unref;
 
-        //Position of brick could have changed and it could be local now.
-        //Compute the position again
-        pos_data = GF_CALLOC (1, sizeof (*pos_data), gf_afr_mt_pos_data_t);
-        if (!pos_data)
-                goto out;
-        pos_data->this = this;
-        pos_data->child = child;
-        ret = synctask_new (this->ctx->env, afr_syncop_find_child_position,
-                            afr_handle_child_up, NULL, pos_data);
-        if (ret)
-                goto out;
-out:
-        return;
+    ret = _afr_shd_ta_get_xattrs(this, loc, &post_crawl_xdata);
+    if (ret)
+        goto unref;
+
+    if (!are_dicts_equal(pre_crawl_xdata, post_crawl_xdata, NULL, NULL)) {
+        ret = -1;
+        goto unref;
+    }
+
+    ret = afr_shd_ta_unset_xattrs(this, loc, &post_crawl_xdata, healer->subvol);
+
+unref:
+    if (post_crawl_xdata) {
+        dict_unref(post_crawl_xdata);
+        post_crawl_xdata = NULL;
+    }
+
+    if (ret || ret_lock)
+        healer->rerun = 1;
+
+    if (!ret_lock)
+        afr_ta_post_op_unlock(this, loc);
 }
 
-static int
-get_pathinfo_host (char *pathinfo, char *hostname, size_t size)
+gf_boolean_t
+afr_bricks_available_for_heal(afr_private_t *priv)
 {
-        char    *start = NULL;
-        char    *end = NULL;
-        int     ret  = -1;
-        int     i    = 0;
+    int up_children = 0;
 
-        if (!pathinfo)
-                goto out;
+    up_children = __afr_get_up_children_count(priv);
+    if (up_children < 2) {
+        return _gf_false;
+    }
+    return _gf_true;
+}
 
-        start = strchr (pathinfo, ':');
-        if (!start)
-                goto out;
-        end = strrchr (pathinfo, ':');
-        if (start == end)
-                goto out;
+static gf_boolean_t
+afr_shd_ta_needs_heal(xlator_t *this, struct subvol_healer *healer)
+{
+    dict_t *xdata = NULL;
+    afr_private_t *priv = NULL;
+    loc_t loc = {
+        0,
+    };
+    int ret = -1;
+    int i = 0;
+    gf_boolean_t need_heal = _gf_false;
+
+    priv = this->private;
+
+    ret = afr_shd_fill_ta_loc(this, &loc);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed to populate thin-arbiter loc for: %s.", loc.name);
+        healer->rerun = 1;
+        goto out;
+    }
+
+    if (_afr_shd_ta_get_xattrs(this, &loc, &xdata)) {
+        healer->rerun = 1;
+        goto out;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (afr_ta_dict_contains_pending_xattr(xdata, priv, i)) {
+            need_heal = _gf_true;
+            break;
+        }
+    }
 
-        memset (hostname, 0, size);
-        i = 0;
-        while (++start != end)
-                hostname[i++] = *start;
-        ret = 0;
 out:
-        return ret;
+    if (xdata)
+        dict_unref(xdata);
+    loc_wipe(&loc);
+
+    return need_heal;
 }
 
-int
-afr_local_pathinfo (char *pathinfo, gf_boolean_t *local)
+static int
+afr_shd_anon_inode_cleaner(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent,
+                           void *data)
 {
-        int             ret   = 0;
-        char            pathinfohost[1024] = {0};
-        char            localhost[1024] = {0};
-        xlator_t        *this = THIS;
-
-        *local = _gf_false;
-        ret = get_pathinfo_host (pathinfo, pathinfohost, sizeof (pathinfohost));
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "Invalid pathinfo: %s",
-                        pathinfo);
+    struct subvol_healer *healer = data;
+    afr_private_t *priv = healer->this->private;
+    call_frame_t *frame = NULL;
+    afr_local_t *local = NULL;
+    int ret = 0;
+    loc_t loc = {0};
+    int count = 0;
+    int i = 0;
+    int op_errno = 0;
+    struct iatt *iatt = NULL;
+    gf_boolean_t multiple_links = _gf_false;
+    unsigned char *gfid_present = alloca0(priv->child_count);
+    unsigned char *entry_present = alloca0(priv->child_count);
+    char *type = "file";
+
+    frame = afr_frame_create(healer->this, &ret);
+    if (!frame) {
+        ret = -ret;
+        goto out;
+    }
+    local = frame->local;
+    if (AFR_COUNT(local->child_up, priv->child_count) != priv->child_count) {
+        gf_msg_debug(healer->this->name, 0,
+                     "Not all bricks are up. Skipping "
+                     "cleanup of %s on %s",
+                     entry->d_name, subvol->name);
+        ret = 0;
+        goto out;
+    }
+
+    loc.inode = inode_new(parent->inode->table);
+    if (!loc.inode) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    ret = gf_uuid_parse(entry->d_name, loc.gfid);
+    if (ret) {
+        ret = 0;
+        goto out;
+    }
+    AFR_ONLIST(local->child_up, frame, afr_selfheal_discover_cbk, lookup, &loc,
+               NULL);
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->replies[i].op_ret == 0) {
+            count++;
+            gfid_present[i] = 1;
+            iatt = &local->replies[i].poststat;
+            if (iatt->ia_type == IA_IFDIR) {
+                type = "dir";
+            }
+
+            if (i == healer->subvol) {
+                if (local->replies[i].poststat.ia_nlink > 1) {
+                    multiple_links = _gf_true;
+                }
+            }
+        } else if (local->replies[i].op_errno != ENOENT &&
+                   local->replies[i].op_errno != ESTALE) {
+            /*We don't have complete view. Skip the entry*/
+            gf_msg_debug(healer->this->name, local->replies[i].op_errno,
+                         "Skipping cleanup of %s on %s", entry->d_name,
+                         subvol->name);
+            ret = 0;
+            goto out;
+        }
+    }
+
+    /*Inode is deleted from subvol*/
+    if (count == 1 || (iatt->ia_type != IA_IFDIR && multiple_links)) {
+        gf_msg(healer->this->name, GF_LOG_WARNING, 0,
+               AFR_MSG_EXPUNGING_FILE_OR_DIR, "expunging %s %s/%s on %s", type,
+               priv->anon_inode_name, entry->d_name, subvol->name);
+        ret = afr_shd_entry_purge(subvol, parent->inode, entry->d_name,
+                                  iatt->ia_type);
+        if (ret == -ENOENT || ret == -ESTALE)
+            ret = 0;
+    } else if (count > 1) {
+        loc_wipe(&loc);
+        loc.parent = inode_ref(parent->inode);
+        loc.name = entry->d_name;
+        loc.inode = inode_new(parent->inode->table);
+        if (!loc.inode) {
+            ret = -ENOMEM;
+            goto out;
+        }
+        AFR_ONLIST(local->child_up, frame, afr_selfheal_discover_cbk, lookup,
+                   &loc, NULL);
+        count = 0;
+        for (i = 0; i < priv->child_count; i++) {
+            if (local->replies[i].op_ret == 0) {
+                count++;
+                entry_present[i] = 1;
+                iatt = &local->replies[i].poststat;
+            } else if (local->replies[i].op_errno != ENOENT &&
+                       local->replies[i].op_errno != ESTALE) {
+                /*We don't have complete view. Skip the entry*/
+                gf_msg_debug(healer->this->name, local->replies[i].op_errno,
+                             "Skipping cleanup of %s on %s", entry->d_name,
+                             subvol->name);
+                ret = 0;
                 goto out;
+            }
         }
-
-        ret = gethostname (localhost, sizeof (localhost));
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "gethostname() failed, "
-                        "reason: %s", strerror (errno));
+        for (i = 0; i < priv->child_count; i++) {
+            if (gfid_present[i] && !entry_present[i]) {
+                /*Entry is not anonymous on at least one subvol*/
+                gf_msg_debug(healer->this->name, 0,
+                             "Valid entry present on %s "
+                             "Skipping cleanup of %s on %s",
+                             priv->children[i]->name, entry->d_name,
+                             subvol->name);
+                ret = 0;
                 goto out;
+            }
         }
 
-        if (!strcmp (localhost, pathinfohost))
-                *local = _gf_true;
+        gf_msg(healer->this->name, GF_LOG_WARNING, 0,
+               AFR_MSG_EXPUNGING_FILE_OR_DIR,
+               "expunging %s %s/%s on all subvols", type, priv->anon_inode_name,
+               entry->d_name);
+        ret = 0;
+        for (i = 0; i < priv->child_count; i++) {
+            op_errno = -afr_shd_entry_purge(priv->children[i], loc.parent,
+                                            entry->d_name, iatt->ia_type);
+            if (op_errno != ENOENT && op_errno != ESTALE) {
+                ret |= -op_errno;
+            }
+        }
+    }
+
 out:
-        return ret;
+    if (frame)
+        AFR_STACK_DESTROY(frame);
+    loc_wipe(&loc);
+    return ret;
 }
 
-int
-afr_crawl_build_start_loc (xlator_t *this, afr_crawl_data_t *crawl_data,
-                           loc_t *dirloc)
+static void
+afr_cleanup_anon_inode_dir(struct subvol_healer *healer)
 {
-        afr_private_t *priv = NULL;
-        dict_t        *xattr = NULL;
-        void          *index_gfid = NULL;
-        loc_t         rootloc = {0};
-        struct iatt   iattr = {0};
-        struct iatt   parent = {0};
-        int           ret = 0;
-        xlator_t      *readdir_xl = crawl_data->readdir_xl;
-
-        priv = this->private;
-        if (crawl_data->crawl == FULL) {
-                afr_build_root_loc (this, dirloc);
-        } else {
-                afr_build_root_loc (this, &rootloc);
-                ret = syncop_getxattr (readdir_xl, &rootloc, &xattr,
-                                       GF_XATTROP_INDEX_GFID);
-                if (ret < 0)
-                        goto out;
-                ret = dict_get_ptr (xattr, GF_XATTROP_INDEX_GFID, &index_gfid);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_ERROR, "failed to get index "
-                                "dir gfid on %s", readdir_xl->name);
-                        goto out;
-                }
-                if (!index_gfid) {
-                        gf_log (this->name, GF_LOG_ERROR, "index gfid empty "
-                                "on %s", readdir_xl->name);
-                        ret = -1;
-                        goto out;
-                }
-                uuid_copy (dirloc->gfid, index_gfid);
-                dirloc->path = "";
-                dirloc->inode = inode_new (priv->root_inode->table);
-                ret = syncop_lookup (readdir_xl, dirloc, NULL,
-                                     &iattr, NULL, &parent);
-                if (ret < 0) {
-                        if (errno != ENOENT) {
-                                gf_log (this->name, GF_LOG_ERROR, "lookup "
-                                        "failed on index dir on %s - (%s)",
-                                        readdir_xl->name, strerror (errno));
-                        }
-                        goto out;
-                }
-                ret = _link_inode_update_loc (this, dirloc, &iattr);
-                if (ret)
-                        goto out;
-        }
-        ret = 0;
+    int ret = 0;
+    call_frame_t *frame = NULL;
+    afr_private_t *priv = healer->this->private;
+    loc_t loc = {0};
+
+    ret = afr_anon_inode_create(healer->this, healer->subvol, &loc.inode);
+    if (ret)
+        goto out;
+
+    frame = afr_frame_create(healer->this, &ret);
+    if (!frame) {
+        ret = -ret;
+        goto out;
+    }
+
+    ret = syncop_mt_dir_scan(frame, priv->children[healer->subvol], &loc,
+                             GF_CLIENT_PID_SELF_HEALD, healer,
+                             afr_shd_anon_inode_cleaner, NULL,
+                             priv->shd.max_threads, priv->shd.wait_qlength);
 out:
-        if (xattr)
-                dict_unref (xattr);
-        loc_wipe (&rootloc);
-        return ret;
+    if (frame)
+        AFR_STACK_DESTROY(frame);
+    loc_wipe(&loc);
+    return;
 }
 
-int
-afr_crawl_opendir (xlator_t *this, afr_crawl_data_t *crawl_data, fd_t **dirfd,
-                   loc_t *dirloc)
+void *
+afr_shd_index_healer(void *data)
 {
-        fd_t          *fd   = NULL;
-        int           ret = 0;
-
-        if (crawl_data->crawl == FULL) {
-                fd = fd_create (dirloc->inode, crawl_data->pid);
-                if (!fd) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Failed to create fd for %s", dirloc->path);
-                        ret = -1;
-                        goto out;
-                }
+    struct subvol_healer *healer = NULL;
+    xlator_t *this = NULL;
+    int ret = 0;
+    afr_private_t *priv = NULL;
+    dict_t *pre_crawl_xdata = NULL;
+    loc_t loc = {
+        0,
+    };
+
+    healer = data;
+    THIS = this = healer->this;
+    priv = this->private;
+
+    for (;;) {
+        afr_shd_healer_wait(healer);
+
+        if (!afr_bricks_available_for_heal(priv))
+            continue;
+
+        ASSERT_LOCAL(this, healer);
+        priv->local[healer->subvol] = healer->local;
+
+        if (priv->thin_arbiter_count) {
+            if (afr_shd_ta_needs_heal(this, healer))
+                afr_shd_ta_get_xattrs(this, &loc, healer, &pre_crawl_xdata);
+        }
 
-                ret = syncop_opendir (crawl_data->readdir_xl, dirloc, fd);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "opendir failed on %s", dirloc->path);
-                        goto out;
-                }
-        } else {
-                fd = fd_anonymous (dirloc->inode);
+        do {
+            gf_msg_debug(this->name, 0, "starting index sweep on subvol %s",
+                         afr_subvol_name(this, healer->subvol));
+
+            afr_shd_sweep_prepare(healer);
+
+            ret = afr_shd_index_sweep_all(healer);
+
+            afr_shd_sweep_done(healer);
+            /*
+              As long as at least one gfid was
+              healed, keep retrying. We may have
+              just healed a directory and thereby
+              created entries for other gfids which
+              could not be healed thus far.
+            */
+
+            gf_msg_debug(this->name, 0, "finished index sweep on subvol %s",
+                         afr_subvol_name(this, healer->subvol));
+            /*
+              Give a pause before retrying to avoid a busy loop
+              in case the only entry in index is because of
+              an ongoing I/O.
+            */
+            sleep(1);
+        } while (ret > 0);
+
+        if (ret == 0) {
+            afr_cleanup_anon_inode_dir(healer);
         }
-        ret = 0;
-out:
-        if (!ret)
-                *dirfd = fd;
-        return ret;
+
+        if (ret == 0 && pre_crawl_xdata &&
+            !healer->crawl_event.heal_failed_count) {
+            afr_shd_ta_check_and_unset_xattrs(this, &loc, healer,
+                                              pre_crawl_xdata);
+        }
+
+        if (pre_crawl_xdata) {
+            dict_unref(pre_crawl_xdata);
+            pre_crawl_xdata = NULL;
+        }
+    }
+
+    return NULL;
 }
 
-xlator_t*
-afr_crawl_readdir_xl_get (xlator_t *this, afr_crawl_data_t *crawl_data)
+void *
+afr_shd_full_healer(void *data)
 {
-        afr_private_t *priv = this->private;
+    struct subvol_healer *healer = NULL;
+    xlator_t *this = NULL;
+    int run = 0;
 
-        if (crawl_data->crawl == FULL) {
-                return this;
-        } else {
-                return priv->children[crawl_data->child];
+    healer = data;
+    THIS = this = healer->this;
+
+    for (;;) {
+        pthread_mutex_lock(&healer->mutex);
+        {
+            run = __afr_shd_healer_wait(healer);
+            if (!run)
+                healer->running = _gf_false;
         }
-        return NULL;
+        pthread_mutex_unlock(&healer->mutex);
+
+        if (!run)
+            break;
+
+        ASSERT_LOCAL(this, healer);
+
+        gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_SELF_HEAL_INFO,
+               "starting full sweep on subvol %s",
+               afr_subvol_name(this, healer->subvol));
+
+        afr_shd_sweep_prepare(healer);
+
+        afr_shd_full_sweep(healer, this->itable->root);
+
+        afr_shd_sweep_done(healer);
+
+        gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_SELF_HEAL_INFO,
+               "finished full sweep on subvol %s",
+               afr_subvol_name(this, healer->subvol));
+    }
+
+    return NULL;
 }
 
 int
-afr_crawl_build_child_loc (xlator_t *this, loc_t *child, loc_t *parent,
-                           gf_dirent_t *entry, afr_crawl_data_t *crawl_data)
+afr_shd_healer_init(xlator_t *this, struct subvol_healer *healer)
 {
-        int           ret = -1;
-        afr_private_t *priv = NULL;
+    int ret = 0;
 
-        priv = this->private;
-        if (crawl_data->crawl == FULL) {
-                ret = afr_build_child_loc (this, child, parent, entry->d_name);
-        } else {
-                child->inode = inode_new (priv->root_inode->table);
-                if (!child->inode)
-                        goto out;
-                uuid_parse (entry->d_name, child->gfid);
-                ret = _loc_assign_gfid_path (child);
-        }
+    ret = pthread_mutex_init(&healer->mutex, NULL);
+    if (ret)
+        goto out;
+
+    ret = pthread_cond_init(&healer->cond, NULL);
+    if (ret)
+        goto out;
+
+    healer->this = this;
+    healer->running = _gf_false;
+    healer->rerun = _gf_false;
+    healer->local = _gf_false;
 out:
-        return ret;
+    return ret;
 }
 
-static int
-_process_entries (xlator_t *this, loc_t *parentloc, gf_dirent_t *entries,
-                  off_t *offset, afr_crawl_data_t *crawl_data)
+int
+afr_shd_healer_spawn(xlator_t *this, struct subvol_healer *healer,
+                     void *(threadfn)(void *))
 {
-        gf_dirent_t      *entry = NULL;
-        gf_dirent_t      *tmp = NULL;
-        int              ret = 0;
-        loc_t            entry_loc = {0};
-        fd_t             *fd = NULL;
-        struct iatt      iattr = {0};
-
-        list_for_each_entry_safe (entry, tmp, &entries->list, list) {
-                if (!_crawl_proceed (this, crawl_data->child,
-                                     crawl_data->crawl_flags, NULL)) {
-                        ret = -1;
-                        goto out;
-                }
-                *offset = entry->d_off;
-                if (IS_ENTRY_CWD (entry->d_name) ||
-                    IS_ENTRY_PARENT (entry->d_name))
-                        continue;
-                if ((crawl_data->crawl == FULL) &&
-                     uuid_is_null (entry->d_stat.ia_gfid)) {
-                        gf_log (this->name, GF_LOG_WARNING, "%s/%s: No "
-                                "gfid present skipping",
-                                parentloc->path, entry->d_name);
-                        continue;
-                }
+    int ret = 0;
 
-                loc_wipe (&entry_loc);
-                ret = afr_crawl_build_child_loc (this, &entry_loc, parentloc,
-                                                 entry, crawl_data);
-                if (ret)
-                        goto out;
-
-                ret = crawl_data->process_entry (this, crawl_data, entry,
-                                                 &entry_loc, parentloc, &iattr);
-
-                if (ret)
-                        continue;
-
-                if (crawl_data->crawl == INDEX)
-                        continue;
-
-                if (!IA_ISDIR (iattr.ia_type))
-                        continue;
-                fd = NULL;
-                ret = afr_crawl_opendir (this, crawl_data, &fd, &entry_loc);
-                if (ret)
-                        continue;
-                ret = _crawl_directory (fd, &entry_loc, crawl_data);
-                if (fd)
-                        fd_unref (fd);
+    pthread_mutex_lock(&healer->mutex);
+    {
+        if (healer->running) {
+            pthread_cond_signal(&healer->cond);
+        } else {
+            ret = gf_thread_create(&healer->thread, NULL, threadfn, healer,
+                                   "shdheal");
+            if (ret)
+                goto unlock;
+            healer->running = 1;
         }
-        ret = 0;
-out:
-        loc_wipe (&entry_loc);
-        return ret;
-}
 
-static int
-_crawl_directory (fd_t *fd, loc_t *loc, afr_crawl_data_t *crawl_data)
-{
-        xlator_t        *this = NULL;
-        off_t           offset   = 0;
-        gf_dirent_t     entries;
-        int             ret = 0;
-        gf_boolean_t    free_entries = _gf_false;
-        xlator_t        *readdir_xl = crawl_data->readdir_xl;
-
-        INIT_LIST_HEAD (&entries.list);
-        this = THIS;
-
-        GF_ASSERT (loc->inode);
-
-        if (crawl_data->crawl == FULL)
-                gf_log (this->name, GF_LOG_DEBUG, "crawling %s", loc->path);
-        else
-                gf_log (this->name, GF_LOG_DEBUG, "crawling INDEX %s",
-                        uuid_utoa (loc->gfid));
-
-        while (1) {
-                if (crawl_data->crawl == FULL)
-                        ret = syncop_readdirp (readdir_xl, fd, 131072, offset,
-                                               NULL, &entries);
-                else
-                        ret = syncop_readdir (readdir_xl, fd, 131072, offset,
-                                              &entries);
-                if (ret <= 0)
-                        break;
-                ret = 0;
-                free_entries = _gf_true;
+        healer->rerun = 1;
+    }
+unlock:
+    pthread_mutex_unlock(&healer->mutex);
 
-                if (!_crawl_proceed (this, crawl_data->child,
-                                     crawl_data->crawl_flags, NULL)) {
-                        ret = -1;
-                        goto out;
-                }
-                if (list_empty (&entries.list))
-                        goto out;
+    return ret;
+}
 
-                ret = _process_entries (this, loc, &entries, &offset,
-                                        crawl_data);
-                gf_dirent_free (&entries);
-                free_entries = _gf_false;
-        }
-        ret = 0;
-out:
-        if (free_entries)
-                gf_dirent_free (&entries);
-        return ret;
+int
+afr_shd_full_healer_spawn(xlator_t *this, int subvol)
+{
+    return afr_shd_healer_spawn(this, NTH_FULL_HEALER(this, subvol),
+                                afr_shd_full_healer);
 }
 
-static char*
-position_str_get (afr_child_pos_t pos)
+int
+afr_shd_index_healer_spawn(xlator_t *this, int subvol)
 {
-        switch (pos) {
-        case AFR_POS_UNKNOWN:
-                return "unknown";
-        case AFR_POS_LOCAL:
-                return "local";
-        case AFR_POS_REMOTE:
-                return "remote";
-        }
-        return NULL;
+    return afr_shd_healer_spawn(this, NTH_INDEX_HEALER(this, subvol),
+                                afr_shd_index_healer);
 }
 
 int
-afr_find_child_position (xlator_t *this, int child, afr_child_pos_t *pos)
+afr_shd_dict_add_crawl_event(xlator_t *this, dict_t *output,
+                             crawl_event_t *crawl_event)
 {
-        afr_private_t    *priv = NULL;
-        afr_self_heald_t *shd  = NULL;
-        dict_t           *xattr_rsp = NULL;
-        loc_t            loc = {0};
-        int              ret = 0;
-        char             *node_uuid = NULL;
-
-        priv = this->private;
-        shd  = &priv->shd;
-
-        afr_build_root_loc (this, &loc);
-
-        ret = syncop_getxattr (priv->children[child], &loc, &xattr_rsp,
-                               GF_XATTR_NODE_UUID_KEY);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_ERROR, "getxattr failed on %s - "
-                        "(%s)", priv->children[child]->name, strerror (errno));
-                goto out;
-        }
+    int ret = 0;
+    uint64_t count = 0;
+    char key[128] = {0};
+    int keylen = 0;
+    char suffix[64] = {0};
+    int xl_id = 0;
+    uint64_t healed_count = 0;
+    uint64_t split_brain_count = 0;
+    uint64_t heal_failed_count = 0;
+    char *start_time_str = 0;
+    char *end_time_str = NULL;
+    char *crawl_type = NULL;
+    int progress = -1;
+    int child = -1;
+
+    child = crawl_event->child;
+    healed_count = crawl_event->healed_count;
+    split_brain_count = crawl_event->split_brain_count;
+    heal_failed_count = crawl_event->heal_failed_count;
+    crawl_type = crawl_event->crawl_type;
+
+    if (!crawl_event->start_time)
+        goto out;
+
+    start_time_str = gf_strdup(ctime(&crawl_event->start_time));
+
+    if (crawl_event->end_time)
+        end_time_str = gf_strdup(ctime(&crawl_event->end_time));
+
+    ret = dict_get_int32(output, this->name, &xl_id);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_GET_FAILED,
+               "xl does not have id");
+        goto out;
+    }
+
+    snprintf(key, sizeof(key), "statistics-%d-%d-count", xl_id, child);
+    ret = dict_get_uint64(output, key, &count);
+
+    snprintf(suffix, sizeof(suffix), "%d-%d-%" PRIu64, xl_id, child, count);
+    snprintf(key, sizeof(key), "statistics_healed_cnt-%s", suffix);
+    ret = dict_set_uint64(output, key, healed_count);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+               "Could not add statistics_healed_count to output");
+        goto out;
+    }
+
+    snprintf(key, sizeof(key), "statistics_sb_cnt-%s", suffix);
+    ret = dict_set_uint64(output, key, split_brain_count);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+               "Could not add statistics_split_brain_count to output");
+        goto out;
+    }
+
+    keylen = snprintf(key, sizeof(key), "statistics_crawl_type-%s", suffix);
+    ret = dict_set_strn(output, key, keylen, crawl_type);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+               "Could not add statistics_crawl_type to output");
+        goto out;
+    }
+
+    snprintf(key, sizeof(key), "statistics_heal_failed_cnt-%s", suffix);
+    ret = dict_set_uint64(output, key, heal_failed_count);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+               "Could not add statistics_healed_failed_count to output");
+        goto out;
+    }
+
+    keylen = snprintf(key, sizeof(key), "statistics_strt_time-%s", suffix);
+    ret = dict_set_dynstrn(output, key, keylen, start_time_str);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+               "Could not add statistics_crawl_start_time to output");
+        goto out;
+    } else {
+        start_time_str = NULL;
+    }
+
+    if (!end_time_str)
+        progress = 1;
+    else
+        progress = 0;
+
+    keylen = snprintf(key, sizeof(key), "statistics_end_time-%s", suffix);
+    if (!end_time_str)
+        end_time_str = gf_strdup("Could not determine the end time");
+    ret = dict_set_dynstrn(output, key, keylen, end_time_str);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+               "Could not add statistics_crawl_end_time to output");
+        goto out;
+    } else {
+        end_time_str = NULL;
+    }
+
+    keylen = snprintf(key, sizeof(key), "statistics_inprogress-%s", suffix);
+
+    ret = dict_set_int32n(output, key, keylen, progress);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+               "Could not add statistics_inprogress to output");
+        goto out;
+    }
+
+    snprintf(key, sizeof(key), "statistics-%d-%d-count", xl_id, child);
+    ret = dict_set_uint64(output, key, count + 1);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+               "Could not increment the counter.");
+        goto out;
+    }
+out:
+    GF_FREE(start_time_str);
+    GF_FREE(end_time_str);
+    return ret;
+}
 
-        ret = dict_get_str (xattr_rsp, GF_XATTR_NODE_UUID_KEY, &node_uuid);
+int
+afr_shd_dict_add_path(xlator_t *this, dict_t *output, int child, char *path,
+                      struct timeval *tv)
+{
+    int ret = -1;
+    uint64_t count = 0;
+    char key[64] = {0};
+    int keylen = 0;
+    char xl_id_child_str[32] = {0};
+    int xl_id = 0;
+
+    ret = dict_get_int32(output, this->name, &xl_id);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_GET_FAILED,
+               "xl does not have id");
+        goto out;
+    }
+
+    snprintf(xl_id_child_str, sizeof(xl_id_child_str), "%d-%d", xl_id, child);
+    snprintf(key, sizeof(key), "%s-count", xl_id_child_str);
+    ret = dict_get_uint64(output, key, &count);
+
+    keylen = snprintf(key, sizeof(key), "%s-%" PRIu64, xl_id_child_str, count);
+    ret = dict_set_dynstrn(output, key, keylen, path);
+
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+               "%s: Could not add to output", path);
+        goto out;
+    }
+
+    if (tv) {
+        snprintf(key, sizeof(key), "%s-%" PRIu64 "-time", xl_id_child_str,
+                 count);
+        ret = dict_set_uint32(output, key, tv->tv_sec);
         if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "node-uuid key not found on "
-                        "child %s", priv->children[child]->name);
-                goto out;
+            gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+                   "%s: Could not set time", path);
+            goto out;
         }
+    }
+
+    snprintf(key, sizeof(key), "%s-count", xl_id_child_str);
 
-        if (!strcmp (node_uuid, shd->node_uuid))
-                *pos = AFR_POS_LOCAL;
-        else
-                *pos = AFR_POS_REMOTE;
+    ret = dict_set_uint64(output, key, count + 1);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+               "Could not increment count");
+        goto out;
+    }
 
-        gf_log (this->name, GF_LOG_DEBUG, "child %s is %s",
-                priv->children[child]->name, position_str_get (*pos));
+    ret = 0;
 out:
-        if (ret)
-                *pos = AFR_POS_UNKNOWN;
-        loc_wipe (&loc);
-        return ret;
+    return ret;
 }
 
 int
-afr_syncop_find_child_position (void *data)
+afr_add_shd_event(circular_buffer_t *cb, void *data)
 {
-        shd_pos_t *pos_data = data;
-        int       ret = 0;
+    dict_t *output = NULL;
+    xlator_t *this = THIS;
+    afr_private_t *priv = NULL;
+    afr_self_heald_t *shd = NULL;
+    shd_event_t *shd_event = NULL;
+    char *path = NULL;
+
+    output = data;
+    priv = this->private;
+    shd = &priv->shd;
+    shd_event = cb->data;
+
+    if (!shd->index_healers[shd_event->child].local)
+        return 0;
 
-        ret = afr_find_child_position (pos_data->this, pos_data->child,
-                                       &pos_data->pos);
-        return ret;
+    path = gf_strdup(shd_event->path);
+    if (!path)
+        return -ENOMEM;
+
+    afr_shd_dict_add_path(this, output, shd_event->child, path, &cb->tv);
+    return 0;
 }
 
-static int
-afr_dir_crawl (void *data)
+int
+afr_add_crawl_event(circular_buffer_t *cb, void *data)
 {
-        xlator_t            *this = NULL;
-        int                 ret = -1;
-        xlator_t            *readdir_xl = NULL;
-        fd_t                *fd = NULL;
-        loc_t               dirloc = {0};
-        afr_crawl_data_t    *crawl_data = data;
-
-        this = THIS;
+    dict_t *output = NULL;
+    xlator_t *this = THIS;
+    afr_private_t *priv = NULL;
+    afr_self_heald_t *shd = NULL;
+    crawl_event_t *crawl_event = NULL;
+
+    output = data;
+    priv = this->private;
+    shd = &priv->shd;
+    crawl_event = cb->data;
+
+    if (!shd->index_healers[crawl_event->child].local)
+        return 0;
 
-        if (!_crawl_proceed (this, crawl_data->child, crawl_data->crawl_flags,
-                             NULL))
-                goto out;
+    afr_shd_dict_add_crawl_event(this, output, crawl_event);
 
-        readdir_xl = afr_crawl_readdir_xl_get (this, crawl_data);
-        if (!readdir_xl)
-                goto out;
-        crawl_data->readdir_xl = readdir_xl;
+    return 0;
+}
 
-        ret = afr_crawl_build_start_loc (this, crawl_data, &dirloc);
+int
+afr_selfheal_daemon_init(xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    afr_self_heald_t *shd = NULL;
+    int ret = -1;
+    int i = 0;
+
+    priv = this->private;
+    shd = &priv->shd;
+
+    shd->index_healers = GF_CALLOC(sizeof(*shd->index_healers),
+                                   priv->child_count,
+                                   gf_afr_mt_subvol_healer_t);
+    if (!shd->index_healers)
+        goto out;
+
+    for (i = 0; i < priv->child_count; i++) {
+        shd->index_healers[i].subvol = i;
+        ret = afr_shd_healer_init(this, &shd->index_healers[i]);
         if (ret)
-                goto out;
-
-        ret = afr_crawl_opendir (this, crawl_data, &fd, &dirloc);
+            goto out;
+    }
+
+    shd->full_healers = GF_CALLOC(sizeof(*shd->full_healers), priv->child_count,
+                                  gf_afr_mt_subvol_healer_t);
+    if (!shd->full_healers)
+        goto out;
+    for (i = 0; i < priv->child_count; i++) {
+        shd->full_healers[i].subvol = i;
+        ret = afr_shd_healer_init(this, &shd->full_healers[i]);
         if (ret)
-                goto out;
-
-        ret = _crawl_directory (fd, &dirloc, crawl_data);
-        if (ret)
-                gf_log (this->name, GF_LOG_ERROR, "Crawl failed on %s",
-                        readdir_xl->name);
-        else
-                gf_log (this->name, GF_LOG_DEBUG, "Crawl completed "
-                        "on %s", readdir_xl->name);
-        if (crawl_data->crawl == INDEX)
-                dirloc.path = NULL;
+            goto out;
+    }
+
+    shd->split_brain = eh_new(AFR_EH_SPLIT_BRAIN_LIMIT, _gf_false,
+                              afr_destroy_shd_event_data);
+    if (!shd->split_brain)
+        goto out;
+
+    shd->statistics = GF_CALLOC(sizeof(eh_t *), priv->child_count,
+                                gf_common_mt_eh_t);
+    if (!shd->statistics)
+        goto out;
+
+    for (i = 0; i < priv->child_count; i++) {
+        shd->statistics[i] = eh_new(AFR_STATISTICS_HISTORY_SIZE, _gf_false,
+                                    afr_destroy_crawl_event_data);
+        if (!shd->statistics[i])
+            goto out;
+        shd->full_healers[i].crawl_event.child = i;
+        shd->full_healers[i].crawl_event.crawl_type = "FULL";
+        shd->index_healers[i].crawl_event.child = i;
+        shd->index_healers[i].crawl_event.crawl_type = "INDEX";
+    }
+
+    ret = 0;
 out:
-        if (fd)
-                fd_unref (fd);
-        if (crawl_data->crawl == INDEX)
-                dirloc.path = NULL;
-        loc_wipe (&dirloc);
-        return ret;
+    return ret;
 }
 
-static int
-afr_dir_exclusive_crawl (void *data)
+void
+afr_selfheal_childup(xlator_t *this, afr_private_t *priv)
 {
-        afr_private_t    *priv = NULL;
-        afr_self_heald_t *shd = NULL;
-        gf_boolean_t     crawl = _gf_false;
-        int              ret = 0;
-        int              child = -1;
-        xlator_t         *this = NULL;
-        afr_crawl_data_t *crawl_data = data;
-
-        this = THIS;
-        priv = this->private;
-        shd = &priv->shd;
-        child = crawl_data->child;
-
-        LOCK (&priv->lock);
-        {
-                if (shd->inprogress[child]) {
-                        if (shd->pending[child] != FULL)
-                                shd->pending[child] = crawl_data->crawl;
-                } else {
-                        shd->inprogress[child] = _gf_true;
-                        crawl = _gf_true;
-                }
-        }
-        UNLOCK (&priv->lock);
+    int subvol = 0;
 
-        if (!crawl) {
-                gf_log (this->name, GF_LOG_INFO, "Another crawl is in progress "
-                        "for %s", priv->children[child]->name);
-                goto out;
-        }
+    if (!priv->shd.iamshd)
+        return;
+    for (subvol = 0; subvol < priv->child_count; subvol++)
+        if (priv->child_up[subvol])
+            afr_shd_index_healer_spawn(this, subvol);
 
-        do {
-                afr_dir_crawl (data);
-                LOCK (&priv->lock);
-                {
-                        if (shd->pending[child] != NONE) {
-                                crawl_data->crawl = shd->pending[child];
-                                shd->pending[child] = NONE;
-                        } else {
-                                shd->inprogress[child] = _gf_false;
-                                crawl = _gf_false;
-                        }
-                }
-                UNLOCK (&priv->lock);
-        } while (crawl);
-out:
-        return ret;
+    return;
 }
 
-void
-afr_start_crawl (xlator_t *this, int idx, afr_crawl_type_t crawl,
-                 process_entry_cbk_t process_entry, void *op_data,
-                 gf_boolean_t exclusive, int crawl_flags,
-                 afr_crawl_done_cbk_t crawl_done)
+int
+afr_shd_get_index_count(xlator_t *this, int i, uint64_t *count)
 {
-        afr_private_t              *priv = NULL;
-        call_frame_t               *frame = NULL;
-        afr_crawl_data_t           *crawl_data = NULL;
-        int                        ret = 0;
-        int (*crawler) (void*) = NULL;
+    afr_private_t *priv = NULL;
+    xlator_t *subvol = NULL;
+    loc_t rootloc = {
+        0,
+    };
+    dict_t *xattr = NULL;
+    int ret = -1;
 
-        priv = this->private;
+    priv = this->private;
+    subvol = priv->children[i];
 
-        frame = create_frame (this, this->ctx->pool);
-        if (!frame)
-                goto out;
+    rootloc.inode = inode_ref(this->itable->root);
+    gf_uuid_copy(rootloc.gfid, rootloc.inode->gfid);
 
-        afr_set_lk_owner (frame, this, frame->root);
-        afr_set_low_priority (frame);
-        crawl_data = GF_CALLOC (1, sizeof (*crawl_data),
-                                gf_afr_mt_crawl_data_t);
-        if (!crawl_data)
-                goto out;
-        crawl_data->process_entry = process_entry;
-        crawl_data->child = idx;
-        crawl_data->pid = frame->root->pid;
-        crawl_data->crawl = crawl;
-        crawl_data->op_data = op_data;
-        crawl_data->crawl_flags = crawl_flags;
-        gf_log (this->name, GF_LOG_DEBUG, "starting crawl %d for %s",
-                crawl_data->crawl, priv->children[idx]->name);
-
-        if (exclusive)
-                crawler = afr_dir_exclusive_crawl;
-        else
-                crawler = afr_dir_crawl;
-        ret = synctask_new (this->ctx->env, crawler,
-                            crawl_done, frame, crawl_data);
-        if (ret)
-                gf_log (this->name, GF_LOG_ERROR, "afr crawl failed for child"
-                        " %d with ret %d", idx, ret);
-out:
-        return;
-}
+    ret = syncop_getxattr(subvol, &rootloc, &xattr, GF_XATTROP_INDEX_COUNT,
+                          NULL, NULL);
+    if (ret < 0)
+        goto out;
 
-void
-afr_build_root_loc (xlator_t *this, loc_t *loc)
-{
-        afr_private_t   *priv = NULL;
+    ret = dict_get_uint64(xattr, GF_XATTROP_INDEX_COUNT, count);
+    if (ret)
+        goto out;
+
+    ret = 0;
+
+out:
+    if (xattr)
+        dict_unref(xattr);
+    loc_wipe(&rootloc);
 
-        priv = this->private;
-        loc->path = gf_strdup ("/");
-        loc->name = "";
-        loc->inode = inode_ref (priv->root_inode);
-        uuid_copy (loc->gfid, loc->inode->gfid);
+    return ret;
 }
 
 int
-afr_set_root_gfid (dict_t *dict)
+afr_xl_op(xlator_t *this, dict_t *input, dict_t *output)
 {
-        uuid_t gfid;
-        int ret = 0;
+    gf_xl_afr_op_t op = GF_SHD_OP_INVALID;
+    int ret = 0;
+    int xl_id = 0;
+    afr_private_t *priv = NULL;
+    afr_self_heald_t *shd = NULL;
+    struct subvol_healer *healer = NULL;
+    int i = 0;
+    char key[64];
+    int keylen = 0;
+    int this_name_len = 0;
+    int op_ret = 0;
+    uint64_t cnt = 0;
+
+#define AFR_SET_DICT_AND_LOG(name, output, key, keylen, dict_str,              \
+                             dict_str_len)                                     \
+    {                                                                          \
+        int ret;                                                               \
+                                                                               \
+        ret = dict_set_nstrn(output, key, keylen, dict_str, dict_str_len);     \
+        if (ret) {                                                             \
+            gf_smsg(name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,         \
+                    "key=%s", key, "value=%s", dict_str, NULL);                \
+        }                                                                      \
+    }
+
+    priv = this->private;
+    shd = &priv->shd;
+
+    ret = dict_get_int32_sizen(input, "xl-op", (int32_t *)&op);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_GET_FAILED,
+                "key=xl-op", NULL);
+        goto out;
+    }
+    this_name_len = strlen(this->name);
+    ret = dict_get_int32n(input, this->name, this_name_len, &xl_id);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_GET_FAILED,
+                "key=%s", this->name, NULL);
+        goto out;
+    }
+    ret = dict_set_int32n(output, this->name, this_name_len, xl_id);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+                "key=%s", this->name, NULL);
+        goto out;
+    }
+    switch (op) {
+        case GF_SHD_OP_HEAL_INDEX:
+            op_ret = 0;
+
+            for (i = 0; i < priv->child_count; i++) {
+                healer = &shd->index_healers[i];
+                keylen = snprintf(key, sizeof(key), "%d-%d-status", xl_id, i);
+
+                if (!priv->child_up[i]) {
+                    AFR_SET_DICT_AND_LOG(this->name, output, key, keylen,
+                                         SBRICK_NOT_CONNECTED,
+                                         SLEN(SBRICK_NOT_CONNECTED));
+                    op_ret = -1;
+                } else if (AFR_COUNT(priv->child_up, priv->child_count) < 2) {
+                    AFR_SET_DICT_AND_LOG(this->name, output, key, keylen,
+                                         SLESS_THAN2_BRICKS_in_REP,
+                                         SLEN(SLESS_THAN2_BRICKS_in_REP));
+                    op_ret = -1;
+                } else if (!afr_shd_is_subvol_local(this, healer->subvol)) {
+                    AFR_SET_DICT_AND_LOG(this->name, output, key, keylen,
+                                         SBRICK_IS_REMOTE,
+                                         SLEN(SBRICK_IS_REMOTE));
+                } else {
+                    AFR_SET_DICT_AND_LOG(this->name, output, key, keylen,
+                                         SSTARTED_SELF_HEAL,
+                                         SLEN(SSTARTED_SELF_HEAL));
 
-        memset (gfid, 0, 16);
-        gfid[15] = 1;
+                    ret = afr_shd_index_healer_spawn(this, i);
 
-        ret = afr_set_dict_gfid (dict, gfid);
+                    if (ret) {
+                        gf_smsg(this->name, GF_LOG_ERROR, -ret,
+                                AFR_MSG_HEALER_SPAWN_FAILED, NULL);
+                    }
+                }
+            }
+            break;
+        case GF_SHD_OP_HEAL_FULL:
+            op_ret = -1;
+
+            for (i = 0; i < priv->child_count; i++) {
+                healer = &shd->full_healers[i];
+                keylen = snprintf(key, sizeof(key), "%d-%d-status", xl_id, i);
+
+                if (!priv->child_up[i]) {
+                    AFR_SET_DICT_AND_LOG(this->name, output, key, keylen,
+                                         SBRICK_NOT_CONNECTED,
+                                         SLEN(SBRICK_NOT_CONNECTED));
+                } else if (AFR_COUNT(priv->child_up, priv->child_count) < 2) {
+                    AFR_SET_DICT_AND_LOG(this->name, output, key, keylen,
+                                         SLESS_THAN2_BRICKS_in_REP,
+                                         SLEN(SLESS_THAN2_BRICKS_in_REP));
+                } else if (!afr_shd_is_subvol_local(this, healer->subvol)) {
+                    AFR_SET_DICT_AND_LOG(this->name, output, key, keylen,
+                                         SBRICK_IS_REMOTE,
+                                         SLEN(SBRICK_IS_REMOTE));
+                } else {
+                    AFR_SET_DICT_AND_LOG(this->name, output, key, keylen,
+                                         SSTARTED_SELF_HEAL,
+                                         SLEN(SSTARTED_SELF_HEAL));
 
-        return ret;
-}
+                    ret = afr_shd_full_healer_spawn(this, i);
 
+                    if (ret) {
+                        gf_smsg(this->name, GF_LOG_ERROR, -ret,
+                                AFR_MSG_HEALER_SPAWN_FAILED, NULL);
+                    }
+                    op_ret = 0;
+                }
+            }
+            break;
+        case GF_SHD_OP_INDEX_SUMMARY:
+            /* this case has been handled in glfs-heal.c */
+            break;
+        case GF_SHD_OP_SPLIT_BRAIN_FILES:
+            eh_dump(shd->split_brain, output, afr_add_shd_event);
+            break;
+        case GF_SHD_OP_STATISTICS:
+            for (i = 0; i < priv->child_count; i++) {
+                eh_dump(shd->statistics[i], output, afr_add_crawl_event);
+                ret = afr_shd_dict_add_crawl_event(
+                    this, output, &shd->index_healers[i].crawl_event);
+                if (ret) {
+                    gf_smsg(this->name, GF_LOG_ERROR, -ret,
+                            AFR_MSG_ADD_CRAWL_EVENT_FAILED, NULL);
+                }
+
+                ret = afr_shd_dict_add_crawl_event(
+                    this, output, &shd->full_healers[i].crawl_event);
+                if (ret) {
+                    gf_smsg(this->name, GF_LOG_ERROR, -ret,
+                            AFR_MSG_ADD_CRAWL_EVENT_FAILED, NULL);
+                }
+            }
+            break;
+        case GF_SHD_OP_STATISTICS_HEAL_COUNT:
+        case GF_SHD_OP_STATISTICS_HEAL_COUNT_PER_REPLICA:
+            op_ret = -1;
+
+            for (i = 0; i < priv->child_count; i++) {
+                if (!priv->child_up[i]) {
+                    keylen = snprintf(key, sizeof(key), "%d-%d-status", xl_id,
+                                      i);
+                    AFR_SET_DICT_AND_LOG(this->name, output, key, keylen,
+                                         SBRICK_NOT_CONNECTED,
+                                         SLEN(SBRICK_NOT_CONNECTED));
+                } else {
+                    snprintf(key, sizeof(key), "%d-%d-hardlinks", xl_id, i);
+                    ret = afr_shd_get_index_count(this, i, &cnt);
+                    if (ret == 0) {
+                        ret = dict_set_uint64(output, key, cnt);
+                    }
+                    if (ret) {
+                        gf_smsg(this->name, GF_LOG_ERROR, -ret,
+                                AFR_MSG_DICT_SET_FAILED, NULL);
+                    }
+                    op_ret = 0;
+                }
+            }
+
+            break;
+
+        default:
+            gf_smsg(this->name, GF_LOG_ERROR, 0, AFR_MSG_INVALID_ARG, "op=%d",
+                    op, NULL);
+            break;
+    }
+out:
+    dict_deln(output, this->name, this_name_len);
+    return op_ret;
+
+#undef AFR_SET_DICT_AND_LOG
+}
diff --git a/xlators/cluster/afr/src/afr-self-heald.h b/xlators/cluster/afr/src/afr-self-heald.h
index 32a8aaca50c..18db728ea7b 100644
--- a/xlators/cluster/afr/src/afr-self-heald.h
+++ b/xlators/cluster/afr/src/afr-self-heald.h
@@ -1,5 +1,5 @@
 /*
-  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
   This file is licensed to you under your choice of the GNU Lesser
@@ -8,45 +8,68 @@
   cases as published by the Free Software Foundation.
 */
 
-#ifndef __AFR_SELF_HEALD_H__
-#define __AFR_SELF_HEALD_H__
-#include "xlator.h"
+#ifndef _AFR_SELF_HEALD_H
+#define _AFR_SELF_HEALD_H
 
-#define IS_ROOT_PATH(path) (!strcmp (path, "/"))
-#define IS_ENTRY_CWD(entry) (!strcmp (entry, "."))
-#define IS_ENTRY_PARENT(entry) (!strcmp (entry, ".."))
-#define AFR_ALL_CHILDREN -1
+#include <pthread.h>
 
-typedef struct afr_crawl_data_ {
-        int                 child;
-        pid_t               pid;
-        afr_crawl_type_t    crawl;
-        xlator_t            *readdir_xl;
-        void                *op_data;
-        int                 crawl_flags;
-        int (*process_entry) (xlator_t *this, struct afr_crawl_data_ *crawl_data,
-                              gf_dirent_t *entry, loc_t *child, loc_t *parent,
-                              struct iatt *iattr);
-} afr_crawl_data_t;
+typedef struct {
+    char *path;
+    int child;
+} shd_event_t;
 
-typedef int (*process_entry_cbk_t) (xlator_t *this, afr_crawl_data_t *crawl_data,
-                              gf_dirent_t *entry, loc_t *child, loc_t *parent,
-                              struct iatt *iattr);
+typedef struct {
+    uint64_t healed_count;
+    uint64_t split_brain_count;
+    uint64_t heal_failed_count;
 
-void afr_build_root_loc (xlator_t *this, loc_t *loc);
+    /* If start_time is 0, it means crawler is not in progress
+       and stats are not valid */
+    time_t start_time;
+    /* If start_time is NOT 0 and end_time is 0, it means
+       cralwer is in progress */
+    time_t end_time;
+    char *crawl_type;
+    int child;
+} crawl_event_t;
 
-int afr_set_root_gfid (dict_t *dict);
+struct subvol_healer {
+    xlator_t *this;
+    crawl_event_t crawl_event;
+    pthread_mutex_t mutex;
+    pthread_cond_t cond;
+    pthread_t thread;
+    int subvol;
+    gf_boolean_t local;
+    gf_boolean_t running;
+    gf_boolean_t rerun;
+};
 
-void
-afr_proactive_self_heal (void *data);
+typedef struct {
+    struct subvol_healer *index_healers;
+    struct subvol_healer *full_healers;
+
+    eh_t *split_brain;
+    eh_t **statistics;
+    int timeout;
+    uint32_t max_threads;
+    uint32_t wait_qlength;
+    uint32_t halo_max_latency_msec;
+    gf_boolean_t iamshd;
+    gf_boolean_t enabled;
+} afr_self_heald_t;
 
 int
-afr_xl_op (xlator_t *this, dict_t *input, dict_t *output);
+afr_selfheal_daemon_init(xlator_t *this);
+
+int
+afr_xl_op(xlator_t *this, dict_t *input, dict_t *output);
+
+int
+afr_shd_gfid_to_path(xlator_t *this, xlator_t *subvol, uuid_t gfid,
+                     char **path_p);
 
-/*
- * In addition to its self-heal use, this is used to find a local default
- * read_child.
- */
 int
-afr_local_pathinfo (char *pathinfo, gf_boolean_t *local);
-#endif /* __AFR_SELF_HEALD_H__ */
+afr_shd_entry_purge(xlator_t *subvol, inode_t *inode, char *name,
+                    ia_type_t type);
+#endif /* !_AFR_SELF_HEALD_H */
diff --git a/xlators/cluster/afr/src/afr-transaction.c b/xlators/cluster/afr/src/afr-transaction.c
index 0b91f466e11..a51f79b1f43 100644
--- a/xlators/cluster/afr/src/afr-transaction.c
+++ b/xlators/cluster/afr/src/afr-transaction.c
@@ -8,1956 +8,2920 @@
   cases as published by the Free Software Foundation.
 */
 
-#include "dict.h"
-#include "byte-order.h"
-#include "common-utils.h"
-#include "timer.h"
+#include <glusterfs/dict.h>
+#include <glusterfs/byte-order.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/timer.h>
 
 #include "afr.h"
 #include "afr-transaction.h"
+#include "afr-self-heal.h"
+#include "afr-messages.h"
 
 #include <signal.h>
 
+typedef enum {
+    AFR_TRANSACTION_PRE_OP,
+    AFR_TRANSACTION_POST_OP,
+} afr_xattrop_type_t;
 
-#define LOCKED_NO       0x0        /* no lock held */
-#define LOCKED_YES      0x1        /* for DATA, METADATA, ENTRY and higher_path
-                                      of RENAME */
-#define LOCKED_LOWER    0x2        /* for lower_path of RENAME */
+static void
+afr_lock_resume_shared(struct list_head *list);
 
-afr_fd_ctx_t *
-__afr_fd_ctx_get (fd_t *fd, xlator_t *this)
-{
-        uint64_t       ctx = 0;
-        int            ret = 0;
-        afr_fd_ctx_t  *fd_ctx = NULL;
-        int            i = 0;
-        afr_private_t *priv = NULL;
+static void
+afr_post_op_handle_success(call_frame_t *frame, xlator_t *this);
 
-        priv = this->private;
+static void
+afr_post_op_handle_failure(call_frame_t *frame, xlator_t *this, int op_errno);
 
-        ret = __fd_ctx_get (fd, this, &ctx);
+void
+__afr_transaction_wake_shared(afr_local_t *local, struct list_head *shared);
 
-        if (ret < 0 && fd_is_anonymous (fd)) {
-                ret = __afr_fd_ctx_set (this, fd);
-                if (ret < 0)
-                        goto out;
+void
+afr_changelog_post_op_do(call_frame_t *frame, xlator_t *this);
 
-                ret = __fd_ctx_get (fd, this, &ctx);
-                if (ret < 0)
-                        goto out;
+int
+afr_changelog_post_op_safe(call_frame_t *frame, xlator_t *this);
 
-                fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-                for (i = 0; i < priv->child_count; i++)
-                        fd_ctx->opened_on[i] = AFR_FD_OPENED;
-        }
+gf_boolean_t
+afr_changelog_pre_op_uninherit(call_frame_t *frame, xlator_t *this);
 
-        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-out:
-        return fd_ctx;
-}
+gf_boolean_t
+afr_changelog_pre_op_update(call_frame_t *frame, xlator_t *this);
 
+int
+afr_changelog_call_count(afr_transaction_type type,
+                         unsigned char *pre_op_subvols,
+                         unsigned char *failed_subvols,
+                         unsigned int child_count);
+int
+afr_changelog_do(call_frame_t *frame, xlator_t *this, dict_t *xattr,
+                 afr_changelog_resume_t changelog_resume,
+                 afr_xattrop_type_t op);
 
-afr_fd_ctx_t *
-afr_fd_ctx_get (fd_t *fd, xlator_t *this)
-{
-        afr_fd_ctx_t  *fd_ctx = NULL;
+static void
+afr_ta_decide_post_op_state(call_frame_t *frame, xlator_t *this);
 
-        LOCK(&fd->lock);
-        {
-                fd_ctx = __afr_fd_ctx_get (fd, this);
-        }
-        UNLOCK(&fd->lock);
+static int
+afr_ta_post_op_do(void *opaque);
 
-        return fd_ctx;
-}
+static int
+afr_ta_post_op_synctask(xlator_t *this, afr_local_t *local);
 
+static int
+afr_changelog_post_op_done(call_frame_t *frame, xlator_t *this);
 
 static void
-afr_save_lk_owner (call_frame_t *frame)
-{
-        afr_local_t * local = NULL;
+afr_changelog_post_op_fail(call_frame_t *frame, xlator_t *this, int op_errno);
 
-        local = frame->local;
-
-        local->saved_lk_owner = frame->root->lk_owner;
+void
+afr_ta_locked_priv_invalidate(afr_private_t *priv)
+{
+    priv->ta_bad_child_index = AFR_CHILD_UNKNOWN;
+    priv->release_ta_notify_dom_lock = _gf_false;
+    priv->ta_notify_dom_lock_offset = 0;
 }
 
-
 static void
-afr_restore_lk_owner (call_frame_t *frame)
+afr_ta_process_waitq(xlator_t *this)
 {
-        afr_local_t * local = NULL;
-
-        local = frame->local;
-
-        frame->root->lk_owner = local->saved_lk_owner;
+    afr_local_t *entry = NULL;
+    afr_private_t *priv = this->private;
+    struct list_head waitq = {
+        0,
+    };
+
+    INIT_LIST_HEAD(&waitq);
+    LOCK(&priv->lock);
+    list_splice_init(&priv->ta_waitq, &waitq);
+    UNLOCK(&priv->lock);
+    list_for_each_entry(entry, &waitq, ta_waitq)
+    {
+        afr_ta_decide_post_op_state(entry->transaction.frame, this);
+    }
 }
 
-static void
-__mark_all_pending (int32_t *pending[], int child_count,
-                    afr_transaction_type type)
+int
+afr_ta_lock_release_done(int ret, call_frame_t *ta_frame, void *opaque)
 {
-        int i = 0;
-        int j = 0;
+    afr_ta_process_waitq(ta_frame->this);
+    STACK_DESTROY(ta_frame->root);
+    return 0;
+}
 
-        for (i = 0; i < child_count; i++) {
-                j = afr_index_for_transaction_type (type);
-                pending[i][j] = hton32 (1);
-        }
+int
+afr_release_notify_lock_for_ta(void *opaque)
+{
+    xlator_t *this = NULL;
+    afr_private_t *priv = NULL;
+    loc_t loc = {
+        0,
+    };
+    struct gf_flock flock = {
+        0,
+    };
+    int ret = -1;
+
+    this = (xlator_t *)opaque;
+    priv = this->private;
+    ret = afr_fill_ta_loc(this, &loc, _gf_true);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed to populate loc for thin-arbiter.");
+        goto out;
+    }
+    flock.l_type = F_UNLCK;
+    flock.l_start = priv->ta_notify_dom_lock_offset;
+    flock.l_len = 1;
+    ret = syncop_inodelk(priv->children[THIN_ARBITER_BRICK_INDEX],
+                         AFR_TA_DOM_NOTIFY, &loc, F_SETLK, &flock, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed to unlock AFR_TA_DOM_NOTIFY lock.");
+    }
+
+    LOCK(&priv->lock);
+    {
+        afr_ta_locked_priv_invalidate(priv);
+    }
+    UNLOCK(&priv->lock);
+out:
+    loc_wipe(&loc);
+    return ret;
 }
 
+void
+afr_zero_fill_stat(afr_local_t *local)
+{
+    if (!local)
+        return;
+    if (local->transaction.type == AFR_DATA_TRANSACTION ||
+        local->transaction.type == AFR_METADATA_TRANSACTION) {
+        gf_zero_fill_stat(&local->cont.inode_wfop.prebuf);
+        gf_zero_fill_stat(&local->cont.inode_wfop.postbuf);
+    } else if (local->transaction.type == AFR_ENTRY_TRANSACTION ||
+               local->transaction.type == AFR_ENTRY_RENAME_TRANSACTION) {
+        gf_zero_fill_stat(&local->cont.dir_fop.buf);
+        gf_zero_fill_stat(&local->cont.dir_fop.preparent);
+        gf_zero_fill_stat(&local->cont.dir_fop.postparent);
+        if (local->transaction.type == AFR_ENTRY_TRANSACTION)
+            return;
+        gf_zero_fill_stat(&local->cont.dir_fop.prenewparent);
+        gf_zero_fill_stat(&local->cont.dir_fop.postnewparent);
+    }
+}
 
-static void
-__mark_child_dead (int32_t *pending[], int child_count, int child,
-                   afr_transaction_type type)
+/* In case of errors afr needs to choose which xdata from lower xlators it needs
+ * to unwind with. The way it is done is by checking if there are
+ * any good subvols which failed. Give preference to errnos other than
+ * ENOTCONN even if the child is source */
+void
+afr_pick_error_xdata(afr_local_t *local, afr_private_t *priv, inode_t *inode1,
+                     unsigned char *readable1, inode_t *inode2,
+                     unsigned char *readable2)
 {
-        int j = 0;
+    int s = -1; /*selection*/
+    int i = 0;
+    unsigned char *readable = NULL;
+
+    if (local->xdata_rsp) {
+        dict_unref(local->xdata_rsp);
+        local->xdata_rsp = NULL;
+    }
+
+    readable = alloca0(priv->child_count * sizeof(*readable));
+    if (inode2 && readable2) { /*rename fop*/
+        AFR_INTERSECT(readable, readable1, readable2, priv->child_count);
+    } else {
+        memcpy(readable, readable1, sizeof(*readable) * priv->child_count);
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
+
+        if (local->replies[i].op_ret >= 0)
+            continue;
+
+        if (local->replies[i].op_errno == ENOTCONN)
+            continue;
+
+        /*Order is important in the following condition*/
+        if ((s < 0) || (!readable[s] && readable[i]))
+            s = i;
+    }
+
+    if (s != -1 && local->replies[s].xdata) {
+        local->xdata_rsp = dict_ref(local->replies[s].xdata);
+    } else if (s == -1) {
+        for (i = 0; i < priv->child_count; i++) {
+            if (!local->replies[i].valid)
+                continue;
 
-        j = afr_index_for_transaction_type (type);
+            if (local->replies[i].op_ret >= 0)
+                continue;
 
-        pending[child][j] = 0;
+            if (!local->replies[i].xdata)
+                continue;
+            local->xdata_rsp = dict_ref(local->replies[i].xdata);
+            break;
+        }
+    }
 }
 
+gf_boolean_t
+afr_needs_changelog_update(afr_local_t *local)
+{
+    if (local->transaction.type == AFR_DATA_TRANSACTION)
+        return _gf_true;
+    if (!local->optimistic_change_log)
+        return _gf_true;
+    return _gf_false;
+}
 
-static void
-__mark_pre_op_done_on_fd (call_frame_t *frame, xlator_t *this, int child_index)
+gf_boolean_t
+afr_changelog_has_quorum(afr_local_t *local, xlator_t *this)
 {
-        afr_local_t   *local = NULL;
-        afr_fd_ctx_t  *fd_ctx = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    unsigned char *success_children = NULL;
 
-        local = frame->local;
+    priv = this->private;
+    success_children = alloca0(priv->child_count);
 
-        if (!local->fd)
-                return;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->transaction.failed_subvols[i]) {
+            success_children[i] = 1;
+        }
+    }
 
-        fd_ctx = afr_fd_ctx_get (local->fd, this);
+    if (afr_has_quorum(success_children, this, NULL)) {
+        return _gf_true;
+    }
 
-        if (!fd_ctx)
-                goto out;
+    return _gf_false;
+}
 
-        LOCK (&local->fd->lock);
-        {
-                if (local->transaction.type == AFR_DATA_TRANSACTION)
-                        fd_ctx->pre_op_done[child_index]++;
-        }
-        UNLOCK (&local->fd->lock);
-out:
-        return;
+gf_boolean_t
+afr_is_write_subvol_valid(call_frame_t *frame, xlator_t *this)
+{
+    int i = 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    uint64_t write_subvol = 0;
+    unsigned char *writable = NULL;
+    uint16_t datamap = 0;
+
+    local = frame->local;
+    priv = this->private;
+    writable = alloca0(priv->child_count);
+
+    write_subvol = afr_write_subvol_get(frame, this);
+    datamap = (write_subvol & 0x00000000ffff0000) >> 16;
+    for (i = 0; i < priv->child_count; i++) {
+        if (datamap & (1 << i))
+            writable[i] = 1;
+
+        if (writable[i] && !local->transaction.failed_subvols[i])
+            return _gf_true;
+    }
+
+    return _gf_false;
 }
 
-static void
-__mark_non_participant_children (int32_t *pending[], int child_count,
-                                 unsigned char *participants,
-                                 afr_transaction_type type)
+int
+afr_transaction_fop(call_frame_t *frame, xlator_t *this)
 {
-        int i = 0;
-        int j = 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int call_count = -1;
+    unsigned char *failed_subvols = NULL;
+    int i = 0;
+
+    local = frame->local;
+    priv = this->private;
+
+    failed_subvols = local->transaction.failed_subvols;
+    call_count = priv->child_count -
+                 AFR_COUNT(failed_subvols, priv->child_count);
+    /* Fail if pre-op did not succeed on quorum no. of bricks. */
+    if (!afr_changelog_has_quorum(local, this) || !call_count) {
+        local->op_ret = -1;
+        /* local->op_errno is already captured in changelog cbk. */
+        afr_transaction_resume(frame, this);
+        return 0;
+    }
+
+    /* Fail if at least one writeable brick isn't up.*/
+    if (local->transaction.type == AFR_DATA_TRANSACTION &&
+        !afr_is_write_subvol_valid(frame, this)) {
+        local->op_ret = -1;
+        local->op_errno = EIO;
+        afr_transaction_resume(frame, this);
+        return 0;
+    }
 
-        j = afr_index_for_transaction_type (type);
-        for (i = 0; i < child_count; i++) {
-                if (!participants[i])
-                        pending[i][j] = 0;
+    local->call_count = call_count;
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->transaction.pre_op[i] && !failed_subvols[i]) {
+            local->transaction.wind(frame, this, i);
+
+            if (!--call_count)
+                break;
         }
-}
+    }
 
+    return 0;
+}
 
-void
-__mark_all_success (int32_t *pending[], int child_count,
-                    afr_transaction_type type)
+int
+afr_transaction_done(call_frame_t *frame, xlator_t *this)
 {
-        int i;
-        int j;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    gf_boolean_t unwind = _gf_false;
+    afr_lock_t *lock = NULL;
+    afr_local_t *lock_local = NULL;
+
+    priv = this->private;
+    local = frame->local;
 
-        for (i = 0; i < child_count; i++) {
-                j = afr_index_for_transaction_type (type);
-                pending[i][j] = hton32 (-1);
+    if (priv->consistent_metadata) {
+        LOCK(&frame->lock);
+        {
+            unwind = (local->transaction.main_frame != NULL);
+        }
+        UNLOCK(&frame->lock);
+        if (unwind) /*It definitely did post-op*/
+            afr_zero_fill_stat(local);
+    }
+
+    if (local->transaction.do_eager_unlock) {
+        lock = &local->inode_ctx->lock[local->transaction.type];
+        LOCK(&local->inode->lock);
+        {
+            lock->acquired = _gf_false;
+            lock->release = _gf_false;
+            list_splice_init(&lock->frozen, &lock->waiting);
+            if (list_empty(&lock->waiting))
+                goto unlock;
+            lock_local = list_entry(lock->waiting.next, afr_local_t,
+                                    transaction.wait_list);
+            list_del_init(&lock_local->transaction.wait_list);
+            list_add(&lock_local->transaction.owner_list, &lock->owners);
         }
+    unlock:
+        UNLOCK(&local->inode->lock);
+    }
+    if (lock_local) {
+        afr_lock(lock_local->transaction.frame,
+                 lock_local->transaction.frame->this);
+    }
+    local->transaction.unwind(frame, this);
+
+    GF_ASSERT(list_empty(&local->transaction.owner_list));
+    GF_ASSERT(list_empty(&local->transaction.wait_list));
+    AFR_STACK_DESTROY(frame);
+
+    return 0;
 }
 
-void
-_set_all_child_errno (int *child_errno, unsigned int child_count)
+static void
+afr_lock_fail_shared(afr_local_t *local, struct list_head *list)
 {
-        int     i = 0;
-
-        for (i = 0; i < child_count; i++)
-                if (child_errno[i] == 0)
-                        child_errno[i] = ENOTCONN;
+    afr_local_t *each = NULL;
+
+    while (!list_empty(list)) {
+        each = list_entry(list->next, afr_local_t, transaction.wait_list);
+        list_del_init(&each->transaction.wait_list);
+        each->op_ret = -1;
+        each->op_errno = local->op_errno;
+        afr_transaction_done(each->transaction.frame,
+                             each->transaction.frame->this);
+    }
 }
 
-void
-afr_transaction_perform_fop (call_frame_t *frame, xlator_t *this)
+static void
+afr_handle_lock_acquire_failure(afr_local_t *local)
 {
-        afr_local_t     *local = NULL;
-        afr_private_t   *priv = NULL;
-        fd_t            *fd   = NULL;
+    struct list_head shared;
+    afr_lock_t *lock = NULL;
 
-        local = frame->local;
-        priv  = this->private;
-        fd    = local->fd;
+    if (!local->transaction.eager_lock_on)
+        goto out;
 
-        __mark_all_success (local->pending, priv->child_count,
-                            local->transaction.type);
+    lock = &local->inode_ctx->lock[local->transaction.type];
 
-        _set_all_child_errno (local->child_errno, priv->child_count);
+    INIT_LIST_HEAD(&shared);
+    LOCK(&local->inode->lock);
+    {
+        lock->release = _gf_true;
+        list_splice_init(&lock->waiting, &shared);
+    }
+    UNLOCK(&local->inode->lock);
 
-        /*  Perform fops with the lk-owner from top xlator.
-         *  Eg: lk-owner of posix-lk and flush should be same,
-         *  flush cant clear the  posix-lks without that lk-owner.
-         */
-        afr_save_lk_owner (frame);
-        frame->root->lk_owner =
-                local->transaction.main_frame->root->lk_owner;
-
-
-        /* The wake up needs to happen independent of
-           what type of fop arrives here. If it was
-           a write, then it has already inherited the
-           lock and changelog. If it was not a write,
-           then the presumption of the optimization (of
-           optimizing for successive write operations)
-           fails.
-        */
-        if (fd)
-                afr_delayed_changelog_wake_up (this, fd);
-        local->transaction.fop (frame, this);
+    afr_lock_fail_shared(local, &shared);
+    local->transaction.do_eager_unlock = _gf_true;
+out:
+    local->internal_lock.lock_cbk = afr_transaction_done;
+    afr_unlock(local->transaction.frame, local->transaction.frame->this);
 }
 
-
-static int
-__changelog_enabled (afr_private_t *priv, afr_transaction_type type)
+call_frame_t *
+afr_transaction_detach_fop_frame(call_frame_t *frame)
 {
-        int ret = 0;
+    afr_local_t *local = NULL;
+    call_frame_t *fop_frame = NULL;
 
-        switch (type) {
-        case AFR_DATA_TRANSACTION:
-                if (priv->data_change_log)
-                        ret = 1;
+    local = frame->local;
 
-                break;
+    afr_handle_inconsistent_fop(frame, &local->op_ret, &local->op_errno);
+    LOCK(&frame->lock);
+    {
+        fop_frame = local->transaction.main_frame;
+        local->transaction.main_frame = NULL;
+    }
+    UNLOCK(&frame->lock);
 
-        case AFR_METADATA_TRANSACTION:
-                if (priv->metadata_change_log)
-                        ret = 1;
+    return fop_frame;
+}
 
-                break;
+static void
+afr_save_lk_owner(call_frame_t *frame)
+{
+    afr_local_t *local = NULL;
 
-        case AFR_ENTRY_TRANSACTION:
-        case AFR_ENTRY_RENAME_TRANSACTION:
-                if (priv->entry_change_log)
-                        ret = 1;
+    local = frame->local;
 
-                break;
-        }
-
-        return ret;
+    local->saved_lk_owner = frame->root->lk_owner;
 }
 
-
-static int
-__fop_changelog_needed (call_frame_t *frame, xlator_t *this)
+static void
+afr_restore_lk_owner(call_frame_t *frame)
 {
-        afr_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
-        int op_ret = 0;
-        afr_transaction_type type = -1;
-
-        priv  = this->private;
-        local = frame->local;
-        type  = local->transaction.type;
+    afr_local_t *local = NULL;
 
-        if (__changelog_enabled (priv, type)) {
-                switch (local->op) {
+    local = frame->local;
 
-                case GF_FOP_WRITE:
-                case GF_FOP_FTRUNCATE:
-                        op_ret = 1;
-                        break;
+    frame->root->lk_owner = local->saved_lk_owner;
+}
 
-                case GF_FOP_FLUSH:
-                        op_ret = 0;
-                        break;
+void
+__mark_all_success(call_frame_t *frame, xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int i;
 
-                default:
-                        op_ret = 1;
-                }
-        }
+    local = frame->local;
+    priv = this->private;
 
-        return op_ret;
+    for (i = 0; i < priv->child_count; i++) {
+        local->transaction.failed_subvols[i] = 0;
+    }
 }
 
-int
-afr_set_pending_dict (afr_private_t *priv, dict_t *xattr, int32_t **pending,
-                      int child, afr_xattrop_type_t op)
+void
+afr_compute_pre_op_sources(call_frame_t *frame, xlator_t *this)
 {
-        int i = 0;
-        int ret = 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    afr_transaction_type type = -1;
+    dict_t *xdata = NULL;
+    int **matrix = NULL;
+    int idx = -1;
+    int i = 0;
+    int j = 0;
+
+    priv = this->private;
+    local = frame->local;
+    type = local->transaction.type;
+    idx = afr_index_for_transaction_type(type);
+    matrix = ALLOC_MATRIX(priv->child_count, int);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->transaction.changelog_xdata[i])
+            continue;
+        xdata = local->transaction.changelog_xdata[i];
+        afr_selfheal_fill_matrix(this, matrix, i, idx, xdata);
+    }
+
+    memset(local->transaction.pre_op_sources, 1, priv->child_count);
+
+    /*If lock or pre-op failed on a brick, it is not a source. */
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->transaction.failed_subvols[i])
+            local->transaction.pre_op_sources[i] = 0;
+    }
+
+    /* If brick is blamed by others, it is not a source. */
+    for (i = 0; i < priv->child_count; i++)
+        for (j = 0; j < priv->child_count; j++)
+            if (matrix[i][j] != 0)
+                local->transaction.pre_op_sources[j] = 0;
+}
 
-        if (op == LOCAL_FIRST) {
-                ret = dict_set_static_bin (xattr, priv->pending_key[child],
-                                           pending[child],
-                                   AFR_NUM_CHANGE_LOGS * sizeof (int32_t));
-                if (ret)
-                        goto out;
-        }
-        for (i = 0; i < priv->child_count; i++) {
-                if (i == child)
-                        continue;
-                ret = dict_set_static_bin (xattr, priv->pending_key[i],
-                                           pending[i],
-                                   AFR_NUM_CHANGE_LOGS * sizeof (int32_t));
-                /* 3 = data+metadata+entry */
+void
+afr_txn_arbitrate_fop(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int pre_op_sources_count = 0;
+    int i = 0;
+
+    priv = this->private;
+    local = frame->local;
+
+    afr_compute_pre_op_sources(frame, this);
+    pre_op_sources_count = AFR_COUNT(local->transaction.pre_op_sources,
+                                     priv->child_count);
+
+    /* If arbiter is the only source, do not proceed. */
+    if (pre_op_sources_count < 2 &&
+        local->transaction.pre_op_sources[ARBITER_BRICK_INDEX]) {
+        local->op_ret = -1;
+        local->op_errno = ENOTCONN;
+        for (i = 0; i < priv->child_count; i++)
+            local->transaction.failed_subvols[i] = 1;
+    }
+
+    afr_transaction_fop(frame, this);
+
+    return;
+}
 
-                if (ret < 0)
-                        goto out;
+int
+afr_transaction_perform_fop(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int ret = 0;
+    int failure_count = 0;
+    struct list_head shared;
+    afr_lock_t *lock = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    INIT_LIST_HEAD(&shared);
+    if (local->transaction.type == AFR_DATA_TRANSACTION &&
+        !local->transaction.inherited) {
+        ret = afr_write_subvol_set(frame, this);
+        if (ret) {
+            /*act as if operation failed on all subvols*/
+            local->op_ret = -1;
+            local->op_errno = -ret;
+            for (i = 0; i < priv->child_count; i++)
+                local->transaction.failed_subvols[i] = 1;
         }
-        if (op == LOCAL_LAST) {
-                ret = dict_set_static_bin (xattr, priv->pending_key[child],
-                                           pending[child],
-                                   AFR_NUM_CHANGE_LOGS * sizeof (int32_t));
-                if (ret)
-                        goto out;
+    }
+
+    if (local->pre_op_compat)
+        /* old mode, pre-op was done as afr_changelog_do()
+           just now, before OP */
+        afr_changelog_pre_op_update(frame, this);
+
+    if (!local->transaction.eager_lock_on || local->transaction.inherited)
+        goto fop;
+    failure_count = AFR_COUNT(local->transaction.failed_subvols,
+                              priv->child_count);
+    if (failure_count == priv->child_count) {
+        afr_handle_lock_acquire_failure(local);
+        return 0;
+    } else {
+        lock = &local->inode_ctx->lock[local->transaction.type];
+        LOCK(&local->inode->lock);
+        {
+            lock->acquired = _gf_true;
+            __afr_transaction_wake_shared(local, &shared);
         }
-out:
-        return ret;
+        UNLOCK(&local->inode->lock);
+    }
+
+fop:
+    /*  Perform fops with the lk-owner from top xlator.
+     *  Eg: lk-owner of posix-lk and flush should be same,
+     *  flush cant clear the  posix-lks without that lk-owner.
+     */
+    afr_save_lk_owner(frame);
+    frame->root->lk_owner = local->transaction.main_frame->root->lk_owner;
+
+    if (priv->arbiter_count == 1) {
+        afr_txn_arbitrate_fop(frame, this);
+    } else {
+        afr_transaction_fop(frame, this);
+    }
+
+    afr_lock_resume_shared(&shared);
+    return 0;
 }
 
 int
-afr_lock_server_count (afr_private_t *priv, afr_transaction_type type)
+afr_set_pending_dict(afr_private_t *priv, dict_t *xattr, int **pending)
 {
-        int ret = 0;
+    int i = 0;
+    int ret = 0;
 
-        switch (type) {
-        case AFR_DATA_TRANSACTION:
-                ret = priv->child_count;
-                break;
+    for (i = 0; i < priv->child_count; i++) {
+        ret = dict_set_static_bin(xattr, priv->pending_key[i], pending[i],
+                                  AFR_NUM_CHANGE_LOGS * sizeof(int));
+        /* 3 = data+metadata+entry */
 
-        case AFR_METADATA_TRANSACTION:
-                ret = priv->child_count;
-                break;
+        if (ret)
+            break;
+    }
 
-        case AFR_ENTRY_TRANSACTION:
-        case AFR_ENTRY_RENAME_TRANSACTION:
-                ret = priv->child_count;
-                break;
-        }
-
-        return ret;
+    return ret;
 }
 
-/* {{{ pending */
-
-int32_t
-afr_changelog_post_op_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                           int32_t op_ret, int32_t op_errno, dict_t *xattr,
-                           dict_t *xdata)
+static void
+afr_ta_dom_lock_check_and_release(afr_ta_fop_state_t fop_state, xlator_t *this)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_private_t       *priv     = NULL;
-        afr_local_t         *local    = NULL;
-        int                  call_count = -1;
-
-        priv     = this->private;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        LOCK (&frame->lock);
-        {
-                call_count = --local->call_count;
+    afr_private_t *priv = this->private;
+    unsigned int inmem_count = 0;
+    unsigned int onwire_count = 0;
+    gf_boolean_t release = _gf_false;
+
+    LOCK(&priv->lock);
+    {
+        /*Once we get notify lock release upcall notification,
+         if any of the fop state counters are non-zero, we will
+         not release the lock.
+         */
+        onwire_count = priv->ta_on_wire_txn_count;
+        inmem_count = priv->ta_in_mem_txn_count;
+        switch (fop_state) {
+            case TA_GET_INFO_FROM_TA_FILE:
+                onwire_count = --priv->ta_on_wire_txn_count;
+                break;
+            case TA_INFO_IN_MEMORY_SUCCESS:
+            case TA_INFO_IN_MEMORY_FAILED:
+                inmem_count = --priv->ta_in_mem_txn_count;
+                break;
+            case TA_WAIT_FOR_NOTIFY_LOCK_REL:
+                GF_ASSERT(0);
+                break;
+            case TA_SUCCESS:
+                break;
         }
-        UNLOCK (&frame->lock);
-
-        if (call_count == 0) {
-                if (local->transaction.resume_stub) {
-			call_resume (local->transaction.resume_stub);
-                        local->transaction.resume_stub = NULL;
-                }
+        release = priv->release_ta_notify_dom_lock;
+    }
+    UNLOCK(&priv->lock);
 
-                if (afr_lock_server_count (priv, local->transaction.type) == 0) {
-                        local->transaction.done (frame, this);
-                } else {
-                        int_lock->lock_cbk = local->transaction.done;
-                        afr_unlock (frame, this);
-                }
-        }
+    if (inmem_count != 0 || release == _gf_false || onwire_count != 0)
+        return;
 
-        return 0;
+    afr_ta_lock_release_synctask(this);
 }
 
-
-void
-afr_transaction_rm_stale_children (call_frame_t *frame, xlator_t *this,
-                                   inode_t *inode, afr_transaction_type type)
-{
-        int             i = -1;
-        int             count = 0;
-        int             read_child = -1;
-        afr_private_t   *priv = NULL;
-        afr_local_t     *local = NULL;
-        int             **pending = NULL;
-        int             idx = 0;
-        int32_t         *stale_children = NULL;
-        int32_t         *fresh_children = NULL;
-        gf_boolean_t    rm_stale_children = _gf_false;
-
-        idx = afr_index_for_transaction_type (type);
-
-        priv = this->private;
-        local = frame->local;
-        pending = local->pending;
-
-        if (local->op_ret < 0)
-                goto out;
-        fresh_children = local->fresh_children;
-        read_child = afr_inode_get_read_ctx (this, inode, fresh_children);
-        if (read_child < 0) {
-                gf_log (this->name, GF_LOG_DEBUG, "Possible split-brain "
-                        "for %s", uuid_utoa (inode->gfid));
-                goto out;
+static void
+afr_ta_process_onwireq(afr_ta_fop_state_t fop_state, xlator_t *this)
+{
+    afr_private_t *priv = this->private;
+    afr_local_t *entry = NULL;
+    int bad_child = AFR_CHILD_UNKNOWN;
+
+    struct list_head onwireq = {
+        0,
+    };
+    INIT_LIST_HEAD(&onwireq);
+
+    LOCK(&priv->lock);
+    {
+        bad_child = priv->ta_bad_child_index;
+        if (bad_child == AFR_CHILD_UNKNOWN) {
+            /*The previous on-wire ta_post_op was a failure. Just dequeue
+             *one element to wind on-wire again. */
+            entry = list_entry(priv->ta_onwireq.next, afr_local_t, ta_onwireq);
+            list_del_init(&entry->ta_onwireq);
+        } else {
+            /* Prepare to process all fops based on bad_child_index. */
+            list_splice_init(&priv->ta_onwireq, &onwireq);
         }
+    }
+    UNLOCK(&priv->lock);
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (!afr_is_child_present (fresh_children,
-                                           priv->child_count, i))
-                        continue;
-                if (pending[i][idx])
-                        continue;
-                /* child is down or op failed on it */
-                if (!stale_children)
-                        stale_children = afr_children_create (priv->child_count);
-                if (!stale_children)
-                        goto out;
-
-                rm_stale_children = _gf_true;
-                stale_children[count++] = i;
-                gf_log (this->name, GF_LOG_DEBUG, "Removing stale child "
-                        "%d for %s", i, uuid_utoa (inode->gfid));
+    if (entry) {
+        afr_ta_post_op_synctask(this, entry);
+        return;
+    } else {
+        while (!list_empty(&onwireq)) {
+            entry = list_entry(onwireq.next, afr_local_t, ta_onwireq);
+            list_del_init(&entry->ta_onwireq);
+            if (entry->ta_failed_subvol == bad_child) {
+                afr_post_op_handle_success(entry->transaction.frame, this);
+            } else {
+                afr_post_op_handle_failure(entry->transaction.frame, this, EIO);
+            }
         }
+    }
+}
 
-        if (!rm_stale_children)
-                goto out;
-
-        afr_inode_rm_stale_children (this, inode, stale_children);
-out:
-        GF_FREE (stale_children);
-        return;
+int
+afr_changelog_post_op_done(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_internal_lock_t *int_lock = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+    int_lock = &local->internal_lock;
+
+    if (priv->thin_arbiter_count) {
+        /*fop should not come here with TA_WAIT_FOR_NOTIFY_LOCK_REL state */
+        afr_ta_dom_lock_check_and_release(local->fop_state, this);
+    }
+
+    /* Fail the FOP if post-op did not succeed on quorum no. of bricks. */
+    if (!afr_changelog_has_quorum(local, this)) {
+        local->op_ret = -1;
+        /*local->op_errno is already captured in changelog cbk*/
+    }
+
+    if (local->transaction.resume_stub) {
+        call_resume(local->transaction.resume_stub);
+        local->transaction.resume_stub = NULL;
+    }
+
+    int_lock->lock_cbk = afr_transaction_done;
+    afr_unlock(frame, this);
+
+    return 0;
 }
 
-afr_inodelk_t*
-afr_get_inodelk (afr_internal_lock_t *int_lock, char *dom)
+static void
+afr_changelog_post_op_fail(call_frame_t *frame, xlator_t *this, int op_errno)
 {
-        afr_inodelk_t *inodelk = NULL;
-        int           i = 0;
+    afr_local_t *local = frame->local;
+    local->op_ret = -1;
+    local->op_errno = op_errno;
 
-        for (i = 0; int_lock->inodelk[i].domain; i++) {
-                inodelk = &int_lock->inodelk[i];
-                if (strcmp (dom, inodelk->domain) == 0)
-                        return inodelk;
-        }
-        return NULL;
+    gf_msg(this->name, GF_LOG_ERROR, op_errno, AFR_MSG_THIN_ARB,
+           "Failing %s for gfid %s. Fop state is:%d", gf_fop_list[local->op],
+           uuid_utoa(local->inode->gfid), local->fop_state);
+
+    afr_changelog_post_op_done(frame, this);
 }
 
-unsigned char*
-afr_locked_nodes_get (afr_transaction_type type, afr_internal_lock_t *int_lock)
+unsigned char *
+afr_locked_nodes_get(afr_transaction_type type, afr_internal_lock_t *int_lock)
 {
-        unsigned char *locked_nodes = NULL;
-        afr_inodelk_t *inodelk = NULL;
-        switch (type) {
-        case AFR_DATA_TRANSACTION:
-        case AFR_METADATA_TRANSACTION:
-                inodelk = afr_get_inodelk (int_lock, int_lock->domain);
-                locked_nodes = inodelk->locked_nodes;
-        break;
-
-        case AFR_ENTRY_TRANSACTION:
-        case AFR_ENTRY_RENAME_TRANSACTION:
-                /*Because same set of subvols participate in all lockee
-                 * entities*/
-                locked_nodes = int_lock->lockee[0].locked_nodes;
-        break;
-        }
-        return locked_nodes;
+    /*Because same set of subvols participate in all lockee
+     * entities*/
+    return int_lock->lockee[0].locked_nodes;
 }
 
 int
-afr_changelog_pre_op_call_count (afr_transaction_type type,
-                                 afr_internal_lock_t *int_lock,
-                                 unsigned int child_count)
+afr_changelog_call_count(afr_transaction_type type,
+                         unsigned char *pre_op_subvols,
+                         unsigned char *failed_subvols,
+                         unsigned int child_count)
 {
-        int           call_count = 0;
-        unsigned char *locked_nodes = NULL;
+    int i = 0;
+    int call_count = 0;
 
-        locked_nodes = afr_locked_nodes_get (type, int_lock);
-        GF_ASSERT (locked_nodes);
+    for (i = 0; i < child_count; i++) {
+        if (pre_op_subvols[i] && !failed_subvols[i]) {
+            call_count++;
+        }
+    }
 
-        call_count = afr_locked_children_count (locked_nodes, child_count);
-        if (type == AFR_ENTRY_RENAME_TRANSACTION)
-                call_count *= 2;
+    if (type == AFR_ENTRY_RENAME_TRANSACTION)
+        call_count *= 2;
 
-        return call_count;
+    return call_count;
 }
 
-int
-afr_changelog_post_op_call_count (afr_transaction_type type,
-                                  unsigned char *pre_op,
-                                  unsigned int child_count)
+gf_boolean_t
+afr_txn_nothing_failed(call_frame_t *frame, xlator_t *this)
 {
-        int           call_count = 0;
-
-        call_count = afr_pre_op_done_children_count (pre_op, child_count);
-        if (type == AFR_ENTRY_RENAME_TRANSACTION)
-                call_count *= 2;
-
-        return call_count;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int i = 0;
+
+    local = frame->local;
+    priv = this->private;
+
+    if (priv->thin_arbiter_count) {
+        /* We need to perform post-op even if 1 data brick was down
+         * before the txn started.*/
+        if (AFR_COUNT(local->transaction.failed_subvols, priv->child_count))
+            return _gf_false;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->transaction.pre_op[i] &&
+            local->transaction.failed_subvols[i])
+            return _gf_false;
+    }
+
+    return _gf_true;
 }
 
 void
-afr_compute_txn_changelog (afr_local_t *local, afr_private_t *priv)
+afr_handle_symmetric_errors(call_frame_t *frame, xlator_t *this)
 {
-        int             i = 0;
-        int             index = 0;
-        int32_t         postop = 0;
-        int32_t         preop = 1;
-        int32_t         **txn_changelog = NULL;
+    if (afr_is_symmetric_error(frame, this))
+        __mark_all_success(frame, this);
+}
 
-        txn_changelog = local->transaction.txn_changelog;
-        index = afr_index_for_transaction_type (local->transaction.type);
-        for (i = 0; i < priv->child_count; i++) {
-                postop = ntoh32 (local->pending[i][index]);
-                txn_changelog[i][index] = hton32 (postop + preop);
+gf_boolean_t
+afr_has_quorum(unsigned char *subvols, xlator_t *this, call_frame_t *frame)
+{
+    unsigned int quorum_count = 0;
+    afr_private_t *priv = NULL;
+    unsigned int up_children_count = 0;
+
+    priv = this->private;
+    up_children_count = AFR_COUNT(subvols, priv->child_count);
+
+    if (afr_lookup_has_quorum(frame, up_children_count))
+        return _gf_true;
+
+    if (priv->quorum_count == AFR_QUORUM_AUTO) {
+        /*
+         * Special case for auto-quorum with an even number of nodes.
+         *
+         * A replica set with even count N can only handle the same
+         * number of failures as odd N-1 before losing "vanilla"
+         * quorum, and the probability of more simultaneous failures is
+         * actually higher.  For example, with a 1% chance of failure
+         * we'd have a 0.03% chance of two simultaneous failures with
+         * N=3 but a 0.06% chance with N=4.  However, the special case
+         * is necessary for N=2 because there's no real quorum in that
+         * case (i.e. can't normally survive *any* failures).  In that
+         * case, we treat the first node as a tie-breaker, allowing
+         * quorum to be retained in some cases while still honoring the
+         * all-important constraint that there can not simultaneously
+         * be two partitioned sets of nodes each believing they have
+         * quorum.  Of two equally sized sets, the one without that
+         * first node will lose.
+         *
+         * It turns out that the special case is beneficial for higher
+         * values of N as well.  Continuing the example above, the
+         * probability of losing quorum with N=4 and this type of
+         * quorum is (very) slightly lower than with N=3 and vanilla
+         * quorum.  The difference becomes even more pronounced with
+         * higher N.  Therefore, even though such replica counts are
+         * unlikely to be seen in practice, we might as well use the
+         * "special" quorum then as well.
+         */
+        if ((up_children_count * 2) == priv->child_count) {
+            return subvols[0];
         }
+    }
+
+    if (priv->quorum_count == AFR_QUORUM_AUTO) {
+        quorum_count = priv->child_count / 2 + 1;
+    } else {
+        quorum_count = priv->quorum_count;
+    }
+
+    if (up_children_count >= quorum_count)
+        return _gf_true;
+
+    return _gf_false;
 }
 
-afr_xattrop_type_t
-afr_get_postop_xattrop_type (int32_t **pending, int optimized, int child,
-                             afr_transaction_type type)
+static gf_boolean_t
+afr_has_fop_quorum(call_frame_t *frame)
 {
-        int                     index = 0;
-        afr_xattrop_type_t      op = LOCAL_LAST;
+    xlator_t *this = frame->this;
+    afr_local_t *local = frame->local;
+    unsigned char *locked_nodes = NULL;
 
-        index = afr_index_for_transaction_type (type);
-        if (optimized && !pending[child][index])
-                op = LOCAL_FIRST;
-        return op;
+    locked_nodes = afr_locked_nodes_get(local->transaction.type,
+                                        &local->internal_lock);
+    return afr_has_quorum(locked_nodes, this, NULL);
 }
 
-void
-afr_set_postop_dict (afr_local_t *local, xlator_t *this, dict_t *xattr,
-                     int optimized, int child)
-{
-        int32_t                 **txn_changelog = NULL;
-        int32_t                 **changelog = NULL;
-        afr_private_t           *priv = NULL;
-        int                     ret = 0;
-        afr_xattrop_type_t      op = LOCAL_LAST;
-
-        priv = this->private;
-        txn_changelog = local->transaction.txn_changelog;
-        op = afr_get_postop_xattrop_type (local->pending, optimized, child,
-                                          local->transaction.type);
-        if (optimized)
-                changelog = txn_changelog;
-        else
-                changelog = local->pending;
-        ret = afr_set_pending_dict (priv, xattr, changelog, child, op);
-        if (ret < 0)
-                gf_log (this->name, GF_LOG_INFO,
-                        "failed to set pending entry");
+static gf_boolean_t
+afr_has_fop_cbk_quorum(call_frame_t *frame)
+{
+    afr_local_t *local = frame->local;
+    xlator_t *this = frame->this;
+    afr_private_t *priv = this->private;
+    unsigned char *success = alloca0(priv->child_count);
+    int i = 0;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->transaction.pre_op[i])
+            if (!local->transaction.failed_subvols[i])
+                success[i] = 1;
+    }
+
+    return afr_has_quorum(success, this, NULL);
 }
 
-
 gf_boolean_t
-afr_txn_nothing_failed (call_frame_t *frame, xlator_t *this)
+afr_need_dirty_marking(call_frame_t *frame, xlator_t *this)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t *local = NULL;
-        int index = -1;
-        int i = 0;
+    afr_private_t *priv = this->private;
+    afr_local_t *local = NULL;
+    gf_boolean_t need_dirty = _gf_false;
 
-        local = frame->local;
-        priv = this->private;
+    local = frame->local;
 
-        index = afr_index_for_transaction_type (local->transaction.type);
+    if (!priv->quorum_count || !local->optimistic_change_log)
+        return _gf_false;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->pending[i][index] == 0)
-                        return _gf_false;
-        }
+    if (local->transaction.type == AFR_DATA_TRANSACTION ||
+        local->transaction.type == AFR_METADATA_TRANSACTION)
+        return _gf_false;
 
-        return _gf_true;
+    if (AFR_COUNT(local->transaction.failed_subvols, priv->child_count) ==
+        priv->child_count)
+        return _gf_false;
+
+    if (!afr_has_fop_cbk_quorum(frame))
+        need_dirty = _gf_true;
+
+    return need_dirty;
 }
 
-static void
-afr_dir_fop_handle_all_fop_failures (call_frame_t *frame)
+void
+afr_handle_quorum(call_frame_t *frame, xlator_t *this)
 {
-        xlator_t        *this = NULL;
-        afr_local_t     *local = NULL;
-        afr_private_t   *priv = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    const char *file = NULL;
+    uuid_t gfid = {0};
 
-        this = frame->this;
-        local = frame->local;
-        priv = this->private;
+    local = frame->local;
+    priv = frame->this->private;
 
-        if ((local->transaction.type != AFR_ENTRY_TRANSACTION) &&
-            (local->transaction.type != AFR_ENTRY_RENAME_TRANSACTION))
-                return;
+    if (priv->quorum_count == 0)
+        return;
 
-        if (local->op_ret >= 0)
-                goto out;
+    /* If the fop already failed return right away to preserve errno */
+    if (local->op_ret == -1)
+        return;
 
-        __mark_all_success (local->pending, priv->child_count,
-                            local->transaction.type);
-out:
+    /*
+     * Network split may happen just after the fops are unwound, so check
+     * if the fop succeeded in a way it still follows quorum. If it doesn't,
+     * mark the fop as failure, mark the changelogs so it reflects that
+     * failure.
+     *
+     * Scenario:
+     * There are 3 mounts on 3 machines(node1, node2, node3) all writing to
+     * single file. Network split happened in a way that node1 can't see
+     * node2, node3. Node2, node3 both of them can't see node1. Now at the
+     * time of sending write all the bricks are up. Just after write fop is
+     * wound on node1, network split happens. Node1 thinks write fop failed
+     * on node2, node3 so marks pending changelog for those 2 extended
+     * attributes on node1. Node2, node3 thinks writes failed on node1 so
+     * they mark pending changelog for node1. When the network is stable
+     * again the file already is in split-brain. These checks prevent
+     * marking pending changelog on other subvolumes if the fop doesn't
+     * succeed in a way it is still following quorum. So with this fix what
+     * is happening is, node1 will have all pending changelog(FOOL) because
+     * the write succeeded only on node1 but failed on node2, node3 so
+     * instead of marking pending changelogs on node2, node3 it just treats
+     * the fop as failure and goes into DIRTY state. Where as node2, node3
+     * say they are sources and have pending changelog to node1 so there is
+     * no split-brain with the fix. The problem is eliminated completely.
+     */
+
+    if (afr_has_fop_cbk_quorum(frame))
         return;
+
+    if (afr_need_dirty_marking(frame, this))
+        goto set_response;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->transaction.pre_op[i])
+            afr_transaction_fop_failed(frame, frame->this, i);
+    }
+
+set_response:
+    local->op_ret = -1;
+    local->op_errno = afr_final_errno(local, priv);
+    if (local->op_errno == 0)
+        local->op_errno = afr_quorum_errno(priv);
+
+    if (local->fd) {
+        gf_uuid_copy(gfid, local->fd->inode->gfid);
+        file = uuid_utoa(gfid);
+    } else {
+        loc_path(&local->loc, local->loc.name);
+        file = local->loc.path;
+    }
+
+    gf_msg(frame->this->name, GF_LOG_WARNING, local->op_errno,
+           AFR_MSG_QUORUM_FAIL, "%s: Failing %s as quorum is not met", file,
+           gf_fop_list[local->op]);
+
+    switch (local->transaction.type) {
+        case AFR_ENTRY_TRANSACTION:
+        case AFR_ENTRY_RENAME_TRANSACTION:
+            afr_pick_error_xdata(local, priv, local->parent, local->readable,
+                                 local->parent2, local->readable2);
+            break;
+        default:
+            afr_pick_error_xdata(local, priv, local->inode, local->readable,
+                                 NULL, NULL);
+            break;
+    }
 }
 
-static void
-afr_data_handle_quota_errors (call_frame_t *frame, xlator_t *this)
+int
+afr_fill_ta_loc(xlator_t *this, loc_t *loc, gf_boolean_t is_gfid_based_fop)
 {
-        int     i = 0;
-        afr_private_t *priv = NULL;
-        afr_local_t   *local = NULL;
-        gf_boolean_t  all_quota_failures = _gf_false;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+    loc->parent = inode_ref(priv->root_inode);
+    gf_uuid_copy(loc->pargfid, loc->parent->gfid);
+    loc->name = priv->pending_key[THIN_ARBITER_BRICK_INDEX];
+    if (is_gfid_based_fop && gf_uuid_is_null(priv->ta_gfid)) {
+        /* Except afr_ta_id_file_check() which is path based, all other gluster
+         * FOPS need gfid.*/
+        return -EINVAL;
+    }
+    gf_uuid_copy(loc->gfid, priv->ta_gfid);
+    loc->inode = inode_new(loc->parent->table);
+    if (!loc->inode) {
+        loc_wipe(loc);
+        return -ENOMEM;
+    }
+    return 0;
+}
 
-        local = frame->local;
-        priv  = this->private;
-        if (local->transaction.type != AFR_DATA_TRANSACTION)
-                return;
-        /*
-         * Idea is to not leave the file in FOOL-FOOL scenario in case on
-         * all the bricks data transaction failed with EDQUOT to avoid
-         * increasing un-necessary load of self-heals in the system.
-         */
-        all_quota_failures = _gf_true;
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->transaction.pre_op[i] &&
-                    (local->child_errno[i] != EDQUOT)) {
-                        all_quota_failures = _gf_false;
-                        break;
-                }
-        }
-        if (all_quota_failures)
-                __mark_all_success (local->pending, priv->child_count,
-                                    local->transaction.type);
+static int
+afr_ta_post_op_done(int ret, call_frame_t *frame, void *opaque)
+{
+    xlator_t *this = NULL;
+    afr_local_t *local = NULL;
+    call_frame_t *txn_frame = NULL;
+    afr_ta_fop_state_t fop_state;
+
+    local = (afr_local_t *)opaque;
+    fop_state = local->fop_state;
+    txn_frame = local->transaction.frame;
+    this = frame->this;
+
+    if (ret == 0) {
+        /*Mark pending xattrs on the up data brick.*/
+        afr_post_op_handle_success(txn_frame, this);
+    } else {
+        afr_post_op_handle_failure(txn_frame, this, -ret);
+    }
+
+    STACK_DESTROY(frame->root);
+    afr_ta_process_onwireq(fop_state, this);
+
+    return 0;
 }
 
-int
-afr_changelog_post_op_now (call_frame_t *frame, xlator_t *this)
+int **
+afr_set_changelog_xattr(afr_private_t *priv, unsigned char *pending,
+                        dict_t *xattr, afr_local_t *local)
 {
-        afr_private_t * priv = this->private;
-        afr_internal_lock_t *int_lock = NULL;
-        int i          = 0;
-        int call_count = 0;
+    int **changelog = NULL;
+    int idx = 0;
+    int ret = 0;
+    int i;
+
+    if (local->is_new_entry == _gf_true) {
+        changelog = afr_mark_pending_changelog(priv, pending, xattr,
+                                               local->cont.dir_fop.buf.ia_type);
+    } else {
+        idx = afr_index_for_transaction_type(local->transaction.type);
+        changelog = afr_matrix_create(priv->child_count, AFR_NUM_CHANGE_LOGS);
+        if (!changelog) {
+            goto out;
+        }
+        for (i = 0; i < priv->child_count; i++) {
+            if (local->transaction.failed_subvols[i])
+                changelog[i][idx] = hton32(1);
+        }
+        ret = afr_set_pending_dict(priv, xattr, changelog);
+        if (ret < 0) {
+            afr_matrix_cleanup(changelog, priv->child_count);
+            return NULL;
+        }
+    }
 
-        afr_local_t *  local = NULL;
-        afr_fd_ctx_t  *fdctx = NULL;
-        dict_t        **xattr = NULL;
-        int            piggyback = 0;
-        int            nothing_failed = 1;
+out:
+    return changelog;
+}
 
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+static void
+afr_ta_locked_xattrop_validate(afr_private_t *priv, afr_local_t *local,
+                               gf_boolean_t *valid)
+{
+    if (priv->ta_event_gen > local->ta_event_gen) {
+        /* We can't trust the ta's response anymore.*/
+        afr_ta_locked_priv_invalidate(priv);
+        *valid = _gf_false;
+        return;
+    }
+    return;
+}
 
-        __mark_non_participant_children (local->pending, priv->child_count,
-                                         local->transaction.pre_op,
-                                         local->transaction.type);
+static int
+afr_ta_post_op_do(void *opaque)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    xlator_t *this = NULL;
+    dict_t *xattr = NULL;
+    unsigned char *pending = NULL;
+    int **changelog = NULL;
+    int failed_subvol = -1;
+    int success_subvol = -1;
+    loc_t loc = {
+        0,
+    };
+    int i = 0;
+    int ret = 0;
+    gf_boolean_t valid = _gf_true;
+
+    local = (afr_local_t *)opaque;
+    this = local->transaction.frame->this;
+    priv = this->private;
+
+    ret = afr_fill_ta_loc(this, &loc, _gf_true);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed to populate loc for thin-arbiter.");
+        goto out;
+    }
+
+    xattr = dict_new();
+    if (!xattr) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    pending = alloca0(priv->child_count);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->transaction.failed_subvols[i]) {
+            pending[i] = 1;
+            failed_subvol = i;
+        } else {
+            success_subvol = i;
+        }
+    }
+
+    changelog = afr_set_changelog_xattr(priv, pending, xattr, local);
+
+    if (!changelog) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    ret = afr_ta_post_op_lock(this, &loc);
+    if (ret)
+        goto out;
+
+    ret = syncop_xattrop(priv->children[THIN_ARBITER_BRICK_INDEX], &loc,
+                         GF_XATTROP_ADD_ARRAY, xattr, NULL, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Post-op on thin-arbiter id file %s failed for gfid %s.",
+               priv->pending_key[THIN_ARBITER_BRICK_INDEX],
+               uuid_utoa(local->inode->gfid));
+    }
+    LOCK(&priv->lock);
+    {
+        if (ret == 0) {
+            priv->ta_bad_child_index = failed_subvol;
+        } else if (ret == -EINVAL) {
+            priv->ta_bad_child_index = success_subvol;
+            ret = -EIO; /* TA failed the fop. Return EIO to application. */
+        }
 
-        afr_data_handle_quota_errors (frame, this);
-        afr_dir_fop_handle_all_fop_failures (frame);
+        afr_ta_locked_xattrop_validate(priv, local, &valid);
+    }
+    UNLOCK(&priv->lock);
+    if (valid == _gf_false) {
+        gf_msg(this->name, GF_LOG_ERROR, EIO, AFR_MSG_THIN_ARB,
+               "Post-op on thin-arbiter id file %s for gfid %s invalidated due "
+               "to event-gen mismatch.",
+               priv->pending_key[THIN_ARBITER_BRICK_INDEX],
+               uuid_utoa(local->inode->gfid));
+        ret = -EIO;
+    }
+
+    afr_ta_post_op_unlock(this, &loc);
+out:
+    if (xattr)
+        dict_unref(xattr);
 
-        if (local->fd)
-                afr_transaction_rm_stale_children (frame, this,
-                                                   local->fd->inode,
-                                                   local->transaction.type);
+    if (changelog)
+        afr_matrix_cleanup(changelog, priv->child_count);
 
-        xattr = alloca (priv->child_count * sizeof (*xattr));
-        memset (xattr, 0, (priv->child_count * sizeof (*xattr)));
-        for (i = 0; i < priv->child_count; i++) {
-                xattr[i] = dict_new ();
-        }
+    loc_wipe(&loc);
 
-        call_count = afr_changelog_post_op_call_count (local->transaction.type,
-                                                       local->transaction.pre_op,
-                                                       priv->child_count);
-        local->call_count = call_count;
+    return ret;
+}
 
-        if (local->fd)
-                fdctx = afr_fd_ctx_get (local->fd, this);
+static int
+afr_ta_post_op_synctask(xlator_t *this, afr_local_t *local)
+{
+    call_frame_t *ta_frame = NULL;
+    int ret = 0;
+
+    ta_frame = afr_ta_frame_create(this);
+    if (!ta_frame) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_THIN_ARB,
+               "Failed to create ta_frame");
+        goto err;
+    }
+    ret = synctask_new(this->ctx->env, afr_ta_post_op_do, afr_ta_post_op_done,
+                       ta_frame, local);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_THIN_ARB,
+               "Failed to launch post-op on thin arbiter for gfid %s",
+               uuid_utoa(local->inode->gfid));
+        STACK_DESTROY(ta_frame->root);
+        goto err;
+    }
+
+    return ret;
+err:
+    afr_changelog_post_op_fail(local->transaction.frame, this, ENOMEM);
+    return ret;
+}
 
-        if (call_count == 0) {
-                /* no child is up */
-                int_lock->lock_cbk = local->transaction.done;
-                afr_unlock (frame, this);
-                goto out;
+static void
+afr_ta_set_fop_state(afr_private_t *priv, afr_local_t *local,
+                     int *on_wire_count)
+{
+    LOCK(&priv->lock);
+    {
+        if (priv->release_ta_notify_dom_lock == _gf_true) {
+            /* Put the fop in waitq until notify dom lock is released.*/
+            local->fop_state = TA_WAIT_FOR_NOTIFY_LOCK_REL;
+            list_add_tail(&local->ta_waitq, &priv->ta_waitq);
+        } else if (priv->ta_bad_child_index == AFR_CHILD_UNKNOWN) {
+            /* Post-op on thin-arbiter to decide success/failure. */
+            local->fop_state = TA_GET_INFO_FROM_TA_FILE;
+            *on_wire_count = ++priv->ta_on_wire_txn_count;
+            if (*on_wire_count > 1) {
+                /*Avoid sending multiple on-wire post-ops on TA*/
+                list_add_tail(&local->ta_onwireq, &priv->ta_onwireq);
+            }
+        } else if (local->ta_failed_subvol == priv->ta_bad_child_index) {
+            /* Post-op on TA not needed as the fop failed on the in-memory bad
+             * brick. Just mark pending xattrs on the good data brick.*/
+            local->fop_state = TA_INFO_IN_MEMORY_SUCCESS;
+            priv->ta_in_mem_txn_count++;
+        } else {
+            /* Post-op on TA not needed as the fop succeeded only on the
+             * in-memory bad data brick and not the good one. Fail the fop.*/
+            local->fop_state = TA_INFO_IN_MEMORY_FAILED;
+            priv->ta_in_mem_txn_count++;
         }
+    }
+    UNLOCK(&priv->lock);
+}
 
-        nothing_failed = afr_txn_nothing_failed (frame, this);
+static void
+afr_ta_fill_failed_subvol(afr_private_t *priv, afr_local_t *local)
+{
+    int i = 0;
 
-        afr_compute_txn_changelog (local , priv);
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->transaction.failed_subvols[i]) {
+            local->ta_failed_subvol = i;
+            break;
+        }
+    }
+}
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (!local->transaction.pre_op[i])
-                        continue;
+static void
+afr_post_op_handle_success(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
 
-                if (local->transaction.type != AFR_DATA_TRANSACTION)
-                        afr_set_postop_dict (local, this, xattr[i],
-                                             local->optimistic_change_log, i);
-                switch (local->transaction.type) {
-                case AFR_DATA_TRANSACTION:
-                {
-                        if (!fdctx) {
-                                afr_set_postop_dict (local, this, xattr[i],
-                                                     0, i);
-                                STACK_WIND (frame, afr_changelog_post_op_cbk,
-                                            priv->children[i],
-                                            priv->children[i]->fops->xattrop,
-                                            &local->loc,
-                                            GF_XATTROP_ADD_ARRAY, xattr[i],
-                                            NULL);
-                                break;
-                        }
-
-                        /* local->transaction.postop_piggybacked[] was
-                           precomputed in is_piggyback_postop() when called from
-                           afr_changelog_post_op_safe()
-                        */
-
-                        piggyback = 0;
-                        if (local->transaction.postop_piggybacked[i])
-                                piggyback = 1;
-
-                        afr_set_postop_dict (local, this, xattr[i],
-                                             piggyback, i);
-
-                        if (nothing_failed && piggyback) {
-                                afr_changelog_post_op_cbk (frame, (void *)(long)i,
-                                                           this, 1, 0, xattr[i], NULL);
-                        } else {
-                                STACK_WIND_COOKIE (frame,
-                                                   afr_changelog_post_op_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->fxattrop,
-                                                   local->fd,
-                                                   GF_XATTROP_ADD_ARRAY, xattr[i],
-                                                   NULL);
-                        }
-                }
-                break;
-                case AFR_METADATA_TRANSACTION:
-                {
-                        if (nothing_failed && local->optimistic_change_log) {
-                                afr_changelog_post_op_cbk (frame, (void *)(long)i,
-                                                           this, 1, 0, xattr[i],
-                                                           NULL);
-                                break;
-                        }
-
-                        if (local->fd)
-                                STACK_WIND (frame, afr_changelog_post_op_cbk,
-                                            priv->children[i],
-                                            priv->children[i]->fops->fxattrop,
-                                            local->fd,
-                                            GF_XATTROP_ADD_ARRAY, xattr[i],
-                                            NULL);
-                        else
-                                STACK_WIND (frame, afr_changelog_post_op_cbk,
-                                            priv->children[i],
-                                            priv->children[i]->fops->xattrop,
-                                            &local->loc,
-                                            GF_XATTROP_ADD_ARRAY, xattr[i],
-                                            NULL);
-                }
-                break;
+    local = frame->local;
+    if (local->is_new_entry == _gf_true) {
+        afr_mark_new_entry_changelog(frame, this);
+    }
+    afr_changelog_post_op_do(frame, this);
 
-                case AFR_ENTRY_RENAME_TRANSACTION:
-                {
-                        if (nothing_failed && local->optimistic_change_log) {
-                                afr_changelog_post_op_cbk (frame, (void *)(long)i,
-                                                           this, 1, 0, xattr[i],
-                                                           NULL);
-                        } else {
-                                STACK_WIND_COOKIE (frame, afr_changelog_post_op_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->xattrop,
-                                                   &local->transaction.new_parent_loc,
-                                                   GF_XATTROP_ADD_ARRAY, xattr[i],
-                                                   NULL);
-                        }
-                        call_count--;
-                }
+    return;
+}
 
-                /*
-                  set it again because previous stack_wind
-                  might have already returned (think of case
-                  where subvolume is posix) and would have
-                  used the dict as placeholder for return
-                  value
-                */
+static void
+afr_post_op_handle_failure(call_frame_t *frame, xlator_t *this, int op_errno)
+{
+    afr_changelog_post_op_fail(frame, this, op_errno);
 
-                afr_set_postop_dict (local, this, xattr[i],
-                                     local->optimistic_change_log, i);
+    return;
+}
 
-                /* fall through */
+static void
+afr_ta_decide_post_op_state(call_frame_t *frame, xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int on_wire_count = 0;
+
+    priv = this->private;
+    local = frame->local;
+
+    afr_ta_set_fop_state(priv, local, &on_wire_count);
+
+    switch (local->fop_state) {
+        case TA_GET_INFO_FROM_TA_FILE:
+            if (on_wire_count == 1)
+                afr_ta_post_op_synctask(this, local);
+            /*else, fop is queued in ta_onwireq.*/
+            break;
+        case TA_WAIT_FOR_NOTIFY_LOCK_REL:
+            /*Post releasing the notify lock, we will act on this queue*/
+            break;
+        case TA_INFO_IN_MEMORY_SUCCESS:
+            afr_post_op_handle_success(frame, this);
+            break;
+        case TA_INFO_IN_MEMORY_FAILED:
+            afr_post_op_handle_failure(frame, this, EIO);
+            break;
+        default:
+            break;
+    }
+    return;
+}
 
-                case AFR_ENTRY_TRANSACTION:
-                {
-                        if (nothing_failed && local->optimistic_change_log) {
-                                afr_changelog_post_op_cbk (frame, (void *)(long)i,
-                                                           this, 1, 0, xattr[i],
-                                                           NULL);
-                                break;
-                        }
-
-                        if (local->fd)
-                                STACK_WIND (frame, afr_changelog_post_op_cbk,
-                                            priv->children[i],
-                                            priv->children[i]->fops->fxattrop,
-                                            local->fd,
-                                            GF_XATTROP_ADD_ARRAY, xattr[i],
-                                            NULL);
-                        else
-                                STACK_WIND (frame, afr_changelog_post_op_cbk,
-                                            priv->children[i],
-                                            priv->children[i]->fops->xattrop,
-                                            &local->transaction.parent_loc,
-                                            GF_XATTROP_ADD_ARRAY, xattr[i],
-                                            NULL);
-                }
-                break;
-                }
+static void
+afr_handle_failure_using_thin_arbiter(call_frame_t *frame, xlator_t *this)
+{
+    afr_private_t *priv = this->private;
+    afr_local_t *local = frame->local;
+
+    afr_ta_fill_failed_subvol(priv, local);
+    gf_msg_debug(this->name, 0,
+                 "Fop failed on data brick (%s) for gfid=%s. "
+                 "ta info needed to decide fop result.",
+                 priv->children[local->ta_failed_subvol]->name,
+                 uuid_utoa(local->inode->gfid));
+    afr_ta_decide_post_op_state(frame, this);
+}
 
-                if (!--call_count)
-                        break;
+void
+afr_changelog_post_op_do(call_frame_t *frame, xlator_t *this)
+{
+    afr_private_t *priv = this->private;
+    afr_local_t *local = NULL;
+    dict_t *xattr = NULL;
+    int i = 0;
+    int ret = 0;
+    int idx = 0;
+    int nothing_failed = 1;
+    gf_boolean_t need_undirty = _gf_false;
+
+    afr_handle_quorum(frame, this);
+    local = frame->local;
+    idx = afr_index_for_transaction_type(local->transaction.type);
+
+    xattr = dict_new();
+    if (!xattr) {
+        afr_changelog_post_op_fail(frame, this, ENOMEM);
+        goto out;
+    }
+
+    nothing_failed = afr_txn_nothing_failed(frame, this);
+
+    if (afr_changelog_pre_op_uninherit(frame, this))
+        need_undirty = _gf_false;
+    else
+        need_undirty = _gf_true;
+
+    if (local->op_ret < 0 && !nothing_failed) {
+        if (afr_need_dirty_marking(frame, this)) {
+            local->dirty[idx] = hton32(1);
+            goto set_dirty;
         }
 
+        afr_changelog_post_op_done(frame, this);
+        goto out;
+    }
+
+    if (nothing_failed && !need_undirty) {
+        afr_changelog_post_op_done(frame, this);
+        goto out;
+    }
+
+    if (local->transaction.in_flight_sb) {
+        afr_changelog_post_op_fail(frame, this,
+                                   local->transaction.in_flight_sb_errno);
+        goto out;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->transaction.failed_subvols[i])
+            local->pending[i][idx] = hton32(1);
+    }
+
+    ret = afr_set_pending_dict(priv, xattr, local->pending);
+    if (ret < 0) {
+        afr_changelog_post_op_fail(frame, this, ENOMEM);
+        goto out;
+    }
+
+    if (need_undirty)
+        local->dirty[idx] = hton32(-1);
+    else
+        local->dirty[idx] = hton32(0);
+
+set_dirty:
+    ret = dict_set_static_bin(xattr, AFR_DIRTY, local->dirty,
+                              sizeof(int) * AFR_NUM_CHANGE_LOGS);
+    if (ret) {
+        afr_changelog_post_op_fail(frame, this, ENOMEM);
+        goto out;
+    }
+
+    afr_changelog_do(frame, this, xattr, afr_changelog_post_op_done,
+                     AFR_TRANSACTION_POST_OP);
 out:
-        for (i = 0; i < priv->child_count; i++) {
-                dict_unref (xattr[i]);
-        }
+    if (xattr)
+        dict_unref(xattr);
 
-        return 0;
+    return;
 }
 
-
-int32_t
-afr_changelog_pre_op_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                          int32_t op_ret, int32_t op_errno, dict_t *xattr,
-                          dict_t *xdata)
+static int
+afr_changelog_post_op_now(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = this->private;
-        int call_count  = -1;
-        int child_index = (long) cookie;
-
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                switch (op_ret) {
-                case 0:
-                        __mark_pre_op_done_on_fd (frame, this, child_index);
-                        //fallthrough we need to mark the pre_op
-                case 1:
-                        local->transaction.pre_op[child_index] = 1;
-                        /* special op_ret for piggyback */
-                        break;
-                case -1:
-                        if (op_errno == ENOTSUP) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "xattrop not supported by %s",
-                                        priv->children[child_index]->name);
-                                local->op_ret = -1;
-
-                        } else if (!child_went_down (op_ret, op_errno)) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "xattrop failed on child %s: %s",
-                                        priv->children[child_index]->name,
-                                        strerror (op_errno));
-                        }
-                        local->op_errno = op_errno;
-                        break;
-                }
-
-                call_count = --local->call_count;
-        }
-        UNLOCK (&frame->lock);
-
-        if (call_count == 0) {
-                if ((local->op_ret == -1) &&
-                    (local->op_errno == ENOTSUP)) {
-                        local->transaction.resume (frame, this);
-                } else {
-                        afr_transaction_perform_fop (frame, this);
-                }
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int failed_count = 0;
+
+    priv = this->private;
+    local = frame->local;
+
+    if (priv->thin_arbiter_count) {
+        failed_count = AFR_COUNT(local->transaction.failed_subvols,
+                                 priv->child_count);
+        if (failed_count == 1) {
+            afr_handle_failure_using_thin_arbiter(frame, this);
+            return 0;
+        } else {
+            /* Txn either succeeded or failed on both data bricks. Let
+             * post_op_do handle it as the case might be. */
         }
+    }
 
-        return 0;
+    afr_changelog_post_op_do(frame, this);
+    return 0;
 }
 
-int
-afr_changelog_pre_op (call_frame_t *frame, xlator_t *this)
+gf_boolean_t
+afr_changelog_pre_op_uninherit(call_frame_t *frame, xlator_t *this)
 {
-        afr_private_t * priv = this->private;
-        int i = 0;
-        int ret = 0;
-        int call_count = 0;
-        dict_t **xattr = NULL;
-        afr_fd_ctx_t *fdctx = NULL;
-        afr_local_t *local = NULL;
-        int          piggyback = 0;
-        afr_internal_lock_t *int_lock = NULL;
-        unsigned char       *locked_nodes = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    afr_inode_ctx_t *ctx = NULL;
+    int i = 0;
+    gf_boolean_t ret = _gf_false;
+    int type = 0;
+
+    local = frame->local;
+    priv = this->private;
+    ctx = local->inode_ctx;
+
+    type = afr_index_for_transaction_type(local->transaction.type);
+    if (type != AFR_DATA_TRANSACTION)
+        return !local->transaction.dirtied;
+
+    if (local->transaction.no_uninherit)
+        return _gf_false;
 
-        local = frame->local;
-        int_lock = &local->internal_lock;
+    /* This function must be idempotent. So check if we
+       were called before and return the same answer again.
 
-        xattr = alloca (priv->child_count * sizeof (*xattr));
-        memset (xattr, 0, (priv->child_count * sizeof (*xattr)));
+       It is important to keep this function idempotent for
+       the call in afr_changelog_post_op_safe() to not have
+       side effects on the call from afr_changelog_post_op_now()
+    */
+    if (local->transaction.uninherit_done)
+        return local->transaction.uninherit_value;
 
+    LOCK(&local->inode->lock);
+    {
         for (i = 0; i < priv->child_count; i++) {
-                xattr[i] = dict_new ();
+            if (local->transaction.pre_op[i] != ctx->pre_op_done[type][i]) {
+                ret = !local->transaction.dirtied;
+                goto unlock;
+            }
         }
 
-        call_count = afr_changelog_pre_op_call_count (local->transaction.type,
-                                                      int_lock,
-                                                      priv->child_count);
-        if (call_count == 0) {
-                local->internal_lock.lock_cbk =
-                        local->transaction.done;
-                afr_unlock (frame, this);
-                goto out;
+        if (ctx->inherited[type]) {
+            ret = _gf_true;
+            ctx->inherited[type]--;
+        } else if (ctx->on_disk[type]) {
+            ret = _gf_false;
+            ctx->on_disk[type]--;
+        } else {
+            /* ASSERT */
+            ret = _gf_false;
         }
 
-        local->call_count = call_count;
-
-        __mark_all_pending (local->pending, priv->child_count,
-                            local->transaction.type);
-
-        if (local->fd)
-                fdctx = afr_fd_ctx_get (local->fd, this);
-
-        locked_nodes = afr_locked_nodes_get (local->transaction.type, int_lock);
-        for (i = 0; i < priv->child_count; i++) {
-                if (!locked_nodes[i])
-                        continue;
-                ret = afr_set_pending_dict (priv, xattr[i], local->pending,
-                                            i, LOCAL_FIRST);
-
-                if (ret < 0)
-                        gf_log (this->name, GF_LOG_INFO,
-                                "failed to set pending entry");
-
-
-                switch (local->transaction.type) {
-                case AFR_DATA_TRANSACTION:
-                {
-                        if (!fdctx) {
-                                STACK_WIND_COOKIE (frame,
-                                                   afr_changelog_pre_op_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->xattrop,
-                                                   &(local->loc),
-                                                   GF_XATTROP_ADD_ARRAY, xattr[i],
-                                                   NULL);
-                                break;
-                        }
-
-                        LOCK (&local->fd->lock);
-                        {
-                                piggyback = 0;
-                                if (fdctx->pre_op_done[i]) {
-                                        fdctx->pre_op_piggyback[i]++;
-                                        piggyback = 1;
-                                        fdctx->hit++;
-                                } else {
-                                        fdctx->miss++;
-                                }
-                        }
-                        UNLOCK (&local->fd->lock);
-
-                        afr_set_delayed_post_op (frame, this);
-
-                        if (piggyback)
-                                afr_changelog_pre_op_cbk (frame, (void *)(long)i,
-                                                          this, 1, 0, xattr[i],
-                                                          NULL);
-                        else
-                                STACK_WIND_COOKIE (frame,
-                                                   afr_changelog_pre_op_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->fxattrop,
-                                                   local->fd,
-                                                   GF_XATTROP_ADD_ARRAY, xattr[i],
-                                                   NULL);
-                }
-                break;
-                case AFR_METADATA_TRANSACTION:
-                {
-                        if (local->optimistic_change_log) {
-                                afr_changelog_pre_op_cbk (frame, (void *)(long)i,
-                                                          this, 1, 0, xattr[i],
-                                                          NULL);
-                                break;
-                        }
-
-                        if (local->fd)
-                                STACK_WIND_COOKIE (frame,
-                                                   afr_changelog_pre_op_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->fxattrop,
-                                                   local->fd,
-                                                   GF_XATTROP_ADD_ARRAY, xattr[i],
-                                                   NULL);
-                        else
-                                STACK_WIND_COOKIE (frame,
-                                                   afr_changelog_pre_op_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->xattrop,
-                                                   &(local->loc),
-                                                   GF_XATTROP_ADD_ARRAY, xattr[i],
-                                                   NULL);
-                }
-                break;
-
-                case AFR_ENTRY_RENAME_TRANSACTION:
-                {
-                        if (local->optimistic_change_log) {
-                                afr_changelog_pre_op_cbk (frame, (void *)(long)i,
-                                                          this, 1, 0, xattr[i],
-                                                          NULL);
-                        } else {
-                                STACK_WIND_COOKIE (frame,
-                                                   afr_changelog_pre_op_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->xattrop,
-                                                   &local->transaction.new_parent_loc,
-                                                   GF_XATTROP_ADD_ARRAY, xattr[i],
-                                                   NULL);
-                        }
-
-                        call_count--;
-                }
+        if (!ctx->inherited[type] && !ctx->on_disk[type]) {
+            for (i = 0; i < priv->child_count; i++)
+                ctx->pre_op_done[type][i] = 0;
+        }
+    }
+unlock:
+    UNLOCK(&local->inode->lock);
 
+    local->transaction.uninherit_done = _gf_true;
+    local->transaction.uninherit_value = ret;
 
-                /*
-                  set it again because previous stack_wind
-                  might have already returned (think of case
-                  where subvolume is posix) and would have
-                  used the dict as placeholder for return
-                  value
-                */
+    return ret;
+}
 
-                ret = afr_set_pending_dict (priv, xattr[i], local->pending,
-                                            i, LOCAL_FIRST);
+gf_boolean_t
+afr_changelog_pre_op_inherit(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    gf_boolean_t ret = _gf_false;
+    int type = 0;
 
-                if (ret < 0)
-                        gf_log (this->name, GF_LOG_INFO,
-                                "failed to set pending entry");
+    local = frame->local;
+    priv = this->private;
 
-                /* fall through */
+    if (local->transaction.type != AFR_DATA_TRANSACTION)
+        return _gf_false;
 
-                case AFR_ENTRY_TRANSACTION:
-                {
-                        if (local->optimistic_change_log) {
-                                afr_changelog_pre_op_cbk (frame, (void *)(long)i,
-                                                          this, 1, 0, xattr[i],
-                                                          NULL);
-                                break;
-                        }
-
-                        if (local->fd)
-                                STACK_WIND_COOKIE (frame,
-                                                   afr_changelog_pre_op_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->fxattrop,
-                                                   local->fd,
-                                                   GF_XATTROP_ADD_ARRAY, xattr[i],
-                                                   NULL);
-                        else
-                                STACK_WIND_COOKIE (frame,
-                                                   afr_changelog_pre_op_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->xattrop,
-                                                   &local->transaction.parent_loc,
-                                                   GF_XATTROP_ADD_ARRAY, xattr[i],
-                                                   NULL);
-                }
-                break;
-                }
+    type = afr_index_for_transaction_type(local->transaction.type);
 
-                if (!--call_count)
-                        break;
+    LOCK(&local->inode->lock);
+    {
+        if (!local->inode_ctx->on_disk[type]) {
+            /* nothing to inherit yet */
+            ret = _gf_false;
+            goto unlock;
         }
-out:
+
         for (i = 0; i < priv->child_count; i++) {
-                dict_unref (xattr[i]);
+            if (local->transaction.pre_op[i] !=
+                local->inode_ctx->pre_op_done[type][i]) {
+                /* either inherit exactly, or don't */
+                ret = _gf_false;
+                goto unlock;
+            }
         }
 
-        return 0;
-}
+        local->inode_ctx->inherited[type]++;
 
+        ret = _gf_true;
 
-int
-afr_post_blocking_inodelk_cbk (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "Blocking inodelks failed.");
-                local->transaction.done (frame, this);
-        } else {
-
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Blocking inodelks done. Proceeding to FOP");
-                afr_internal_lock_finish (frame, this);
-        }
+        local->transaction.inherited = _gf_true;
+    }
+unlock:
+    UNLOCK(&local->inode->lock);
 
-        return 0;
+    return ret;
 }
 
-
-int
-afr_post_nonblocking_inodelk_cbk (call_frame_t *frame, xlator_t *this)
+gf_boolean_t
+afr_changelog_pre_op_update(call_frame_t *frame, xlator_t *this)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    gf_boolean_t ret = _gf_false;
+    int type = 0;
 
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+    local = frame->local;
+    priv = this->private;
 
-        /* Initiate blocking locks if non-blocking has failed */
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Non blocking inodelks failed. Proceeding to blocking");
-                int_lock->lock_cbk = afr_post_blocking_inodelk_cbk;
-                afr_blocking_lock (frame, this);
-        } else {
+    if (local->transaction.type == AFR_ENTRY_TRANSACTION ||
+        local->transaction.type == AFR_ENTRY_RENAME_TRANSACTION)
+        return _gf_false;
 
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Non blocking inodelks done. Proceeding to FOP");
-                afr_internal_lock_finish (frame, this);
-        }
+    if (local->transaction.inherited)
+        /* was already inherited in afr_changelog_pre_op */
+        return _gf_false;
 
-        return 0;
-}
+    if (!local->transaction.dirtied)
+        return _gf_false;
 
+    if (!afr_txn_nothing_failed(frame, this))
+        return _gf_false;
 
-int
-afr_post_blocking_entrylk_cbk (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
+    type = afr_index_for_transaction_type(local->transaction.type);
 
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+    ret = _gf_false;
 
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "Blocking entrylks failed.");
-                local->transaction.done (frame, this);
+    LOCK(&local->inode->lock);
+    {
+        if (!local->inode_ctx->on_disk[type]) {
+            for (i = 0; i < priv->child_count; i++)
+                local->inode_ctx->pre_op_done[type][i] =
+                    (!local->transaction.failed_subvols[i]);
         } else {
-
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Blocking entrylks done. Proceeding to FOP");
-                afr_internal_lock_finish (frame, this);
+            for (i = 0; i < priv->child_count; i++)
+                if (local->inode_ctx->pre_op_done[type][i] !=
+                    (!local->transaction.failed_subvols[i])) {
+                    local->transaction.no_uninherit = 1;
+                    goto unlock;
+                }
         }
+        local->inode_ctx->on_disk[type]++;
 
-        return 0;
-}
+        ret = _gf_true;
+    }
+unlock:
+    UNLOCK(&local->inode->lock);
 
+    return ret;
+}
 
 int
-afr_post_nonblocking_entrylk_cbk (call_frame_t *frame, xlator_t *this)
+afr_changelog_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                  int op_errno, dict_t *xattr, dict_t *xdata)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
+    afr_local_t *local = NULL;
+    int call_count = -1;
+    int child_index = -1;
 
-        local = frame->local;
-        int_lock = &local->internal_lock;
+    local = frame->local;
+    child_index = (long)cookie;
 
-        /* Initiate blocking locks if non-blocking has failed */
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Non blocking entrylks failed. Proceeding to blocking");
-                int_lock->lock_cbk = afr_post_blocking_entrylk_cbk;
-                afr_blocking_lock (frame, this);
-        } else {
+    if (op_ret == -1) {
+        local->op_errno = op_errno;
+        afr_transaction_fop_failed(frame, this, child_index);
+    }
 
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Non blocking entrylks done. Proceeding to FOP");
-                afr_internal_lock_finish (frame, this);
-        }
+    if (xattr)
+        local->transaction.changelog_xdata[child_index] = dict_ref(xattr);
 
-        return 0;
-}
+    call_count = afr_frame_return(frame);
 
+    if (call_count == 0) {
+        local->transaction.changelog_resume(frame, this);
+    }
 
-int
-afr_post_blocking_rename_cbk (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+    return 0;
+}
 
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "Blocking entrylks failed.");
-                local->transaction.done (frame, this);
-        } else {
+void
+afr_changelog_populate_xdata(call_frame_t *frame, afr_xattrop_type_t op,
+                             dict_t **xdata, dict_t **newloc_xdata)
+{
+    int i = 0;
+    int ret = 0;
+    char *key = NULL;
+    int keylen = 0;
+    const char *name = NULL;
+    dict_t *xdata1 = NULL;
+    dict_t *xdata2 = NULL;
+    xlator_t *this = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    gf_boolean_t need_entry_key_set = _gf_true;
+
+    local = frame->local;
+    this = THIS;
+    priv = this->private;
+
+    if (local->transaction.type == AFR_DATA_TRANSACTION ||
+        local->transaction.type == AFR_METADATA_TRANSACTION)
+        goto out;
+
+    if (!priv->esh_granular)
+        goto out;
+
+    xdata1 = dict_new();
+    if (!xdata1)
+        goto out;
+
+    name = local->loc.name;
+    if (local->op == GF_FOP_LINK)
+        name = local->newloc.name;
+
+    switch (op) {
+        case AFR_TRANSACTION_PRE_OP:
+            key = GF_XATTROP_ENTRY_IN_KEY;
+            break;
+        case AFR_TRANSACTION_POST_OP:
+            if (afr_txn_nothing_failed(frame, this)) {
+                key = GF_XATTROP_ENTRY_OUT_KEY;
+                for (i = 0; i < priv->child_count; i++) {
+                    if (!local->transaction.failed_subvols[i])
+                        continue;
+                    need_entry_key_set = _gf_false;
+                    break;
+                }
+                /* If the transaction itself did not fail and there
+                 * are no failed subvolumes, check whether the fop
+                 * failed due to a symmetric error. If it did, do
+                 * not set the ENTRY_OUT xattr which would end up
+                 * deleting a name index which was created possibly by
+                 * an earlier entry txn that may have failed on some
+                 * of the sub-volumes.
+                 */
+                if (local->op_ret)
+                    need_entry_key_set = _gf_false;
+            } else {
+                key = GF_XATTROP_ENTRY_IN_KEY;
+            }
+            break;
+    }
+
+    if (need_entry_key_set) {
+        keylen = strlen(key);
+        ret = dict_set_strn(xdata1, key, keylen, (char *)name);
+        if (ret)
+            gf_msg(THIS->name, GF_LOG_ERROR, 0, AFR_MSG_DICT_SET_FAILED,
+                   "%s/%s: Could not set %s key during xattrop",
+                   uuid_utoa(local->loc.pargfid), local->loc.name, key);
+        if (local->transaction.type == AFR_ENTRY_RENAME_TRANSACTION) {
+            xdata2 = dict_new();
+            if (!xdata2)
+                goto out;
 
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Blocking entrylks done. Proceeding to FOP");
-                afr_internal_lock_finish (frame, this);
+            ret = dict_set_strn(xdata2, key, keylen,
+                                (char *)local->newloc.name);
+            if (ret)
+                gf_msg(THIS->name, GF_LOG_ERROR, 0, AFR_MSG_DICT_SET_FAILED,
+                       "%s/%s: Could not set %s key during "
+                       "xattrop",
+                       uuid_utoa(local->newloc.pargfid), local->newloc.name,
+                       key);
         }
-        return 0;
-}
+    }
 
+    *xdata = xdata1;
+    *newloc_xdata = xdata2;
+    xdata1 = xdata2 = NULL;
+out:
+    if (xdata1)
+        dict_unref(xdata1);
+    return;
+}
 
 int
-afr_post_lower_unlock_cbk (call_frame_t *frame, xlator_t *this)
+afr_changelog_prepare(xlator_t *this, call_frame_t *frame, int *call_count,
+                      afr_changelog_resume_t changelog_resume,
+                      afr_xattrop_type_t op, dict_t **xdata,
+                      dict_t **newloc_xdata)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
 
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+    local = frame->local;
+    priv = this->private;
 
-        GF_ASSERT (!int_lock->higher_locked);
+    *call_count = afr_changelog_call_count(
+        local->transaction.type, local->transaction.pre_op,
+        local->transaction.failed_subvols, priv->child_count);
 
-        int_lock->lock_cbk = afr_post_blocking_rename_cbk;
-        afr_blocking_lock (frame, this);
+    if (*call_count == 0) {
+        changelog_resume(frame, this);
+        return -1;
+    }
 
-        return 0;
-}
+    afr_changelog_populate_xdata(frame, op, xdata, newloc_xdata);
+    local->call_count = *call_count;
 
+    local->transaction.changelog_resume = changelog_resume;
+    return 0;
+}
 
 int
-afr_set_transaction_flock (afr_local_t *local)
+afr_changelog_do(call_frame_t *frame, xlator_t *this, dict_t *xattr,
+                 afr_changelog_resume_t changelog_resume, afr_xattrop_type_t op)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_inodelk_t       *inodelk  = NULL;
-
-        int_lock = &local->internal_lock;
-        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    dict_t *xdata = NULL;
+    dict_t *newloc_xdata = NULL;
+    int i = 0;
+    int call_count = 0;
+    int ret = 0;
+
+    local = frame->local;
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->transaction.changelog_xdata[i]) {
+            dict_unref(local->transaction.changelog_xdata[i]);
+            local->transaction.changelog_xdata[i] = NULL;
+        }
+    }
 
-        inodelk->flock.l_len   = local->transaction.len;
-        inodelk->flock.l_start = local->transaction.start;
-        inodelk->flock.l_type  = F_WRLCK;
+    ret = afr_changelog_prepare(this, frame, &call_count, changelog_resume, op,
+                                &xdata, &newloc_xdata);
 
+    if (ret)
         return 0;
-}
-
-int
-afr_lock_rec (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
 
-        int_lock->transaction_lk_type = AFR_TRANSACTION_LK;
-        int_lock->domain = this->name;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->transaction.pre_op[i] ||
+            local->transaction.failed_subvols[i])
+            continue;
 
         switch (local->transaction.type) {
-        case AFR_DATA_TRANSACTION:
-        case AFR_METADATA_TRANSACTION:
-                afr_set_transaction_flock (local);
-
-                int_lock->lock_cbk = afr_post_nonblocking_inodelk_cbk;
-
-                afr_nonblocking_inodelk (frame, this);
+            case AFR_DATA_TRANSACTION:
+            case AFR_METADATA_TRANSACTION:
+                if (!local->fd) {
+                    STACK_WIND_COOKIE(
+                        frame, afr_changelog_cbk, (void *)(long)i,
+                        priv->children[i], priv->children[i]->fops->xattrop,
+                        &local->loc, GF_XATTROP_ADD_ARRAY, xattr, xdata);
+                } else {
+                    STACK_WIND_COOKIE(
+                        frame, afr_changelog_cbk, (void *)(long)i,
+                        priv->children[i], priv->children[i]->fops->fxattrop,
+                        local->fd, GF_XATTROP_ADD_ARRAY, xattr, xdata);
+                }
                 break;
+            case AFR_ENTRY_RENAME_TRANSACTION:
 
-        case AFR_ENTRY_RENAME_TRANSACTION:
+                STACK_WIND_COOKIE(frame, afr_changelog_cbk, (void *)(long)i,
+                                  priv->children[i],
+                                  priv->children[i]->fops->xattrop,
+                                  &local->transaction.new_parent_loc,
+                                  GF_XATTROP_ADD_ARRAY, xattr, newloc_xdata);
+                call_count--;
 
-                int_lock->lock_cbk = afr_post_nonblocking_entrylk_cbk;
-                afr_nonblocking_entrylk (frame, this);
-                break;
+                /* fall through */
 
-        case AFR_ENTRY_TRANSACTION:
-                int_lock->lk_basename = local->transaction.basename;
-                if (&local->transaction.parent_loc)
-                        int_lock->lk_loc = &local->transaction.parent_loc;
+            case AFR_ENTRY_TRANSACTION:
+                if (local->fd)
+                    STACK_WIND_COOKIE(
+                        frame, afr_changelog_cbk, (void *)(long)i,
+                        priv->children[i], priv->children[i]->fops->fxattrop,
+                        local->fd, GF_XATTROP_ADD_ARRAY, xattr, xdata);
                 else
-                        GF_ASSERT (local->fd);
-
-                int_lock->lock_cbk = afr_post_nonblocking_entrylk_cbk;
-                afr_nonblocking_entrylk (frame, this);
+                    STACK_WIND_COOKIE(frame, afr_changelog_cbk, (void *)(long)i,
+                                      priv->children[i],
+                                      priv->children[i]->fops->xattrop,
+                                      &local->transaction.parent_loc,
+                                      GF_XATTROP_ADD_ARRAY, xattr, xdata);
                 break;
         }
 
-        return 0;
-}
+        if (!--call_count)
+            break;
+    }
 
+    if (xdata)
+        dict_unref(xdata);
+    if (newloc_xdata)
+        dict_unref(newloc_xdata);
+    return 0;
+}
 
-int
-afr_lock (call_frame_t *frame, xlator_t *this)
+static void
+afr_init_optimistic_changelog_for_txn(xlator_t *this, afr_local_t *local)
 {
-        afr_set_lock_number (frame, this);
+    int locked_count = 0;
+    afr_private_t *priv = NULL;
 
-        return afr_lock_rec (frame, this);
-}
+    priv = this->private;
 
+    locked_count = AFR_COUNT(local->transaction.pre_op, priv->child_count);
+    if (priv->optimistic_change_log && locked_count == priv->child_count)
+        local->optimistic_change_log = 1;
 
-/* }}} */
+    return;
+}
 
 int
-afr_internal_lock_finish (call_frame_t *frame, xlator_t *this)
+afr_changelog_pre_op(call_frame_t *frame, xlator_t *this)
 {
-        if (__fop_changelog_needed (frame, this)) {
-                afr_changelog_pre_op (frame, this);
+    afr_private_t *priv = this->private;
+    int i = 0;
+    int ret = 0;
+    int call_count = 0;
+    int op_errno = 0;
+    afr_local_t *local = NULL;
+    afr_internal_lock_t *int_lock = NULL;
+    unsigned char *locked_nodes = NULL;
+    int idx = -1;
+    gf_boolean_t pre_nop = _gf_true;
+    dict_t *xdata_req = NULL;
+
+    local = frame->local;
+    int_lock = &local->internal_lock;
+    idx = afr_index_for_transaction_type(local->transaction.type);
+
+    locked_nodes = afr_locked_nodes_get(local->transaction.type, int_lock);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (locked_nodes[i]) {
+            local->transaction.pre_op[i] = 1;
+            call_count++;
         } else {
-                afr_transaction_perform_fop (frame, this);
+            local->transaction.failed_subvols[i] = 1;
+        }
+    }
+
+    afr_init_optimistic_changelog_for_txn(this, local);
+
+    if (afr_changelog_pre_op_inherit(frame, this))
+        goto next;
+
+    /* This condition should not be met with present code, as
+     * transaction.done will be called if locks are not acquired on even a
+     * single node.
+     */
+    if (call_count == 0) {
+        op_errno = ENOTCONN;
+        goto err;
+    }
+
+    /* Check if the fop can be performed on at least
+     * quorum number of nodes.
+     */
+    if (priv->quorum_count && !afr_has_fop_quorum(frame)) {
+        op_errno = int_lock->lock_op_errno;
+        if (op_errno == 0)
+            op_errno = afr_quorum_errno(priv);
+        goto err;
+    }
+
+    xdata_req = dict_new();
+    if (!xdata_req) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    if (call_count < priv->child_count)
+        pre_nop = _gf_false;
+
+    /* Set an all-zero pending changelog so that in the cbk, we can get the
+     * current on-disk values. In a replica 3 volume with arbiter enabled,
+     * these values are needed to arrive at a go/ no-go of the fop phase to
+     * avoid ending up in split-brain.*/
+
+    ret = afr_set_pending_dict(priv, xdata_req, local->pending);
+    if (ret < 0) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    if (afr_needs_changelog_update(local)) {
+        local->dirty[idx] = hton32(1);
+
+        ret = dict_set_static_bin(xdata_req, AFR_DIRTY, local->dirty,
+                                  sizeof(int) * AFR_NUM_CHANGE_LOGS);
+        if (ret) {
+            op_errno = ENOMEM;
+            goto err;
         }
 
-        return 0;
-}
+        pre_nop = _gf_false;
+        local->transaction.dirtied = 1;
+    }
 
+    if (pre_nop)
+        goto next;
 
-void
-afr_set_delayed_post_op (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t    *local = NULL;
-        afr_private_t  *priv = NULL;
+    if (!local->pre_op_compat) {
+        dict_copy(xdata_req, local->xdata_req);
+        goto next;
+    }
 
-        /* call this function from any of the related optimizations
-           which benefit from delaying post op are enabled, namely:
+    afr_changelog_do(frame, this, xdata_req, afr_transaction_perform_fop,
+                     AFR_TRANSACTION_PRE_OP);
 
-           - changelog piggybacking
-           - eager locking
-        */
+    if (xdata_req)
+        dict_unref(xdata_req);
 
-        priv = this->private;
-        if (!priv)
-                return;
+    return 0;
+next:
+    afr_transaction_perform_fop(frame, this);
 
-        if (!priv->post_op_delay_secs)
-                return;
+    if (xdata_req)
+        dict_unref(xdata_req);
 
-        local = frame->local;
-        if (!local->transaction.eager_lock_on)
-                return;
+    return 0;
+err:
+    local->internal_lock.lock_cbk = afr_transaction_done;
+    local->op_ret = -1;
+    local->op_errno = op_errno;
 
-        if (!local)
-                return;
+    afr_handle_lock_acquire_failure(local);
 
-        if (!local->fd)
-                return;
+    if (xdata_req)
+        dict_unref(xdata_req);
 
-        if (local->op == GF_FOP_WRITE)
-                local->delayed_post_op = _gf_true;
+    return 0;
 }
 
-gf_boolean_t
-afr_are_multiple_fds_opened (inode_t *inode, xlator_t *this)
+int
+afr_post_nonblocking_lock_cbk(call_frame_t *frame, xlator_t *this)
 {
-        afr_inode_ctx_t *ictx = NULL;
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
+
+    local = frame->local;
+    int_lock = &local->internal_lock;
+
+    /* Initiate blocking locks if non-blocking has failed */
+    if (int_lock->lock_op_ret < 0) {
+        gf_msg_debug(this->name, 0,
+                     "Non blocking locks failed. Proceeding to blocking");
+        int_lock->lock_cbk = afr_internal_lock_finish;
+        afr_blocking_lock(frame, this);
+    } else {
+        gf_msg_debug(this->name, 0,
+                     "Non blocking locks done. Proceeding to FOP");
+
+        afr_internal_lock_finish(frame, this);
+    }
+
+    return 0;
+}
 
-        if (!inode) {
-                /* If false is returned, it may keep on taking eager-lock
-                 * which may lead to starvation, so return true to avoid that.
-                 */
-                gf_log_callingfn (this->name, GF_LOG_ERROR, "Invalid inode");
-                return _gf_true;
-        }
-        /* Lets say mount1 has eager-lock(full-lock) and after the eager-lock
-         * is taken mount2 opened the same file, it won't be able to
-         * perform any data operations until mount1 releases eager-lock.
-         * To avoid such scenario do not enable eager-lock for this transaction
-         * if open-fd-count is > 1
-         */
+int
+afr_post_blocking_rename_cbk(call_frame_t *frame, xlator_t *this)
+{
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
 
-        ictx = afr_inode_ctx_get (inode, this);
-        if (!ictx)
-                return _gf_true;
+    local = frame->local;
+    int_lock = &local->internal_lock;
 
-        if (ictx->open_fd_count > 1)
-                return _gf_true;
+    if (int_lock->lock_op_ret < 0) {
+        gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_INTERNAL_LKS_FAILED,
+               "Blocking entrylks failed.");
 
-        return _gf_false;
+        afr_transaction_done(frame, this);
+    } else {
+        gf_msg_debug(this->name, 0,
+                     "Blocking entrylks done. Proceeding to FOP");
+
+        afr_internal_lock_finish(frame, this);
+    }
+    return 0;
 }
 
-gf_boolean_t
-afr_any_fops_failed (afr_local_t *local, afr_private_t *priv)
+int
+afr_post_lower_unlock_cbk(call_frame_t *frame, xlator_t *this)
 {
-        if (local->success_count != priv->child_count)
-                return _gf_true;
-        return _gf_false;
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
+
+    local = frame->local;
+    int_lock = &local->internal_lock;
+
+    GF_ASSERT(!int_lock->higher_locked);
+
+    int_lock->lock_cbk = afr_post_blocking_rename_cbk;
+    afr_blocking_lock(frame, this);
+
+    return 0;
 }
 
-gf_boolean_t
-is_afr_delayed_changelog_post_op_needed (call_frame_t *frame, xlator_t *this)
+int
+afr_set_transaction_flock(xlator_t *this, afr_local_t *local,
+                          afr_lockee_t *lockee)
 {
-        afr_local_t      *local = NULL;
-        gf_boolean_t      res = _gf_false;
-        afr_private_t    *priv  = NULL;
+    afr_private_t *priv = NULL;
+    struct gf_flock *flock = NULL;
+
+    priv = this->private;
+    flock = &lockee->flock;
+
+    if ((priv->arbiter_count || local->transaction.eager_lock_on ||
+         priv->full_lock) &&
+        local->transaction.type == AFR_DATA_TRANSACTION) {
+        /*Lock entire file to avoid network split brains.*/
+        flock->l_len = 0;
+        flock->l_start = 0;
+    } else {
+        flock->l_len = local->transaction.len;
+        flock->l_start = local->transaction.start;
+    }
+    flock->l_type = F_WRLCK;
+
+    return 0;
+}
 
-        priv  = this->private;
+int
+afr_lock(call_frame_t *frame, xlator_t *this)
+{
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
+    int i = 0;
 
-        local = frame->local;
-        if (!local)
-                goto out;
+    local = frame->local;
+    int_lock = &local->internal_lock;
 
-        if (!local->delayed_post_op)
-                goto out;
+    int_lock->lock_cbk = afr_post_nonblocking_lock_cbk;
+    int_lock->domain = this->name;
 
-        //Mark pending changelog ASAP
-        if (afr_any_fops_failed (local, priv))
-                goto out;
+    switch (local->transaction.type) {
+        case AFR_DATA_TRANSACTION:
+        case AFR_METADATA_TRANSACTION:
+            for (i = 0; i < int_lock->lockee_count; i++) {
+                afr_set_transaction_flock(this, local, &int_lock->lockee[i]);
+            }
 
-        if (local->fd && afr_are_multiple_fds_opened (local->fd->inode, this))
-                goto out;
+            break;
 
-        res = _gf_true;
-out:
-        return res;
-}
+        case AFR_ENTRY_TRANSACTION:
+            int_lock->lk_basename = local->transaction.basename;
+            if (local->transaction.parent_loc.path)
+                int_lock->lk_loc = &local->transaction.parent_loc;
+            else
+                GF_ASSERT(local->fd);
+            break;
+        case AFR_ENTRY_RENAME_TRANSACTION:
+            break;
+    }
+    afr_lock_nonblocking(frame, this);
 
+    return 0;
+}
 
-void
-afr_delayed_changelog_post_op (xlator_t *this, call_frame_t *frame, fd_t *fd,
-                               call_stub_t *stub);
+static gf_boolean_t
+afr_locals_overlap(afr_local_t *local1, afr_local_t *local2)
+{
+    uint64_t start1 = local1->transaction.start;
+    uint64_t start2 = local2->transaction.start;
+    uint64_t end1 = 0;
+    uint64_t end2 = 0;
+
+    if (local1->transaction.len)
+        end1 = start1 + local1->transaction.len - 1;
+    else
+        end1 = ULLONG_MAX;
+
+    if (local2->transaction.len)
+        end2 = start2 + local2->transaction.len - 1;
+    else
+        end2 = ULLONG_MAX;
+
+    return ((end1 >= start2) && (end2 >= start1));
+}
 
-void
-afr_delayed_changelog_wake_up_cbk (void *data)
+gf_boolean_t
+afr_has_lock_conflict(afr_local_t *local, gf_boolean_t waitlist_check)
 {
-        fd_t           *fd = NULL;
+    afr_local_t *each = NULL;
+    afr_lock_t *lock = NULL;
+
+    lock = &local->inode_ctx->lock[local->transaction.type];
+    /*
+     * Once full file lock is acquired in eager-lock phase, overlapping
+     * writes do not compete for inode-locks, instead are transferred to the
+     * next writes. Because of this overlapping writes are not ordered.
+     * This can cause inconsistencies in replication.
+     * Example:
+     * Two overlapping writes w1, w2 are sent in parallel on same fd
+     * in two threads t1, t2.
+     * Both threads can execute afr_writev_wind in the following manner.
+     * t1 winds w1 on brick-0
+     * t2 winds w2 on brick-0
+     * t2 winds w2 on brick-1
+     * t1 winds w1 on brick-1
+     *
+     * This check makes sure the locks are not transferred for
+     * overlapping writes.
+     */
+    list_for_each_entry(each, &lock->owners, transaction.owner_list)
+    {
+        if (afr_locals_overlap(each, local)) {
+            return _gf_true;
+        }
+    }
 
-        fd = data;
+    if (!waitlist_check)
+        return _gf_false;
+    list_for_each_entry(each, &lock->waiting, transaction.wait_list)
+    {
+        if (afr_locals_overlap(each, local)) {
+            return _gf_true;
+        }
+    }
+    return _gf_false;
+}
 
-        afr_delayed_changelog_wake_up (THIS, fd);
+/* }}} */
+static void
+afr_copy_inodelk_vars(afr_internal_lock_t *dst, afr_internal_lock_t *src,
+                      xlator_t *this, int lockee_num)
+{
+    afr_private_t *priv = this->private;
+    afr_lockee_t *sl = &src->lockee[lockee_num];
+    afr_lockee_t *dl = &dst->lockee[lockee_num];
+
+    dst->domain = src->domain;
+    dl->flock.l_len = sl->flock.l_len;
+    dl->flock.l_start = sl->flock.l_start;
+    dl->flock.l_type = sl->flock.l_type;
+    dl->locked_count = sl->locked_count;
+    memcpy(dl->locked_nodes, sl->locked_nodes,
+           priv->child_count * sizeof(*dl->locked_nodes));
 }
 
+void
+__afr_transaction_wake_shared(afr_local_t *local, struct list_head *shared)
+{
+    gf_boolean_t conflict = _gf_false;
+    afr_local_t *each = NULL;
+    afr_lock_t *lock = &local->inode_ctx->lock[local->transaction.type];
+
+    while (!conflict) {
+        if (list_empty(&lock->waiting))
+            return;
+        each = list_entry(lock->waiting.next, afr_local_t,
+                          transaction.wait_list);
+        if (afr_has_lock_conflict(each, _gf_false)) {
+            conflict = _gf_true;
+        }
+        if (conflict && !list_empty(&lock->owners))
+            return;
+        afr_copy_inodelk_vars(&each->internal_lock, &local->internal_lock,
+                              each->transaction.frame->this, 0);
+        list_move_tail(&each->transaction.wait_list, shared);
+        list_add_tail(&each->transaction.owner_list, &lock->owners);
+    }
+}
 
-/*
-  Check if the frame is destined to get optimized away
-  with changelog piggybacking
-*/
-static gf_boolean_t
-is_piggyback_post_op (call_frame_t *frame, fd_t *fd)
+static void
+afr_lock_resume_shared(struct list_head *list)
 {
-        afr_fd_ctx_t *fdctx = NULL;
-        afr_local_t *local = NULL;
-        gf_boolean_t piggyback = _gf_true;
-        afr_private_t *priv = NULL;
-        int i = 0;
+    afr_local_t *each = NULL;
+
+    while (!list_empty(list)) {
+        each = list_entry(list->next, afr_local_t, transaction.wait_list);
+        list_del_init(&each->transaction.wait_list);
+        afr_changelog_pre_op(each->transaction.frame,
+                             each->transaction.frame->this);
+    }
+}
 
-        priv = frame->this->private;
-        local = frame->local;
-        fdctx = afr_fd_ctx_get (fd, frame->this);
+int
+afr_internal_lock_finish(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = frame->local;
+    afr_lock_t *lock = NULL;
+
+    local->internal_lock.lock_cbk = NULL;
+    if (!local->transaction.eager_lock_on) {
+        if (local->internal_lock.lock_op_ret < 0) {
+            afr_transaction_done(frame, this);
+            return 0;
+        }
+        afr_changelog_pre_op(frame, this);
+    } else {
+        lock = &local->inode_ctx->lock[local->transaction.type];
+        if (local->internal_lock.lock_op_ret < 0) {
+            afr_handle_lock_acquire_failure(local);
+        } else {
+            lock->event_generation = local->event_generation;
+            afr_changelog_pre_op(frame, this);
+        }
+    }
 
-        LOCK(&fd->lock);
-        {
-                piggyback = _gf_true;
+    return 0;
+}
 
-                for (i = 0; i < priv->child_count; i++) {
-                        if (!local->transaction.pre_op[i])
-                                continue;
-                        if (fdctx->pre_op_piggyback[i]) {
-                                fdctx->pre_op_piggyback[i]--;
-                                local->transaction.postop_piggybacked[i] = 1;
-                        } else {
-                                /* For at least _one_ subvolume we cannot
-                                   piggyback on the changelog, and have to
-                                   perform a hard POST-OP and therefore fsync
-                                   if necesssary
-                                */
-                                piggyback = _gf_false;
-                                GF_ASSERT (fdctx->pre_op_done[i]);
-                                fdctx->pre_op_done[i]--;
-                        }
-                }
+gf_boolean_t
+afr_are_conflicting_ops_waiting(afr_local_t *local, xlator_t *this)
+{
+    afr_lock_t *lock = NULL;
+    lock = &local->inode_ctx->lock[local->transaction.type];
+
+    /* Lets say mount1 has eager-lock(full-lock) and after the eager-lock
+     * is taken mount2 opened the same file, it won't be able to
+     * perform any {meta,}data operations until mount1 releases eager-lock.
+     * To avoid such scenario do not enable eager-lock for this transaction
+     * if open-fd-count is > 1 for metadata transactions and if num-inodelks > 1
+     * for data transactions
+     */
+
+    if (local->transaction.type == AFR_METADATA_TRANSACTION) {
+        if (local->inode_ctx->open_fd_count > 1) {
+            return _gf_true;
         }
-        UNLOCK(&fd->lock);
-
-        if (!afr_txn_nothing_failed (frame, frame->this)) {
-                /* something failed in this transaction,
-                   we will be performing a hard post-op
-                */
-                return _gf_false;
+    } else if (local->transaction.type == AFR_DATA_TRANSACTION) {
+        if (lock->num_inodelks > 1) {
+            return _gf_true;
         }
+    }
 
-        return piggyback;
+    return _gf_false;
 }
 
+gf_boolean_t
+afr_is_delayed_changelog_post_op_needed(call_frame_t *frame, xlator_t *this,
+                                        int delay)
+{
+    afr_local_t *local = NULL;
+    afr_lock_t *lock = NULL;
+    gf_boolean_t res = _gf_false;
+
+    local = frame->local;
+    lock = &local->inode_ctx->lock[local->transaction.type];
+
+    if (!afr_txn_nothing_failed(frame, this)) {
+        lock->release = _gf_true;
+        goto out;
+    }
+
+    if (afr_are_conflicting_ops_waiting(local, this)) {
+        lock->release = _gf_true;
+        goto out;
+    }
+
+    if (!list_empty(&lock->owners))
+        goto out;
+    else
+        GF_ASSERT(list_empty(&lock->waiting));
+
+    if (lock->release) {
+        goto out;
+    }
+
+    if (!delay) {
+        goto out;
+    }
+
+    if (local->transaction.disable_delayed_post_op) {
+        goto out;
+    }
+
+    if ((local->op != GF_FOP_WRITE) && (local->op != GF_FOP_FXATTROP) &&
+        (local->op != GF_FOP_FSYNC)) {
+        /*Only allow writes/fsyncs but shard does [f]xattrops on writes, so
+         * they are fine too*/
+        goto out;
+    }
+
+    res = _gf_true;
+out:
+    return res;
+}
+
+void
+afr_delayed_changelog_wake_up_cbk(void *data)
+{
+    afr_lock_t *lock = NULL;
+    afr_local_t *local = data;
+    afr_local_t *timer_local = NULL;
+    struct list_head shared;
+
+    INIT_LIST_HEAD(&shared);
+    lock = &local->inode_ctx->lock[local->transaction.type];
+    LOCK(&local->inode->lock);
+    {
+        timer_local = list_entry(lock->post_op.next, afr_local_t,
+                                 transaction.owner_list);
+        if (list_empty(&lock->owners) && (local == timer_local)) {
+            GF_ASSERT(list_empty(&lock->waiting));
+            /*Last owner*/
+            lock->release = _gf_true;
+            lock->delay_timer = NULL;
+        }
+    }
+    UNLOCK(&local->inode->lock);
+    afr_changelog_post_op_now(local->transaction.frame,
+                              local->transaction.frame->this);
+}
 
 /* SET operation */
 int
-afr_fd_report_unstable_write (xlator_t *this, fd_t *fd)
+afr_fd_report_unstable_write(xlator_t *this, afr_local_t *local)
 {
-        afr_fd_ctx_t *fdctx = NULL;
-
-        fdctx = afr_fd_ctx_get (fd, this);
+    LOCK(&local->inode->lock);
+    {
+        local->inode_ctx->witnessed_unstable_write = _gf_true;
+    }
+    UNLOCK(&local->inode->lock);
 
-        LOCK(&fd->lock);
-        {
-                fdctx->witnessed_unstable_write = _gf_true;
-        }
-        UNLOCK(&fd->lock);
-
-        return 0;
+    return 0;
 }
 
 /* TEST and CLEAR operation */
 gf_boolean_t
-afr_fd_has_witnessed_unstable_write (xlator_t *this, fd_t *fd)
+afr_fd_has_witnessed_unstable_write(xlator_t *this, inode_t *inode)
 {
-        afr_fd_ctx_t *fdctx = NULL;
-        gf_boolean_t witness = _gf_false;
+    afr_inode_ctx_t *ctx = NULL;
+    gf_boolean_t witness = _gf_false;
 
-	fdctx = afr_fd_ctx_get (fd, this);
-        if (!fdctx)
-                return _gf_true;
+    LOCK(&inode->lock);
+    {
+        (void)__afr_inode_ctx_get(this, inode, &ctx);
 
-        LOCK(&fd->lock);
-        {
-                if (fdctx->witnessed_unstable_write) {
-                        witness = _gf_true;
-                        fdctx->witnessed_unstable_write = _gf_false;
-                }
+        if (ctx->witnessed_unstable_write) {
+            witness = _gf_true;
+            ctx->witnessed_unstable_write = _gf_false;
         }
-        UNLOCK (&fd->lock);
+    }
+    UNLOCK(&inode->lock);
 
-        return witness;
+    return witness;
 }
 
-
 int
-afr_changelog_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int op_ret, int op_errno, struct iatt *pre,
-                         struct iatt *post, dict_t *xdata)
-{
-        afr_private_t *priv = NULL;
-        int child_index = (long) cookie;
-        int call_count = -1;
-        afr_local_t *local = NULL;
-
-        priv = this->private;
-        local = frame->local;
-
-        if (afr_fop_failed (op_ret, op_errno)) {
-                /* Failure of fsync() is as good as failure of previous
-                   write(). So treat it like one.
-                */
-                gf_log (this->name, GF_LOG_WARNING,
-                        "fsync(%s) failed on subvolume %s. Transaction was %s",
-                        uuid_utoa (local->fd->inode->gfid),
-                        priv->children[child_index]->name,
-                        gf_fop_list[local->op]);
-
-                afr_transaction_fop_failed (frame, this, child_index);
-        }
+afr_changelog_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, struct iatt *pre,
+                        struct iatt *post, dict_t *xdata)
+{
+    afr_private_t *priv = NULL;
+    int child_index = (long)cookie;
+    int call_count = -1;
+    afr_local_t *local = NULL;
 
-        call_count = afr_frame_return (frame);
+    priv = this->private;
+    local = frame->local;
 
-        if (call_count == 0)
-                afr_changelog_post_op_now (frame, this);
+    if (op_ret != 0) {
+        /* Failure of fsync() is as good as failure of previous
+           write(). So treat it like one.
+        */
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, AFR_MSG_FSYNC_FAILED,
+               "fsync(%s) failed on subvolume %s. Transaction was %s",
+               uuid_utoa(local->fd->inode->gfid),
+               priv->children[child_index]->name, gf_fop_list[local->op]);
 
-        return 0;
-}
+        afr_transaction_fop_failed(frame, this, child_index);
+    }
+
+    call_count = afr_frame_return(frame);
 
+    if (call_count == 0)
+        afr_changelog_post_op_now(frame, this);
+
+    return 0;
+}
 
 int
-afr_changelog_fsync (call_frame_t *frame, xlator_t *this)
+afr_changelog_fsync(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *local = NULL;
-        int i = 0;
-        int call_count = 0;
-        afr_private_t *priv = NULL;
- 	dict_t *xdata = NULL;
- 	GF_UNUSED int ret = -1;
+    afr_local_t *local = NULL;
+    int i = 0;
+    int call_count = 0;
+    afr_private_t *priv = NULL;
+    dict_t *xdata = NULL;
+    GF_UNUSED int ret = -1;
 
-        local = frame->local;
-        priv = this->private;
+    local = frame->local;
+    priv = this->private;
 
-        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
-                                                     priv->child_count);
+    call_count = AFR_COUNT(local->transaction.pre_op, priv->child_count);
 
-        if (!call_count) {
-                /* will go straight to unlock */
-                afr_changelog_post_op_now (frame, this);
-                return 0;
-        }
+    if (!call_count) {
+        /* will go straight to unlock */
+        afr_changelog_post_op_now(frame, this);
+        return 0;
+    }
 
-        local->call_count = call_count;
+    local->call_count = call_count;
 
-	xdata = dict_new();
-	if (xdata)
-		ret = dict_set_int32 (xdata, "batch-fsync", 1);
+    xdata = dict_new();
+    if (xdata) {
+        ret = dict_set_int32_sizen(xdata, "batch-fsync", 1);
+        ret = dict_set_str(xdata, GLUSTERFS_INTERNAL_FOP_KEY, "yes");
+    }
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (!local->transaction.pre_op[i])
-                        continue;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->transaction.pre_op[i])
+            continue;
 
-                STACK_WIND_COOKIE (frame, afr_changelog_fsync_cbk,
-                                (void *) (long) i, priv->children[i],
-                                priv->children[i]->fops->fsync, local->fd,
-                                1, xdata);
-                if (!--call_count)
-                        break;
-        }
+        STACK_WIND_COOKIE(frame, afr_changelog_fsync_cbk, (void *)(long)i,
+                          priv->children[i], priv->children[i]->fops->fsync,
+                          local->fd, 1, xdata);
+        if (!--call_count)
+            break;
+    }
 
-	if (xdata)
-		dict_unref (xdata);
+    if (xdata)
+        dict_unref(xdata);
 
-        return 0;
+    return 0;
 }
 
-
-        int
-afr_changelog_post_op_safe (call_frame_t *frame, xlator_t *this)
+int
+afr_changelog_post_op_safe(call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t    *local = NULL;
-        afr_private_t  *priv = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-	local = frame->local;
-        priv = this->private;
-
-        if (!local->fd || local->transaction.type != AFR_DATA_TRANSACTION) {
-                afr_changelog_post_op_now (frame, this);
-                return 0;
-        }
+    local = frame->local;
+    priv = this->private;
 
-        if (is_piggyback_post_op (frame, local->fd)) {
-                /* just detected that this post-op is about to
-                   be optimized away as a new write() has
-                   already piggybacked on this frame's changelog.
-                   */
-                afr_changelog_post_op_now (frame, this);
-                return 0;
-        }
+    if (!local->fd || local->transaction.type != AFR_DATA_TRANSACTION) {
+        afr_changelog_post_op_now(frame, this);
+        return 0;
+    }
 
-        /* Calling afr_changelog_post_op_now() now will result in
-           issuing ->[f]xattrop().
-
-           Performing a hard POST-OP (->[f]xattrop() FOP) is a more
-           responsible operation that what it might appear on the surface.
-
-           The changelog of a file (in the xattr of the file on the server)
-           stores information (pending count) about the state of the file
-           on the OTHER server. This changelog is blindly trusted, and must
-           therefore be updated in such a way it remains trustworthy. This
-           implies that decrementing the pending count (essentially "clearing
-           the dirty flag") must be done STRICTLY after we are sure that the
-           operation on the other server has reached stable storage.
-
-           While the backend filesystem on that server will eventually flush
-           it to stable storage, we (being in userspace) have no mechanism
-           to get notified when the write became "stable".
-
-           This means we need take matter into our own hands and issue an
-           fsync() EVEN IF THE APPLICATION WAS PERFORMING UNSTABLE WRITES,
-           and get an acknowledgement for it. And we need to wait for the
-           fsync() acknowledgement before initiating the hard POST-OP.
-
-           However if the FD itself was opened in O_SYNC or O_DSYNC then
-           we are already guaranteed that the writes were made stable as
-           part of the FOP itself. The same holds true for NFS stable
-           writes which happen on an anonymous FD with O_DSYNC or O_SYNC
-           flag set in the writev() @flags param. For all other write types,
-           mark a flag in the fdctx whenever an unstable write is witnessed.
+    if (afr_changelog_pre_op_uninherit(frame, this) &&
+        afr_txn_nothing_failed(frame, this)) {
+        /* just detected that this post-op is about to
+           be optimized away as a new write() has
+           already piggybacked on this frame's changelog.
            */
-
-        if (!afr_fd_has_witnessed_unstable_write (this, local->fd)) {
-                afr_changelog_post_op_now (frame, this);
-                return 0;
-        }
-
-        /* Check whether users want durability and perform fsync/post-op
-         * accordingly.
-         */
-        if (priv->ensure_durability) {
-                /* Time to fsync() */
-                afr_changelog_fsync (frame, this);
-        } else {
-                afr_changelog_post_op_now (frame, this);
-        }
-
+        afr_changelog_post_op_now(frame, this);
+        return 0;
+    }
+
+    /* Calling afr_changelog_post_op_now() now will result in
+       issuing ->[f]xattrop().
+
+       Performing a hard POST-OP (->[f]xattrop() FOP) is a more
+       responsible operation that what it might appear on the surface.
+
+       The changelog of a file (in the xattr of the file on the server)
+       stores information (pending count) about the state of the file
+       on the OTHER server. This changelog is blindly trusted, and must
+       therefore be updated in such a way it remains trustworthy. This
+       implies that decrementing the pending count (essentially "clearing
+       the dirty flag") must be done STRICTLY after we are sure that the
+       operation on the other server has reached stable storage.
+
+       While the backend filesystem on that server will eventually flush
+       it to stable storage, we (being in userspace) have no mechanism
+       to get notified when the write became "stable".
+
+       This means we need take matter into our own hands and issue an
+       fsync() EVEN IF THE APPLICATION WAS PERFORMING UNSTABLE WRITES,
+       and get an acknowledgement for it. And we need to wait for the
+       fsync() acknowledgement before initiating the hard POST-OP.
+
+       However if the FD itself was opened in O_SYNC or O_DSYNC then
+       we are already guaranteed that the writes were made stable as
+       part of the FOP itself. The same holds true for NFS stable
+       writes which happen on an anonymous FD with O_DSYNC or O_SYNC
+       flag set in the writev() @flags param. For all other write types,
+       mark a flag in the fdctx whenever an unstable write is witnessed.
+       */
+
+    if (!afr_fd_has_witnessed_unstable_write(this, local->inode)) {
+        afr_changelog_post_op_now(frame, this);
         return 0;
+    }
+
+    /* Check whether users want durability and perform fsync/post-op
+     * accordingly.
+     */
+    if (priv->ensure_durability) {
+        /* Time to fsync() */
+        afr_changelog_fsync(frame, this);
+    } else {
+        afr_changelog_post_op_now(frame, this);
+    }
+
+    return 0;
 }
 
-
 void
-afr_delayed_changelog_post_op (xlator_t *this, call_frame_t *frame, fd_t *fd,
-                               call_stub_t *stub)
+afr_changelog_post_op(call_frame_t *frame, xlator_t *this)
 {
-	afr_fd_ctx_t      *fd_ctx = NULL;
-	call_frame_t      *prev_frame = NULL;
-	struct timeval     delta = {0, };
-	afr_private_t     *priv = NULL;
-	afr_local_t       *local = NULL;
-
-	priv = this->private;
-
-	fd_ctx = afr_fd_ctx_get (fd, this);
-	if (!fd_ctx)
-                goto out;
+    struct timespec delta = {
+        0,
+    };
+    afr_private_t *priv = NULL;
+    afr_local_t *local = frame->local;
+    afr_lock_t *lock = NULL;
+    gf_boolean_t post_op = _gf_true;
+    struct list_head shared;
+
+    priv = this->private;
+    delta.tv_sec = priv->post_op_delay_secs;
+    delta.tv_nsec = 0;
+
+    INIT_LIST_HEAD(&shared);
+    if (!local->transaction.eager_lock_on)
+        goto out;
+
+    lock = &local->inode_ctx->lock[local->transaction.type];
+    LOCK(&local->inode->lock);
+    {
+        list_del_init(&local->transaction.owner_list);
+        list_add(&local->transaction.owner_list, &lock->post_op);
+        __afr_transaction_wake_shared(local, &shared);
+
+        if (!afr_is_delayed_changelog_post_op_needed(frame, this,
+                                                     delta.tv_sec)) {
+            if (list_empty(&lock->owners))
+                lock->release = _gf_true;
+            goto unlock;
+        }
 
-	delta.tv_sec = priv->post_op_delay_secs;
-	delta.tv_usec = 0;
-
-	pthread_mutex_lock (&fd_ctx->delay_lock);
-	{
-		prev_frame = fd_ctx->delay_frame;
-		fd_ctx->delay_frame = NULL;
-		if (fd_ctx->delay_timer)
-			gf_timer_call_cancel (this->ctx, fd_ctx->delay_timer);
-		fd_ctx->delay_timer = NULL;
-		if (!frame)
-			goto unlock;
-		fd_ctx->delay_timer = gf_timer_call_after (this->ctx, delta,
-							   afr_delayed_changelog_wake_up_cbk,
-							   fd);
-		fd_ctx->delay_frame = frame;
-	}
+        GF_ASSERT(lock->delay_timer == NULL);
+        lock->delay_timer = gf_timer_call_after(
+            this->ctx, delta, afr_delayed_changelog_wake_up_cbk, local);
+        if (!lock->delay_timer) {
+            lock->release = _gf_true;
+        } else {
+            post_op = _gf_false;
+        }
+    }
 unlock:
-	pthread_mutex_unlock (&fd_ctx->delay_lock);
+    UNLOCK(&local->inode->lock);
+
+    if (!list_empty(&shared)) {
+        afr_lock_resume_shared(&shared);
+    }
 
 out:
-	if (prev_frame) {
-		local = prev_frame->local;
-		local->transaction.resume_stub = stub;
-		afr_changelog_post_op_safe (prev_frame, this);
-	} else if (stub) {
-		call_resume (stub);
-	}
+    if (post_op) {
+        if (!local->transaction.eager_lock_on || lock->release) {
+            afr_changelog_post_op_safe(frame, this);
+        } else {
+            afr_changelog_post_op_now(frame, this);
+        }
+    }
 }
 
-
-void
-afr_changelog_post_op (call_frame_t *frame, xlator_t *this)
+int
+afr_transaction_resume(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t  *local = NULL;
+    afr_local_t *local = NULL;
 
-        local = frame->local;
+    local = frame->local;
 
-        if (is_afr_delayed_changelog_post_op_needed (frame, this))
-                afr_delayed_changelog_post_op (this, frame, local->fd, NULL);
-        else
-                afr_changelog_post_op_safe (frame, this);
-}
+    afr_restore_lk_owner(frame);
 
+    afr_handle_symmetric_errors(frame, this);
 
+    if (!local->pre_op_compat)
+        /* new mode, pre-op was done along
+           with OP */
+        afr_changelog_pre_op_update(frame, this);
 
-/* Wake up the sleeping/delayed post-op, and also register
-   a stub to have it resumed after this transaction
-   completely finishes.
+    afr_changelog_post_op(frame, this);
 
-   The @stub gets saved in @local and gets resumed in
-   afr_local_cleanup()
-   */
-void
-afr_delayed_changelog_wake_resume (xlator_t *this, fd_t *fd, call_stub_t *stub)
-{
-        afr_delayed_changelog_post_op (this, NULL, fd, stub);
+    return 0;
 }
 
+/**
+ * afr_transaction_fop_failed - inform that an fop failed
+ */
 
 void
-afr_delayed_changelog_wake_up (xlator_t *this, fd_t *fd)
+afr_transaction_fop_failed(call_frame_t *frame, xlator_t *this, int child_index)
 {
-        afr_delayed_changelog_post_op (this, NULL, fd, NULL);
-}
+    afr_local_t *local = NULL;
+
+    local = frame->local;
 
+    local->transaction.failed_subvols[child_index] = 1;
+}
 
-        int
-afr_transaction_resume (call_frame_t *frame, xlator_t *this)
+static gf_boolean_t
+__need_previous_lock_unlocked(afr_local_t *local)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_private_t       *priv     = NULL;
+    afr_lock_t *lock = NULL;
 
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        priv     = this->private;
+    lock = &local->inode_ctx->lock[local->transaction.type];
+    if (!lock->acquired)
+        return _gf_false;
+    if (lock->acquired && lock->event_generation != local->event_generation)
+        return _gf_true;
+    return _gf_false;
+}
 
-        if (local->transaction.eager_lock_on) {
-                /* We don't need to retain "local" in the
-                   fd list anymore, writes to all subvols
-                   are finished by now */
-                LOCK (&local->fd->lock);
-                {
-                        list_del_init (&local->transaction.eager_locked);
-                }
-                UNLOCK (&local->fd->lock);
+void
+__afr_eager_lock_handle(afr_local_t *local, gf_boolean_t *take_lock,
+                        gf_boolean_t *do_pre_op, afr_local_t **timer_local)
+{
+    afr_lock_t *lock = NULL;
+    afr_local_t *owner_local = NULL;
+    xlator_t *this = local->transaction.frame->this;
+
+    local->transaction.eager_lock_on = _gf_true;
+    afr_set_lk_owner(local->transaction.frame, this, local->inode);
+
+    lock = &local->inode_ctx->lock[local->transaction.type];
+    if (__need_previous_lock_unlocked(local)) {
+        if (!list_empty(&lock->owners)) {
+            lock->release = _gf_true;
+        } else if (lock->delay_timer) {
+            lock->release = _gf_true;
+            if (gf_timer_call_cancel(this->ctx, lock->delay_timer)) {
+                /* It will be put in frozen list
+                 * in the code flow below*/
+            } else {
+                *timer_local = list_entry(lock->post_op.next, afr_local_t,
+                                          transaction.owner_list);
+                lock->delay_timer = NULL;
+            }
         }
-
-        afr_restore_lk_owner (frame);
-
-        if (__fop_changelog_needed (frame, this)) {
-                afr_changelog_post_op (frame, this);
+    }
+
+    if (lock->release) {
+        list_add_tail(&local->transaction.wait_list, &lock->frozen);
+        *take_lock = _gf_false;
+        goto out;
+    }
+
+    if (lock->delay_timer) {
+        *take_lock = _gf_false;
+        if (gf_timer_call_cancel(this->ctx, lock->delay_timer)) {
+            list_add_tail(&local->transaction.wait_list, &lock->frozen);
         } else {
-                if (afr_lock_server_count (priv, local->transaction.type) == 0) {
-                        local->transaction.done (frame, this);
-                } else {
-                        int_lock->lock_cbk = local->transaction.done;
-                        afr_unlock (frame, this);
-                }
+            *timer_local = list_entry(lock->post_op.next, afr_local_t,
+                                      transaction.owner_list);
+            afr_copy_inodelk_vars(&local->internal_lock,
+                                  &(*timer_local)->internal_lock, this, 0);
+            lock->delay_timer = NULL;
+            *do_pre_op = _gf_true;
+            list_add_tail(&local->transaction.owner_list, &lock->owners);
         }
+        goto out;
+    }
 
-        return 0;
+    if (!list_empty(&lock->owners)) {
+        if (!lock->acquired || afr_has_lock_conflict(local, _gf_true)) {
+            list_add_tail(&local->transaction.wait_list, &lock->waiting);
+            *take_lock = _gf_false;
+            goto out;
+        }
+        owner_local = list_entry(lock->owners.next, afr_local_t,
+                                 transaction.owner_list);
+        afr_copy_inodelk_vars(&local->internal_lock,
+                              &owner_local->internal_lock, this, 0);
+        *take_lock = _gf_false;
+        *do_pre_op = _gf_true;
+    }
+
+    if (lock->acquired)
+        GF_ASSERT(!(*take_lock));
+    list_add_tail(&local->transaction.owner_list, &lock->owners);
+out:
+    return;
 }
 
-
-/**
- * afr_transaction_fop_failed - inform that an fop failed
- */
-
 void
-afr_transaction_fop_failed (call_frame_t *frame, xlator_t *this,
-                            int child_index)
+afr_transaction_start(afr_local_t *local, xlator_t *this)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
-
-        local = frame->local;
-        priv  = this->private;
-
-        __mark_child_dead (local->pending, priv->child_count,
-                        child_index, local->transaction.type);
+    afr_private_t *priv = NULL;
+    gf_boolean_t take_lock = _gf_true;
+    gf_boolean_t do_pre_op = _gf_false;
+    afr_local_t *timer_local = NULL;
+
+    priv = this->private;
+
+    if (local->transaction.type != AFR_DATA_TRANSACTION &&
+        local->transaction.type != AFR_METADATA_TRANSACTION)
+        goto lock_phase;
+
+    if (!priv->eager_lock)
+        goto lock_phase;
+
+    LOCK(&local->inode->lock);
+    {
+        __afr_eager_lock_handle(local, &take_lock, &do_pre_op, &timer_local);
+    }
+    UNLOCK(&local->inode->lock);
+lock_phase:
+    if (!local->transaction.eager_lock_on) {
+        afr_set_lk_owner(local->transaction.frame, this,
+                         local->transaction.frame->root);
+    }
+
+    if (take_lock) {
+        afr_lock(local->transaction.frame, this);
+    } else if (do_pre_op) {
+        afr_changelog_pre_op(local->transaction.frame, this);
+    }
+    /*Always call delayed_changelog_wake_up_cbk after calling pre-op above
+     * so that any inheriting can happen*/
+    if (timer_local)
+        afr_delayed_changelog_wake_up_cbk(timer_local);
 }
 
-
-
-        static gf_boolean_t
-afr_locals_overlap (afr_local_t *local1, afr_local_t *local2)
+int
+afr_write_txn_refresh_done(call_frame_t *frame, xlator_t *this, int err)
 {
-        uint64_t start1 = local1->transaction.start;
-        uint64_t start2 = local2->transaction.start;
-        uint64_t end1 = 0;
-        uint64_t end2 = 0;
-
-        if (local1->transaction.len)
-                end1 = start1 + local1->transaction.len - 1;
-        else
-                end1 = ULLONG_MAX;
-
-        if (local2->transaction.len)
-                end2 = start2 + local2->transaction.len - 1;
-        else
-                end2 = ULLONG_MAX;
-
-        return ((end1 >= start2) && (end2 >= start1));
+    afr_local_t *local = frame->local;
+
+    if (err) {
+        AFR_SET_ERROR_AND_CHECK_SPLIT_BRAIN(-1, err);
+        goto fail;
+    }
+
+    afr_transaction_start(local, this);
+    return 0;
+fail:
+    local->transaction.unwind(frame, this);
+    AFR_STACK_DESTROY(frame);
+    return 0;
 }
 
-void
-afr_transaction_eager_lock_init (afr_local_t *local, xlator_t *this)
+int
+afr_transaction_lockee_init(call_frame_t *frame)
 {
-        afr_private_t *priv = NULL;
-        afr_fd_ctx_t  *fdctx = NULL;
-        afr_local_t   *each = NULL;
-
-        priv = this->private;
+    afr_local_t *local = frame->local;
+    afr_internal_lock_t *int_lock = &local->internal_lock;
+    afr_private_t *priv = frame->this->private;
+    int ret = 0;
 
-        if (!local->fd)
-                return;
-
-        if (local->transaction.type != AFR_DATA_TRANSACTION)
-                return;
-
-        if (!priv->eager_lock)
-                return;
-
-        fdctx = afr_fd_ctx_get (local->fd, this);
-        if (!fdctx)
-                return;
+    switch (local->transaction.type) {
+        case AFR_DATA_TRANSACTION:
+        case AFR_METADATA_TRANSACTION:
+            ret = afr_add_inode_lockee(local, priv->child_count);
+            break;
 
-        if (afr_are_multiple_fds_opened (local->fd->inode, this))
-                return;
-        /*
-         * Once full file lock is acquired in eager-lock phase, overlapping
-         * writes do not compete for inode-locks, instead are transferred to the
-         * next writes. Because of this overlapping writes are not ordered.
-         * This can cause inconsistencies in replication.
-         * Example:
-         * Two overlapping writes w1, w2 are sent in parallel on same fd
-         * in two threads t1, t2.
-         * Both threads can execute afr_writev_wind in the following manner.
-         * t1 winds w1 on brick-0
-         * t2 winds w2 on brick-0
-         * t2 winds w2 on brick-1
-         * t1 winds w1 on brick-1
-         *
-         * This check makes sure the locks are not transferred for
-         * overlapping writes.
-         */
-        LOCK (&local->fd->lock);
-        {
-                list_for_each_entry (each, &fdctx->eager_locked,
-                                     transaction.eager_locked) {
-                        if (afr_locals_overlap (each, local)) {
-                                local->transaction.eager_lock_on = _gf_false;
-                                goto unlock;
-                        }
+        case AFR_ENTRY_TRANSACTION:
+        case AFR_ENTRY_RENAME_TRANSACTION:
+            ret = afr_add_entry_lockee(local, &local->transaction.parent_loc,
+                                       local->transaction.basename,
+                                       priv->child_count);
+            if (ret) {
+                goto out;
+            }
+            if (local->op == GF_FOP_RENAME) {
+                ret = afr_add_entry_lockee(
+                    local, &local->transaction.new_parent_loc,
+                    local->transaction.new_basename, priv->child_count);
+                if (ret) {
+                    goto out;
                 }
 
-                local->transaction.eager_lock_on = _gf_true;
-                list_add_tail (&local->transaction.eager_locked,
-                               &fdctx->eager_locked);
-        }
-unlock:
-        UNLOCK (&local->fd->lock);
+                if (local->newloc.inode &&
+                    IA_ISDIR(local->newloc.inode->ia_type)) {
+                    ret = afr_add_entry_lockee(local, &local->newloc, NULL,
+                                               priv->child_count);
+                    if (ret) {
+                        goto out;
+                    }
+                }
+            } else if (local->op == GF_FOP_RMDIR) {
+                ret = afr_add_entry_lockee(local, &local->loc, NULL,
+                                           priv->child_count);
+                if (ret) {
+                    goto out;
+                }
+            }
+
+            if (int_lock->lockee_count > 1) {
+                qsort(int_lock->lockee, int_lock->lockee_count,
+                      sizeof(*int_lock->lockee), afr_entry_lockee_cmp);
+            }
+            break;
+    }
+out:
+    return ret;
 }
 
-
 int
-afr_transaction (call_frame_t *frame, xlator_t *this, afr_transaction_type type)
+afr_transaction(call_frame_t *frame, xlator_t *this, afr_transaction_type type)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
-        fd_t            *fd   = NULL;
-        int             ret   = -1;
-
-        local = frame->local;
-        priv  = this->private;
-
-        local->transaction.resume = afr_transaction_resume;
-        local->transaction.type   = type;
-
-        ret = afr_transaction_local_init (local, this);
-        if (ret < 0)
-            goto out;
-
-        afr_transaction_eager_lock_init (local, this);
-
-        if (local->fd && local->transaction.eager_lock_on)
-                afr_set_lk_owner (frame, this, local->fd);
-        else
-                afr_set_lk_owner (frame, this, frame->root);
-
-        if (!local->transaction.eager_lock_on && local->loc.inode) {
-                fd = fd_lookup (local->loc.inode, frame->root->pid);
-                if (fd == NULL)
-                        fd = fd_lookup_anonymous (local->loc.inode);
-
-                if (fd) {
-                        afr_delayed_changelog_wake_up (this, fd);
-                        fd_unref (fd);
-                }
-        }
-
-        if (afr_lock_server_count (priv, local->transaction.type) == 0) {
-                afr_internal_lock_finish (frame, this);
-        } else {
-                afr_lock (frame, this);
-        }
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int ret = -1;
+    int event_generation = 0;
+
+    local = frame->local;
+    priv = this->private;
+    local->transaction.frame = frame;
+
+    local->transaction.type = type;
+
+    if (priv->quorum_count && !afr_has_quorum(local->child_up, this, NULL)) {
+        ret = -afr_quorum_errno(priv);
+        goto out;
+    }
+
+    if (!afr_is_consistent_io_possible(local, priv, &ret)) {
+        ret = -ret; /*op_errno to ret conversion*/
+        goto out;
+    }
+
+    if (priv->thin_arbiter_count && !afr_ta_has_quorum(priv, local)) {
+        ret = -afr_quorum_errno(priv);
+        goto out;
+    }
+
+    ret = afr_transaction_local_init(local, this);
+    if (ret < 0)
+        goto out;
+
+    ret = afr_transaction_lockee_init(frame);
+    if (ret)
+        goto out;
+
+    if (type != AFR_METADATA_TRANSACTION) {
+        goto txn_start;
+    }
+
+    ret = afr_inode_get_readable(frame, local->inode, this, local->readable,
+                                 &event_generation, type);
+    if (ret < 0 ||
+        afr_is_inode_refresh_reqd(local->inode, this, priv->event_generation,
+                                  event_generation)) {
+        afr_inode_refresh(frame, this, local->inode, local->loc.gfid,
+                          afr_write_txn_refresh_done);
         ret = 0;
+        goto out;
+    }
+
+txn_start:
+    ret = 0;
+    afr_transaction_start(local, this);
 out:
-        return ret;
+    return ret;
 }
diff --git a/xlators/cluster/afr/src/afr-transaction.h b/xlators/cluster/afr/src/afr-transaction.h
index fa626fd0d6e..beefa26f4a6 100644
--- a/xlators/cluster/afr/src/afr-transaction.h
+++ b/xlators/cluster/afr/src/afr-transaction.h
@@ -11,41 +11,65 @@
 #ifndef __TRANSACTION_H__
 #define __TRANSACTION_H__
 
-typedef enum {
-        LOCAL_FIRST = 1,
-        LOCAL_LAST = 2
-} afr_xattrop_type_t;
+#include "afr.h"
 
 void
-afr_transaction_fop_failed (call_frame_t *frame, xlator_t *this,
-			    int child_index);
+afr_transaction_fop_failed(call_frame_t *frame, xlator_t *this,
+                           int child_index);
+
+int32_t
+afr_transaction(call_frame_t *frame, xlator_t *this, afr_transaction_type type);
 
 int
-afr_lock_server_count (afr_private_t *priv, afr_transaction_type type);
+afr_set_pending_dict(afr_private_t *priv, dict_t *xattr, int32_t **pending);
+
+void
+afr_delayed_changelog_wake_up(xlator_t *this, fd_t *fd);
 
-afr_inodelk_t*
-afr_get_inodelk (afr_internal_lock_t *int_lock, char *dom);
+void
+__mark_all_success(call_frame_t *frame, xlator_t *this);
 
-int32_t
-afr_transaction (call_frame_t *frame, xlator_t *this, afr_transaction_type type);
+gf_boolean_t
+afr_txn_nothing_failed(call_frame_t *frame, xlator_t *this);
 
-afr_fd_ctx_t *
-afr_fd_ctx_get (fd_t *fd, xlator_t *this);
 int
-afr_set_pending_dict (afr_private_t *priv, dict_t *xattr, int32_t **pending,
-                      int child, afr_xattrop_type_t op);
-void
-afr_set_delayed_post_op (call_frame_t *frame, xlator_t *this);
+afr_read_txn(call_frame_t *frame, xlator_t *this, inode_t *inode,
+             afr_read_txn_wind_t readfn, afr_transaction_type type);
+
+int
+afr_read_txn_continue(call_frame_t *frame, xlator_t *this, int subvol);
 
 void
-afr_delayed_changelog_wake_up (xlator_t *this, fd_t *fd);
+afr_pending_read_increment(afr_private_t *priv, int child_index);
 
 void
-__mark_all_success (int32_t *pending[], int child_count,
-                    afr_transaction_type type);
-gf_boolean_t
-afr_any_fops_failed (afr_local_t *local, afr_private_t *priv);
+afr_pending_read_decrement(afr_private_t *priv, int child_index);
 
+call_frame_t *
+afr_transaction_detach_fop_frame(call_frame_t *frame);
+gf_boolean_t
+afr_has_quorum(unsigned char *subvols, xlator_t *this, call_frame_t *frame);
 gf_boolean_t
-afr_txn_nothing_failed (call_frame_t *frame, xlator_t *this);
+afr_needs_changelog_update(afr_local_t *local);
+void
+afr_zero_fill_stat(afr_local_t *local);
+
+void
+afr_pick_error_xdata(afr_local_t *local, afr_private_t *priv, inode_t *inode1,
+                     unsigned char *readable1, inode_t *inode2,
+                     unsigned char *readable2);
+int
+afr_transaction_resume(call_frame_t *frame, xlator_t *this);
+
+int
+afr_lock(call_frame_t *frame, xlator_t *this);
+
+void
+afr_delayed_changelog_wake_up_cbk(void *data);
+
+int
+afr_release_notify_lock_for_ta(void *opaque);
+
+int
+afr_ta_lock_release_done(int ret, call_frame_t *ta_frame, void *opaque);
 #endif /* __TRANSACTION_H__ */
diff --git a/xlators/cluster/afr/src/afr.c b/xlators/cluster/afr/src/afr.c
index 4338b799aa7..df7366f0a65 100644
--- a/xlators/cluster/afr/src/afr.c
+++ b/xlators/cluster/afr/src/afr.c
@@ -15,772 +15,1330 @@
 #include <stdlib.h>
 #include <signal.h>
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
 #include "afr-common.c"
-
-#define SHD_INODE_LRU_LIMIT          2048
-#define AFR_EH_HEALED_LIMIT          1024
-#define AFR_EH_HEAL_FAIL_LIMIT       1024
-#define AFR_EH_SPLIT_BRAIN_LIMIT     1024
+#include "afr-messages.h"
 
 struct volume_options options[];
 
+static char *afr_favorite_child_policies[AFR_FAV_CHILD_POLICY_MAX + 1] = {
+    [AFR_FAV_CHILD_NONE] = "none",
+    [AFR_FAV_CHILD_BY_SIZE] = "size",
+    [AFR_FAV_CHILD_BY_CTIME] = "ctime",
+    [AFR_FAV_CHILD_BY_MTIME] = "mtime",
+    [AFR_FAV_CHILD_BY_MAJORITY] = "majority",
+    [AFR_FAV_CHILD_POLICY_MAX] = NULL,
+};
+
 int32_t
-notify (xlator_t *this, int32_t event,
-        void *data, ...)
+notify(xlator_t *this, int32_t event, void *data, ...)
 {
-        int ret = -1;
-        va_list         ap;
-        void *data2 = NULL;
+    int ret = -1;
+    va_list ap;
+    void *data2 = NULL;
 
-        va_start (ap, data);
-        data2 = va_arg (ap, dict_t*);
-        va_end (ap);
-        ret = afr_notify (this, event, data, data2);
+    va_start(ap, data);
+    data2 = va_arg(ap, dict_t *);
+    va_end(ap);
+    ret = afr_notify(this, event, data, data2);
 
-        return ret;
+    return ret;
 }
 
 int32_t
-mem_acct_init (xlator_t *this)
+mem_acct_init(xlator_t *this)
 {
-        int     ret = -1;
+    int ret = -1;
 
-        if (!this)
-                return ret;
-
-        ret = xlator_mem_acct_init (this, gf_afr_mt_end + 1);
+    if (!this)
+        return ret;
 
-        if (ret != 0) {
-                gf_log(this->name, GF_LOG_ERROR, "Memory accounting init"
-                       "failed");
-                return ret;
-        }
+    ret = xlator_mem_acct_init(this, gf_afr_mt_end + 1);
 
+    if (ret != 0) {
         return ret;
-}
+    }
 
+    return ret;
+}
 
 int
-xlator_subvolume_index (xlator_t *this, xlator_t *subvol)
+xlator_subvolume_index(xlator_t *this, xlator_t *subvol)
 {
-        int index = -1;
-        int i = 0;
-        xlator_list_t *list = NULL;
-
-        list = this->children;
-
-        while (list) {
-                if (subvol == list->xlator ||
-                    strcmp (subvol->name, list->xlator->name) == 0) {
-                        index = i;
-                        break;
-                }
-                list = list->next;
-                i++;
+    int index = -1;
+    int i = 0;
+    xlator_list_t *list = NULL;
+
+    list = this->children;
+
+    while (list) {
+        if (subvol == list->xlator ||
+            strcmp(subvol->name, list->xlator->name) == 0) {
+            index = i;
+            break;
         }
+        list = list->next;
+        i++;
+    }
 
-        return index;
+    return index;
 }
 
-void
-fix_quorum_options (xlator_t *this, afr_private_t *priv, char *qtype)
+static void
+fix_quorum_options(xlator_t *this, afr_private_t *priv, char *qtype,
+                   dict_t *options)
 {
-        if (priv->quorum_count && strcmp(qtype,"fixed")) {
-                gf_log(this->name,GF_LOG_WARNING,
-                       "quorum-type %s overriding quorum-count %u",
-                       qtype, priv->quorum_count);
-        }
-        if (!strcmp(qtype,"none")) {
-                priv->quorum_count = 0;
-        }
-        else if (!strcmp(qtype,"auto")) {
-                priv->quorum_count = AFR_QUORUM_AUTO;
-        }
+    if (dict_get_sizen(options, "quorum-type") == NULL) {
+        /* If user doesn't configure anything enable auto-quorum if the
+         * replica has more than two subvolumes */
+        if (priv->child_count > 2)
+            qtype = "auto";
+    }
+
+    if (priv->quorum_count && strcmp(qtype, "fixed")) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_QUORUM_OVERRIDE,
+               "quorum-type %s overriding quorum-count %u", qtype,
+               priv->quorum_count);
+    }
+
+    if (!strcmp(qtype, "none")) {
+        priv->quorum_count = 0;
+    } else if (!strcmp(qtype, "auto")) {
+        priv->quorum_count = AFR_QUORUM_AUTO;
+    }
 }
 
 int
-reconfigure (xlator_t *this, dict_t *options)
+afr_set_favorite_child_policy(afr_private_t *priv, char *policy)
 {
-        afr_private_t *priv        = NULL;
-        xlator_t      *read_subvol = NULL;
-        int            read_subvol_index = -1;
-        int            ret         = -1;
-        int            index       = -1;
-        char          *qtype       = NULL;
+    int index = -1;
 
-        priv = this->private;
+    index = gf_get_index_by_elem(afr_favorite_child_policies, policy);
+    if (index < 0 || index >= AFR_FAV_CHILD_POLICY_MAX)
+        return -1;
 
-        GF_OPTION_RECONF ("background-self-heal-count",
-                          priv->background_self_heal_count, options, uint32,
-                          out);
+    priv->fav_child_policy = index;
 
-        GF_OPTION_RECONF ("metadata-self-heal",
-                          priv->metadata_self_heal, options, bool, out);
+    return 0;
+}
 
-        GF_OPTION_RECONF ("data-self-heal", priv->data_self_heal, options, str,
-                          out);
+static void
+set_data_self_heal_algorithm(afr_private_t *priv, char *algo)
+{
+    if (!algo) {
+        priv->data_self_heal_algorithm = AFR_SELFHEAL_DATA_DYNAMIC;
+    } else if (strcmp(algo, "full") == 0) {
+        priv->data_self_heal_algorithm = AFR_SELFHEAL_DATA_FULL;
+    } else if (strcmp(algo, "diff") == 0) {
+        priv->data_self_heal_algorithm = AFR_SELFHEAL_DATA_DIFF;
+    } else {
+        priv->data_self_heal_algorithm = AFR_SELFHEAL_DATA_DYNAMIC;
+    }
+}
 
-        GF_OPTION_RECONF ("entry-self-heal", priv->entry_self_heal, options,
-                          bool, out);
+void
+afr_handle_anon_inode_options(afr_private_t *priv, dict_t *options)
+{
+    char *volfile_id_str = NULL;
+    uuid_t anon_inode_gfid = {0};
+
+    /*If volume id is not present don't enable anything*/
+    if (dict_get_str(options, "volume-id", &volfile_id_str))
+        return;
+    GF_ASSERT(strlen(AFR_ANON_DIR_PREFIX) + strlen(volfile_id_str) <= NAME_MAX);
+    /*anon_inode_name is not supposed to change once assigned*/
+    if (!priv->anon_inode_name[0]) {
+        snprintf(priv->anon_inode_name, sizeof(priv->anon_inode_name), "%s-%s",
+                 AFR_ANON_DIR_PREFIX, volfile_id_str);
+        gf_uuid_parse(volfile_id_str, anon_inode_gfid);
+        /*Flip a bit to make sure volfile-id and anon-gfid are not same*/
+        anon_inode_gfid[0] ^= 1;
+        uuid_utoa_r(anon_inode_gfid, priv->anon_gfid_str);
+    }
+}
 
-        GF_OPTION_RECONF ("strict-readdir", priv->strict_readdir, options, bool,
-                          out);
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+    afr_private_t *priv = NULL;
+    xlator_t *read_subvol = NULL;
+    int read_subvol_index = -1;
+    int timeout_old = 0;
+    int ret = -1;
+    int index = -1;
+    char *qtype = NULL;
+    char *fav_child_policy = NULL;
+    char *data_self_heal = NULL;
+    char *data_self_heal_algorithm = NULL;
+    char *locking_scheme = NULL;
+    gf_boolean_t consistent_io = _gf_false;
+    gf_boolean_t choose_local_old = _gf_false;
+    gf_boolean_t enabled_old = _gf_false;
 
-        GF_OPTION_RECONF ("data-self-heal-window-size",
-                          priv->data_self_heal_window_size, options,
-                          uint32, out);
+    priv = this->private;
 
-        GF_OPTION_RECONF ("data-change-log", priv->data_change_log, options,
-                          bool, out);
+    GF_OPTION_RECONF("metadata-splitbrain-forced-heal",
+                     priv->metadata_splitbrain_forced_heal, options, bool, out);
 
-        GF_OPTION_RECONF ("metadata-change-log",
-                          priv->metadata_change_log, options, bool, out);
+    GF_OPTION_RECONF("background-self-heal-count",
+                     priv->background_self_heal_count, options, uint32, out);
 
-        GF_OPTION_RECONF ("entry-change-log", priv->entry_change_log, options,
-                          bool, out);
+    GF_OPTION_RECONF("heal-wait-queue-length", priv->heal_wait_qlen, options,
+                     uint32, out);
 
-        GF_OPTION_RECONF ("data-self-heal-algorithm",
-                          priv->data_self_heal_algorithm, options, str, out);
+    GF_OPTION_RECONF("metadata-self-heal", priv->metadata_self_heal, options,
+                     bool, out);
 
-        GF_OPTION_RECONF ("self-heal-daemon", priv->shd.enabled, options, bool, out);
+    GF_OPTION_RECONF("data-self-heal", data_self_heal, options, str, out);
+    if (gf_string2boolean(data_self_heal, &priv->data_self_heal) == -1)
+        goto out;
 
-        GF_OPTION_RECONF ("read-subvolume", read_subvol, options, xlator, out);
+    GF_OPTION_RECONF("entry-self-heal", priv->entry_self_heal, options, bool,
+                     out);
 
-        GF_OPTION_RECONF ("read-hash-mode", priv->hash_mode,
-                          options, uint32, out);
+    GF_OPTION_RECONF("data-self-heal-window-size",
+                     priv->data_self_heal_window_size, options, uint32, out);
 
-        if (read_subvol) {
-                index = xlator_subvolume_index (this, read_subvol);
-                if (index == -1) {
-                        gf_log (this->name, GF_LOG_ERROR, "%s not a subvolume",
-                                read_subvol->name);
-                        goto out;
-                }
-                priv->read_child = index;
-        }
+    GF_OPTION_RECONF("data-self-heal-algorithm", data_self_heal_algorithm,
+                     options, str, out);
+    set_data_self_heal_algorithm(priv, data_self_heal_algorithm);
 
-        GF_OPTION_RECONF ("read-subvolume-index",read_subvol_index, options,int32,out);
+    GF_OPTION_RECONF("halo-enabled", priv->halo_enabled, options, bool, out);
 
-        if (read_subvol_index >-1) {
-                index=read_subvol_index;
-                if (index >= priv->child_count) {
-                        gf_log (this->name, GF_LOG_ERROR, "%d not a subvolume-index",
-                                index);
-                        goto out;
-                }
-                priv->read_child = index;
-        }
+    GF_OPTION_RECONF("halo-shd-max-latency", priv->shd.halo_max_latency_msec,
+                     options, uint32, out);
 
-        GF_OPTION_RECONF ("eager-lock", priv->eager_lock, options, bool, out);
-        GF_OPTION_RECONF ("quorum-type", qtype, options, str, out);
-        GF_OPTION_RECONF ("quorum-count", priv->quorum_count, options,
-                          uint32, out);
-        fix_quorum_options(this,priv,qtype);
-        GF_OPTION_RECONF ("heal-timeout", priv->shd.timeout, options,
-                          int32, out);
-
-	GF_OPTION_RECONF ("post-op-delay-secs", priv->post_op_delay_secs, options,
-			  uint32, out);
-
-        GF_OPTION_RECONF (AFR_SH_READDIR_SIZE_KEY, priv->sh_readdir_size,
-                          options, size, out);
-        /* Reset this so we re-discover in case the topology changed.  */
-        GF_OPTION_RECONF ("readdir-failover", priv->readdir_failover, options,
-                          bool, out);
-        GF_OPTION_RECONF ("ensure-durability", priv->ensure_durability, options,
-                          bool, out);
-        priv->did_discovery = _gf_false;
+    GF_OPTION_RECONF("halo-nfsd-max-latency", priv->nfsd.halo_max_latency_msec,
+                     options, uint32, out);
 
-        ret = 0;
-out:
-        return ret;
+    GF_OPTION_RECONF("halo-max-latency", priv->halo_max_latency_msec, options,
+                     uint32, out);
 
-}
+    GF_OPTION_RECONF("halo-max-replicas", priv->halo_max_replicas, options,
+                     uint32, out);
 
+    GF_OPTION_RECONF("halo-min-replicas", priv->halo_min_replicas, options,
+                     uint32, out);
 
-static const char *favorite_child_warning_str = "You have specified subvolume '%s' "
-        "as the 'favorite child'. This means that if a discrepancy in the content "
-        "or attributes (ownership, permission, etc.) of a file is detected among "
-        "the subvolumes, the file on '%s' will be considered the definitive "
-        "version and its contents will OVERWRITE the contents of the file on other "
-        "subvolumes. All versions of the file except that on '%s' "
-        "WILL BE LOST.";
+    GF_OPTION_RECONF("read-subvolume", read_subvol, options, xlator, out);
 
+    choose_local_old = priv->choose_local;
+    GF_OPTION_RECONF("choose-local", priv->choose_local, options, bool, out);
 
-int32_t
-init (xlator_t *this)
-{
-        afr_private_t *priv        = NULL;
-        int            child_count = 0;
-        xlator_list_t *trav        = NULL;
-        int            i           = 0;
-        int            ret         = -1;
-        GF_UNUSED int  op_errno    = 0;
-        xlator_t      *read_subvol = NULL;
-        int            read_subvol_index = -1;
-        xlator_t      *fav_child   = NULL;
-        char          *qtype       = NULL;
-
-        if (!this->children) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "replicate translator needs more than one "
-                        "subvolume defined.");
-                return -1;
+    if (choose_local_old != priv->choose_local) {
+        priv->read_child = -1;
+        if (choose_local_old == _gf_false)
+            priv->did_discovery = _gf_false;
+    }
+
+    GF_OPTION_RECONF("read-hash-mode", priv->hash_mode, options, uint32, out);
+
+    if (read_subvol) {
+        index = xlator_subvolume_index(this, read_subvol);
+        if (index == -1) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_INVALID_SUBVOL,
+                   "%s not a subvolume", read_subvol->name);
+            goto out;
         }
-
-        if (!this->parents) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "Volume is dangling.");
+        priv->read_child = index;
+    }
+
+    GF_OPTION_RECONF("read-subvolume-index", read_subvol_index, options, int32,
+                     out);
+
+    if (read_subvol_index > -1) {
+        index = read_subvol_index;
+        if (index >= priv->child_count) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_INVALID_SUBVOL,
+                   "%d not a subvolume-index", index);
+            goto out;
         }
+        priv->read_child = index;
+    }
+
+    GF_OPTION_RECONF("pre-op-compat", priv->pre_op_compat, options, bool, out);
+    GF_OPTION_RECONF("locking-scheme", locking_scheme, options, str, out);
+    priv->granular_locks = (strcmp(locking_scheme, "granular") == 0);
+    GF_OPTION_RECONF("full-lock", priv->full_lock, options, bool, out);
+    GF_OPTION_RECONF("granular-entry-heal", priv->esh_granular, options, bool,
+                     out);
+
+    GF_OPTION_RECONF("eager-lock", priv->eager_lock, options, bool, out);
+    GF_OPTION_RECONF("optimistic-change-log", priv->optimistic_change_log,
+                     options, bool, out);
+    GF_OPTION_RECONF("quorum-type", qtype, options, str, out);
+    GF_OPTION_RECONF("quorum-count", priv->quorum_count, options, uint32, out);
+    fix_quorum_options(this, priv, qtype, options);
+    if (priv->quorum_count && !afr_has_quorum(priv->child_up, this, NULL))
+        gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_QUORUM_FAIL,
+               "Client-quorum is not met");
+
+    GF_OPTION_RECONF("post-op-delay-secs", priv->post_op_delay_secs, options,
+                     uint32, out);
+
+    GF_OPTION_RECONF(AFR_SH_READDIR_SIZE_KEY, priv->sh_readdir_size, options,
+                     size_uint64, out);
+    /* Reset this so we re-discover in case the topology changed.  */
+    GF_OPTION_RECONF("ensure-durability", priv->ensure_durability, options,
+                     bool, out);
+
+    enabled_old = priv->shd.enabled;
+    GF_OPTION_RECONF("self-heal-daemon", priv->shd.enabled, options, bool, out);
+
+    GF_OPTION_RECONF("iam-self-heal-daemon", priv->shd.iamshd, options, bool,
+                     out);
+
+    timeout_old = priv->shd.timeout;
+    GF_OPTION_RECONF("heal-timeout", priv->shd.timeout, options, int32, out);
+
+    GF_OPTION_RECONF("consistent-metadata", priv->consistent_metadata, options,
+                     bool, out);
+
+    GF_OPTION_RECONF("shd-max-threads", priv->shd.max_threads, options, uint32,
+                     out);
+
+    GF_OPTION_RECONF("shd-wait-qlength", priv->shd.wait_qlength, options,
+                     uint32, out);
+
+    GF_OPTION_RECONF("favorite-child-policy", fav_child_policy, options, str,
+                     out);
+    if (afr_set_favorite_child_policy(priv, fav_child_policy) == -1)
+        goto out;
+
+    priv->did_discovery = _gf_false;
+
+    GF_OPTION_RECONF("consistent-io", consistent_io, options, bool, out);
+    if (priv->quorum_count != 0)
+        consistent_io = _gf_false;
+    priv->consistent_io = consistent_io;
+
+    afr_handle_anon_inode_options(priv, options);
+
+    GF_OPTION_RECONF("use-anonymous-inode", priv->use_anon_inode, options, bool,
+                     out);
+    if (priv->shd.enabled) {
+        if ((priv->shd.enabled != enabled_old) ||
+            (timeout_old != priv->shd.timeout))
+            afr_selfheal_childup(this, priv);
+    }
+
+    ret = 0;
+out:
+    return ret;
+}
 
-	this->private = GF_CALLOC (1, sizeof (afr_private_t),
-                                   gf_afr_mt_afr_private_t);
-        if (!this->private)
-                goto out;
+static int
+afr_pending_xattrs_init(afr_private_t *priv, xlator_t *this)
+{
+    int ret = -1;
+    int i = 0;
+    char *ptr = NULL;
+    char *ptr1 = NULL;
+    char *xattrs_list = NULL;
+    xlator_list_t *trav = NULL;
+    int child_count = -1;
+
+    trav = this->children;
+    child_count = priv->child_count;
+    if (priv->thin_arbiter_count) {
+        /* priv->pending_key[THIN_ARBITER_BRICK_INDEX] is used as the
+         * name of the thin arbiter file for persistence across add/
+         * removal of DHT subvols.*/
+        child_count++;
+    }
+
+    GF_OPTION_INIT("afr-pending-xattr", xattrs_list, str, out);
+    priv->pending_key = GF_CALLOC(sizeof(*priv->pending_key), child_count,
+                                  gf_afr_mt_char);
+    if (!priv->pending_key) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    if (!xattrs_list) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_NO_CHANGELOG,
+               "Unable to fetch afr-pending-xattr option from volfile."
+               " Falling back to using client translator names. ");
 
-        priv = this->private;
-        LOCK_INIT (&priv->lock);
-        LOCK_INIT (&priv->read_child_lock);
-        //lock recovery is not done in afr
-        pthread_mutex_init (&priv->mutex, NULL);
-        INIT_LIST_HEAD (&priv->saved_fds);
+        while (i < child_count) {
+            ret = gf_asprintf(&priv->pending_key[i], "%s.%s", AFR_XATTR_PREFIX,
+                              trav->xlator->name);
+            if (ret == -1) {
+                ret = -ENOMEM;
+                goto out;
+            }
+            trav = trav->next;
+            i++;
+        }
+        ret = 0;
+        goto out;
+    }
 
-        child_count = xlator_subvolume_count (this);
+    ptr = ptr1 = gf_strdup(xattrs_list);
+    if (!ptr) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    for (i = 0, ptr = strtok(ptr, ","); ptr; ptr = strtok(NULL, ",")) {
+        ret = gf_asprintf(&priv->pending_key[i], "%s.%s", AFR_XATTR_PREFIX,
+                          ptr);
+        if (ret == -1) {
+            ret = -ENOMEM;
+            goto out;
+        }
+        i++;
+    }
+    ret = 0;
 
-        priv->child_count = child_count;
+out:
+    GF_FREE(ptr1);
+    return ret;
+}
 
-        priv->read_child = -1;
+void
+afr_ta_init(afr_private_t *priv)
+{
+    priv->thin_arbiter_count = 1;
+    priv->child_count--;
+    priv->ta_child_up = 0;
+    priv->ta_bad_child_index = AFR_CHILD_UNKNOWN;
+    priv->ta_notify_dom_lock_offset = 0;
+    priv->ta_in_mem_txn_count = 0;
+    priv->ta_on_wire_txn_count = 0;
+    priv->release_ta_notify_dom_lock = _gf_false;
+    INIT_LIST_HEAD(&priv->ta_waitq);
+    INIT_LIST_HEAD(&priv->ta_onwireq);
+    gf_uuid_clear(priv->ta_gfid);
+}
 
-        GF_OPTION_INIT ("read-subvolume", read_subvol, xlator, out);
-        if (read_subvol) {
-                priv->read_child = xlator_subvolume_index (this, read_subvol);
-                if (priv->read_child == -1) {
-                        gf_log (this->name, GF_LOG_ERROR, "%s not a subvolume",
-                                read_subvol->name);
-                        goto out;
-                }
-        }
-        GF_OPTION_INIT ("read-subvolume-index",read_subvol_index,int32,out);
-        if (read_subvol_index > -1) {
-                if (read_subvol_index >= priv->child_count) {
-                        gf_log (this->name, GF_LOG_ERROR, "%d not a subvolume-index",
-                                read_subvol_index);
-                        goto out;
-                }
-                priv->read_child = read_subvol_index;
+int32_t
+init(xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    int child_count = 0;
+    xlator_list_t *trav = NULL;
+    int i = 0;
+    int ret = -1;
+    GF_UNUSED int op_errno = 0;
+    xlator_t *read_subvol = NULL;
+    int read_subvol_index = -1;
+    char *qtype = NULL;
+    char *fav_child_policy = NULL;
+    char *thin_arbiter = NULL;
+    char *data_self_heal = NULL;
+    char *locking_scheme = NULL;
+    char *data_self_heal_algorithm = NULL;
+
+    if (!this->children) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_CHILD_MISCONFIGURED,
+               "replicate translator needs more than one "
+               "subvolume defined.");
+        return -1;
+    }
+
+    if (!this->parents) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_VOL_MISCONFIGURED,
+               "Volume is dangling.");
+    }
+
+    this->private = GF_CALLOC(1, sizeof(afr_private_t),
+                              gf_afr_mt_afr_private_t);
+    if (!this->private)
+        goto out;
+
+    priv = this->private;
+    INIT_LIST_HEAD(&priv->saved_locks);
+    INIT_LIST_HEAD(&priv->lk_healq);
+    LOCK_INIT(&priv->lock);
+
+    child_count = xlator_subvolume_count(this);
+
+    priv->child_count = child_count;
+
+    priv->read_child = -1;
+
+    GF_OPTION_INIT("arbiter-count", priv->arbiter_count, uint32, out);
+    GF_OPTION_INIT("thin-arbiter", thin_arbiter, str, out);
+    if (thin_arbiter && strlen(thin_arbiter) > 0) {
+        afr_ta_init(priv);
+    }
+    INIT_LIST_HEAD(&priv->healing);
+    INIT_LIST_HEAD(&priv->heal_waiting);
+
+    priv->spb_choice_timeout = AFR_DEFAULT_SPB_CHOICE_TIMEOUT;
+
+    GF_OPTION_INIT("afr-dirty-xattr", priv->afr_dirty, str, out);
+
+    GF_OPTION_INIT("metadata-splitbrain-forced-heal",
+                   priv->metadata_splitbrain_forced_heal, bool, out);
+
+    GF_OPTION_INIT("read-subvolume", read_subvol, xlator, out);
+    if (read_subvol) {
+        priv->read_child = xlator_subvolume_index(this, read_subvol);
+        if (priv->read_child == -1) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_INVALID_SUBVOL,
+                   "%s not a subvolume", read_subvol->name);
+            goto out;
         }
-        GF_OPTION_INIT ("choose-local", priv->choose_local, bool, out);
-
-        GF_OPTION_INIT ("read-hash-mode", priv->hash_mode, uint32, out);
-
-        priv->favorite_child = -1;
-        GF_OPTION_INIT ("favorite-child", fav_child, xlator, out);
-        if (fav_child) {
-                priv->favorite_child = xlator_subvolume_index (this, fav_child);
-                if (priv->favorite_child == -1) {
-                        gf_log (this->name, GF_LOG_ERROR, "%s not a subvolume",
-                                fav_child->name);
-                        goto out;
-                }
-                gf_log (this->name, GF_LOG_WARNING,
-                        favorite_child_warning_str, fav_child->name,
-                        fav_child->name, fav_child->name);
+    }
+    GF_OPTION_INIT("read-subvolume-index", read_subvol_index, int32, out);
+    if (read_subvol_index > -1) {
+        if (read_subvol_index >= priv->child_count) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_INVALID_SUBVOL,
+                   "%d not a subvolume-index", read_subvol_index);
+            goto out;
         }
+        priv->read_child = read_subvol_index;
+    }
+    GF_OPTION_INIT("choose-local", priv->choose_local, bool, out);
 
+    priv->pending_reads = GF_CALLOC(sizeof(*priv->pending_reads),
+                                    priv->child_count, gf_afr_mt_atomic_t);
 
-        GF_OPTION_INIT ("background-self-heal-count",
-                        priv->background_self_heal_count, uint32, out);
+    GF_OPTION_INIT("read-hash-mode", priv->hash_mode, uint32, out);
 
-        GF_OPTION_INIT ("data-self-heal", priv->data_self_heal, str, out);
+    priv->favorite_child = -1;
 
-        GF_OPTION_INIT ("data-self-heal-algorithm",
-                        priv->data_self_heal_algorithm, str, out);
+    GF_OPTION_INIT("favorite-child-policy", fav_child_policy, str, out);
+    if (afr_set_favorite_child_policy(priv, fav_child_policy) == -1)
+        goto out;
 
-        GF_OPTION_INIT ("data-self-heal-window-size",
-                        priv->data_self_heal_window_size, uint32, out);
+    GF_OPTION_INIT("shd-max-threads", priv->shd.max_threads, uint32, out);
 
-        GF_OPTION_INIT ("metadata-self-heal", priv->metadata_self_heal, bool,
-                        out);
+    GF_OPTION_INIT("shd-wait-qlength", priv->shd.wait_qlength, uint32, out);
 
-        GF_OPTION_INIT ("entry-self-heal", priv->entry_self_heal, bool, out);
+    GF_OPTION_INIT("background-self-heal-count",
+                   priv->background_self_heal_count, uint32, out);
 
-        GF_OPTION_INIT ("self-heal-daemon", priv->shd.enabled, bool, out);
+    GF_OPTION_INIT("heal-wait-queue-length", priv->heal_wait_qlen, uint32, out);
 
-        GF_OPTION_INIT ("iam-self-heal-daemon", priv->shd.iamshd, bool, out);
+    GF_OPTION_INIT("data-self-heal", data_self_heal, str, out);
+    if (gf_string2boolean(data_self_heal, &priv->data_self_heal) == -1)
+        goto out;
 
-        GF_OPTION_INIT ("data-change-log", priv->data_change_log, bool, out);
+    GF_OPTION_INIT("data-self-heal-algorithm", data_self_heal_algorithm, str,
+                   out);
+    set_data_self_heal_algorithm(priv, data_self_heal_algorithm);
 
-        GF_OPTION_INIT ("metadata-change-log", priv->metadata_change_log, bool,
-                        out);
+    GF_OPTION_INIT("data-self-heal-window-size",
+                   priv->data_self_heal_window_size, uint32, out);
 
-        GF_OPTION_INIT ("entry-change-log", priv->entry_change_log, bool, out);
+    GF_OPTION_INIT("metadata-self-heal", priv->metadata_self_heal, bool, out);
 
-        GF_OPTION_INIT ("optimistic-change-log", priv->optimistic_change_log,
-                        bool, out);
+    GF_OPTION_INIT("entry-self-heal", priv->entry_self_heal, bool, out);
 
-        GF_OPTION_INIT ("inodelk-trace", priv->inodelk_trace, bool, out);
+    GF_OPTION_INIT("halo-shd-max-latency", priv->shd.halo_max_latency_msec,
+                   uint32, out);
 
-        GF_OPTION_INIT ("entrylk-trace", priv->entrylk_trace, bool, out);
+    GF_OPTION_INIT("halo-max-latency", priv->halo_max_latency_msec, uint32,
+                   out);
+    GF_OPTION_INIT("halo-max-replicas", priv->halo_max_replicas, uint32, out);
+    GF_OPTION_INIT("halo-min-replicas", priv->halo_min_replicas, uint32, out);
 
-        GF_OPTION_INIT ("strict-readdir", priv->strict_readdir, bool, out);
+    GF_OPTION_INIT("halo-enabled", priv->halo_enabled, bool, out);
 
-        GF_OPTION_INIT ("eager-lock", priv->eager_lock, bool, out);
-        GF_OPTION_INIT ("quorum-type", qtype, str, out);
-        GF_OPTION_INIT ("quorum-count", priv->quorum_count, uint32, out);
-        GF_OPTION_INIT (AFR_SH_READDIR_SIZE_KEY, priv->sh_readdir_size, size,
-                        out);
-        fix_quorum_options(this,priv,qtype);
+    GF_OPTION_INIT("halo-nfsd-max-latency", priv->nfsd.halo_max_latency_msec,
+                   uint32, out);
 
-	GF_OPTION_INIT ("post-op-delay-secs", priv->post_op_delay_secs, uint32, out);
-        GF_OPTION_INIT ("readdir-failover", priv->readdir_failover, bool, out);
-        GF_OPTION_INIT ("ensure-durability", priv->ensure_durability, bool,
-                        out);
+    GF_OPTION_INIT("iam-nfs-daemon", priv->nfsd.iamnfsd, bool, out);
 
-        priv->wait_count = 1;
+    GF_OPTION_INIT("optimistic-change-log", priv->optimistic_change_log, bool,
+                   out);
 
-        priv->child_up = GF_CALLOC (sizeof (unsigned char), child_count,
-                                    gf_afr_mt_char);
-        if (!priv->child_up) {
-                ret = -ENOMEM;
-                goto out;
-        }
+    GF_OPTION_INIT("pre-op-compat", priv->pre_op_compat, bool, out);
+    GF_OPTION_INIT("locking-scheme", locking_scheme, str, out);
+    priv->granular_locks = (strcmp(locking_scheme, "granular") == 0);
+    GF_OPTION_INIT("full-lock", priv->full_lock, bool, out);
+    GF_OPTION_INIT("granular-entry-heal", priv->esh_granular, bool, out);
 
-        for (i = 0; i < child_count; i++)
-                priv->child_up[i] = -1; /* start with unknown state.
-                                           this initialization needed
-                                           for afr_notify() to work
-                                           reliably
-                                        */
+    GF_OPTION_INIT("eager-lock", priv->eager_lock, bool, out);
+    GF_OPTION_INIT("quorum-type", qtype, str, out);
+    GF_OPTION_INIT("quorum-count", priv->quorum_count, uint32, out);
+    GF_OPTION_INIT(AFR_SH_READDIR_SIZE_KEY, priv->sh_readdir_size, size_uint64,
+                   out);
+    fix_quorum_options(this, priv, qtype, this->options);
 
-        priv->children = GF_CALLOC (sizeof (xlator_t *), child_count,
-                                    gf_afr_mt_xlator_t);
-        if (!priv->children) {
-                ret = -ENOMEM;
-                goto out;
-        }
+    GF_OPTION_INIT("post-op-delay-secs", priv->post_op_delay_secs, uint32, out);
+    GF_OPTION_INIT("ensure-durability", priv->ensure_durability, bool, out);
 
-        priv->pending_key = GF_CALLOC (sizeof (*priv->pending_key),
-                                       child_count,
-                                       gf_afr_mt_char);
-        if (!priv->pending_key) {
-                ret = -ENOMEM;
-                goto out;
-        }
+    GF_OPTION_INIT("self-heal-daemon", priv->shd.enabled, bool, out);
 
-        trav = this->children;
-        i = 0;
-        while (i < child_count) {
-                priv->children[i] = trav->xlator;
-
-                ret = gf_asprintf (&priv->pending_key[i], "%s.%s",
-                                   AFR_XATTR_PREFIX,
-                                   trav->xlator->name);
-                if (-1 == ret) {
-                        ret = -ENOMEM;
-                        goto out;
-                }
-
-                trav = trav->next;
-                i++;
-        }
+    GF_OPTION_INIT("iam-self-heal-daemon", priv->shd.iamshd, bool, out);
+    GF_OPTION_INIT("heal-timeout", priv->shd.timeout, int32, out);
 
-        ret = gf_asprintf (&priv->sh_domain, "%s-self-heal", this->name);
-        if (-1 == ret) {
-                ret = -ENOMEM;
-                goto out;
-        }
+    GF_OPTION_INIT("consistent-metadata", priv->consistent_metadata, bool, out);
+    GF_OPTION_INIT("consistent-io", priv->consistent_io, bool, out);
+    afr_handle_anon_inode_options(priv, this->options);
 
-        priv->last_event = GF_CALLOC (child_count, sizeof (*priv->last_event),
-                                      gf_afr_mt_int32_t);
-        if (!priv->last_event) {
-                ret = -ENOMEM;
-                goto out;
-        }
+    GF_OPTION_INIT("use-anonymous-inode", priv->use_anon_inode, bool, out);
+    if (priv->quorum_count != 0)
+        priv->consistent_io = _gf_false;
 
-        /* keep more local here as we may need them for self-heal etc */
-        this->local_pool = mem_pool_new (afr_local_t, 512);
-        if (!this->local_pool) {
-                ret = -1;
-                gf_log (this->name, GF_LOG_ERROR,
-                        "failed to create local_t's memory pool");
-                goto out;
-        }
+    priv->wait_count = 1;
 
-        priv->first_lookup = 1;
-        priv->root_inode = NULL;
+    priv->local = GF_CALLOC(sizeof(unsigned char), child_count, gf_afr_mt_char);
+    if (!priv->local) {
+        ret = -ENOMEM;
+        goto out;
+    }
 
-        if (!priv->shd.iamshd) {
-                ret = 0;
-                goto out;
-        }
+    priv->anon_inode = GF_CALLOC(sizeof(unsigned char), child_count,
+                                 gf_afr_mt_char);
 
-        ret = -ENOMEM;
-        priv->shd.pos = GF_CALLOC (sizeof (*priv->shd.pos), child_count,
-                                   gf_afr_mt_brick_pos_t);
-        if (!priv->shd.pos)
-                goto out;
+    priv->child_up = GF_CALLOC(sizeof(unsigned char), child_count,
+                               gf_afr_mt_char);
 
-        priv->shd.pending = GF_CALLOC (sizeof (*priv->shd.pending), child_count,
-                                       gf_afr_mt_int32_t);
-        if (!priv->shd.pending)
-                goto out;
+    priv->child_latency = GF_MALLOC(sizeof(*priv->child_latency) * child_count,
+                                    gf_afr_mt_child_latency_t);
+    priv->halo_child_up = GF_CALLOC(sizeof(unsigned char), child_count,
+                                    gf_afr_mt_char);
 
-        priv->shd.inprogress = GF_CALLOC (sizeof (*priv->shd.inprogress),
-                                          child_count, gf_afr_mt_shd_bool_t);
-        if (!priv->shd.inprogress)
-                goto out;
-        priv->shd.timer = GF_CALLOC (sizeof (*priv->shd.timer), child_count,
-                                     gf_afr_mt_shd_timer_t);
-        if (!priv->shd.timer)
-                goto out;
+    if (!priv->child_up || !priv->child_latency || !priv->halo_child_up ||
+        !priv->anon_inode) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    /*Initialize to -ve ping timeout so that they are not considered
+     * in child-up events until ping-event comes*/
+    for (i = 0; i < child_count; i++)
+        priv->child_latency[i] = -1;
+
+    priv->children = GF_CALLOC(sizeof(xlator_t *), child_count,
+                               gf_afr_mt_xlator_t);
+    if (!priv->children) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    ret = afr_pending_xattrs_init(priv, this);
+    if (ret)
+        goto out;
+
+    trav = this->children;
+    i = 0;
+    while (i < child_count) {
+        priv->children[i] = trav->xlator;
+        trav = trav->next;
+        i++;
+    }
+
+    ret = gf_asprintf(&priv->sh_domain, AFR_SH_DATA_DOMAIN_FMT, this->name);
+    if (-1 == ret) {
+        ret = -ENOMEM;
+        goto out;
+    }
 
-        priv->shd.healed = eh_new (AFR_EH_HEALED_LIMIT, _gf_false);
-        if (!priv->shd.healed)
-                goto out;
+    priv->last_event = GF_CALLOC(child_count, sizeof(*priv->last_event),
+                                 gf_afr_mt_int32_t);
+    if (!priv->last_event) {
+        ret = -ENOMEM;
+        goto out;
+    }
 
-        priv->shd.heal_failed = eh_new (AFR_EH_HEAL_FAIL_LIMIT, _gf_false);
-        if (!priv->shd.heal_failed)
-                goto out;
+    this->itable = inode_table_new(SHD_INODE_LRU_LIMIT, this);
+    if (!this->itable) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    if (priv->shd.iamshd) {
+        ret = afr_selfheal_daemon_init(this);
+        if (ret) {
+            ret = -ENOMEM;
+            goto out;
+        }
+    }
 
-        priv->shd.split_brain = eh_new (AFR_EH_SPLIT_BRAIN_LIMIT, _gf_false);
-        if (!priv->shd.split_brain)
-                goto out;
+    /* keep more local here as we may need them for self-heal etc */
+    this->local_pool = mem_pool_new(afr_local_t, 512);
+    if (!this->local_pool) {
+        ret = -1;
+        goto out;
+    }
 
-        this->itable = inode_table_new (SHD_INODE_LRU_LIMIT, this);
-        if (!this->itable)
-                goto out;
-        priv->root_inode = inode_ref (this->itable->root);
-        GF_OPTION_INIT ("node-uuid", priv->shd.node_uuid, str, out);
-        GF_OPTION_INIT ("heal-timeout", priv->shd.timeout, int32, out);
+    priv->root_inode = NULL;
 
-        ret = 0;
+    ret = 0;
 out:
-        return ret;
+    return ret;
+}
+void
+afr_destroy_healer_object(xlator_t *this, struct subvol_healer *healer)
+{
+    int ret = -1;
+
+    if (!healer)
+        return;
+
+    if (healer->running) {
+        /*
+         * If there are any resources to cleanup, We need
+         * to do that gracefully using pthread_cleanup_push
+         */
+        ret = gf_thread_cleanup_xint(healer->thread);
+        if (ret)
+            gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_SELF_HEAL_FAILED,
+                   "Failed to clean up healer threads.");
+        healer->thread = 0;
+    }
+    pthread_cond_destroy(&healer->cond);
+    pthread_mutex_destroy(&healer->mutex);
 }
 
-
-int
-fini (xlator_t *this)
+void
+afr_selfheal_daemon_fini(xlator_t *this)
+{
+    struct subvol_healer *healer = NULL;
+    afr_self_heald_t *shd = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+
+    priv = this->private;
+    if (!priv)
+        return;
+
+    shd = &priv->shd;
+    if (!shd->iamshd)
+        return;
+
+    for (i = 0; i < priv->child_count; i++) {
+        healer = &shd->index_healers[i];
+        afr_destroy_healer_object(this, healer);
+
+        healer = &shd->full_healers[i];
+        afr_destroy_healer_object(this, healer);
+
+        if (shd->statistics[i])
+            eh_destroy(shd->statistics[i]);
+    }
+    GF_FREE(shd->index_healers);
+    GF_FREE(shd->full_healers);
+    GF_FREE(shd->statistics);
+    if (shd->split_brain)
+        eh_destroy(shd->split_brain);
+}
+void
+fini(xlator_t *this)
 {
-        afr_private_t *priv = NULL;
+    afr_private_t *priv = NULL;
 
-        priv = this->private;
-        this->private = NULL;
-        afr_priv_destroy (priv);
-        if (this->itable);//I dont see any destroy func
+    priv = this->private;
 
-        return 0;
-}
+    afr_selfheal_daemon_fini(this);
+    GF_ASSERT(list_empty(&priv->saved_locks));
+
+    LOCK(&priv->lock);
+    if (priv->timer != NULL) {
+        gf_timer_call_cancel(this->ctx, priv->timer);
+        priv->timer = NULL;
+    }
+    UNLOCK(&priv->lock);
+
+    if (this->local_pool != NULL) {
+        mem_pool_destroy(this->local_pool);
+        this->local_pool = NULL;
+    }
 
+    this->private = NULL;
+    afr_priv_destroy(priv);
+    if (this->itable) {
+        inode_table_destroy(this->itable);
+        this->itable = NULL;
+    }
+
+    return;
+}
 
 struct xlator_fops fops = {
-        .lookup      = afr_lookup,
-        .open        = afr_open,
-        .lk          = afr_lk,
-        .flush       = afr_flush,
-        .statfs      = afr_statfs,
-        .fsync       = afr_fsync,
-        .fsyncdir    = afr_fsyncdir,
-        .xattrop     = afr_xattrop,
-        .fxattrop    = afr_fxattrop,
-        .inodelk     = afr_inodelk,
-        .finodelk    = afr_finodelk,
-        .entrylk     = afr_entrylk,
-        .fentrylk    = afr_fentrylk,
-	.fallocate   = afr_fallocate,
-	.discard     = afr_discard,
-
-        /* inode read */
-        .access      = afr_access,
-        .stat        = afr_stat,
-        .fstat       = afr_fstat,
-        .readlink    = afr_readlink,
-        .getxattr    = afr_getxattr,
-        .fgetxattr   = afr_fgetxattr,
-        .readv       = afr_readv,
-
-        /* inode write */
-        .writev      = afr_writev,
-        .truncate    = afr_truncate,
-        .ftruncate   = afr_ftruncate,
-        .setxattr    = afr_setxattr,
-        .fsetxattr   = afr_fsetxattr,
-        .setattr     = afr_setattr,
-        .fsetattr    = afr_fsetattr,
-        .removexattr = afr_removexattr,
-        .fremovexattr = afr_fremovexattr,
-
-        /* dir read */
-        .opendir     = afr_opendir,
-        .readdir     = afr_readdir,
-        .readdirp    = afr_readdirp,
-
-        /* dir write */
-        .create      = afr_create,
-        .mknod       = afr_mknod,
-        .mkdir       = afr_mkdir,
-        .unlink      = afr_unlink,
-        .rmdir       = afr_rmdir,
-        .link        = afr_link,
-        .symlink     = afr_symlink,
-        .rename      = afr_rename,
+    .lookup = afr_lookup,
+    .lk = afr_lk,
+    .flush = afr_flush,
+    .statfs = afr_statfs,
+    .fsyncdir = afr_fsyncdir,
+    .inodelk = afr_inodelk,
+    .finodelk = afr_finodelk,
+    .entrylk = afr_entrylk,
+    .fentrylk = afr_fentrylk,
+    .ipc = afr_ipc,
+    .lease = afr_lease,
+
+    /* inode read */
+    .access = afr_access,
+    .stat = afr_stat,
+    .fstat = afr_fstat,
+    .readlink = afr_readlink,
+    .getxattr = afr_getxattr,
+    .fgetxattr = afr_fgetxattr,
+    .readv = afr_readv,
+    .seek = afr_seek,
+
+    /* inode write */
+    .writev = afr_writev,
+    .truncate = afr_truncate,
+    .ftruncate = afr_ftruncate,
+    .setxattr = afr_setxattr,
+    .fsetxattr = afr_fsetxattr,
+    .setattr = afr_setattr,
+    .fsetattr = afr_fsetattr,
+    .removexattr = afr_removexattr,
+    .fremovexattr = afr_fremovexattr,
+    .fallocate = afr_fallocate,
+    .discard = afr_discard,
+    .zerofill = afr_zerofill,
+    .xattrop = afr_xattrop,
+    .fxattrop = afr_fxattrop,
+    .fsync = afr_fsync,
+
+    /*inode open*/
+    .opendir = afr_opendir,
+    .open = afr_open,
+
+    /* dir read */
+    .readdir = afr_readdir,
+    .readdirp = afr_readdirp,
+
+    /* dir write */
+    .create = afr_create,
+    .mknod = afr_mknod,
+    .mkdir = afr_mkdir,
+    .unlink = afr_unlink,
+    .rmdir = afr_rmdir,
+    .link = afr_link,
+    .symlink = afr_symlink,
+    .rename = afr_rename,
 };
 
-
 struct xlator_dumpops dumpops = {
-        .priv       = afr_priv_dump,
+    .priv = afr_priv_dump,
 };
 
-
 struct xlator_cbks cbks = {
-        .release     = afr_release,
-        .releasedir  = afr_releasedir,
-        .forget      = afr_forget,
+    .release = afr_release,
+    .releasedir = afr_releasedir,
+    .forget = afr_forget,
 };
 
-
 struct volume_options options[] = {
-        { .key  = {"read-subvolume" },
-          .type = GF_OPTION_TYPE_XLATOR,
-          .description = "inode-read fops happen only on one of the bricks in "
-                         "replicate. Afr will prefer the one specified using "
-                         "this option if it is not stale. Option value must be "
-                         "one of the xlator names of the children. "
-                         "Ex: <volname>-client-0 till "
-                         "<volname>-client-<number-of-bricks - 1>"
-        },
-        { .key  = {"read-subvolume-index" },
-          .type = GF_OPTION_TYPE_INT,
-          .default_value = "-1",
-          .description = "inode-read fops happen only on one of the bricks in "
-                         "replicate. AFR will prefer the one specified using "
-                         "this option if it is not stale. allowed options"
-                         " include -1 till replica-count - 1"
-        },
-        { .key = {"read-hash-mode" },
-          .type = GF_OPTION_TYPE_INT,
-          .min = 0,
-          .max = 2,
-          .default_value = "0",
-          .description = "inode-read fops happen only on one of the bricks in "
-                         "replicate. AFR will prefer the one computed using "
-                         "the method specified using this option"
-                         "0 = first responder, "
-                         "1 = hash by GFID of file (all clients use "
-                                                    "same subvolume), "
-                         "2 = hash by GFID of file and client PID",
-        },
-        { .key  = {"choose-local" },
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "true",
-          .description = "Choose a local subvolume(i.e. Brick) to read from if "
-                         "read-subvolume is not explicitly set.",
-        },
-        { .key  = {"favorite-child"},
-          .type = GF_OPTION_TYPE_XLATOR,
-          .description = "If a split-brain happens choose subvol/brick set by "
-                         "this option as source."
-        },
-        { .key  = {"background-self-heal-count"},
-          .type = GF_OPTION_TYPE_INT,
-          .min  = 0,
-          .default_value = "16",
-          .validate = GF_OPT_VALIDATE_MIN,
-          .description = "This specifies the number of self-heals that can be "
-                         " performed in background without blocking the fop"
-        },
-        { .key  = {"data-self-heal"},
-          .type = GF_OPTION_TYPE_STR,
-          .value = {"1", "on", "yes", "true", "enable",
-                    "0", "off", "no", "false", "disable",
-                    "open"},
-          .default_value = "on",
-          .description   = "Using this option we can enable/disable data "
-                           "self-heal on the file. \"open\" means data "
-                           "self-heal action will only be triggered by file "
-                           "open operations."
-        },
-        { .key  = {"data-self-heal-algorithm"},
-          .type = GF_OPTION_TYPE_STR,
-          .description   = "Select between \"full\", \"diff\". The "
-                           "\"full\" algorithm copies the entire file from "
-                           "source to sink. The \"diff\" algorithm copies to "
-                           "sink only those blocks whose checksums don't match "
-                           "with those of source. If no option is configured "
-                           "the option is chosen dynamically as follows: "
-                           "If the file does not exist on one of the sinks "
-                           "or empty file exists or if the source file size is "
-                           "about the same as page size the entire file will "
-                           "be read and written i.e \"full\" algo, "
-                           "otherwise \"diff\" algo is chosen.",
-          .value = { "diff", "full"}
-        },
-        { .key  = {"data-self-heal-window-size"},
-          .type = GF_OPTION_TYPE_INT,
-          .min  = 1,
-          .max  = 1024,
-          .default_value = "1",
-          .description = "Maximum number blocks per file for which self-heal "
-                         "process would be applied simultaneously."
-        },
-        { .key  = {"metadata-self-heal"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "on",
-          .description = "Using this option we can enable/disable metadata "
-                         "i.e. Permissions, ownerships, xattrs self-heal on "
-                         "the file/directory."
-        },
-        { .key  = {"entry-self-heal"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "on",
-          .description = "Using this option we can enable/disable entry "
-                         "self-heal on the directory."
-        },
-        { .key  = {"data-change-log"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "on",
-          .description = "Data fops like write/truncate will not perform "
-                         "pre/post fop changelog operations in afr transaction "
-                         "if this option is disabled"
-        },
-        { .key  = {"metadata-change-log"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "on",
-          .description = "Metadata fops like setattr/setxattr will not perform "
-                         "pre/post fop changelog operations in afr transaction "
-                         "if this option is disabled"
-        },
-        { .key  = {"entry-change-log"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "on",
-          .description = "Entry fops like create/unlink will not perform "
-                         "pre/post fop changelog operations in afr transaction "
-                         "if this option is disabled"
-        },
-        { .key  = {"optimistic-change-log"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "on",
-          .description = "Entry/Metadata fops will not perform "
-                         "pre fop changelog operations in afr transaction "
-                         "if this option is enabled."
-        },
-        { .key  = {"strict-readdir"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "off",
-        },
-        { .key = {"inodelk-trace"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "off",
-          .description = "Enabling this option logs inode lock/unlocks"
-        },
-        { .key = {"entrylk-trace"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "off",
-          .description = "Enabling this option logs entry lock/unlocks"
-        },
-        { .key = {"eager-lock"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "on",
-          .description = "Lock phase of a transaction has two sub-phases. "
-                         "First is an attempt to acquire locks in parallel by "
-                         "broadcasting non-blocking lock requests. If lock "
-                         "aquistion fails on any server, then the held locks "
-                         "are unlocked and revert to a blocking locked mode "
-                         "sequentially on one server after another.  If this "
-                         "option is enabled the initial broadcasting lock "
-                         "request attempt to acquire lock on the entire file. "
-                         "If this fails, we revert back to the sequential "
-                         "\"regional\" blocking lock as before. In the case "
-                         "where such an \"eager\" lock is granted in the "
-                         "non-blocking phase, it gives rise to an opportunity "
-                         "for optimization. i.e, if the next write transaction "
-                         "on the same FD arrives before the unlock phase of "
-                         "the first transaction, it \"takes over\" the full "
-                         "file lock. Similarly if yet another data transaction "
-                         "arrives before the unlock phase of the \"optimized\" "
-                         "transaction, that in turn \"takes over\" the lock as "
-                         "well. The actual unlock now happens at the end of "
-                         "the last \"optimzed\" transaction."
-
-        },
-        { .key = {"self-heal-daemon"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "off",
-          .description = "This option applies to only self-heal-daemon. "
-                         "Index directory crawl and automatic healing of files"
-                         "will not be performed if this option is turned off."
-        },
-        { .key = {"iam-self-heal-daemon"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "off",
-          .description = "This option differentiates if the replicate "
-                         "translator is running as part of self-heal-daemon "
-                         "or not."
-        },
-        { .key = {"quorum-type"},
-          .type = GF_OPTION_TYPE_STR,
-          .value = { "none", "auto", "fixed"},
-          .default_value = "none",
-          .description = "If value is \"fixed\" only allow writes if "
-                         "quorum-count bricks are present.  If value is "
-                         "\"auto\" only allow writes if more than half of "
-                         "bricks, or exactly half including the first, are "
-                         "present.",
-        },
-        { .key = {"quorum-count"},
-          .type = GF_OPTION_TYPE_INT,
-          .min = 1,
-          .max = INT_MAX,
-          .default_value = 0,
-          .description = "If quorum-type is \"fixed\" only allow writes if "
-                         "this many bricks or present.  Other quorum types "
-                         "will OVERWRITE this value.",
-        },
-        { .key  = {"node-uuid"},
-          .type = GF_OPTION_TYPE_STR,
-          .description = "Local glusterd uuid string, used in starting "
-                         "self-heal-daemon so that it can crawl only on "
-                         "local index directories.",
-        },
-        { .key  = {"heal-timeout"},
-          .type = GF_OPTION_TYPE_INT,
-          .min  = 60,
-          .max  = INT_MAX,
-          .default_value = "600",
-          .description = "time interval for checking the need to self-heal "
-                         "in self-heal-daemon"
-        },
-        { .key  = {"post-op-delay-secs"},
-          .type = GF_OPTION_TYPE_INT,
-          .min  = 0,
-          .max  = INT_MAX,
-          .default_value = "1",
-          .description = "Time interval induced artificially before "
-	                 "post-operation phase of the transaction to "
-                         "enhance overlap of adjacent write operations.",
-        },
-        { .key = {AFR_SH_READDIR_SIZE_KEY},
-          .type = GF_OPTION_TYPE_SIZET,
-          .description = "readdirp size for performing entry self-heal",
-          .min = 1024,
-          .max = 131072,
-          .default_value = "1KB",
-        },
-        { .key = {"readdir-failover"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .description = "readdir(p) will not failover if this option is off",
-          .default_value = "on",
-        },
-        { .key = {"ensure-durability"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .description = "Afr performs fsyncs for transactions if this "
-                         "option is on to make sure the changelogs/data is "
-                         "written to the disk",
-          .default_value = "on",
-        },
-        { .key  = {NULL} },
+    {.key = {"read-subvolume"},
+     .type = GF_OPTION_TYPE_XLATOR,
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "inode-read fops happen only on one of the bricks in "
+                    "replicate. Afr will prefer the one specified using "
+                    "this option if it is not stale. Option value must be "
+                    "one of the xlator names of the children. "
+                    "Ex: <volname>-client-0 till "
+                    "<volname>-client-<number-of-bricks - 1>"},
+    {.key = {"read-subvolume-index"},
+     .type = GF_OPTION_TYPE_INT,
+     .default_value = "-1",
+     .op_version = {2},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "inode-read fops happen only on one of the bricks in "
+                    "replicate. AFR will prefer the one specified using "
+                    "this option if it is not stale. allowed options"
+                    " include -1 till replica-count - 1"},
+    {.key = {"read-hash-mode"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 0,
+     .max = 5,
+     .default_value = "1",
+     .op_version = {2},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description =
+         "inode-read fops happen only on one of the bricks in "
+         "replicate. AFR will prefer the one computed using "
+         "the method specified using this option.\n"
+         "0 = first readable child of AFR, starting from 1st child.\n"
+         "1 = hash by GFID of file (all clients use "
+         "same subvolume).\n"
+         "2 = hash by GFID of file and client PID.\n"
+         "3 = brick having the least outstanding read requests.\n"
+         "4 = brick having the least network ping latency.\n"
+         "5 = Hybrid mode between 3 and 4, ie least value among "
+         "network-latency multiplied by outstanding-read-requests."},
+    {
+        .key = {"choose-local"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "true",
+        .op_version = {2},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"replicate"},
+        .description = "Choose a local subvolume (i.e. Brick) to read from"
+                       " if read-subvolume is not explicitly set.",
+    },
+    {.key = {"background-self-heal-count"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 0,
+     .max = 256,
+     .default_value = "8",
+     .validate = GF_OPT_VALIDATE_MIN,
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "This specifies the number of per client self-heal "
+                    "jobs that can perform parallel heals in the "
+                    "background."},
+    {.key = {"halo-shd-max-latency"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = 99999,
+     .default_value = "99999",
+     .op_version = {GD_OP_VERSION_3_11_0},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate", "halo"},
+     .description = "Maximum latency for shd halo replication in msec."},
+    {.key = {"halo-enabled"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "False",
+     .op_version = {GD_OP_VERSION_3_11_0},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate", "halo"},
+     .description = "Enable Halo (geo) replication mode."},
+    {.key = {"halo-nfsd-max-latency"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = 99999,
+     .default_value = "5",
+     .op_version = {GD_OP_VERSION_3_11_0},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate", "halo"},
+     .description = "Maximum latency for nfsd halo replication in msec."},
+    {.key = {"halo-max-latency"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = AFR_HALO_MAX_LATENCY,
+     .default_value = "5",
+     .op_version = {GD_OP_VERSION_3_11_0},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate", "halo"},
+     .description = "Maximum latency for halo replication in msec."},
+    {.key = {"halo-max-replicas"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = 99999,
+     .default_value = "99999",
+     .op_version = {GD_OP_VERSION_3_11_0},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate", "halo"},
+     .description = "The maximum number of halo replicas; replicas"
+                    " beyond this value will be written asynchronously"
+                    "via the SHD."},
+    {.key = {"halo-min-replicas"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = 99999,
+     .default_value = "2",
+     .op_version = {GD_OP_VERSION_3_11_0},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate", "halo"},
+     .description = "The minimmum number of halo replicas, before adding "
+                    "out of region replicas."},
+    {.key = {"heal-wait-queue-length"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 0,
+     .max = 10000, /*Around 100MB with sizeof(afr_local_t)= 10496 bytes*/
+     .default_value = "128",
+     .validate = GF_OPT_VALIDATE_MIN,
+     .op_version = {GD_OP_VERSION_3_7_10},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "This specifies the number of heals that can be queued"
+                    " for the parallel background self heal jobs."},
+    {.key = {"data-self-heal"},
+     .type = GF_OPTION_TYPE_STR,
+     .value = {"1", "on", "yes", "true", "enable", "0", "off", "no", "false",
+               "disable", "open"},
+     .default_value = "off",
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "Using this option we can enable/disable data "
+                    "self-heal on the file. \"open\" means data "
+                    "self-heal action will only be triggered by file "
+                    "open operations."},
+    {.key = {"data-self-heal-algorithm"},
+     .type = GF_OPTION_TYPE_STR,
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "Select between \"full\", \"diff\". The "
+                    "\"full\" algorithm copies the entire file from "
+                    "source to sink. The \"diff\" algorithm copies to "
+                    "sink only those blocks whose checksums don't match "
+                    "with those of source. If no option is configured "
+                    "the option is chosen dynamically as follows: "
+                    "If the file does not exist on one of the sinks "
+                    "or empty file exists or if the source file size is "
+                    "about the same as page size the entire file will "
+                    "be read and written i.e \"full\" algo, "
+                    "otherwise \"diff\" algo is chosen.",
+     .value = {"diff", "full"}},
+    {.key = {"data-self-heal-window-size"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = 1024,
+     .default_value = "1",
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "Maximum number blocks per file for which self-heal "
+                    "process would be applied simultaneously."},
+    {.key = {"metadata-self-heal"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     /*.validate_fn = validate_replica*/
+     .description = "Using this option we can enable/disable metadata "
+                    "i.e. Permissions, ownerships, xattrs self-heal on "
+                    "the file/directory."},
+    {.key = {"entry-self-heal"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     /*.validate_fn = validate_replica*/
+     .description = "Using this option we can enable/disable entry "
+                    "self-heal on the directory."},
+    {.key = {"data-change-log"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "This option exists only for backward compatibility "
+                    "and configuring it doesn't have any effect"},
+    {.key = {"metadata-change-log"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "This option exists only for backward compatibility "
+                    "and configuring it doesn't have any effect"},
+    {.key = {"entry-change-log"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "This option exists only for backward compatibility "
+                    "and configuring it doesn't have any effect"},
+    {.key = {"optimistic-change-log"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .description = "Entry/Metadata fops will not perform "
+                    "pre fop changelog operations in afr transaction "
+                    "if this option is enabled."},
+    {.key = {"inodelk-trace"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description = "Enabling this option logs inode lock/unlocks"},
+    {.key = {"entrylk-trace"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description = "Enabling this option logs entry lock/unlocks"},
+    {.key = {"pre-op-compat"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .description = "Use separate pre-op xattrop() FOP rather than "
+                    "overloading xdata of the OP"},
+    {.key = {"eager-lock"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description =
+         "Enable/Disable eager lock for replica volume. "
+         "Lock phase of a transaction has two sub-phases. "
+         "First is an attempt to acquire locks in parallel by "
+         "broadcasting non-blocking lock requests. If lock "
+         "acquisition fails on any server, then the held locks "
+         "are unlocked and we revert to a blocking locks mode "
+         "sequentially on one server after another.  If this "
+         "option is enabled the initial broadcasting lock "
+         "request attempts to acquire a full lock on the entire file. "
+         "If this fails, we revert back to the sequential "
+         "\"regional\" blocking locks as before. In the case "
+         "where such an \"eager\" lock is granted in the "
+         "non-blocking phase, it gives rise to an opportunity "
+         "for optimization. i.e, if the next write transaction "
+         "on the same FD arrives before the unlock phase of "
+         "the first transaction, it \"takes over\" the full "
+         "file lock. Similarly if yet another data transaction "
+         "arrives before the unlock phase of the \"optimized\" "
+         "transaction, that in turn \"takes over\" the lock as "
+         "well. The actual unlock now happens at the end of "
+         "the last \"optimized\" transaction."
+
+    },
+    {.key = {"self-heal-daemon"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE,
+     .tags = {"replicate"},
+     /*.validate_fn   = validate_replica_heal_enable_disable*/
+     .description = "This option applies to only self-heal-daemon. "
+                    "Index directory crawl and automatic healing of files "
+                    "will not be performed if this option is turned off."},
+    {.key = {"iam-self-heal-daemon"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description = "This option differentiates if the replicate "
+                    "translator is running as part of self-heal-daemon "
+                    "or not."},
+    {.key = {"iam-nfs-daemon"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description = "This option differentiates if the replicate "
+                    "translator is running as part of an NFS daemon "
+                    "or not."},
+    {
+        .key = {"quorum-type"},
+        .type = GF_OPTION_TYPE_STR,
+        .value = {"none", "auto", "fixed"},
+        .default_value = "none",
+        .op_version = {1},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"replicate"},
+        /*.option = quorum-type*/
+        .description = "If value is \"fixed\" only allow writes if "
+                       "quorum-count bricks are present.  If value is "
+                       "\"auto\" only allow writes if more than half of "
+                       "bricks, or exactly half including the first, are "
+                       "present.",
+    },
+    {
+        .key = {"quorum-count"},
+        .type = GF_OPTION_TYPE_INT,
+        .min = 1,
+        .max = INT_MAX,
+        .default_value = 0,
+        .op_version = {1},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"replicate"},
+        /*.option = quorum-count*/
+        /*.validate_fn = validate_quorum_count*/
+        .description = "If quorum-type is \"fixed\" only allow writes if "
+                       "this many bricks are present.  Other quorum types "
+                       "will OVERWRITE this value.",
+    },
+    {
+        .key = {"quorum-reads"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "no",
+        .op_version = {GD_OP_VERSION_3_7_0},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"replicate"},
+        .description = "This option has been removed. Reads are not allowed "
+                       "if quorum is not met.",
+    },
+    {
+        .key = {"node-uuid"},
+        .type = GF_OPTION_TYPE_STR,
+        .description = "Local glusterd uuid string, used in starting "
+                       "self-heal-daemon so that it can crawl only on "
+                       "local index directories.",
+    },
+    {
+        .key = {"post-op-delay-secs"},
+        .type = GF_OPTION_TYPE_INT,
+        .min = 0,
+        .max = INT_MAX,
+        .default_value = "1",
+        .op_version = {2},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"replicate"},
+        .description = "Time interval induced artificially before "
+                       "post-operation phase of the transaction to "
+                       "enhance overlap of adjacent write operations.",
+    },
+    {
+        .key = {AFR_SH_READDIR_SIZE_KEY},
+        .type = GF_OPTION_TYPE_SIZET,
+        .description = "readdirp size for performing entry self-heal",
+        .min = 1024,
+        .max = 131072,
+        .op_version = {2},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE,
+        .tags = {"replicate"},
+        .default_value = "1KB",
+    },
+    {
+        .key = {"ensure-durability"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .op_version = {3},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"replicate"},
+        .description = "Afr performs fsyncs for transactions if this "
+                       "option is on to make sure the changelogs/data is "
+                       "written to the disk",
+        .default_value = "on",
+    },
+    {
+        .key = {"afr-dirty-xattr"},
+        .type = GF_OPTION_TYPE_STR,
+        .default_value = AFR_DIRTY_DEFAULT,
+    },
+    {.key = {"afr-pending-xattr"},
+     .type = GF_OPTION_TYPE_STR,
+     .description = "Comma separated list of xattrs that are used to  "
+                    "capture information on pending heals."},
+    {
+        .key = {"metadata-splitbrain-forced-heal"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "off",
+    },
+    {.key = {"heal-timeout"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 5,
+     .max = INT_MAX,
+     .default_value = "600",
+     .op_version = {2},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "time interval for checking the need to self-heal "
+                    "in self-heal-daemon"},
+    {
+        .key = {"consistent-metadata"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "no",
+        .op_version = {GD_OP_VERSION_3_7_0},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"replicate"},
+        .description = "If this option is enabled, readdirp will force "
+                       "lookups on those entries read whose read child is "
+                       "not the same as that of the parent. This will "
+                       "guarantee that all read operations on a file serve "
+                       "attributes from the same subvol as long as it holds "
+                       " a good copy of the file/dir.",
+    },
+    {.key = {"arbiter-count"},
+     .type = GF_OPTION_TYPE_INT,
+     .description = "subset of child_count. Has to be 0 or 1."},
+    {
+        .key = {"thin-arbiter"},
+        .type = GF_OPTION_TYPE_STR,
+        .op_version = {GD_OP_VERSION_4_1_0},
+        .flags = OPT_FLAG_SETTABLE,
+        .tags = {"replicate"},
+        .description = "contains host:path of thin abriter brick",
+    },
+    {.key = {"shd-max-threads"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = 64,
+     .default_value = "1",
+     .op_version = {GD_OP_VERSION_3_7_12},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "Maximum number of parallel heals SHD can do per "
+                    "local brick. This can substantially lower heal times"
+                    ", but can also crush your bricks if you don't have "
+                    "the storage hardware to support this."},
+    {
+        .key = {"shd-wait-qlength"},
+        .type = GF_OPTION_TYPE_INT,
+        .min = 1,
+        .max = 655536,
+        .default_value = "1024",
+        .op_version = {GD_OP_VERSION_3_7_12},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"replicate"},
+        .description = "This option can be used to control number of heals"
+                       " that can wait in SHD per subvolume",
+    },
+    {
+        .key = {"locking-scheme"},
+        .type = GF_OPTION_TYPE_STR,
+        .value = {"full", "granular"},
+        .default_value = "full",
+        .op_version = {GD_OP_VERSION_3_7_12},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"replicate"},
+        .description = "If this option is set to granular, self-heal will "
+                       "stop being compatible with afr-v1, which helps afr "
+                       "be more granular while self-healing",
+    },
+    {.key = {"full-lock"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "yes",
+     .op_version = {GD_OP_VERSION_3_13_2},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE,
+     .tags = {"replicate"},
+     .description = "If this option is disabled, then the IOs will take "
+                    "range locks same as versions till 3.13.1."},
+    {
+        .key = {"granular-entry-heal"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "no",
+        .op_version = {GD_OP_VERSION_3_8_0},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"replicate"},
+        .description = "If this option is enabled, self-heal will resort to "
+                       "granular way of recording changelogs and doing entry "
+                       "self-heal.",
+    },
+    {
+        .key = {"favorite-child-policy"},
+        .type = GF_OPTION_TYPE_STR,
+        .value = {"none", "size", "ctime", "mtime", "majority"},
+        .default_value = "none",
+        .op_version = {GD_OP_VERSION_3_7_12},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"replicate"},
+        .description = "This option can be used to automatically resolve "
+                       "split-brains using various policies without user "
+                       "intervention. \"size\" picks the file with the "
+                       "biggest size as the source. \"ctime\" and \"mtime\" "
+                       "pick the file with the latest ctime and mtime "
+                       "respectively as the source. \"majority\" picks a file"
+                       " with identical mtime and size in more than half the "
+                       "number of bricks in the replica.",
+    },
+    {
+        .key = {"consistent-io"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "no",
+        .description = "If this option is enabled, i/o will fail even if "
+                       "one of the bricks is down in the replicas",
+    },
+    {.key = {"use-compound-fops"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "no",
+     .op_version = {GD_OP_VERSION_3_8_4},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "This option exists only for backward compatibility "
+                    "and configuring it doesn't have any effect"},
+    {.key = {"use-anonymous-inode"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "no",
+     .op_version = {GD_OP_VERSION_8_0},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE,
+     .tags = {"replicate"},
+     .description = "Setting this option heals directory renames efficiently"},
+
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .dumpops = &dumpops,
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "replicate",
+    .category = GF_MAINTAINED,
 };
diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h
index d49cb6ccb56..d62f9a9caf2 100644
--- a/xlators/cluster/afr/src/afr.h
+++ b/xlators/cluster/afr/src/afr.h
@@ -8,1169 +8,1416 @@
   cases as published by the Free Software Foundation.
 */
 
-
 #ifndef __AFR_H__
 #define __AFR_H__
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "call-stub.h"
-#include "compat-errno.h"
+#include <glusterfs/call-stub.h>
+#include <glusterfs/compat-errno.h>
 #include "afr-mem-types.h"
-#include "afr-self-heal-algorithm.h"
 
 #include "libxlator.h"
-#include "timer.h"
+#include <glusterfs/timer.h>
+#include <glusterfs/syncop.h>
+
+#include "afr-self-heald.h"
+#include "afr-messages.h"
 
-#define AFR_XATTR_PREFIX "trusted.afr"
+#define SHD_INODE_LRU_LIMIT 1
 #define AFR_PATHINFO_HEADER "REPLICATE:"
 #define AFR_SH_READDIR_SIZE_KEY "self-heal-readdir-size"
+#define AFR_SH_DATA_DOMAIN_FMT "%s:self-heal"
+#define AFR_DIRTY_DEFAULT AFR_XATTR_PREFIX ".dirty"
+#define AFR_DIRTY (((afr_private_t *)(THIS->private))->afr_dirty)
+
+#define AFR_LOCKEE_COUNT_MAX 3
+#define AFR_DOM_COUNT_MAX 3
+#define AFR_NUM_CHANGE_LOGS 3              /*data + metadata + entry*/
+#define AFR_DEFAULT_SPB_CHOICE_TIMEOUT 300 /*in seconds*/
+
+#define ARBITER_BRICK_INDEX 2
+#define THIN_ARBITER_BRICK_INDEX 2
+#define AFR_TA_DOM_NOTIFY "afr.ta.dom-notify"
+#define AFR_TA_DOM_MODIFY "afr.ta.dom-modify"
+
+#define AFR_LK_HEAL_DOM "afr.lock-heal.domain"
+
+#define AFR_HALO_MAX_LATENCY 99999
+#define AFR_ANON_DIR_PREFIX ".glusterfs-anonymous-inode"
+
+#define PFLAG_PENDING (1 << 0)
+#define PFLAG_SBRAIN (1 << 1)
+
+typedef int (*afr_lock_cbk_t)(call_frame_t *frame, xlator_t *this);
+
+typedef int (*afr_read_txn_wind_t)(call_frame_t *frame, xlator_t *this,
+                                   int subvol);
+
+typedef int (*afr_inode_refresh_cbk_t)(call_frame_t *frame, xlator_t *this,
+                                       int err);
+
+typedef int (*afr_changelog_resume_t)(call_frame_t *frame, xlator_t *this);
+
+#define AFR_COUNT(array, max)                                                  \
+    ({                                                                         \
+        int __i;                                                               \
+        int __res = 0;                                                         \
+        for (__i = 0; __i < max; __i++)                                        \
+            if (array[__i])                                                    \
+                __res++;                                                       \
+        __res;                                                                 \
+    })
+#define AFR_INTERSECT(dst, src1, src2, max)                                    \
+    ({                                                                         \
+        int __i;                                                               \
+        for (__i = 0; __i < max; __i++)                                        \
+            dst[__i] = src1[__i] && src2[__i];                                 \
+    })
+#define AFR_CMP(a1, a2, len)                                                   \
+    ({                                                                         \
+        int __cmp = 0;                                                         \
+        int __i;                                                               \
+        for (__i = 0; __i < len; __i++)                                        \
+            if (a1[__i] != a2[__i]) {                                          \
+                __cmp = 1;                                                     \
+                break;                                                         \
+            }                                                                  \
+        __cmp;                                                                 \
+    })
+#define AFR_IS_ARBITER_BRICK(priv, index)                                      \
+    ((priv->arbiter_count == 1) && (index == ARBITER_BRICK_INDEX))
+
+#define AFR_SET_ERROR_AND_CHECK_SPLIT_BRAIN(ret, errnum)                       \
+    do {                                                                       \
+        local->op_ret = ret;                                                   \
+        local->op_errno = errnum;                                              \
+        if (local->op_errno == EIO)                                            \
+            gf_msg(this->name, GF_LOG_ERROR, local->op_errno,                  \
+                   AFR_MSG_SPLIT_BRAIN,                                        \
+                   "Failing %s on gfid %s: "                                   \
+                   "split-brain observed.",                                    \
+                   gf_fop_list[local->op], uuid_utoa(local->inode->gfid));     \
+    } while (0)
+
+#define AFR_ERROR_OUT_IF_FDCTX_INVALID(__fd, __this, __error, __label)         \
+    do {                                                                       \
+        afr_fd_ctx_t *__fd_ctx = NULL;                                         \
+        __fd_ctx = afr_fd_ctx_get(__fd, __this);                               \
+        if (__fd_ctx && __fd_ctx->is_fd_bad) {                                 \
+            __error = EBADF;                                                   \
+            goto __label;                                                      \
+        }                                                                      \
+    } while (0)
 
-#define AFR_LOCKEE_COUNT_MAX    3
-#define AFR_DOM_COUNT_MAX    3
-
-struct _pump_private;
-
-typedef int (*afr_expunge_done_cbk_t) (call_frame_t *frame, xlator_t *this,
-                                       int child, int32_t op_error,
-                                       int32_t op_errno);
-
-typedef int (*afr_impunge_done_cbk_t) (call_frame_t *frame, xlator_t *this,
-                                       int32_t op_error, int32_t op_errno);
-typedef int (*afr_post_remove_call_t) (call_frame_t *frame, xlator_t *this);
-
-typedef int (*afr_lock_cbk_t) (call_frame_t *frame, xlator_t *this);
-typedef void (*afr_lookup_done_cbk_t) (call_frame_t *frame, xlator_t *this,
-                                      int32_t op_ret, int32_t op_errno);
+typedef enum {
+    AFR_READ_POLICY_FIRST_UP,
+    AFR_READ_POLICY_GFID_HASH,
+    AFR_READ_POLICY_GFID_PID_HASH,
+    AFR_READ_POLICY_LESS_LOAD,
+    AFR_READ_POLICY_LEAST_LATENCY,
+    AFR_READ_POLICY_LOAD_LATENCY_HYBRID,
+} afr_read_hash_mode_t;
 
 typedef enum {
-        AFR_POS_UNKNOWN,
-        AFR_POS_LOCAL,
-        AFR_POS_REMOTE
-} afr_child_pos_t;
+    AFR_FAV_CHILD_NONE,
+    AFR_FAV_CHILD_BY_SIZE,
+    AFR_FAV_CHILD_BY_CTIME,
+    AFR_FAV_CHILD_BY_MTIME,
+    AFR_FAV_CHILD_BY_MAJORITY,
+    AFR_FAV_CHILD_POLICY_MAX,
+} afr_favorite_child_policy;
 
 typedef enum {
-        SPLIT_BRAIN = 1,
-        ALL_FOOLS = 2
-} afr_subvol_status_t;
+    AFR_SELFHEAL_DATA_FULL = 0,
+    AFR_SELFHEAL_DATA_DIFF,
+    AFR_SELFHEAL_DATA_DYNAMIC,
+} afr_data_self_heal_type_t;
 
 typedef enum {
-        AFR_INODE_SET_READ_CTX = 1,
-        AFR_INODE_RM_STALE_CHILDREN,
-        AFR_INODE_SET_OPENDIR_DONE,
-        AFR_INODE_GET_READ_CTX,
-        AFR_INODE_GET_OPENDIR_DONE,
-} afr_inode_op_t;
-
-typedef struct afr_inode_params_ {
-        afr_inode_op_t op;
-        union {
-                gf_boolean_t value;
-                struct {
-                        int32_t read_child;
-                        int32_t *children;
-                } read_ctx;
-        } u;
-} afr_inode_params_t;
-
-typedef enum afr_spb_state {
-        DONT_KNOW,
-        SPB,
-        NO_SPB
-} afr_spb_state_t;
-
-typedef struct afr_inode_ctx_ {
-        uint64_t masks;
-        int32_t  *fresh_children;//increasing order of latency
-        afr_spb_state_t mdata_spb;
-        afr_spb_state_t data_spb;
-        uint32_t        open_fd_count;
-} afr_inode_ctx_t;
+    AFR_CHILD_UNKNOWN = -1,
+    AFR_CHILD_ZERO,
+    AFR_CHILD_ONE,
+    AFR_CHILD_THIN_ARBITER,
+} afr_child_index;
 
 typedef enum {
-        NONE,
-        INDEX,
-        FULL,
-} afr_crawl_type_t;
-
-typedef struct afr_self_heald_ {
-        gf_boolean_t     enabled;
-        gf_boolean_t     iamshd;
-        afr_crawl_type_t *pending;
-        gf_boolean_t     *inprogress;
-        afr_child_pos_t  *pos;
-        gf_timer_t       **timer;
-        eh_t             *healed;
-        eh_t             *heal_failed;
-        eh_t             *split_brain;
-        char             *node_uuid;
-        int              timeout;
-} afr_self_heald_t;
+    TA_WAIT_FOR_NOTIFY_LOCK_REL, /*FOP came after notify domain lock upcall
+                                   notification and waiting for its release.*/
+    TA_GET_INFO_FROM_TA_FILE,    /*FOP needs post-op on ta file to get
+                                  *info about which brick is bad.*/
+    TA_INFO_IN_MEMORY_SUCCESS,   /*Bad brick info is in memory and fop failed
+                                  *on BAD brick - Success*/
+    TA_INFO_IN_MEMORY_FAILED,    /*Bad brick info is in memory and fop failed
+                                  *on GOOD brick - Failed*/
+    TA_SUCCESS,                  /*FOP succeeded on both data bricks.*/
+} afr_ta_fop_state_t;
+
+struct afr_nfsd {
+    uint32_t halo_max_latency_msec;
+    gf_boolean_t iamnfsd;
+};
+
+typedef struct _afr_lk_heal_info {
+    fd_t *fd;
+    int32_t cmd;
+    struct gf_flock flock;
+    dict_t *xdata_req;
+    unsigned char *locked_nodes;
+    struct list_head pos;
+    gf_lkowner_t lk_owner;
+    pid_t pid;
+    int32_t *child_up_event_gen;
+    int32_t *child_down_event_gen;
+} afr_lk_heal_info_t;
 
 typedef struct _afr_private {
-        gf_lock_t lock;               /* to guard access to child_count, etc */
-        unsigned int child_count;     /* total number of children   */
+    gf_lock_t lock;             /* to guard access to child_count, etc */
+    unsigned int child_count;   /* total number of children   */
+    unsigned int arbiter_count; /*subset of child_count.
+                                  Has to be 0 or 1.*/
+
+    xlator_t **children;
+
+    inode_t *root_inode;
+
+    int favorite_child; /* subvolume to be preferred in resolving
+                                    split-brain cases */
+    /* For thin-arbiter. */
+    uuid_t ta_gfid;
+    unsigned int thin_arbiter_count; /* 0 or 1 at the moment.*/
+    int ta_bad_child_index;
+    int ta_event_gen;
+    unsigned int ta_in_mem_txn_count;
+    unsigned int ta_on_wire_txn_count;
+    struct list_head ta_waitq;
+    struct list_head ta_onwireq;
+
+    unsigned char *anon_inode;
+    unsigned char *child_up;
+    unsigned char *halo_child_up;
+    int64_t *child_latency;
+    unsigned char *local;
+
+    char **pending_key;
+
+    afr_data_self_heal_type_t data_self_heal_algorithm;
+    unsigned int data_self_heal_window_size; /* max number of pipelined
+                                                read/writes */
+
+    struct list_head heal_waiting; /*queue for files that need heal*/
+    uint32_t heal_wait_qlen; /*configurable queue length for heal_waiting*/
+    int32_t heal_waiters;    /* No. of elements currently in wait queue.*/
+
+    struct list_head healing;            /* queue for files that are undergoing
+                                            background heal*/
+    uint32_t background_self_heal_count; /*configurable queue length for
+                                           healing queue*/
+    int32_t healers; /* No. of elements currently undergoing background
+                      heal*/
+
+    gf_boolean_t release_ta_notify_dom_lock;
+
+    gf_boolean_t metadata_self_heal; /* on/off */
+    gf_boolean_t entry_self_heal;    /* on/off */
+
+    gf_boolean_t metadata_splitbrain_forced_heal; /* on/off */
+    int read_child;                               /* read-subvolume */
+    gf_atomic_t *pending_reads; /*No. of pending read cbks per child.*/
+
+    gf_timer_t *timer; /* launched when parent up is received */
+
+    unsigned int wait_count; /* # of servers to wait for success */
+
+    unsigned char ta_child_up;
+    gf_boolean_t optimistic_change_log;
+    gf_boolean_t eager_lock;
+    gf_boolean_t pre_op_compat; /* on/off */
+    uint32_t post_op_delay_secs;
+    unsigned int quorum_count;
+
+    off_t ta_notify_dom_lock_offset;
+    afr_favorite_child_policy fav_child_policy; /*Policy to use for automatic
+                                              resolution of split-brains.*/
+    afr_read_hash_mode_t hash_mode; /* for when read_child is not set */
+
+    int32_t *last_event;
+
+    /* @event_generation: Keeps count of number of events received which can
+       potentially impact consistency decisions. The events are CHILD_UP
+       and CHILD_DOWN, when we have to recalculate the freshness/staleness
+       of copies to detect if changes had happened while the other server
+       was down. CHILD_DOWN and CHILD_UP can also be received on network
+       disconnect/reconnects and not necessarily server going down/up.
+       Recalculating freshness/staleness on network events is equally
+       important as we might have had a network split brain.
+    */
+    uint32_t event_generation;
+    char vol_uuid[UUID_SIZE + 1];
+
+    gf_boolean_t choose_local;
+    gf_boolean_t did_discovery;
+    gf_boolean_t ensure_durability;
+    gf_boolean_t halo_enabled;
+    gf_boolean_t consistent_metadata;
+    gf_boolean_t need_heal;
+    gf_boolean_t granular_locks;
+    uint64_t sh_readdir_size;
+    char *sh_domain;
+    char *afr_dirty;
+
+    uint64_t spb_choice_timeout;
+
+    afr_self_heald_t shd;
+    struct afr_nfsd nfsd;
+
+    uint32_t halo_max_latency_msec;
+    uint32_t halo_max_replicas;
+    uint32_t halo_min_replicas;
+
+    gf_boolean_t full_lock;
+    gf_boolean_t esh_granular;
+    gf_boolean_t consistent_io;
+    gf_boolean_t data_self_heal; /* on/off */
+    gf_boolean_t use_anon_inode;
+
+    /*For lock healing.*/
+    struct list_head saved_locks;
+    struct list_head lk_healq;
+
+    /*For anon-inode handling */
+    char anon_inode_name[NAME_MAX + 1];
+    char anon_gfid_str[UUID_SIZE + 1];
+} afr_private_t;
 
-        unsigned int read_child_rr;   /* round-robin index of the read_child */
-        gf_lock_t read_child_lock;    /* lock to protect above */
+typedef enum {
+    AFR_DATA_TRANSACTION,         /* truncate, write, ... */
+    AFR_METADATA_TRANSACTION,     /* chmod, chown, ... */
+    AFR_ENTRY_TRANSACTION,        /* create, rmdir, ... */
+    AFR_ENTRY_RENAME_TRANSACTION, /* rename */
+} afr_transaction_type;
 
-        xlator_t **children;
+/*
+  xattr format: trusted.afr.volume = [x y z]
+  x - data pending
+  y - metadata pending
+  z - entry pending
+*/
 
-        int first_lookup;
-        inode_t *root_inode;
+static inline int
+afr_index_for_transaction_type(afr_transaction_type type)
+{
+    switch (type) {
+        case AFR_DATA_TRANSACTION:
+            return 0;
 
-        unsigned char *child_up;
+        case AFR_METADATA_TRANSACTION:
+            return 1;
 
-        char **pending_key;
+        case AFR_ENTRY_TRANSACTION:
+        case AFR_ENTRY_RENAME_TRANSACTION:
+            return 2;
+    }
 
-        char         *data_self_heal;              /* on/off/open */
-        char *       data_self_heal_algorithm;    /* name of algorithm */
-        unsigned int data_self_heal_window_size;  /* max number of pipelined
-                                                     read/writes */
+    return -1; /* make gcc happy */
+}
 
-        unsigned int background_self_heal_count;
-        unsigned int background_self_heals_started;
-        gf_boolean_t metadata_self_heal;   /* on/off */
-        gf_boolean_t entry_self_heal;      /* on/off */
+static inline int
+afr_index_from_ia_type(ia_type_t type)
+{
+    switch (type) {
+        case IA_IFDIR:
+            return afr_index_for_transaction_type(AFR_ENTRY_TRANSACTION);
+        case IA_IFREG:
+            return afr_index_for_transaction_type(AFR_DATA_TRANSACTION);
+        default:
+            return -1;
+    }
+}
 
-        gf_boolean_t data_change_log;       /* on/off */
-        gf_boolean_t metadata_change_log;   /* on/off */
-        gf_boolean_t entry_change_log;      /* on/off */
+typedef struct {
+    struct gf_flock flock;
+    loc_t loc;
+    fd_t *fd;
+    char *basename;
+    unsigned char *locked_nodes;
+    int locked_count;
 
-        int read_child;               /* read-subvolume */
-        unsigned int hash_mode;       /* for when read_child is not set */
-        int favorite_child;  /* subvolume to be preferred in resolving
-                                         split-brain cases */
+} afr_lockee_t;
 
-        gf_boolean_t inodelk_trace;
-        gf_boolean_t entrylk_trace;
+int
+afr_entry_lockee_cmp(const void *l1, const void *l2);
 
-        gf_boolean_t strict_readdir;
+typedef struct {
+    loc_t *lk_loc;
 
-        unsigned int wait_count;      /* # of servers to wait for success */
+    afr_lockee_t lockee[AFR_LOCKEE_COUNT_MAX];
 
-        uint64_t up_count;      /* number of CHILD_UPs we have seen */
-        uint64_t down_count;    /* number of CHILD_DOWNs we have seen */
+    const char *lk_basename;
+    const char *lower_basename;
+    const char *higher_basename;
 
-        struct _pump_private *pump_private; /* Set if we are loaded as pump */
-        int                   use_afr_in_pump;
+    unsigned char *lower_locked_nodes;
 
-        pthread_mutex_t  mutex;
-        struct list_head saved_fds;   /* list of fds on which locks have succeeded */
-        gf_boolean_t      optimistic_change_log;
-        gf_boolean_t      eager_lock;
-	uint32_t          post_op_delay_secs;
-        unsigned int      quorum_count;
+    afr_lock_cbk_t lock_cbk;
 
-        char                   vol_uuid[UUID_SIZE + 1];
-        int32_t                *last_event;
-        afr_self_heald_t       shd;
-        gf_boolean_t           choose_local;
-        gf_boolean_t           did_discovery;
-        gf_boolean_t           readdir_failover;
-        uint64_t               sh_readdir_size;
-        gf_boolean_t           ensure_durability;
-        char                   *sh_domain;
-} afr_private_t;
+    int lockee_count;
+
+    int32_t lk_call_count;
+    int32_t lk_expected_count;
+    int32_t lk_attempted_count;
+
+    int32_t lock_op_ret;
+    int32_t lock_op_errno;
+    char *domain; /* Domain on which inode/entry lock/unlock in progress.*/
+    int32_t lock_count;
+    char lower_locked;
+    char higher_locked;
+} afr_internal_lock_t;
+
+struct afr_reply {
+    int valid;
+    int32_t op_ret;
+    dict_t *xattr; /*For xattrop*/
+    dict_t *xdata;
+    struct iatt poststat;
+    struct iatt postparent;
+    struct iatt prestat;
+    struct iatt preparent;
+    struct iatt preparent2;
+    struct iatt postparent2;
+    int32_t op_errno;
+    /* For rchecksum */
+    uint8_t checksum[SHA256_DIGEST_LENGTH];
+    gf_boolean_t buf_has_zeroes;
+    gf_boolean_t fips_mode_rchecksum;
+    /* For lookup */
+    int8_t need_heal;
+};
 
 typedef enum {
-        AFR_SELF_HEAL_NOT_ATTEMPTED,
-        AFR_SELF_HEAL_STARTED,
-        AFR_SELF_HEAL_FAILED,
-        AFR_SELF_HEAL_SYNC_BEGIN,
-} afr_self_heal_status;
+    AFR_FD_NOT_OPENED,
+    AFR_FD_OPENED,
+    AFR_FD_OPENING
+} afr_fd_open_status_t;
 
 typedef struct {
-        afr_self_heal_status gfid_or_missing_entry_self_heal;
-        afr_self_heal_status metadata_self_heal;
-        afr_self_heal_status data_self_heal;
-        afr_self_heal_status entry_self_heal;
-} afr_sh_status_for_all_type;
+    afr_fd_open_status_t *opened_on; /* which subvolumes the fd is open on */
+    int flags;
+
+    /* the subvolume on which the latest sequence of readdirs (starting
+       at offset 0) has begun. Till the next readdir request with 0 offset
+       arrives, we continue to read off this subvol.
+    */
+    int readdir_subvol;
+    /* lock-healing related members. */
+    gf_boolean_t is_fd_bad;
+    afr_lk_heal_info_t *lk_heal_info;
 
-typedef enum {
-        AFR_SELF_HEAL_ENTRY,
-        AFR_SELF_HEAL_METADATA,
-        AFR_SELF_HEAL_DATA,
-        AFR_SELF_HEAL_GFID_OR_MISSING_ENTRY,
-        AFR_SELF_HEAL_INVALID = -1,
-} afr_self_heal_type;
+} afr_fd_ctx_t;
 
 typedef enum {
-        AFR_CHECK_ALL,
-        AFR_CHECK_SPECIFIC,
-} afr_sh_fail_check_type;
-
-struct afr_self_heal_ {
-        /* External interface: These are variables (some optional) that
-           are set by whoever has triggered self-heal */
-
-        gf_boolean_t do_data_self_heal;
-        gf_boolean_t do_metadata_self_heal;
-        gf_boolean_t do_entry_self_heal;
-        gf_boolean_t do_gfid_self_heal;
-        gf_boolean_t do_missing_entry_self_heal;
-        gf_boolean_t force_confirm_spb; /* Check for split-brains even when
-                                           self-heal is turned off */
-
-        gf_boolean_t forced_merge;        /* Is this a self-heal triggered to
-                                             forcibly merge the directories? */
-
-        gf_boolean_t background;          /* do self-heal in background
-                                             if possible */
-        ia_type_t type;                   /* st_mode of the entry we're doing
-                                             self-heal on */
-        inode_t   *inode;                 /* inode on which the self-heal is
-                                             performed on */
-        uuid_t  sh_gfid_req;                 /* gfid self-heal needs to be done
-                                             with this gfid if it is not null */
-
-        /* Function to call to unwind. If self-heal is being done in the
-           background, this function will be called as soon as possible. */
-
-        int (*unwind) (call_frame_t *frame, xlator_t *this, int32_t op_ret,
-                       int32_t op_errno, int32_t sh_failed);
-
-        /* End of external interface members */
-
-
-        /* array of stat's, one for each child */
-        struct iatt *buf;
-        struct iatt *parentbufs;
-        struct iatt parentbuf;
-        struct iatt entrybuf;
-
-        afr_expunge_done_cbk_t expunge_done;
-        afr_impunge_done_cbk_t impunge_done;
-
-        /* array of xattr's, one for each child */
-        dict_t **xattr;
-
-        /* array containing if the lookups succeeded in the order of response
-         */
-        int32_t *success_children;
-        int     success_count;
-        /* array containing the fresh children found in the self-heal process */
-        int32_t *fresh_children;
-        /* array containing the fresh children found in the parent lookup */
-        int32_t *fresh_parent_dirs;
-        /* array of errno's, one for each child */
-        int *child_errno;
-        /*loc used for lookup*/
-        loc_t lookup_loc;
-        int32_t lookup_flags;
-        afr_lookup_done_cbk_t lookup_done;
-
-        int32_t **pending_matrix;
-        int32_t **delta_matrix;
-
-        int32_t op_ret;
-        int32_t op_errno;
-
-        int *sources;
-        int source;
-        int active_source;
-        int active_sinks;
-        unsigned char *success;
-        unsigned char *locked_nodes;
-        int lock_count;
-
-        const char *linkname;
-        gf_boolean_t entries_skipped;
-
-        gf_boolean_t actual_sh_started;
-        gf_boolean_t sync_done;
-        gf_boolean_t data_lock_held;
-        gf_boolean_t sh_dom_lock_held;
-        gf_boolean_t eof_reached;
-        fd_t  *healing_fd;
-        int   file_has_holes;
-        blksize_t block_size;
-        off_t file_size;
-        off_t offset;
-        unsigned char *write_needed;
-        uint8_t *checksum;
-        afr_post_remove_call_t post_remove_call;
-
-        char    *data_sh_info;
-        char    *metadata_sh_info;
+    AFR_FOP_LOCK_PARALLEL,
+    AFR_FOP_LOCK_SERIAL,
+    AFR_FOP_LOCK_QUORUM_FAILED,
+} afr_fop_lock_state_t;
+
+typedef struct _afr_inode_lock_t {
+    /* @num_inodelks:
+       Number of inodelks queried from the server, as queried through
+       xdata in FOPs. Currently, used to decide if eager-locking must be
+       temporarily disabled.
+    */
+    int32_t num_inodelks;
+    unsigned int event_generation;
+    gf_timer_t *delay_timer;
+    struct list_head owners;  /*Transactions that are performing fop*/
+    struct list_head post_op; /*Transactions that are done with the fop
+                               *So can not conflict with the fops*/
+    struct list_head waiting; /*Transaction that are waiting for
+                               *conflicting transactions to complete*/
+    struct list_head frozen;  /*Transactions that need to go as part of
+                               * next batch of eager-lock*/
+    gf_boolean_t release;
+    gf_boolean_t acquired;
+} afr_lock_t;
+
+typedef struct _afr_inode_ctx {
+    uint64_t read_subvol;
+    uint64_t write_subvol;
+    int lock_count;
+    int spb_choice;
+    gf_timer_t *timer;
+    unsigned int *pre_op_done[AFR_NUM_CHANGE_LOGS];
+    int inherited[AFR_NUM_CHANGE_LOGS];
+    int on_disk[AFR_NUM_CHANGE_LOGS];
+    /*Only 2 types of transactions support eager-locks now. DATA/METADATA*/
+    afr_lock_t lock[2];
+
+    /* @open_fd_count:
+       Number of open FDs queried from the server, as queried through
+       xdata in FOPs. Currently, used to decide if eager-locking must be
+       temporarily disabled.
+    */
+    uint32_t open_fd_count;
+    gf_boolean_t need_refresh;
+
+    /* set if any write on this fd was a non stable write
+       (i.e, without O_SYNC or O_DSYNC)
+    */
+    gf_boolean_t witnessed_unstable_write;
+} afr_inode_ctx_t;
 
-        loc_t parent_loc;
-        call_frame_t *orig_frame;
-        call_frame_t *old_loop_frame;
-        gf_boolean_t unwound;
-
-        afr_sh_algo_private_t *private;
-        afr_sh_status_for_all_type  afr_all_sh_status;
-        afr_self_heal_type       sh_type_in_action;
-
-        struct afr_sh_algorithm  *algo;
-        afr_lock_cbk_t data_lock_success_handler;
-        afr_lock_cbk_t data_lock_failure_handler;
-	gf_boolean_t data_lock_block;
-        int (*completion_cbk) (call_frame_t *frame, xlator_t *this);
-        int (*sh_data_algo_start) (call_frame_t *frame, xlator_t *this);
-        int (*algo_completion_cbk) (call_frame_t *frame, xlator_t *this);
-        int (*algo_abort_cbk) (call_frame_t *frame, xlator_t *this);
-        void (*gfid_sh_success_cbk) (call_frame_t *sh_frame, xlator_t *this);
-
-        call_frame_t *sh_frame;
-};
+typedef struct _afr_local {
+    glusterfs_fop_t op;
+    unsigned int call_count;
 
-typedef struct afr_self_heal_ afr_self_heal_t;
+    /* @event_generation: copy of priv->event_generation taken at the
+       time of starting the transaction. The copy is made so that we
+       have a stable value through the various phases of the transaction.
+    */
+    unsigned int event_generation;
 
-typedef enum {
-        AFR_DATA_TRANSACTION,          /* truncate, write, ... */
-        AFR_METADATA_TRANSACTION,      /* chmod, chown, ... */
-        AFR_ENTRY_TRANSACTION,         /* create, rmdir, ... */
-        AFR_ENTRY_RENAME_TRANSACTION,  /* rename */
-} afr_transaction_type;
+    uint32_t open_fd_count;
+    int32_t num_inodelks;
 
-typedef enum {
-        AFR_TRANSACTION_LK,
-        AFR_SELFHEAL_LK,
-} transaction_lk_type_t;
+    int32_t op_ret;
+    int32_t op_errno;
 
-typedef enum {
-        AFR_LOCK_OP,
-        AFR_UNLOCK_OP,
-} afr_lock_op_type_t;
+    int dirty[AFR_NUM_CHANGE_LOGS];
 
-typedef enum {
-        AFR_DATA_SELF_HEAL_LK,
-        AFR_METADATA_SELF_HEAL_LK,
-        AFR_ENTRY_SELF_HEAL_LK,
-}selfheal_lk_type_t;
+    int32_t **pending;
 
-typedef enum {
-        AFR_INODELK_TRANSACTION,
-        AFR_INODELK_NB_TRANSACTION,
-        AFR_ENTRYLK_TRANSACTION,
-        AFR_ENTRYLK_NB_TRANSACTION,
-        AFR_INODELK_SELFHEAL,
-        AFR_INODELK_NB_SELFHEAL,
-        AFR_ENTRYLK_SELFHEAL,
-        AFR_ENTRYLK_NB_SELFHEAL,
-} afr_lock_call_type_t;
+    loc_t loc;
+    loc_t newloc;
 
-/*
-  xattr format: trusted.afr.volume = [x y z]
-  x - data pending
-  y - metadata pending
-  z - entry pending
-*/
+    fd_t *fd;
+    afr_fd_ctx_t *fd_ctx;
 
-static inline int
-afr_index_for_transaction_type (afr_transaction_type type)
-{
-        switch (type) {
+    /* @child_up: copy of priv->child_up taken at the time of transaction
+       start. The copy is taken so that we have a stable child_up array
+       through the phases of the transaction as priv->child_up[i] can keep
+       changing through time.
+    */
+    unsigned char *child_up;
 
-        case AFR_DATA_TRANSACTION:
-                return 0;
+    /* @read_attempted:
+       array of flags representing subvolumes where read operations of
+       the read transaction have already been attempted. The array is
+       first pre-filled with down subvolumes, and as reads are performed
+       on other subvolumes, those are set as well. This way if the read
+       operation fails we do not retry on that subvolume again.
+    */
+    unsigned char *read_attempted;
 
-        case AFR_METADATA_TRANSACTION:
-                return 1;
+    /* @readfn:
 
-        case AFR_ENTRY_TRANSACTION:
-        case AFR_ENTRY_RENAME_TRANSACTION:
-                return 2;
-        }
+       pointer to function which will perform the read operation on a given
+       subvolume. Used in read transactions.
+    */
 
-        return -1;  /* make gcc happy */
-}
+    afr_read_txn_wind_t readfn;
 
-typedef struct {
-        loc_t                   loc;
-        char                    *basename;
-        unsigned char           *locked_nodes;
-        int                     locked_count;
+    /* @inode:
 
-} afr_entry_lockee_t;
+       the inode on which the read txn is performed on. ref'ed and copied
+       from either fd->inode or loc.inode
+    */
 
-int
-afr_entry_lockee_cmp (const void *l1, const void *l2);
+    inode_t *inode;
 
-typedef struct {
-        char    *domain; /* Domain on which inodelk is taken */
-        struct gf_flock flock;
-        unsigned char *locked_nodes;
-        int32_t lock_count;
-} afr_inodelk_t;
+    /* @parent[2]:
 
-typedef struct {
-        loc_t *lk_loc;
+       parent inode[s] on which directory transactions are performed.
+    */
 
-        int                     lockee_count;
-        afr_entry_lockee_t      lockee[AFR_LOCKEE_COUNT_MAX];
+    inode_t *parent;
+    inode_t *parent2;
 
-        afr_inodelk_t       inodelk[AFR_DOM_COUNT_MAX];
-        const char *lk_basename;
-        const char *lower_basename;
-        const char *higher_basename;
-        char lower_locked;
-        char higher_locked;
+    /* @readable:
 
-        unsigned char *locked_nodes;
-        unsigned char *lower_locked_nodes;
+       array of flags representing servers from which a read can be
+       performed. This is the output of afr_inode_refresh()
+    */
+    unsigned char *readable;
+    unsigned char *readable2; /*For rename transaction*/
 
-        selfheal_lk_type_t selfheal_lk_type;
-        transaction_lk_type_t transaction_lk_type;
+    afr_inode_refresh_cbk_t refreshfn;
 
-        int32_t lock_count;
-        int32_t entrylk_lock_count;
+    /* @refreshinode:
 
-        uint64_t lock_number;
-        int32_t lk_call_count;
-        int32_t lk_expected_count;
-        int32_t lk_attempted_count;
+       Inode currently getting refreshed.
+    */
+    inode_t *refreshinode;
 
-        int32_t lock_op_ret;
-        int32_t lock_op_errno;
-        afr_lock_cbk_t lock_cbk;
-        char *domain; /* Domain on which inode/entry lock/unlock in progress.*/
-} afr_internal_lock_t;
+    dict_t *xattr_req;
 
-typedef struct _afr_locked_fd {
-        fd_t  *fd;
-        struct list_head list;
-} afr_locked_fd_t;
+    dict_t *dict;
 
-struct afr_reply {
-	int	valid;
-	int32_t	op_ret;
-	int32_t	op_errno;
-};
+    int read_subvol; /* Current read subvolume */
 
-typedef struct _afr_local {
-        int     uid;
-        int     gid;
-        unsigned int call_count;
-        unsigned int success_count;
-        unsigned int enoent_count;
-        uint32_t     open_fd_count;
-        gf_boolean_t update_open_fd_count;
+    int optimistic_change_log;
 
+    afr_internal_lock_t internal_lock;
 
-        unsigned int unhealable;
+    /*To handle setattr/setxattr on yet to be linked inode from dht*/
+    uuid_t refreshgfid;
 
-        unsigned int read_child_index;
-        unsigned char read_child_returned;
-        unsigned int first_up_child;
+    /* @refreshed:
 
-	gf_lkowner_t  saved_lk_owner;
+       the inode was "refreshed" (i.e, pending xattrs from all subvols
+       freshly inspected and inode ctx updated accordingly) as part of
+       this transaction already.
+    */
+    gf_boolean_t refreshed;
 
-        int32_t op_ret;
-        int32_t op_errno;
+    gf_boolean_t update_num_inodelks;
+    gf_boolean_t update_open_fd_count;
 
-        int32_t **pending;
+    /*
+      @pre_op_compat:
 
-        loc_t loc;
-        loc_t newloc;
+      compatibility mode of pre-op. send a separate pre-op and
+      op operations as part of transaction, rather than combining
+    */
 
-        fd_t *fd;
+    gf_boolean_t pre_op_compat;
 
-        glusterfs_fop_t fop;
+    /* Is the current writev() going to perform a stable write?
+       i.e, is fd->flags or @flags writev param have O_SYNC or
+       O_DSYNC?
+    */
+    gf_boolean_t stable_write;
 
-        unsigned char *child_up;
-        int32_t       *fresh_children; //in the order of response
+    /* This write appended to the file. Nnot necessarily O_APPEND,
+       just means the offset of write was at the end of file.
+    */
+    gf_boolean_t append_write;
 
-        int32_t *child_errno;
+    /*
+      This struct contains the arguments for the "continuation"
+      (scheme-like) of fops
+    */
 
-        dict_t  *xattr_req;
+    struct {
+        struct {
+            struct statvfs buf;
+            unsigned char buf_set;
+        } statfs;
 
-        int32_t  inodelk_count;
-        int32_t  entrylk_count;
+        struct {
+            fd_t *fd;
+            int32_t flags;
+        } open;
 
-        afr_internal_lock_t internal_lock;
+        struct {
+            struct gf_flock user_flock;
+            struct gf_flock ret_flock;
+            unsigned char *locked_nodes;
+            int32_t cmd;
+            /*For lock healing only.*/
+            unsigned char *dom_locked_nodes;
+            int32_t *dom_lock_op_ret;
+            int32_t *dom_lock_op_errno;
+            struct gf_flock *getlk_rsp;
+        } lk;
+
+        /* inode read */
 
-        afr_locked_fd_t *locked_fd;
-        int32_t          source_child;
-        int32_t          lock_recovery_child;
+        struct {
+            int32_t mask;
+            int last_index; /* index of the child we tried previously */
+        } access;
 
-        dict_t  *dict;
-        int      optimistic_change_log;
-	gf_boolean_t      delayed_post_op;
+        struct {
+            int last_index;
+        } stat;
 
-	/* Is the current writev() going to perform a stable write?
-	   i.e, is fd->flags or @flags writev param have O_SYNC or
-	   O_DSYNC?
-	*/
-	gf_boolean_t      stable_write;
+        struct {
+            int last_index;
+        } fstat;
 
-	/* This write appended to the file. Nnot necessarily O_APPEND,
-	   just means the offset of write was at the end of file.
-	*/
-	gf_boolean_t      append_write;
+        struct {
+            size_t size;
+            int last_index;
+        } readlink;
 
-        /*
-          This struct contains the arguments for the "continuation"
-          (scheme-like) of fops
-        */
+        struct {
+            char *name;
+            long xattr_len;
+            int last_index;
+        } getxattr;
+
+        struct {
+            size_t size;
+            off_t offset;
+            int last_index;
+            uint32_t flags;
+        } readv;
+
+        /* dir read */
+
+        struct {
+            uint32_t *checksum;
+            int success_count;
+            int32_t op_ret;
+            int32_t op_errno;
+        } opendir;
+
+        struct {
+            int32_t op_ret;
+            int32_t op_errno;
+            size_t size;
+            off_t offset;
+            dict_t *dict;
+            int last_index;
+            gf_boolean_t failed;
+        } readdir;
+        /* inode write */
+
+        struct {
+            struct iatt prebuf;
+            struct iatt postbuf;
+        } inode_wfop;  // common structure for all inode-write-fops
+
+        struct {
+            struct iovec *vector;
+            struct iobref *iobref;
+            off_t offset;
+            int32_t op_ret;
+            int32_t count;
+            uint32_t flags;
+        } writev;
+
+        struct {
+            off_t offset;
+        } truncate;
+
+        struct {
+            off_t offset;
+        } ftruncate;
+
+        struct {
+            struct iatt in_buf;
+            int32_t valid;
+        } setattr;
 
-        int   op;
         struct {
-                struct {
-                        unsigned char buf_set;
-                        struct statvfs buf;
-                } statfs;
-
-                struct {
-                        uint32_t parent_entrylk;
-                        uuid_t  gfid_req;
-                        inode_t *inode;
-                        struct iatt buf;
-                        struct iatt postparent;
-                        dict_t **xattrs;
-                        dict_t *xattr;
-                        struct iatt *postparents;
-                        struct iatt *bufs;
-                        int32_t read_child;
-                        int32_t *sources;
-                        int32_t *success_children;
-                        int32_t **pending_matrix;
-                        gf_boolean_t fresh_lookup;
-                        gf_boolean_t possible_spb;
-                } lookup;
-
-                struct {
-                        int32_t flags;
-                } open;
-
-                struct {
-                        int32_t cmd;
-                        struct gf_flock user_flock;
-                        struct gf_flock ret_flock;
-                        unsigned char *locked_nodes;
-                } lk;
-
-                /* inode read */
-
-                struct {
-                        int32_t mask;
-                        int last_index;  /* index of the child we tried previously */
-                } access;
-
-                struct {
-                        int last_index;
-                } stat;
-
-                struct {
-                        int last_index;
-                } fstat;
-
-                struct {
-                        size_t size;
-                        int last_index;
-                } readlink;
-
-                struct {
-                        char *name;
-                        int last_index;
-                        long xattr_len;
-                } getxattr;
-
-                struct {
-                        size_t size;
-                        off_t offset;
-                        int last_index;
-                        uint32_t flags;
-                } readv;
-
-                /* dir read */
-
-                struct {
-                        int success_count;
-                        int32_t op_ret;
-                        int32_t op_errno;
-
-                        uint32_t *checksum;
-                } opendir;
-
-                struct {
-                        int32_t op_ret;
-                        int32_t op_errno;
-                        size_t size;
-                        off_t offset;
-                        dict_t *dict;
-                        gf_boolean_t failed;
-                        int last_index;
-                } readdir;
-                /* inode write */
-
-                struct {
-                        struct iatt prebuf;
-                        struct iatt postbuf;
-                } inode_wfop; //common structure for all inode-write-fops
-
-                struct {
-                        int32_t op_ret;
-
-                        struct iovec *vector;
-                        struct iobref *iobref;
-                        int32_t count;
-                        off_t offset;
-                        uint32_t flags;
-                } writev;
-
-                struct {
-                        off_t offset;
-                } truncate;
-
-                struct {
-                        off_t offset;
-                } ftruncate;
-
-                struct {
-                        struct iatt in_buf;
-                        int32_t valid;
-                } setattr;
-
-                struct {
-                        struct iatt in_buf;
-                        int32_t valid;
-                } fsetattr;
-
-                struct {
-                        dict_t *dict;
-                        int32_t flags;
-                } setxattr;
-
-                struct {
-                        dict_t *dict;
-                        int32_t flags;
-                } fsetxattr;
-
-                struct {
-                        char *name;
-                } removexattr;
-
-                struct {
-                        dict_t *xattr;
-                } xattrop;
-
-                struct {
-                        dict_t *xattr;
-                } fxattrop;
-
-                /* dir write */
-
-                struct {
-                        inode_t *inode;
-                        struct iatt buf;
-                        struct iatt preparent;
-                        struct iatt postparent;
-                        struct iatt prenewparent;
-                        struct iatt postnewparent;
-                } dir_fop; //common structure for all dir fops
-
-                struct {
-                        fd_t *fd;
-                        dict_t *params;
-                        int32_t flags;
-                        mode_t mode;
-                } create;
-
-                struct {
-                        dev_t dev;
-                        mode_t mode;
-                        dict_t *params;
-                } mknod;
-
-                struct {
-                        int32_t mode;
-                        dict_t *params;
-                } mkdir;
-
-                struct {
-                        int flags;
-                } rmdir;
-
-                struct {
-                        dict_t *params;
-                        char *linkpath;
-                } symlink;
-
-		struct {
-			int32_t mode;
-			off_t offset;
-			size_t len;
-		} fallocate;
-
-		struct {
-			off_t offset;
-			size_t len;
-		} discard;
-
-        } cont;
+            struct iatt in_buf;
+            int32_t valid;
+        } fsetattr;
 
         struct {
-                off_t start, len;
+            dict_t *dict;
+            int32_t flags;
+        } setxattr;
 
-                gf_boolean_t    eager_lock_on;
-                int *eager_lock;
+        struct {
+            dict_t *dict;
+            int32_t flags;
+        } fsetxattr;
 
-                char *basename;
-                char *new_basename;
+        struct {
+            char *name;
+        } removexattr;
 
-                loc_t parent_loc;
-                loc_t new_parent_loc;
+        struct {
+            dict_t *xattr;
+            gf_xattrop_flags_t optype;
+        } xattrop;
 
-                afr_transaction_type type;
+        /* dir write */
 
-		/* pre-compute the post piggyback status before
-		   entering POST-OP phase
-		*/
-		int              *postop_piggybacked;
+        struct {
+            inode_t *inode;
+            struct iatt buf;
+            struct iatt preparent;
+            struct iatt postparent;
+            struct iatt prenewparent;
+            struct iatt postnewparent;
+        } dir_fop;  // common structure for all dir fops
 
-		/* stub to resume on destruction
-		   of the transaction frame */
-		call_stub_t      *resume_stub;
+        struct {
+            fd_t *fd;
+            dict_t *params;
+            int32_t flags;
+            mode_t mode;
+        } create;
 
-		struct list_head  eager_locked;
+        struct {
+            dict_t *params;
+            dev_t dev;
+            mode_t mode;
+        } mknod;
 
-                int32_t         **txn_changelog;//changelog after pre+post ops
-                unsigned char   *pre_op;
+        struct {
+            dict_t *params;
+            int32_t mode;
+        } mkdir;
 
-                call_frame_t *main_frame;
+        struct {
+            dict_t *params;
+            char *linkpath;
+        } symlink;
 
-                int (*fop) (call_frame_t *frame, xlator_t *this);
+        struct {
+            off_t offset;
+            size_t len;
+            int32_t mode;
+        } fallocate;
 
-                int (*done) (call_frame_t *frame, xlator_t *this);
+        struct {
+            off_t offset;
+            size_t len;
+        } discard;
 
-                int (*resume) (call_frame_t *frame, xlator_t *this);
+        struct {
+            off_t offset;
+            off_t len;
+            struct iatt prebuf;
+            struct iatt postbuf;
+        } zerofill;
 
-                int (*unwind) (call_frame_t *frame, xlator_t *this);
+        struct {
+            char *volume;
+            int32_t cmd;
+            int32_t in_cmd;
+            struct gf_flock in_flock;
+            struct gf_flock flock;
+            void *xdata;
+        } inodelk;
 
-                /* post-op hook */
-        } transaction;
+        struct {
+            char *volume;
+            char *basename;
+            void *xdata;
+            entrylk_cmd in_cmd;
+            entrylk_cmd cmd;
+            entrylk_type type;
+        } entrylk;
 
-        afr_self_heal_t self_heal;
+        struct {
+            off_t offset;
+            gf_seek_what_t what;
+        } seek;
 
-        struct marker_str     marker;
+        struct {
+            struct gf_lease user_lease;
+            struct gf_lease ret_lease;
+            unsigned char *locked_nodes;
+        } lease;
 
-        /* extra data for fops */
-        dict_t         *xdata_req;
-        dict_t         *xdata_rsp;
+        struct {
+            int flags;
+        } rmdir;
 
-        mode_t          umask;
-        int             xflag;
-        gf_boolean_t    do_discovery;
-	struct afr_reply *replies;
-} afr_local_t;
+        struct {
+            int32_t datasync;
+        } fsync;
 
-typedef enum {
-        AFR_FD_NOT_OPENED,
-        AFR_FD_OPENED,
-        AFR_FD_OPENING
-} afr_fd_open_status_t;
+        struct {
+            uuid_t gfid_req;
+            gf_boolean_t needs_fresh_lookup;
+        } lookup;
 
-typedef struct {
-        unsigned int *pre_op_done;
-        afr_fd_open_status_t *opened_on; /* which subvolumes the fd is open on */
-        unsigned int *pre_op_piggyback;
+    } cont;
 
-        unsigned int *lock_piggyback;
-        unsigned int *lock_acquired;
+    struct {
+        char *basename;
+        char *new_basename;
 
-        int flags;
-        uint64_t up_count;   /* number of CHILD_UPs this fd has seen */
-        uint64_t down_count; /* number of CHILD_DOWNs this fd has seen */
+        loc_t parent_loc;
+        loc_t new_parent_loc;
 
-        int32_t last_tried;
+        /* stub to resume on destruction
+           of the transaction frame */
+        call_stub_t *resume_stub;
 
-        int  hit, miss;
-        gf_boolean_t failed_over;
-        struct list_head entries; /* needed for readdir failover */
+        struct list_head owner_list;
+        struct list_head wait_list;
 
-        unsigned char *locked_on; /* which subvolumes locks have been successful */
+        unsigned char *pre_op;
 
-	/* used for delayed-post-op optimization */
-	pthread_mutex_t    delay_lock;
-	gf_timer_t        *delay_timer;
-	call_frame_t      *delay_frame;
-        int               call_child;
+        /* Changelog xattr dict for [f]xattrop*/
+        dict_t **changelog_xdata;
+        unsigned char *pre_op_sources;
 
-	/* set if any write on this fd was a non stable write
-	   (i.e, without O_SYNC or O_DSYNC)
-	*/
-	gf_boolean_t      witnessed_unstable_write;
+        /* @failed_subvols: subvolumes on which a pre-op or a
+            FOP failed. */
+        unsigned char *failed_subvols;
 
-	/* list of frames currently in progress */
-	struct list_head  eager_locked;
-} afr_fd_ctx_t;
+        call_frame_t *main_frame; /*Fop frame*/
+        call_frame_t *frame;      /*Transaction frame*/
 
+        int (*wind)(call_frame_t *frame, xlator_t *this, int subvol);
 
-/* try alloc and if it fails, goto label */
-#define AFR_LOCAL_ALLOC_OR_GOTO(var, label) do {                    \
-                var = mem_get0 (THIS->local_pool);                  \
-                if (!var) {                                         \
-                        gf_log (this->name, GF_LOG_ERROR,           \
-                                "out of memory :(");                \
-                        op_errno = ENOMEM;                          \
-                        goto label;                                 \
-                }                                                   \
-        } while (0);
+        int (*unwind)(call_frame_t *frame, xlator_t *this);
 
+        off_t start, len;
 
-/* did a call fail due to a child failing? */
-#define child_went_down(op_ret, op_errno) (((op_ret) < 0) &&            \
-                                           ((op_errno == ENOTCONN) ||   \
-                                            (op_errno == EBADFD)))
+        afr_transaction_type type;
 
-#define afr_fop_failed(op_ret, op_errno) ((op_ret) == -1)
+        int32_t in_flight_sb_errno; /* This is where the cause of the
+                                       failure on the last good copy of
+                                       the file is stored.
+                                       */
 
-/* have we tried all children? */
-#define all_tried(i, count)  ((i) == (count) - 1)
+        /* @changelog_resume: function to be called after changlogging
+           (either pre-op or post-op) is done
+        */
+        afr_changelog_resume_t changelog_resume;
 
-int32_t
-afr_set_dict_gfid (dict_t *dict, uuid_t gfid);
+        gf_boolean_t eager_lock_on;
+        gf_boolean_t do_eager_unlock;
 
-int
-pump_command_reply (call_frame_t *frame, xlator_t *this);
+        /* @dirtied: flag which indicates whether we set dirty flag
+           in the OP. Typically true when we are performing operation
+           on more than one subvol and optimistic changelog is disabled
 
-int32_t
-afr_notify (xlator_t *this, int32_t event, void *data, void *data2);
+           A 'true' value set in @dirtied flag means an 'undirtying'
+           has to be done in POST-OP phase.
+        */
+        gf_boolean_t dirtied;
 
-int
-afr_init_entry_lockee (afr_entry_lockee_t *lockee, afr_local_t *local,
-                       loc_t *loc, char *basename, int child_count);
+        /* @inherited: flag which indicates that the dirty flags
+           of the previous transaction were inherited
+        */
+        gf_boolean_t inherited;
 
-void
-afr_entry_lockee_cleanup (afr_internal_lock_t *int_lock);
+        /*
+          @no_uninherit: flag which indicates that a pre_op_uninherit()
+          must _not_ be attempted (and returned as failure) always. This
+          flag is set when a hard pre-op is performed, but not accounted
+          for it in fd_ctx->on_disk[]. Such transactions are "isolated"
+          from the pre-op piggybacking entirely and therefore uninherit
+          must not be attempted.
+        */
+        gf_boolean_t no_uninherit;
+
+        gf_boolean_t in_flight_sb; /* Indicator for occurrence of
+                                      split-brain while in the middle of
+                                      a txn. */
 
+        /* @uninherit_done:
+           @uninherit_value:
+
+           The above pair variables make pre_op_uninherit() idempotent.
+           Both are FALSE initially. The first call to pre_op_uninherit
+           sets @uninherit_done to TRUE and the return value to
+           @uninherit_value. Further calls will check for @uninherit_done
+           to be TRUE and if so will simply return @uninherit_value.
+        */
+        gf_boolean_t uninherit_done;
+        gf_boolean_t uninherit_value;
+
+        gf_boolean_t disable_delayed_post_op;
+    } transaction;
+
+    syncbarrier_t barrier;
+
+    /* extra data for fops */
+    dict_t *xdata_req;
+    dict_t *xdata_rsp;
+
+    dict_t *xattr_rsp; /*for [f]xattrop*/
+
+    mode_t umask;
+    int xflag;
+    struct afr_reply *replies;
+
+    /* For  client side background heals. */
+    struct list_head healer;
+    call_frame_t *heal_frame;
+
+    afr_inode_ctx_t *inode_ctx;
+
+    /*For thin-arbiter transactions.*/
+    int ta_failed_subvol;
+    int ta_event_gen;
+    struct list_head ta_waitq;
+    struct list_head ta_onwireq;
+    afr_ta_fop_state_t fop_state;
+    afr_fop_lock_state_t fop_lock_state;
+    gf_lkowner_t saved_lk_owner;
+    unsigned char read_txn_query_child;
+    unsigned char ta_child_up;
+    gf_boolean_t do_discovery;
+    gf_boolean_t need_full_crawl;
+    gf_boolean_t is_read_txn;
+    gf_boolean_t is_new_entry;
+} afr_local_t;
+
+typedef struct afr_spbc_timeout {
+    call_frame_t *frame;
+    loc_t *loc;
+    int spb_child_index;
+    gf_boolean_t d_spb;
+    gf_boolean_t m_spb;
+} afr_spbc_timeout_t;
+
+typedef struct afr_spb_status {
+    call_frame_t *frame;
+    loc_t *loc;
+} afr_spb_status_t;
+
+typedef struct afr_empty_brick_args {
+    call_frame_t *frame;
+    char *op_type;
+    loc_t loc;
+    int empty_index;
+} afr_empty_brick_args_t;
+
+typedef struct afr_read_subvol_args {
+    ia_type_t ia_type;
+    uuid_t gfid;
+} afr_read_subvol_args_t;
+
+typedef struct afr_granular_esh_args {
+    fd_t *heal_fd;
+    xlator_t *xl;
+    call_frame_t *frame;
+    gf_boolean_t mismatch; /* flag to represent occurrence of type/gfid
+                              mismatch */
+} afr_granular_esh_args_t;
+
+int
+afr_inode_get_readable(call_frame_t *frame, inode_t *inode, xlator_t *this,
+                       unsigned char *readable, int *event_p, int type);
 int
-afr_attempt_lock_recovery (xlator_t *this, int32_t child_index);
+afr_inode_read_subvol_get(inode_t *inode, xlator_t *this,
+                          unsigned char *data_subvols,
+                          unsigned char *metadata_subvols,
+                          int *event_generation);
+int
+__afr_inode_read_subvol_get(inode_t *inode, xlator_t *this,
+                            unsigned char *data_subvols,
+                            unsigned char *metadata_subvols,
+                            int *event_generation);
 
 int
-afr_save_locked_fd (xlator_t *this, fd_t *fd);
+__afr_inode_read_subvol_set(inode_t *inode, xlator_t *this,
+                            unsigned char *data_subvols,
+                            unsigned char *metadata_subvol,
+                            int event_generation);
+int
+afr_inode_read_subvol_set(inode_t *inode, xlator_t *this,
+                          unsigned char *data_subvols,
+                          unsigned char *metadata_subvols,
+                          int event_generation);
 
 int
-afr_mark_locked_nodes (xlator_t *this, fd_t *fd,
-                       unsigned char *locked_nodes);
+__afr_inode_need_refresh_set(inode_t *inode, xlator_t *this);
 
-void
-afr_set_lk_owner (call_frame_t *frame, xlator_t *this, void *lk_owner);
+int
+afr_inode_need_refresh_set(inode_t *inode, xlator_t *this);
 
 int
-afr_set_lock_number (call_frame_t *frame, xlator_t *this);
+afr_read_subvol_select_by_policy(inode_t *inode, xlator_t *this,
+                                 unsigned char *readable,
+                                 afr_read_subvol_args_t *args);
 
+int
+afr_inode_read_subvol_type_get(inode_t *inode, xlator_t *this,
+                               unsigned char *readable, int *event_p, int type);
+int
+afr_read_subvol_get(inode_t *inode, xlator_t *this, int *subvol_p,
+                    unsigned char *readables, int *event_p,
+                    afr_transaction_type type, afr_read_subvol_args_t *args);
 
-loc_t *
-lower_path (loc_t *l1, const char *b1, loc_t *l2, const char *b2);
+#define afr_data_subvol_get(i, t, s, r, e, a)                                  \
+    afr_read_subvol_get(i, t, s, r, e, AFR_DATA_TRANSACTION, a)
 
-int32_t
-afr_unlock (call_frame_t *frame, xlator_t *this);
+#define afr_metadata_subvol_get(i, t, s, r, e, a)                              \
+    afr_read_subvol_get(i, t, s, r, e, AFR_METADATA_TRANSACTION, a)
 
 int
-afr_nonblocking_entrylk (call_frame_t *frame, xlator_t *this);
+afr_inode_refresh(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                  uuid_t gfid, afr_inode_refresh_cbk_t cbk);
 
-int
-afr_nonblocking_inodelk (call_frame_t *frame, xlator_t *this);
+int32_t
+afr_notify(xlator_t *this, int32_t event, void *data, void *data2);
 
 int
-afr_blocking_lock (call_frame_t *frame, xlator_t *this);
+xattr_is_equal(dict_t *this, char *key1, data_t *value1, void *data);
 
 int
-afr_internal_lock_finish (call_frame_t *frame, xlator_t *this);
+afr_add_entry_lockee(afr_local_t *local, loc_t *loc, char *basename,
+                     int child_count);
 
 int
-afr_lk_transfer_datalock (call_frame_t *dst, call_frame_t *src, char *dom,
-                          unsigned int child_count);
+afr_add_inode_lockee(afr_local_t *local, int child_count);
 
-int pump_start (call_frame_t *frame, xlator_t *this);
+void
+afr_lockees_cleanup(afr_internal_lock_t *int_lock);
 
 int
-__afr_fd_ctx_set (xlator_t *this, fd_t *fd);
+afr_attempt_lock_recovery(xlator_t *this, int32_t child_index);
 
 int
-afr_fd_ctx_set (xlator_t *this, fd_t *fd);
-
-int32_t
-afr_inode_get_read_ctx (xlator_t *this, inode_t *inode, int32_t *fresh_children);
+afr_mark_locked_nodes(xlator_t *this, fd_t *fd, unsigned char *locked_nodes);
 
 void
-afr_inode_set_read_ctx (xlator_t *this, inode_t *inode, int32_t read_child,
-                        int32_t *fresh_children);
+afr_set_lk_owner(call_frame_t *frame, xlator_t *this, void *lk_owner);
 
 int
-afr_build_parent_loc (loc_t *parent, loc_t *child, int32_t *op_errno);
+afr_set_lock_number(call_frame_t *frame, xlator_t *this);
 
-unsigned int
-afr_up_children_count (unsigned char *child_up, unsigned int child_count);
+int32_t
+afr_unlock(call_frame_t *frame, xlator_t *this);
 
-unsigned int
-afr_locked_children_count (unsigned char *children, unsigned int child_count);
+int
+afr_lock_nonblocking(call_frame_t *frame, xlator_t *this);
 
-unsigned int
-afr_pre_op_done_children_count (unsigned char *pre_op,
-                                unsigned int child_count);
+int
+afr_blocking_lock(call_frame_t *frame, xlator_t *this);
 
-gf_boolean_t
-afr_is_fresh_lookup (loc_t *loc, xlator_t *this);
+int
+afr_internal_lock_finish(call_frame_t *frame, xlator_t *this);
 
-void
-afr_update_loc_gfids (loc_t *loc, struct iatt *buf, struct iatt *postparent);
+int
+__afr_fd_ctx_set(xlator_t *this, fd_t *fd);
+
+afr_fd_ctx_t *
+afr_fd_ctx_get(fd_t *fd, xlator_t *this);
 
 int
-afr_locked_nodes_count (unsigned char *locked_nodes, int child_count);
+afr_build_parent_loc(loc_t *parent, loc_t *child, int32_t *op_errno);
 
-void
-afr_local_cleanup (afr_local_t *local, xlator_t *this);
+int
+afr_locked_nodes_count(unsigned char *locked_nodes, int child_count);
 
 int
-afr_frame_return (call_frame_t *frame);
+afr_replies_interpret(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                      gf_boolean_t *start_heal);
 
-gf_boolean_t
-afr_is_split_brain (xlator_t *this, inode_t *inode);
+void
+afr_local_replies_wipe(afr_local_t *local, afr_private_t *priv);
 
 void
-afr_set_split_brain (xlator_t *this, inode_t *inode, afr_spb_state_t mdata_spb,
-                     afr_spb_state_t data_spb);
+afr_local_cleanup(afr_local_t *local, xlator_t *this);
 
 int
-afr_open (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
-          fd_t *fd, dict_t *xdata);
+afr_frame_return(call_frame_t *frame);
 
-void
-afr_set_opendir_done (xlator_t *this, inode_t *inode);
-
-gf_boolean_t
-afr_is_opendir_done (xlator_t *this, inode_t *inode);
+int
+afr_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+         fd_t *fd, dict_t *xdata);
 
 void
-afr_local_transaction_cleanup (afr_local_t *local, xlator_t *this);
+afr_local_transaction_cleanup(afr_local_t *local, xlator_t *this);
 
 int
-afr_cleanup_fd_ctx (xlator_t *this, fd_t *fd);
+afr_cleanup_fd_ctx(xlator_t *this, fd_t *fd);
+
+#define AFR_STACK_UNWIND(fop, frame, op_ret, op_errno, params...)              \
+    do {                                                                       \
+        afr_local_t *__local = NULL;                                           \
+        xlator_t *__this = NULL;                                               \
+        int32_t __op_ret = 0;                                                  \
+        int32_t __op_errno = 0;                                                \
+                                                                               \
+        __op_ret = op_ret;                                                     \
+        __op_errno = op_errno;                                                 \
+        if (frame) {                                                           \
+            __local = frame->local;                                            \
+            __this = frame->this;                                              \
+            afr_handle_inconsistent_fop(frame, &__op_ret, &__op_errno);        \
+            if (__local && __local->is_read_txn)                               \
+                afr_pending_read_decrement(__this->private,                    \
+                                           __local->read_subvol);              \
+            if (__local && __local->xdata_req &&                               \
+                afr_is_lock_mode_mandatory(__local->xdata_req))                \
+                afr_dom_lock_release(frame);                                   \
+            frame->local = NULL;                                               \
+        }                                                                      \
+                                                                               \
+        STACK_UNWIND_STRICT(fop, frame, __op_ret, __op_errno, params);         \
+        if (__local) {                                                         \
+            afr_local_cleanup(__local, __this);                                \
+            mem_put(__local);                                                  \
+        }                                                                      \
+    } while (0)
+
+#define AFR_STACK_DESTROY(frame)                                               \
+    do {                                                                       \
+        afr_local_t *__local = NULL;                                           \
+        xlator_t *__this = NULL;                                               \
+        __local = frame->local;                                                \
+        __this = frame->this;                                                  \
+        frame->local = NULL;                                                   \
+        STACK_DESTROY(frame->root);                                            \
+        if (__local) {                                                         \
+            afr_local_cleanup(__local, __this);                                \
+            mem_put(__local);                                                  \
+        }                                                                      \
+    } while (0);
+
+#define AFR_FRAME_INIT(frame, op_errno)                                        \
+    ({                                                                         \
+        frame->local = mem_get0(THIS->local_pool);                             \
+        if (afr_local_init(frame->local, frame->this->private, &op_errno)) {   \
+            afr_local_cleanup(frame->local, frame->this);                      \
+            mem_put(frame->local);                                             \
+            frame->local = NULL;                                               \
+        };                                                                     \
+        frame->local;                                                          \
+    })
+
+#define AFR_STACK_RESET(frame)                                                 \
+    do {                                                                       \
+        afr_local_t *__local = NULL;                                           \
+        xlator_t *__this = NULL;                                               \
+        __local = frame->local;                                                \
+        __this = frame->this;                                                  \
+        frame->local = NULL;                                                   \
+        int __opr;                                                             \
+        STACK_RESET(frame->root);                                              \
+        if (__local) {                                                         \
+            afr_local_cleanup(__local, __this);                                \
+            mem_put(__local);                                                  \
+        }                                                                      \
+        AFR_FRAME_INIT(frame, __opr);                                          \
+    } while (0)
 
-int
-afr_launch_openfd_self_heal (call_frame_t *frame, xlator_t *this, fd_t *fd);
-
-#define AFR_STACK_UNWIND(fop, frame, params ...)                \
-        do {                                                    \
-                afr_local_t *__local = NULL;                    \
-                xlator_t    *__this = NULL;                     \
-                if (frame) {                                    \
-                        __local = frame->local;                 \
-                        __this = frame->this;                   \
-                        frame->local = NULL;                    \
-                }                                               \
-                STACK_UNWIND_STRICT (fop, frame, params);       \
-                if (__local) {                                  \
-                        afr_local_cleanup (__local, __this);    \
-                        mem_put (__local);                      \
-                }                                               \
-        } while (0)
-
-#define AFR_STACK_DESTROY(frame)                                \
-        do {                                                    \
-                afr_local_t *__local = NULL;                    \
-                xlator_t    *__this = NULL;                     \
-                __local = frame->local;                         \
-                __this = frame->this;                           \
-                frame->local = NULL;                            \
-                STACK_DESTROY (frame->root);                    \
-                if (__local) {                                  \
-                        afr_local_cleanup (__local, __this);    \
-                        mem_put (__local);                      \
-                }                                               \
-        } while (0);
-
-#define AFR_NUM_CHANGE_LOGS            3 /*data + metadata + entry*/
 /* allocate and return a string that is the basename of argument */
 static inline char *
-AFR_BASENAME (const char *str)
+AFR_BASENAME(const char *str)
 {
-        char *__tmp_str = NULL;
-        char *__basename_str = NULL;
-        __tmp_str = gf_strdup (str);
-        __basename_str = gf_strdup (basename (__tmp_str));
-        GF_FREE (__tmp_str);
-        return __basename_str;
+    char *__tmp_str = NULL;
+    char *__basename_str = NULL;
+    __tmp_str = gf_strdup(str);
+    __basename_str = gf_strdup(basename(__tmp_str));
+    GF_FREE(__tmp_str);
+    return __basename_str;
 }
 
+call_frame_t *
+afr_copy_frame(call_frame_t *base);
+
 int
-afr_transaction_local_init (afr_local_t *local, xlator_t *this);
+afr_transaction_local_init(afr_local_t *local, xlator_t *this);
 
 int32_t
-afr_marker_getxattr (call_frame_t *frame, xlator_t *this,
-                     loc_t *loc, const char *name,afr_local_t *local, afr_private_t *priv );
+afr_marker_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                    const char *name, afr_local_t *local, afr_private_t *priv);
 
-int32_t *
-afr_children_create (int32_t child_count);
+int
+afr_local_init(afr_local_t *local, afr_private_t *priv, int32_t *op_errno);
 
 int
-afr_local_init (afr_local_t *local, afr_private_t *priv, int32_t *op_errno);
+afr_internal_lock_init(afr_internal_lock_t *lk, size_t child_count);
 
 int
-afr_internal_lock_init (afr_internal_lock_t *lk, size_t child_count,
-                        transaction_lk_type_t lk_type);
+afr_higher_errno(int32_t old_errno, int32_t new_errno);
 
 int
-afr_first_up_child (unsigned char *child_up, size_t child_count);
+afr_final_errno(afr_local_t *local, afr_private_t *priv);
 
 int
-afr_select_read_child_from_policy (int32_t *fresh_children, int32_t child_count,
-                                   int32_t prev_read_child,
-                                   int32_t config_read_child, int32_t *sources,
-                                   unsigned int hmode, uuid_t gfid);
+afr_xattr_req_prepare(xlator_t *this, dict_t *xattr_req);
 
 void
-afr_set_read_ctx_from_policy (xlator_t *this, inode_t *inode,
-                              int32_t *fresh_children, int32_t prev_read_child,
-                              int32_t config_read_child, uuid_t gfid);
+afr_fix_open(fd_t *fd, xlator_t *this);
 
-int32_t
-afr_get_call_child (xlator_t *this, unsigned char *child_up, int32_t read_child,
-                    int32_t *fresh_children,
-                    int32_t *call_child, int32_t *last_index);
+afr_fd_ctx_t *
+afr_fd_ctx_get(fd_t *fd, xlator_t *this);
 
-int32_t
-afr_next_call_child (int32_t *fresh_children, unsigned char *child_up,
-                     size_t child_count, int32_t *last_index,
-                     int32_t read_child);
-void
-afr_get_fresh_children (int32_t *success_children, int32_t *sources,
-                        int32_t *children, unsigned int child_count);
 void
-afr_children_add_child (int32_t *children, int32_t child,
-                              int32_t child_count);
+afr_set_low_priority(call_frame_t *frame);
+int
+afr_child_fd_ctx_set(xlator_t *this, fd_t *fd, int32_t child, int flags);
+
 void
-afr_children_rm_child (int32_t *children, int32_t child,
-                             int32_t child_count);
+afr_matrix_cleanup(int32_t **pending, unsigned int m);
+
+int32_t **
+afr_matrix_create(unsigned int m, unsigned int n);
+
+int **
+afr_mark_pending_changelog(afr_private_t *priv, unsigned char *pending,
+                           dict_t *xattr, ia_type_t iat);
+
 void
-afr_reset_children (int32_t *children, int32_t child_count);
-int32_t
-afr_most_important_error(int32_t old_errno, int32_t new_errno,
-			 gf_boolean_t eio);
-int
-afr_errno_count (int32_t *children, int *child_errno,
-                 unsigned int child_count, int32_t op_errno);
+afr_filter_xattrs(dict_t *xattr);
+
+/*
+ * Special value indicating we should use the "auto" quorum method instead of
+ * a fixed value (including zero to turn off quorum enforcement).
+ */
+#define AFR_QUORUM_AUTO INT_MAX
+
 int
-afr_get_children_count (int32_t *children, unsigned int child_count);
+afr_fd_report_unstable_write(xlator_t *this, afr_local_t *local);
+
 gf_boolean_t
-afr_is_child_present (int32_t *success_children, int32_t child_count,
-                      int32_t child);
+afr_fd_has_witnessed_unstable_write(xlator_t *this, inode_t *inode);
+
 void
-afr_update_gfid_from_iatts (uuid_t uuid, struct iatt *bufs,
-                            int32_t *success_children,
-                            unsigned int child_count);
+afr_reply_wipe(struct afr_reply *reply);
+
 void
-afr_reset_xattr (dict_t **xattr, unsigned int child_count);
+afr_replies_wipe(struct afr_reply *replies, int count);
+
 gf_boolean_t
-afr_conflicting_iattrs (struct iatt *bufs, int32_t *success_children,
-                        unsigned int child_count, const char *path,
-                        const char *xlator_name);
-unsigned int
-afr_gfid_missing_count (const char *xlator_name, int32_t *children,
-                        struct iatt *bufs, unsigned int child_count,
-                        const char *path);
-void
-afr_xattr_req_prepare (xlator_t *this, dict_t *xattr_req, const char *path);
+afr_xattrs_are_equal(dict_t *dict1, dict_t *dict2);
+
+gf_boolean_t
+afr_is_xattr_ignorable(char *key);
+
+int
+afr_get_heal_info(call_frame_t *frame, xlator_t *this, loc_t *loc);
+
+int
+afr_heal_splitbrain_file(call_frame_t *frame, xlator_t *this, loc_t *loc);
+
+int
+afr_get_split_brain_status(void *opaque);
+
+int
+afr_get_split_brain_status_cbk(int ret, call_frame_t *frame, void *opaque);
+
+int
+afr_inode_split_brain_choice_set(inode_t *inode, xlator_t *this,
+                                 int spb_choice);
+int
+afr_split_brain_read_subvol_get(inode_t *inode, xlator_t *this,
+                                call_frame_t *frame, int *spb_subvol);
+int
+afr_get_child_index_from_name(xlator_t *this, char *name);
+
+int
+afr_is_split_brain(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                   uuid_t gfid, gf_boolean_t *d_spb, gf_boolean_t *m_spb);
+int
+afr_spb_choice_timeout_cancel(xlator_t *this, inode_t *inode);
+
+int
+afr_set_split_brain_choice(int ret, call_frame_t *frame, void *opaque);
+
+gf_boolean_t
+afr_get_need_heal(xlator_t *this);
+
 void
-afr_children_copy (int32_t *dst, int32_t *src, unsigned int child_count);
-afr_transaction_type
-afr_transaction_type_get (ia_type_t ia_type);
+afr_set_need_heal(xlator_t *this, afr_local_t *local);
+
+int
+afr_selfheal_data_open(xlator_t *this, inode_t *inode, fd_t **fd);
+
+int
+afr_get_msg_id(char *op_type);
+
+int
+afr_set_in_flight_sb_status(xlator_t *this, call_frame_t *frame,
+                            inode_t *inode);
+
 int32_t
-afr_resultant_errno_get (int32_t *children,
-                         int *child_errno, unsigned int child_count);
+afr_quorum_errno(afr_private_t *priv);
+
+gf_boolean_t
+afr_is_consistent_io_possible(afr_local_t *local, afr_private_t *priv,
+                              int32_t *op_errno);
+void
+afr_handle_inconsistent_fop(call_frame_t *frame, int32_t *op_ret,
+                            int32_t *op_errno);
+
 void
-afr_inode_rm_stale_children (xlator_t *this, inode_t *inode,
-                             int32_t *stale_children);
+afr_inode_write_fill(call_frame_t *frame, xlator_t *this, int child_index,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata);
 void
-afr_launch_self_heal (call_frame_t *frame, xlator_t *this, inode_t *inode,
-                      gf_boolean_t background, ia_type_t ia_type, char *reason,
-                      void (*gfid_sh_success_cbk) (call_frame_t *sh_frame,
-                                                   xlator_t *this),
-                      int (*unwind) (call_frame_t *frame, xlator_t *this,
-                                     int32_t op_ret, int32_t op_errno,
-                                     int32_t sh_failed));
+afr_process_post_writev(call_frame_t *frame, xlator_t *this);
+
 void
-afr_fix_open (xlator_t *this, fd_t *fd, size_t need_open_count, int *need_open);
+afr_writev_unwind(call_frame_t *frame, xlator_t *this);
 
 void
-afr_open_fd_fix (fd_t *fd, xlator_t *this);
-int
-afr_set_elem_count_get (unsigned char *elems, int child_count);
+afr_writev_copy_outvars(call_frame_t *src_frame, call_frame_t *dst_frame);
 
+void
+afr_update_uninodelk(afr_local_t *local, afr_internal_lock_t *int_lock,
+                     int32_t child_index);
 afr_fd_ctx_t *
-afr_fd_ctx_get (fd_t *fd, xlator_t *this);
+__afr_fd_ctx_get(fd_t *fd, xlator_t *this);
 
 gf_boolean_t
-afr_open_only_data_self_heal (char *data_self_heal);
+afr_is_inode_refresh_reqd(inode_t *inode, xlator_t *this, int event_gen1,
+                          int event_gen2);
 
+int
+afr_serialize_xattrs_with_delimiter(call_frame_t *frame, xlator_t *this,
+                                    char *buf, const char *default_str,
+                                    int32_t *serz_len, char delimiter);
 gf_boolean_t
-afr_data_self_heal_enabled (char *data_self_heal);
+afr_is_symmetric_error(call_frame_t *frame, xlator_t *this);
 
-void
-afr_set_low_priority (call_frame_t *frame);
 int
-afr_child_fd_ctx_set (xlator_t *this, fd_t *fd, int32_t child,
-                      int flags);
+__afr_inode_ctx_get(xlator_t *this, inode_t *inode, afr_inode_ctx_t **ctx);
 
-gf_boolean_t
-afr_have_quorum (char *logname, afr_private_t *priv);
+uint64_t
+afr_write_subvol_get(call_frame_t *frame, xlator_t *this);
 
-void
-afr_matrix_cleanup (int32_t **pending, unsigned int m);
+int
+afr_write_subvol_set(call_frame_t *frame, xlator_t *this);
 
-int32_t**
-afr_matrix_create (unsigned int m, unsigned int n);
+int
+afr_write_subvol_reset(call_frame_t *frame, xlator_t *this);
 
-gf_boolean_t
-afr_is_errno_set (int *child_errno, int child);
+int
+afr_set_inode_local(xlator_t *this, afr_local_t *local, inode_t *inode);
+
+int
+afr_fill_ta_loc(xlator_t *this, loc_t *loc, gf_boolean_t is_gfid_based_fop);
+
+int
+afr_ta_post_op_lock(xlator_t *this, loc_t *loc);
+
+int
+afr_ta_post_op_unlock(xlator_t *this, loc_t *loc);
 
 gf_boolean_t
-afr_is_errno_unset (int *child_errno, int child);
+afr_is_pending_set(xlator_t *this, dict_t *xdata, int type);
+
+int
+__afr_get_up_children_count(afr_private_t *priv);
+
+call_frame_t *
+afr_ta_frame_create(xlator_t *this);
 
 gf_boolean_t
-afr_is_fd_fixable (fd_t *fd);
+afr_ta_has_quorum(afr_private_t *priv, afr_local_t *local);
 
 void
-afr_prepare_new_entry_pending_matrix (int32_t **pending,
-                                      gf_boolean_t (*is_pending) (int *, int),
-                                      int *ctx, struct iatt *buf,
-                                      unsigned int child_count);
+afr_ta_lock_release_synctask(xlator_t *this);
+
 void
-afr_xattr_array_destroy (dict_t **xattr, unsigned int child_count);
-/*
- * Special value indicating we should use the "auto" quorum method instead of
- * a fixed value (including zero to turn off quorum enforcement).
- */
-#define AFR_QUORUM_AUTO INT_MAX
+afr_ta_locked_priv_invalidate(afr_private_t *priv);
 
-/*
- * Having this as a macro will make debugging a bit weirder, but does reduce
- * the probability of functions handling this check inconsistently.
- */
-#define QUORUM_CHECK(_func,_label) do {                                  \
-        if (priv->quorum_count && !afr_have_quorum(this->name,priv)) { \
-                gf_log(this->name,GF_LOG_WARNING,                        \
-                       "failing "#_func" due to lack of quorum");        \
-                op_errno = EROFS;                                        \
-                goto _label;                                             \
-        }                                                                \
-} while (0);
+gf_boolean_t
+afr_lookup_has_quorum(call_frame_t *frame,
+                      const unsigned int up_children_count);
 
+void
+afr_mark_new_entry_changelog(call_frame_t *frame, xlator_t *this);
 
-int
-afr_fd_report_unstable_write (xlator_t *this, fd_t *fd);
+void
+afr_handle_replies_quorum(call_frame_t *frame, xlator_t *this);
 
 gf_boolean_t
-afr_fd_has_witnessed_unstable_write (xlator_t *this, fd_t *fd);
+afr_ta_dict_contains_pending_xattr(dict_t *dict, afr_private_t *priv,
+                                   int child);
 
 void
-afr_delayed_changelog_wake_resume (xlator_t *this, fd_t *fd, call_stub_t *stub);
+afr_selfheal_childup(xlator_t *this, afr_private_t *priv);
 
-int
-afr_inodelk_init (afr_inodelk_t *lk, char *dom, size_t child_count);
+gf_boolean_t
+afr_is_lock_mode_mandatory(dict_t *xdata);
 
 void
-afr_handle_open_fd_count (call_frame_t *frame, xlator_t *this);
+afr_dom_lock_release(call_frame_t *frame);
 
-afr_inode_ctx_t*
-afr_inode_ctx_get (inode_t *inode, xlator_t *this);
+void
+afr_fill_success_replies(afr_local_t *local, afr_private_t *priv,
+                         unsigned char *replies);
 
+gf_boolean_t
+afr_is_private_directory(afr_private_t *priv, uuid_t pargfid, const char *name,
+                         pid_t pid);
 #endif /* __AFR_H__ */
diff --git a/xlators/cluster/afr/src/pump.c b/xlators/cluster/afr/src/pump.c
deleted file mode 100644
index a7f72fb30e9..00000000000
--- a/xlators/cluster/afr/src/pump.c
+++ /dev/null
@@ -1,2641 +0,0 @@
-/*
-  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
-  This file is part of GlusterFS.
-
-  This file is licensed to you under your choice of the GNU Lesser
-  General Public License, version 3 or any later version (LGPLv3 or
-  later), or the GNU General Public License, version 2 (GPLv2), in all
-  cases as published by the Free Software Foundation.
-*/
-
-#include <unistd.h>
-#include <sys/time.h>
-#include <stdlib.h>
-#include <fnmatch.h>
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "afr-common.c"
-#include "defaults.c"
-#include "glusterfs.h"
-
-static uint64_t pump_pid = 0;
-static inline void
-pump_fill_loc_info (loc_t *loc, struct iatt *iatt, struct iatt *parent)
-{
-        afr_update_loc_gfids (loc, iatt, parent);
-        uuid_copy (loc->inode->gfid, iatt->ia_gfid);
-}
-
-static int
-pump_mark_start_pending (xlator_t *this)
-{
-        afr_private_t  *priv      = NULL;
-        pump_private_t *pump_priv = NULL;
-
-        priv      = this->private;
-        pump_priv = priv->pump_private;
-
-        pump_priv->pump_start_pending = 1;
-
-        return 0;
-}
-
-static int
-is_pump_start_pending (xlator_t *this)
-{
-        afr_private_t  *priv      = NULL;
-        pump_private_t *pump_priv = NULL;
-
-        priv      = this->private;
-        pump_priv = priv->pump_private;
-
-        return (pump_priv->pump_start_pending);
-}
-
-static int
-pump_remove_start_pending (xlator_t *this)
-{
-        afr_private_t  *priv      = NULL;
-        pump_private_t *pump_priv = NULL;
-
-        priv      = this->private;
-        pump_priv = priv->pump_private;
-
-        pump_priv->pump_start_pending = 0;
-
-        return 0;
-}
-
-static pump_state_t
-pump_get_state ()
-{
-        xlator_t *this = NULL;
-        afr_private_t *priv = NULL;
-        pump_private_t *pump_priv = NULL;
-
-        pump_state_t ret;
-
-        this = THIS;
-        priv = this->private;
-        pump_priv = priv->pump_private;
-
-        LOCK (&pump_priv->pump_state_lock);
-        {
-                ret = pump_priv->pump_state;
-        }
-        UNLOCK (&pump_priv->pump_state_lock);
-
-        return ret;
-}
-
-int
-pump_change_state (xlator_t *this, pump_state_t state)
-{
-        afr_private_t *priv = NULL;
-        pump_private_t *pump_priv = NULL;
-
-        pump_state_t state_old;
-        pump_state_t state_new;
-
-
-        priv = this->private;
-        pump_priv = priv->pump_private;
-
-        GF_ASSERT (pump_priv);
-
-        LOCK (&pump_priv->pump_state_lock);
-        {
-                state_old = pump_priv->pump_state;
-                state_new = state;
-
-                pump_priv->pump_state = state;
-
-        }
-        UNLOCK (&pump_priv->pump_state_lock);
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "Pump changing state from %d to %d",
-                state_old,
-                state_new);
-
-        return  0;
-}
-
-static int
-pump_set_resume_path (xlator_t *this, const char *path)
-{
-        int ret = 0;
-
-        afr_private_t *priv = NULL;
-        pump_private_t *pump_priv = NULL;
-
-        priv = this->private;
-        pump_priv = priv->pump_private;
-
-        GF_ASSERT (pump_priv);
-
-        LOCK (&pump_priv->resume_path_lock);
-        {
-                strncpy (pump_priv->resume_path, path, strlen (path) + 1);
-        }
-        UNLOCK (&pump_priv->resume_path_lock);
-
-        return ret;
-}
-
-static int
-pump_save_path (xlator_t *this, const char *path)
-{
-        afr_private_t *priv = NULL;
-        pump_state_t state;
-        dict_t *dict = NULL;
-        loc_t  loc = {0};
-        int dict_ret = 0;
-        int ret = -1;
-
-        state = pump_get_state ();
-        if (state == PUMP_STATE_RESUME)
-                return 0;
-
-        priv = this->private;
-
-        GF_ASSERT (priv->root_inode);
-
-        afr_build_root_loc (this, &loc);
-
-        dict = dict_new ();
-        dict_ret = dict_set_str (dict, PUMP_PATH, (char *)path);
-        if (dict_ret)
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to set the key %s", path, PUMP_PATH);
-
-        ret = syncop_setxattr (PUMP_SOURCE_CHILD (this), &loc, dict, 0);
-
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "setxattr failed - could not save path=%s", path);
-        } else {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "setxattr succeeded - saved path=%s", path);
-        }
-
-        dict_unref (dict);
-
-        loc_wipe (&loc);
-        return 0;
-}
-
-static int
-pump_check_and_update_status (xlator_t *this)
-{
-        pump_state_t state;
-        int ret = -1;
-
-        state = pump_get_state ();
-
-        switch (state) {
-
-        case PUMP_STATE_RESUME:
-        case PUMP_STATE_RUNNING:
-        {
-                ret = 0;
-                break;
-        }
-        case PUMP_STATE_PAUSE:
-        {
-                ret = -1;
-                break;
-        }
-        case PUMP_STATE_ABORT:
-        {
-                pump_save_path (this, "/");
-                ret = -1;
-                break;
-        }
-        default:
-        {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Unknown pump state");
-                ret = -1;
-                break;
-        }
-
-        }
-
-        return ret;
-}
-
-static const char *
-pump_get_resume_path (xlator_t *this)
-{
-        afr_private_t *priv = NULL;
-        pump_private_t *pump_priv = NULL;
-
-        const char *resume_path = NULL;
-
-        priv = this->private;
-        pump_priv = priv->pump_private;
-
-        resume_path = pump_priv->resume_path;
-
-        return resume_path;
-}
-
-static int
-pump_update_resume_state (xlator_t *this, const char *path)
-{
-        pump_state_t state;
-        const char *resume_path = NULL;
-
-        state = pump_get_state ();
-
-        if (state == PUMP_STATE_RESUME) {
-                resume_path = pump_get_resume_path (this);
-                if (strcmp (resume_path, "/") == 0) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Reached the resume path (/). Proceeding to change state"
-                                " to running");
-                        pump_change_state (this, PUMP_STATE_RUNNING);
-                } else if (strcmp (resume_path, path) == 0) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Reached the resume path. Proceeding to change state"
-                                " to running");
-                        pump_change_state (this, PUMP_STATE_RUNNING);
-                } else {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Not yet hit the resume path:res-path=%s,path=%s",
-                                resume_path, path);
-                }
-        }
-
-        return 0;
-}
-
-static gf_boolean_t
-is_pump_traversal_allowed (xlator_t *this, const char *path)
-{
-        pump_state_t state;
-        const char *resume_path = NULL;
-        gf_boolean_t ret = _gf_true;
-
-        state = pump_get_state ();
-
-        if (state == PUMP_STATE_RESUME) {
-                resume_path = pump_get_resume_path (this);
-                if (strstr (resume_path, path)) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "On the right path to resumption path");
-                        ret = _gf_true;
-                } else {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Not the right path to resuming=> ignoring traverse");
-                        ret = _gf_false;
-                }
-        }
-
-        return ret;
-}
-
-static int
-pump_save_file_stats (xlator_t *this, const char *path)
-{
-        afr_private_t  *priv        = NULL;
-        pump_private_t *pump_priv   = NULL;
-
-        priv      = this->private;
-        pump_priv = priv->pump_private;
-
-        LOCK (&pump_priv->resume_path_lock);
-        {
-                pump_priv->number_files_pumped++;
-
-                strncpy (pump_priv->current_file, path,
-                         PATH_MAX);
-        }
-        UNLOCK (&pump_priv->resume_path_lock);
-
-        return 0;
-}
-
-static int
-gf_pump_traverse_directory (loc_t *loc)
-{
-        xlator_t        *this              = NULL;
-        fd_t            *fd                = NULL;
-        off_t           offset             = 0;
-        loc_t           entry_loc          = {0};
-        gf_dirent_t     *entry             = NULL;
-        gf_dirent_t     *tmp               = NULL;
-        gf_dirent_t     entries;
-	struct iatt     iatt               = {0};
-        struct iatt     parent             = {0};
-	dict_t          *xattr_rsp         = NULL;
-        int             ret                = 0;
-        gf_boolean_t    is_directory_empty = _gf_true;
-        gf_boolean_t    free_entries       = _gf_false;
-
-        INIT_LIST_HEAD (&entries.list);
-        this = THIS;
-
-        GF_ASSERT (loc->inode);
-
-	fd = fd_create (loc->inode, pump_pid);
-        if (!fd) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Failed to create fd for %s", loc->path);
-                goto out;
-        }
-
-        ret = syncop_opendir (this, loc, fd);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "opendir failed on %s", loc->path);
-                goto out;
-        }
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "pump opendir on %s returned=%d",
-                loc->path, ret);
-
-        while (syncop_readdirp (this, fd, 131072, offset, NULL, &entries)) {
-                free_entries = _gf_true;
-
-                if (list_empty (&entries.list)) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "no more entries in directory");
-                        goto out;
-                }
-
-                list_for_each_entry_safe (entry, tmp, &entries.list, list) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "found readdir entry=%s", entry->d_name);
-
-                        offset = entry->d_off;
-                        if (uuid_is_null (entry->d_stat.ia_gfid)) {
-                                gf_log (this->name, GF_LOG_WARNING, "%s/%s: No "
-                                        "gfid present skipping",
-                                        loc->path, entry->d_name);
-                                continue;
-                        }
-                        loc_wipe (&entry_loc);
-                        ret = afr_build_child_loc (this, &entry_loc, loc,
-                                                   entry->d_name);
-                        if (ret)
-                                goto out;
-
-                        if (!IS_ENTRY_CWD (entry->d_name) &&
-                            !IS_ENTRY_PARENT (entry->d_name)) {
-
-                                    is_directory_empty = _gf_false;
-                                    gf_log (this->name, GF_LOG_DEBUG,
-                                            "lookup %s => %"PRId64,
-                                            entry_loc.path,
-                                            iatt.ia_ino);
-
-                                    ret = syncop_lookup (this, &entry_loc, NULL,
-                                                         &iatt, &xattr_rsp, &parent);
-
-                                    if (ret) {
-                                            gf_log (this->name, GF_LOG_ERROR,
-                                                    "%s: lookup failed",
-                                                    entry_loc.path);
-                                            continue;
-                                    }
-                                    pump_fill_loc_info (&entry_loc, &iatt,
-                                                       &parent);
-
-                                    pump_update_resume_state (this, entry_loc.path);
-
-                                    pump_save_path (this, entry_loc.path);
-                                    pump_save_file_stats (this, entry_loc.path);
-
-                                    ret = pump_check_and_update_status (this);
-                                    if (ret < 0) {
-                                            gf_log (this->name, GF_LOG_DEBUG,
-                                                    "Pump beginning to exit out");
-                                            goto out;
-                                    }
-
-                                    if (IA_ISDIR (iatt.ia_type)) {
-                                            if (is_pump_traversal_allowed (this, entry_loc.path)) {
-                                                    gf_log (this->name, GF_LOG_TRACE,
-                                                            "entering dir=%s",
-                                                            entry->d_name);
-                                                    gf_pump_traverse_directory (&entry_loc);
-                                            }
-                                    }
-                        }
-                }
-
-                gf_dirent_free (&entries);
-                free_entries = _gf_false;
-                gf_log (this->name, GF_LOG_TRACE,
-                        "offset incremented to %d",
-                        (int32_t ) offset);
-
-        }
-
-        ret = syncop_close (fd);
-        if (ret < 0)
-                gf_log (this->name, GF_LOG_DEBUG, "closing the fd failed");
-
-        if (is_directory_empty && IS_ROOT_PATH (loc->path)) {
-               pump_change_state (this, PUMP_STATE_RUNNING);
-               gf_log (this->name, GF_LOG_INFO, "Empty source brick. "
-                                "Nothing to be done.");
-        }
-
-out:
-        if (entry_loc.path)
-                loc_wipe (&entry_loc);
-        if (free_entries)
-                gf_dirent_free (&entries);
-        return 0;
-}
-
-static int
-pump_update_resume_path (xlator_t *this)
-{
-        const char *resume_path = NULL;
-
-        resume_path = pump_get_resume_path (this);
-
-        if (resume_path) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Found a path to resume from: %s",
-                        resume_path);
-
-        }else {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Did not find a path=> setting to '/'");
-                pump_set_resume_path (this, "/");
-        }
-
-        pump_change_state (this, PUMP_STATE_RESUME);
-
-        return 0;
-}
-
-static int32_t
-pump_xattr_cleaner (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-        afr_private_t  *priv      = NULL;
-        loc_t           loc       = {0};
-        int             i         = 0;
-        int             ret       = 0;
-        int             source    = 0;
-        int             sink      = 1;
-
-        priv      = this->private;
-
-        afr_build_root_loc (this, &loc);
-
-        ret = syncop_removexattr (priv->children[source], &loc,
-                                          PUMP_PATH);
-
-        ret = syncop_removexattr (priv->children[sink], &loc,
-                                  PUMP_SINK_COMPLETE);
-
-        for (i = 0; i < priv->child_count; i++) {
-                ret = syncop_removexattr (priv->children[i], &loc,
-                                          PUMP_SOURCE_COMPLETE);
-                if (ret)
-                        gf_log (this->name, GF_LOG_DEBUG, "removexattr "
-                                "failed with %s", strerror (errno));
-        }
-
-        loc_wipe (&loc);
-        return pump_command_reply (frame, this);
-}
-
-static int
-pump_complete_migration (xlator_t *this)
-{
-        afr_private_t *priv = NULL;
-        pump_private_t *pump_priv = NULL;
-        dict_t *dict = NULL;
-        pump_state_t state;
-        loc_t  loc = {0};
-        int dict_ret = 0;
-        int ret = -1;
-
-        priv = this->private;
-        pump_priv = priv->pump_private;
-
-        GF_ASSERT (priv->root_inode);
-
-        afr_build_root_loc (this, &loc);
-
-        dict = dict_new ();
-
-        state = pump_get_state ();
-        if (state == PUMP_STATE_RUNNING) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Pump finished pumping");
-
-                pump_priv->pump_finished = _gf_true;
-
-                dict_ret = dict_set_str (dict, PUMP_SOURCE_COMPLETE, "jargon");
-                if (dict_ret)
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "%s: failed to set the key %s",
-                                loc.path, PUMP_SOURCE_COMPLETE);
-
-                ret = syncop_setxattr (PUMP_SOURCE_CHILD (this), &loc, dict, 0);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "setxattr failed - while  notifying source complete");
-                }
-                dict_ret = dict_set_str (dict, PUMP_SINK_COMPLETE, "jargon");
-                if (dict_ret)
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "%s: failed to set the key %s",
-                                loc.path, PUMP_SINK_COMPLETE);
-
-                ret = syncop_setxattr (PUMP_SINK_CHILD (this), &loc, dict, 0);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "setxattr failed - while notifying sink complete");
-                }
-
-                pump_save_path (this, "/");
-
-        } else if (state == PUMP_STATE_ABORT) {
-                gf_log (this->name, GF_LOG_DEBUG, "Starting cleanup "
-                        "of pump internal xattrs");
-                call_resume (pump_priv->cleaner);
-        }
-
-        loc_wipe (&loc);
-        return 0;
-}
-
-static int
-pump_lookup_sink (loc_t *loc)
-{
-        xlator_t *this = NULL;
-	struct iatt iatt, parent;
-	dict_t *xattr_rsp;
-        dict_t *xattr_req = NULL;
-        int ret = 0;
-
-        this = THIS;
-
-        xattr_req = dict_new ();
-
-        ret = afr_set_root_gfid (xattr_req);
-        if (ret)
-                goto out;
-
-        ret = syncop_lookup (PUMP_SINK_CHILD (this), loc,
-                             xattr_req, &iatt, &xattr_rsp, &parent);
-
-        if (ret) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Lookup on sink child failed");
-                goto out;
-        }
-
-out:
-        if (xattr_req)
-                dict_unref (xattr_req);
-
-        return ret;
-}
-
-static int
-pump_task (void *data)
-{
-	xlator_t *this = NULL;
-        afr_private_t *priv = NULL;
-
-
-        loc_t loc = {0};
-	struct iatt iatt, parent;
-	dict_t *xattr_rsp = NULL;
-        dict_t *xattr_req = NULL;
-
-        int ret = -1;
-
-        this = THIS;
-        priv = this->private;
-
-        GF_ASSERT (priv->root_inode);
-
-        afr_build_root_loc (this, &loc);
-        xattr_req = dict_new ();
-        if (!xattr_req) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Out of memory");
-                ret = -1;
-                goto out;
-        }
-
-        afr_set_root_gfid (xattr_req);
-        ret = syncop_lookup (this, &loc, xattr_req,
-                             &iatt, &xattr_rsp, &parent);
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "lookup: path=%s gfid=%s",
-                loc.path, uuid_utoa (loc.inode->gfid));
-
-        ret = pump_check_and_update_status (this);
-        if (ret < 0) {
-                goto out;
-        }
-
-        pump_update_resume_path (this);
-
-        afr_set_root_gfid (xattr_req);
-        ret = pump_lookup_sink (&loc);
-        if (ret) {
-                pump_update_resume_path (this);
-                goto out;
-        }
-
-        gf_pump_traverse_directory (&loc);
-
-        pump_complete_migration (this);
-out:
-        if (xattr_req)
-                dict_unref (xattr_req);
-
-        loc_wipe (&loc);
-	return 0;
-}
-
-
-static int
-pump_task_completion (int ret, call_frame_t *sync_frame, void *data)
-{
-        xlator_t *this = NULL;
-        afr_private_t *priv = NULL;
-
-        this = THIS;
-
-        priv = this->private;
-
-        inode_unref (priv->root_inode);
-        STACK_DESTROY (sync_frame->root);
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "Pump xlator exiting");
-	return 0;
-}
-
-int
-pump_start (call_frame_t *pump_frame, xlator_t *this)
-{
-	afr_private_t *priv = NULL;
-	pump_private_t *pump_priv = NULL;
-
-	int ret = -1;
-
-	priv = this->private;
-        pump_priv = priv->pump_private;
-
-        afr_set_lk_owner (pump_frame, this, pump_frame->root);
-	pump_pid = (uint64_t) (unsigned long)pump_frame->root;
-
-	ret = synctask_new (pump_priv->env, pump_task,
-                            pump_task_completion,
-                            pump_frame, NULL);
-        if (ret == -1) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "starting pump failed");
-                pump_change_state (this, PUMP_STATE_ABORT);
-                goto out;
-        }
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "setting pump as started lk_owner: %s %"PRIu64,
-                lkowner_utoa (&pump_frame->root->lk_owner), pump_pid);
-
-        priv->use_afr_in_pump = 1;
-out:
-	return ret;
-}
-
-static int
-pump_start_synctask (xlator_t *this)
-{
-        call_frame_t *frame = NULL;
-        int ret = 0;
-
-        frame = create_frame (this, this->ctx->pool);
-        if (!frame) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                ret = -1;
-                goto out;
-        }
-
-        pump_change_state (this, PUMP_STATE_RUNNING);
-
-        ret = pump_start (frame, this);
-
-out:
-        return ret;
-}
-
-int32_t
-pump_cmd_start_setxattr_cbk (call_frame_t *frame,
-                             void *cookie,
-                             xlator_t *this,
-                             int32_t op_ret,
-                             int32_t op_errno, dict_t *xdata)
-
-{
-        call_frame_t *prev = NULL;
-        afr_local_t *local = NULL;
-        int ret = 0;
-
-        local = frame->local;
-
-        if (op_ret < 0) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Could not initiate destination "
-                        "brick connect");
-                ret = op_ret;
-                goto out;
-        }
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "Successfully initiated destination "
-                "brick connect");
-
-        pump_mark_start_pending (this);
-
-        /* send the PARENT_UP as pump is ready now */
-        prev = cookie;
-        if (prev && prev->this)
-                prev->this->notify (prev->this, GF_EVENT_PARENT_UP, this);
-
-out:
-        local->op_ret = ret;
-        pump_command_reply (frame, this);
-
-        return 0;
-}
-
-static int
-pump_initiate_sink_connect (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t   *local     = NULL;
-        afr_private_t *priv      = NULL;
-        dict_t        *dict      = NULL;
-        data_t        *data      = NULL;
-        char          *clnt_cmd  = NULL;
-        loc_t loc = {0};
-
-        int ret = 0;
-
-        priv  = this->private;
-        local = frame->local;
-
-        GF_ASSERT (priv->root_inode);
-
-        afr_build_root_loc (this, &loc);
-
-        data = data_ref (dict_get (local->dict, RB_PUMP_CMD_START));
-        if (!data) {
-                ret = -1;
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Could not get destination brick value");
-                goto out;
-        }
-
-        dict = dict_new ();
-        if (!dict) {
-                ret = -1;
-                goto out;
-        }
-
-        clnt_cmd = GF_CALLOC (1, data->len+1, gf_common_mt_char);
-        if (!clnt_cmd) {
-                ret = -1;
-                goto out;
-        }
-
-        memcpy (clnt_cmd, data->data, data->len);
-        clnt_cmd[data->len] = '\0';
-        gf_log (this->name, GF_LOG_DEBUG, "Got destination brick %s\n",
-                        clnt_cmd);
-
-        ret = dict_set_dynstr (dict, CLIENT_CMD_CONNECT, clnt_cmd);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Could not inititiate destination brick "
-                        "connect");
-                goto out;
-        }
-
-	STACK_WIND (frame,
-		    pump_cmd_start_setxattr_cbk,
-		    PUMP_SINK_CHILD(this),
-		    PUMP_SINK_CHILD(this)->fops->setxattr,
-		    &loc,
-		    dict,
-		    0, NULL);
-
-        ret = 0;
-
-out:
-        if (dict)
-                dict_unref (dict);
-
-        if (data)
-                data_unref (data);
-
-        if (ret && clnt_cmd)
-                GF_FREE (clnt_cmd);
-
-        loc_wipe (&loc);
-        return ret;
-}
-
-static int
-is_pump_aborted (xlator_t *this)
-{
-        pump_state_t state;
-
-        state = pump_get_state ();
-
-        return ((state == PUMP_STATE_ABORT));
-}
-
-int32_t
-pump_cmd_start_getxattr_cbk (call_frame_t *frame,
-                             void *cookie,
-                             xlator_t *this,
-                             int32_t op_ret,
-                             int32_t op_errno,
-                             dict_t *dict, dict_t *xdata)
-{
-        afr_local_t *local = NULL;
-        char *path = NULL;
-
-        pump_state_t state;
-        int ret = 0;
-        int need_unwind = 0;
-        int dict_ret = -1;
-
-        local = frame->local;
-
-        if (op_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "getxattr failed - changing pump "
-                        "state to RUNNING with '/'");
-                path = "/";
-                ret = op_ret;
-        } else {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "getxattr succeeded");
-
-                dict_ret =  dict_get_str (dict, PUMP_PATH, &path);
-                if (dict_ret < 0)
-                        path = "/";
-        }
-
-        state = pump_get_state ();
-        if ((state == PUMP_STATE_RUNNING) ||
-            (state == PUMP_STATE_RESUME)) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Pump is already started");
-                ret = -1;
-                goto out;
-        }
-
-        pump_set_resume_path (this, path);
-
-        if (is_pump_aborted (this))
-                /* We're re-starting pump afresh */
-                ret = pump_initiate_sink_connect (frame, this);
-        else {
-                /* We're re-starting pump from a previous
-                   pause */
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "about to start synctask");
-                ret = pump_start_synctask (this);
-                need_unwind = 1;
-        }
-
-out:
-        if ((ret < 0) || (need_unwind == 1)) {
-                local->op_ret = ret;
-                pump_command_reply (frame, this);
-        }
-	return 0;
-}
-
-int
-pump_execute_status (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t *priv = NULL;
-        pump_private_t *pump_priv = NULL;
-
-        uint64_t number_files = 0;
-
-        char filename[PATH_MAX];
-        char summary[PATH_MAX+256];
-        char *dict_str = NULL;
-
-        int32_t op_ret = 0;
-        int32_t op_errno = 0;
-
-        dict_t *dict = NULL;
-        int ret = -1;
-
-        priv = this->private;
-        pump_priv = priv->pump_private;
-
-        LOCK (&pump_priv->resume_path_lock);
-        {
-                number_files  = pump_priv->number_files_pumped;
-                strncpy (filename, pump_priv->current_file, PATH_MAX);
-        }
-        UNLOCK (&pump_priv->resume_path_lock);
-
-        dict_str     = GF_CALLOC (1, PATH_MAX + 256, gf_afr_mt_char);
-        if (!dict_str) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                op_ret = -1;
-                op_errno = ENOMEM;
-                goto out;
-        }
-
-        if (pump_priv->pump_finished) {
-                snprintf (summary, PATH_MAX+256,
-                          "no_of_files=%"PRIu64, number_files);
-        } else {
-                snprintf (summary, PATH_MAX+256,
-                          "no_of_files=%"PRIu64":current_file=%s",
-                          number_files, filename);
-        }
-        snprintf (dict_str, PATH_MAX+256, "status=%d:%s",
-                  (pump_priv->pump_finished)?1:0, summary);
-
-        dict = dict_new ();
-
-        ret = dict_set_dynstr (dict, RB_PUMP_CMD_STATUS, dict_str);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "dict_set_dynstr returned negative value");
-        } else {
-                dict_str = NULL;
-        }
-
-        op_ret = 0;
-
-out:
-
-        AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict, NULL);
-
-        if (dict)
-                dict_unref (dict);
-
-        GF_FREE (dict_str);
-
-        return 0;
-}
-
-int
-pump_execute_pause (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local = NULL;
-
-        local = frame->local;
-
-        pump_change_state (this, PUMP_STATE_PAUSE);
-
-        local->op_ret = 0;
-        pump_command_reply (frame, this);
-
-        return 0;
-}
-
-int
-pump_execute_start (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t *priv = NULL;
-        afr_local_t   *local = NULL;
-
-        int ret = 0;
-        loc_t loc = {0};
-
-        priv = this->private;
-        local = frame->local;
-
-        if (!priv->root_inode) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Pump xlator cannot be started without an initial "
-                        "lookup");
-                ret = -1;
-                goto out;
-        }
-
-        GF_ASSERT (priv->root_inode);
-
-        afr_build_root_loc (this, &loc);
-
-	STACK_WIND (frame,
-		    pump_cmd_start_getxattr_cbk,
-		    PUMP_SOURCE_CHILD(this),
-		    PUMP_SOURCE_CHILD(this)->fops->getxattr,
-		    &loc,
-		    PUMP_PATH, NULL);
-
-        ret = 0;
-
-out:
-        if (ret < 0) {
-                local->op_ret = ret;
-                pump_command_reply (frame, this);
-        }
-
-        loc_wipe (&loc);
-	return 0;
-}
-
-static int
-pump_cleanup_helper (void *data) {
-        call_frame_t *frame = data;
-
-        pump_xattr_cleaner (frame, 0, frame->this, 0, 0, NULL);
-
-        return 0;
-}
-
-static int
-pump_cleanup_done (int ret, call_frame_t *sync_frame, void *data)
-{
-        STACK_DESTROY (sync_frame->root);
-
-        return 0;
-}
-
-int
-pump_execute_commit (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t  *priv       = NULL;
-        pump_private_t *pump_priv  = NULL;
-        afr_local_t    *local      = NULL;
-        call_frame_t   *sync_frame = NULL;
-        int             ret        = 0;
-
-        priv      = this->private;
-        pump_priv = priv->pump_private;
-        local     = frame->local;
-
-        local->op_ret = 0;
-        if (pump_priv->pump_finished) {
-                pump_change_state (this, PUMP_STATE_COMMIT);
-                sync_frame = create_frame (this, this->ctx->pool);
-                ret = synctask_new (pump_priv->env, pump_cleanup_helper,
-                                    pump_cleanup_done, sync_frame, frame);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_DEBUG, "Couldn't create "
-                                "synctask for cleaning up xattrs.");
-                }
-
-        } else {
-                gf_log (this->name, GF_LOG_ERROR, "Commit can't proceed. "
-                        "Migration in progress");
-                local->op_ret = -1;
-                local->op_errno = EINPROGRESS;
-                pump_command_reply (frame, this);
-        }
-
-        return 0;
-}
-int
-pump_execute_abort (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t  *priv       = NULL;
-        pump_private_t *pump_priv  = NULL;
-        afr_local_t    *local      = NULL;
-        call_frame_t   *sync_frame = NULL;
-        int             ret        = 0;
-
-        priv      = this->private;
-        pump_priv = priv->pump_private;
-        local     = frame->local;
-
-        pump_change_state (this, PUMP_STATE_ABORT);
-
-        LOCK (&pump_priv->resume_path_lock);
-        {
-                pump_priv->number_files_pumped = 0;
-                pump_priv->current_file[0] = '\0';
-        }
-        UNLOCK (&pump_priv->resume_path_lock);
-
-        local->op_ret = 0;
-        if (pump_priv->pump_finished) {
-                sync_frame = create_frame (this, this->ctx->pool);
-                ret = synctask_new (pump_priv->env, pump_cleanup_helper,
-                                    pump_cleanup_done, sync_frame, frame);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_DEBUG, "Couldn't create "
-                                "synctask for cleaning up xattrs.");
-                }
-
-        } else {
-                pump_priv->cleaner = fop_setxattr_cbk_stub (frame,
-                                                            pump_xattr_cleaner,
-                                                            0, 0, NULL);
-        }
-
-        return 0;
-}
-
-gf_boolean_t
-pump_command_status (xlator_t *this, dict_t *dict)
-{
-        char *cmd = NULL;
-        int dict_ret = -1;
-        int ret = _gf_true;
-
-        dict_ret = dict_get_str (dict, RB_PUMP_CMD_STATUS, &cmd);
-        if (dict_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Not a pump status command");
-                ret = _gf_false;
-                goto out;
-        }
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "Hit a pump command - status");
-        ret = _gf_true;
-
-out:
-        return ret;
-
-}
-
-gf_boolean_t
-pump_command_pause (xlator_t *this, dict_t *dict)
-{
-        char *cmd = NULL;
-        int dict_ret = -1;
-        int ret = _gf_true;
-
-        dict_ret = dict_get_str (dict, RB_PUMP_CMD_PAUSE, &cmd);
-        if (dict_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Not a pump pause command");
-                ret = _gf_false;
-                goto out;
-        }
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "Hit a pump command - pause");
-        ret = _gf_true;
-
-out:
-        return ret;
-
-}
-
-gf_boolean_t
-pump_command_commit (xlator_t *this, dict_t *dict)
-{
-        char *cmd = NULL;
-        int dict_ret = -1;
-        int ret = _gf_true;
-
-        dict_ret = dict_get_str (dict, RB_PUMP_CMD_COMMIT, &cmd);
-        if (dict_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Not a pump commit command");
-                ret = _gf_false;
-                goto out;
-        }
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "Hit a pump command - commit");
-        ret = _gf_true;
-
-out:
-        return ret;
-
-}
-
-gf_boolean_t
-pump_command_abort (xlator_t *this, dict_t *dict)
-{
-        char *cmd = NULL;
-        int dict_ret = -1;
-        int ret = _gf_true;
-
-        dict_ret = dict_get_str (dict, RB_PUMP_CMD_ABORT, &cmd);
-        if (dict_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Not a pump abort command");
-                ret = _gf_false;
-                goto out;
-        }
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "Hit a pump command - abort");
-        ret = _gf_true;
-
-out:
-        return ret;
-
-}
-
-gf_boolean_t
-pump_command_start (xlator_t *this, dict_t *dict)
-{
-        char *cmd = NULL;
-        int dict_ret = -1;
-        int ret = _gf_true;
-
-        dict_ret = dict_get_str (dict, RB_PUMP_CMD_START, &cmd);
-        if (dict_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Not a pump start command");
-                ret = _gf_false;
-                goto out;
-        }
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "Hit a pump command - start");
-        ret = _gf_true;
-
-out:
-        return ret;
-
-}
-
-struct _xattr_key {
-        char *key;
-        struct list_head list;
-};
-
-static int
-__gather_xattr_keys (dict_t *dict, char *key, data_t *value,
-                     void *data)
-{
-        struct list_head *  list  = data;
-        struct _xattr_key * xkey  = NULL;
-
-        if (!strncmp (key, AFR_XATTR_PREFIX,
-                      strlen (AFR_XATTR_PREFIX))) {
-
-                xkey = GF_CALLOC (1, sizeof (*xkey), gf_afr_mt_xattr_key);
-                if (!xkey)
-                        return -1;
-
-                xkey->key = key;
-                INIT_LIST_HEAD (&xkey->list);
-
-                list_add_tail (&xkey->list, list);
-        }
-        return 0;
-}
-
-static void
-__filter_xattrs (dict_t *dict)
-{
-        struct list_head keys;
-
-        struct _xattr_key *key;
-        struct _xattr_key *tmp;
-
-        INIT_LIST_HEAD (&keys);
-
-        dict_foreach (dict, __gather_xattr_keys,
-                      (void *) &keys);
-
-        list_for_each_entry_safe (key, tmp, &keys, list) {
-                dict_del (dict, key->key);
-
-                list_del_init (&key->list);
-
-                GF_FREE (key);
-        }
-}
-
-int32_t
-pump_getxattr_cbk (call_frame_t *frame, void *cookie,
-		  xlator_t *this, int32_t op_ret, int32_t op_errno,
-		  dict_t *dict, dict_t *xdata)
-{
-	afr_private_t   *priv           = NULL;
-	afr_local_t     *local          = NULL;
-	xlator_t        **children      = NULL;
-	int             unwind          = 1;
-        int32_t         *last_index     = NULL;
-        int32_t         next_call_child = -1;
-        int32_t         read_child      = -1;
-        int32_t         *fresh_children = NULL;
-
-
-	priv     = this->private;
-	children = priv->children;
-
-	local = frame->local;
-
-        read_child = (long) cookie;
-
-	if (op_ret == -1) {
-		last_index = &local->cont.getxattr.last_index;
-                fresh_children = local->fresh_children;
-                next_call_child = afr_next_call_child (fresh_children,
-                                                       local->child_up,
-                                                       priv->child_count,
-                                                       last_index, read_child);
-                if (next_call_child < 0)
-                        goto out;
-
-		unwind = 0;
-		STACK_WIND_COOKIE (frame, pump_getxattr_cbk,
-				   (void *) (long) read_child,
-				   children[next_call_child],
-				   children[next_call_child]->fops->getxattr,
-				   &local->loc,
-				   local->cont.getxattr.name, NULL);
-	}
-
-out:
-	if (unwind) {
-                if (op_ret >= 0 && dict)
-                        __filter_xattrs (dict);
-
-		AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict, NULL);
-	}
-
-	return 0;
-}
-
-int32_t
-pump_getxattr (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, const char *name, dict_t *xdata)
-{
-	afr_private_t *   priv       = NULL;
-	xlator_t **       children   = NULL;
-	int               call_child = 0;
-	afr_local_t       *local     = NULL;
-	int32_t           ret     = -1;
-	int32_t           op_errno   = 0;
-        uint64_t          read_child = 0;
-
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-	priv     = this->private;
-	VALIDATE_OR_GOTO (priv->children, out);
-
-	children = priv->children;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame, default_getxattr_cbk,
-                            FIRST_CHILD (this),
-                            (FIRST_CHILD (this))->fops->getxattr,
-                            loc, name, xdata);
-                return 0;
-        }
-
-
-	AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
-	local = frame->local;
-
-        ret = afr_local_init (local, priv, &op_errno);
-        if (ret < 0)
-                goto out;
-
-        if (name) {
-                if (!strncmp (name, AFR_XATTR_PREFIX,
-                              strlen (AFR_XATTR_PREFIX))) {
-
-                        op_errno = ENODATA;
-                        goto out;
-                }
-
-                if (!strcmp (name, RB_PUMP_CMD_STATUS)) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Hit pump command - status");
-                        pump_execute_status (frame, this);
-                        ret = 0;
-                        goto out;
-                }
-        }
-
-        local->fresh_children = GF_CALLOC (priv->child_count,
-                                          sizeof (*local->fresh_children),
-                                          gf_afr_mt_int32_t);
-        if (!local->fresh_children) {
-                ret = -1;
-                op_errno = ENOMEM;
-                goto out;
-        }
-
-        read_child = afr_inode_get_read_ctx (this, loc->inode, local->fresh_children);
-        ret = afr_get_call_child (this, local->child_up, read_child,
-                                     local->fresh_children,
-                                     &call_child,
-                                     &local->cont.getxattr.last_index);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
-	loc_copy (&local->loc, loc);
-	if (name)
-	  local->cont.getxattr.name       = gf_strdup (name);
-
-	STACK_WIND_COOKIE (frame, pump_getxattr_cbk,
-			   (void *) (long) call_child,
-			   children[call_child], children[call_child]->fops->getxattr,
-			   loc, name, xdata);
-
-	ret = 0;
-out:
-	if (ret < 0)
-		AFR_STACK_UNWIND (getxattr, frame, -1, op_errno, NULL, NULL);
-	return 0;
-}
-
-static int
-afr_setxattr_unwind (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t *   local = NULL;
-	call_frame_t   *main_frame = NULL;
-
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame)
-			main_frame = local->transaction.main_frame;
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
-
-	if (main_frame) {
-		AFR_STACK_UNWIND (setxattr, main_frame,
-                                  local->op_ret, local->op_errno, NULL);
-	}
-	return 0;
-}
-
-static int
-afr_setxattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-
-	int call_count  = -1;
-	int need_unwind = 0;
-
-	local = frame->local;
-	priv = this->private;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret != -1) {
-			if (local->success_count == 0) {
-				local->op_ret = op_ret;
-			}
-			local->success_count++;
-
-			if (local->success_count == priv->child_count) {
-				need_unwind = 1;
-			}
-		}
-
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
-
-	if (need_unwind)
-		local->transaction.unwind (frame, this);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-	}
-
-	return 0;
-}
-
-static int
-afr_setxattr_wind (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-
-	int call_count = -1;
-	int i = 0;
-
-	local = frame->local;
-	priv = this->private;
-
-	call_count = afr_up_children_count (local->child_up, priv->child_count);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_setxattr_wind_cbk,
-					   (void *) (long) i,
-					   priv->children[i],
-					   priv->children[i]->fops->setxattr,
-					   &local->loc,
-					   local->cont.setxattr.dict,
-					   local->cont.setxattr.flags, NULL);
-
-			if (!--call_count)
-				break;
-		}
-	}
-
-	return 0;
-}
-
-
-static int
-afr_setxattr_done (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t * local = frame->local;
-
-	local->transaction.unwind (frame, this);
-
-	AFR_STACK_DESTROY (frame);
-
-	return 0;
-}
-
-int32_t
-pump_setxattr_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno, dict_t *xdata)
-{
-	AFR_STACK_UNWIND (setxattr, frame, op_ret, op_errno, xdata);
-	return 0;
-}
-
-int
-pump_command_reply (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local = NULL;
-
-        local = frame->local;
-
-        if (local->op_ret < 0)
-                gf_log (this->name, GF_LOG_INFO,
-                        "Command failed");
-        else
-                gf_log (this->name, GF_LOG_INFO,
-                        "Command succeeded");
-
-        AFR_STACK_UNWIND (setxattr,
-                          frame,
-                          local->op_ret,
-                          local->op_errno, NULL);
-
-        return 0;
-}
-
-int
-pump_parse_command (call_frame_t *frame, xlator_t *this,
-                    afr_local_t *local, dict_t *dict)
-{
-
-        int ret = -1;
-
-        if (pump_command_start (this, dict)) {
-                frame->local = local;
-                local->dict = dict_ref (dict);
-                ret = pump_execute_start (frame, this);
-
-        } else if (pump_command_pause (this, dict)) {
-                frame->local = local;
-                local->dict = dict_ref (dict);
-                ret = pump_execute_pause (frame, this);
-
-        } else if (pump_command_abort (this, dict)) {
-                frame->local = local;
-                local->dict = dict_ref (dict);
-                ret = pump_execute_abort (frame, this);
-
-        } else if (pump_command_commit (this, dict)) {
-                frame->local = local;
-                local->dict = dict_ref (dict);
-                ret = pump_execute_commit (frame, this);
-        }
-        return ret;
-}
-
-int
-pump_setxattr (call_frame_t *frame, xlator_t *this,
-               loc_t *loc, dict_t *dict, int32_t flags, dict_t *xdata)
-{
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t   *transaction_frame = NULL;
-	int ret = -1;
-	int op_errno = 0;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-        GF_IF_INTERNAL_XATTR_GOTO ("trusted.glusterfs.pump*", dict,
-                                   op_errno, out);
-
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame, default_setxattr_cbk,
-                            FIRST_CHILD (this),
-                            (FIRST_CHILD (this))->fops->setxattr,
-                            loc, dict, flags, xdata);
-                return 0;
-        }
-
-
-	AFR_LOCAL_ALLOC_OR_GOTO (local, out);
-
-	ret = afr_local_init (local, priv, &op_errno);
-	if (ret < 0) {
-                afr_local_cleanup (local, this);
-		goto out;
-        }
-
-        ret = pump_parse_command (frame, this,
-                                  local, dict);
-        if (ret >= 0) {
-                ret = 0;
-                goto out;
-        }
-
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-                op_errno = ENOMEM;
-                ret = -1;
-                afr_local_cleanup (local, this);
-		goto out;
-	}
-
-	transaction_frame->local = local;
-
-	local->op_ret = -1;
-
-	local->cont.setxattr.dict  = dict_ref (dict);
-	local->cont.setxattr.flags = flags;
-
-	local->transaction.fop    = afr_setxattr_wind;
-	local->transaction.done   = afr_setxattr_done;
-	local->transaction.unwind = afr_setxattr_unwind;
-
-	loc_copy (&local->loc, loc);
-
-	local->transaction.main_frame = frame;
-	local->transaction.start   = LLONG_MAX - 1;
-	local->transaction.len     = 0;
-
-	afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
-
-	ret = 0;
-out:
-	if (ret < 0) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (setxattr, frame, -1, op_errno, NULL);
-	}
-
-	return 0;
-}
-
-/* Defaults */
-static int32_t
-pump_lookup (call_frame_t *frame,
-             xlator_t *this,
-             loc_t *loc,
-             dict_t *xattr_req)
-{
-	afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_lookup_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->lookup,
-                            loc,
-                            xattr_req);
-                return 0;
-        }
-
-        afr_lookup (frame, this, loc, xattr_req);
-        return 0;
-}
-
-
-static int32_t
-pump_truncate (call_frame_t *frame,
-               xlator_t *this,
-               loc_t *loc,
-               off_t offset, dict_t *xdata)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_truncate_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->truncate,
-                            loc,
-                            offset, xdata);
-                return 0;
-        }
-
-        afr_truncate (frame, this, loc, offset, xdata);
-        return 0;
-}
-
-
-static int32_t
-pump_ftruncate (call_frame_t *frame,
-                xlator_t *this,
-                fd_t *fd,
-                off_t offset, dict_t *xdata)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_ftruncate_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->ftruncate,
-                            fd,
-                            offset, xdata);
-                return 0;
-        }
-
-        afr_ftruncate (frame, this, fd, offset, xdata);
-        return 0;
-}
-
-
-
-
-int
-pump_mknod (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, mode_t mode, dev_t rdev, mode_t umask, dict_t *xdata)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame, default_mknod_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->mknod,
-                            loc, mode, rdev, umask, xdata);
-                return 0;
-        }
-        afr_mknod (frame, this, loc, mode, rdev, umask, xdata);
-        return 0;
-
-}
-
-
-
-int
-pump_mkdir (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, mode_t mode, mode_t umask, dict_t *xdata)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame, default_mkdir_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->mkdir,
-                            loc, mode, umask, xdata);
-                return 0;
-        }
-        afr_mkdir (frame, this, loc, mode, umask, xdata);
-        return 0;
-
-}
-
-
-static int32_t
-pump_unlink (call_frame_t *frame,
-             xlator_t *this,
-             loc_t *loc, int xflag, dict_t *xdata)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_unlink_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->unlink,
-                            loc, xflag, xdata);
-                return 0;
-        }
-        afr_unlink (frame, this, loc, xflag, xdata);
-        return 0;
-
-}
-
-
-static int
-pump_rmdir (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, int flags, dict_t *xdata)
-{
-        afr_private_t *priv  = NULL;
-
-	priv = this->private;
-
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame, default_rmdir_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->rmdir,
-                            loc, flags, xdata);
-                return 0;
-        }
-
-        afr_rmdir (frame, this, loc, flags, xdata);
-        return 0;
-
-}
-
-
-
-int
-pump_symlink (call_frame_t *frame, xlator_t *this,
-              const char *linkpath, loc_t *loc, mode_t umask, dict_t *xdata)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame, default_symlink_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->symlink,
-                            linkpath, loc, umask, xdata);
-                return 0;
-        }
-        afr_symlink (frame, this, linkpath, loc, umask, xdata);
-        return 0;
-
-}
-
-
-static int32_t
-pump_rename (call_frame_t *frame,
-             xlator_t *this,
-             loc_t *oldloc,
-             loc_t *newloc, dict_t *xdata)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_rename_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->rename,
-                            oldloc, newloc, xdata);
-                return 0;
-        }
-        afr_rename (frame, this, oldloc, newloc, xdata);
-        return 0;
-
-}
-
-
-static int32_t
-pump_link (call_frame_t *frame,
-           xlator_t *this,
-           loc_t *oldloc,
-           loc_t *newloc, dict_t *xdata)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_link_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->link,
-                            oldloc, newloc, xdata);
-                return 0;
-        }
-        afr_link (frame, this, oldloc, newloc, xdata);
-        return 0;
-
-}
-
-
-static int32_t
-pump_create (call_frame_t *frame, xlator_t *this,
-             loc_t *loc, int32_t flags, mode_t mode,
-             mode_t umask, fd_t *fd, dict_t *xdata)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame, default_create_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->create,
-                            loc, flags, mode, umask, fd, xdata);
-                return 0;
-        }
-        afr_create (frame, this, loc, flags, mode, umask, fd, xdata);
-        return 0;
-
-}
-
-
-static int32_t
-pump_open (call_frame_t *frame,
-           xlator_t *this,
-           loc_t *loc,
-           int32_t flags, fd_t *fd, dict_t *xdata)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_open_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->open,
-                            loc, flags, fd, xdata);
-                return 0;
-        }
-        afr_open (frame, this, loc, flags, fd, xdata);
-        return 0;
-
-}
-
-
-static int32_t
-pump_writev (call_frame_t *frame,
-             xlator_t *this,
-             fd_t *fd,
-             struct iovec *vector,
-             int32_t count,
-             off_t off, uint32_t flags,
-             struct iobref *iobref, dict_t *xdata)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_writev_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->writev,
-                            fd,
-                            vector,
-                            count,
-                            off, flags,
-                            iobref, xdata);
-                return 0;
-        }
-
-        afr_writev (frame, this, fd, vector, count, off, flags, iobref, xdata);
-        return 0;
-}
-
-
-static int32_t
-pump_flush (call_frame_t *frame,
-            xlator_t *this,
-            fd_t *fd, dict_t *xdata)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_flush_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->flush,
-                            fd, xdata);
-                return 0;
-        }
-        afr_flush (frame, this, fd, xdata);
-        return 0;
-
-}
-
-
-static int32_t
-pump_fsync (call_frame_t *frame,
-            xlator_t *this,
-            fd_t *fd,
-            int32_t flags, dict_t *xdata)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_fsync_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->fsync,
-                            fd,
-                            flags, xdata);
-                return 0;
-        }
-        afr_fsync (frame, this, fd, flags, xdata);
-        return 0;
-
-}
-
-
-static int32_t
-pump_opendir (call_frame_t *frame,
-              xlator_t *this,
-              loc_t *loc, fd_t *fd, dict_t *xdata)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_opendir_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->opendir,
-                            loc, fd, xdata);
-                return 0;
-        }
-        afr_opendir (frame, this, loc, fd, xdata);
-        return 0;
-
-}
-
-
-static int32_t
-pump_fsyncdir (call_frame_t *frame,
-               xlator_t *this,
-               fd_t *fd,
-               int32_t flags, dict_t *xdata)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_fsyncdir_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->fsyncdir,
-                            fd,
-                            flags, xdata);
-                return 0;
-        }
-        afr_fsyncdir (frame, this, fd, flags, xdata);
-        return 0;
-
-}
-
-
-static int32_t
-pump_xattrop (call_frame_t *frame,
-              xlator_t *this,
-              loc_t *loc,
-              gf_xattrop_flags_t flags,
-              dict_t *dict, dict_t *xdata)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_xattrop_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->xattrop,
-                            loc,
-                            flags,
-                            dict, xdata);
-                return 0;
-        }
-        afr_xattrop (frame, this, loc, flags, dict, xdata);
-        return 0;
-
-}
-
-static int32_t
-pump_fxattrop (call_frame_t *frame,
-               xlator_t *this,
-               fd_t *fd,
-               gf_xattrop_flags_t flags,
-               dict_t *dict, dict_t *xdata)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_fxattrop_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->fxattrop,
-                            fd,
-                            flags,
-                            dict, xdata);
-                return 0;
-        }
-        afr_fxattrop (frame, this, fd, flags, dict, xdata);
-        return 0;
-
-}
-
-
-static int32_t
-pump_removexattr (call_frame_t *frame,
-                  xlator_t *this,
-                  loc_t *loc,
-                  const char *name, dict_t *xdata)
-{
-        afr_private_t *priv     = NULL;
-        int            op_errno = -1;
-
-        VALIDATE_OR_GOTO (this, out);
-
-        GF_IF_NATIVE_XATTR_GOTO ("trusted.glusterfs.pump*",
-                                 name, op_errno, out);
-
-        op_errno = 0;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_removexattr_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->removexattr,
-                            loc,
-                            name, xdata);
-                return 0;
-        }
-        afr_removexattr (frame, this, loc, name, xdata);
-
- out:
-        if (op_errno)
-                AFR_STACK_UNWIND (removexattr, frame, -1, op_errno, NULL);
-        return 0;
-
-}
-
-
-
-static int32_t
-pump_readdir (call_frame_t *frame,
-              xlator_t *this,
-              fd_t *fd,
-              size_t size,
-              off_t off, dict_t *xdata)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_readdir_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->readdir,
-                            fd, size, off, xdata);
-                return 0;
-        }
-        afr_readdir (frame, this, fd, size, off, xdata);
-        return 0;
-
-}
-
-
-static int32_t
-pump_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd,
-               size_t size, off_t off, dict_t *dict)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_readdirp_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->readdirp,
-                            fd, size, off, dict);
-                return 0;
-        }
-        afr_readdirp (frame, this, fd, size, off, dict);
-        return 0;
-
-}
-
-
-
-static int32_t
-pump_releasedir (xlator_t *this,
-                 fd_t *fd)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (priv->use_afr_in_pump)
-                afr_releasedir (this, fd);
-	return 0;
-
-}
-
-static int32_t
-pump_release (xlator_t *this,
-              fd_t *fd)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (priv->use_afr_in_pump)
-                afr_release (this, fd);
-	return 0;
-
-}
-
-static int32_t
-pump_forget (xlator_t *this, inode_t *inode)
-{
-        afr_private_t  *priv  = NULL;
-
-        priv = this->private;
-        if (priv->use_afr_in_pump)
-                afr_forget (this, inode);
-
-        return 0;
-}
-
-static int32_t
-pump_setattr (call_frame_t *frame,
-              xlator_t *this,
-              loc_t *loc,
-              struct iatt *stbuf,
-              int32_t valid, dict_t *xdata)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_setattr_cbk,
-                            FIRST_CHILD (this),
-                            FIRST_CHILD (this)->fops->setattr,
-                            loc, stbuf, valid, xdata);
-                return 0;
-        }
-        afr_setattr (frame, this, loc, stbuf, valid, xdata);
-        return 0;
-
-}
-
-
-static int32_t
-pump_fsetattr (call_frame_t *frame,
-               xlator_t *this,
-               fd_t *fd,
-               struct iatt *stbuf,
-               int32_t valid, dict_t *xdata)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_fsetattr_cbk,
-                            FIRST_CHILD (this),
-                            FIRST_CHILD (this)->fops->fsetattr,
-                            fd, stbuf, valid, xdata);
-                return 0;
-        }
-        afr_fsetattr (frame, this, fd, stbuf, valid, xdata);
-        return 0;
-
-}
-
-
-/* End of defaults */
-
-
-int32_t
-mem_acct_init (xlator_t *this)
-{
-        int     ret = -1;
-
-        if (!this)
-                return ret;
-
-        ret = xlator_mem_acct_init (this, gf_afr_mt_end + 1);
-
-        if (ret != 0) {
-                gf_log(this->name, GF_LOG_ERROR, "Memory accounting init"
-                                "failed");
-                return ret;
-        }
-
-        return ret;
-}
-
-static int
-is_xlator_pump_sink (xlator_t *child)
-{
-        return (child == PUMP_SINK_CHILD(THIS));
-}
-
-static int
-is_xlator_pump_source (xlator_t *child)
-{
-        return (child == PUMP_SOURCE_CHILD(THIS));
-}
-
-int32_t
-notify (xlator_t *this, int32_t event,
-	void *data, ...)
-{
-        int ret = -1;
-        xlator_t *child_xl = NULL;
-
-        child_xl = (xlator_t *) data;
-
-        ret = afr_notify (this, event, data, NULL);
-
-	switch (event) {
-	case GF_EVENT_CHILD_DOWN:
-                if (is_xlator_pump_source (child_xl))
-                        pump_change_state (this, PUMP_STATE_ABORT);
-                break;
-
-        case GF_EVENT_CHILD_UP:
-                if (is_xlator_pump_sink (child_xl))
-                        if (is_pump_start_pending (this)) {
-                                gf_log (this->name, GF_LOG_DEBUG,
-                                        "about to start synctask");
-                                ret = pump_start_synctask (this);
-                                if (ret < 0)
-                                        gf_log (this->name, GF_LOG_DEBUG,
-                                                "Could not start pump "
-                                                "synctask");
-                                else
-                                        pump_remove_start_pending (this);
-                        }
-        }
-
-        return ret;
-}
-
-int32_t
-init (xlator_t *this)
-{
-	afr_private_t * priv        = NULL;
-        pump_private_t *pump_priv = NULL;
-	int             child_count = 0;
-	xlator_list_t * trav        = NULL;
-	int             i           = 0;
-	int             ret         = -1;
-	GF_UNUSED int   op_errno    = 0;
-
-        int source_child = 0;
-
-	if (!this->children) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"pump translator needs a source and sink"
-                        "subvolumes defined.");
-		return -1;
-	}
-
-	if (!this->parents) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"Volume is dangling.");
-	}
-
-	this->private = GF_CALLOC (1, sizeof (afr_private_t),
-                                   gf_afr_mt_afr_private_t);
-        if (!this->private)
-                goto out;
-
-	priv = this->private;
-        LOCK_INIT (&priv->lock);
-        LOCK_INIT (&priv->read_child_lock);
-        //lock recovery is not done in afr
-        pthread_mutex_init (&priv->mutex, NULL);
-        INIT_LIST_HEAD (&priv->saved_fds);
-
-        child_count = xlator_subvolume_count (this);
-        if (child_count != 2) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "There should be exactly 2 children - one source "
-                        "and one sink");
-                return -1;
-        }
-	priv->child_count = child_count;
-
-        priv->read_child = source_child;
-        priv->favorite_child = source_child;
-        priv->background_self_heal_count = 0;
-
-	priv->data_self_heal     = "on";
-	priv->metadata_self_heal = 1;
-	priv->entry_self_heal    = 1;
-
-        priv->data_self_heal_window_size = 16;
-
-	priv->data_change_log     = 1;
-	priv->metadata_change_log = 1;
-	priv->entry_change_log    = 1;
-        priv->use_afr_in_pump = 1;
-        priv->sh_readdir_size = 65536;
-
-	/* Locking options */
-
-        /* Lock server count infact does not matter. Locks are held
-           on all subvolumes, in this case being the source
-           and the sink.
-        */
-
-	priv->strict_readdir = _gf_false;
-
-	priv->wait_count = 1;
-	priv->child_up = GF_CALLOC (sizeof (unsigned char), child_count,
-                                 gf_afr_mt_char);
-	if (!priv->child_up) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-		op_errno = ENOMEM;
-		goto out;
-	}
-
-	priv->children = GF_CALLOC (sizeof (xlator_t *), child_count,
-                                 gf_afr_mt_xlator_t);
-	if (!priv->children) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-		op_errno = ENOMEM;
-		goto out;
-	}
-
-        priv->pending_key = GF_CALLOC (sizeof (*priv->pending_key),
-                                       child_count,
-                                       gf_afr_mt_char);
-        if (!priv->pending_key) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory.");
-                op_errno = ENOMEM;
-                goto out;
-        }
-
-	trav = this->children;
-	i = 0;
-	while (i < child_count) {
-		priv->children[i] = trav->xlator;
-
-                ret = gf_asprintf (&priv->pending_key[i], "%s.%s", AFR_XATTR_PREFIX,
-                                   trav->xlator->name);
-                if (-1 == ret) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "asprintf failed to set pending key");
-                        op_errno = ENOMEM;
-                        goto out;
-                }
-
-		trav = trav->next;
-		i++;
-	}
-
-        ret = gf_asprintf (&priv->sh_domain, "%s-self-heal", this->name);
-        if (-1 == ret) {
-                op_errno = ENOMEM;
-                goto out;
-        }
-
-        priv->first_lookup = 1;
-        priv->root_inode = NULL;
-
-        priv->last_event = GF_CALLOC (child_count, sizeof (*priv->last_event),
-                                      gf_afr_mt_int32_t);
-        if (!priv->last_event) {
-                ret = -ENOMEM;
-                goto out;
-        }
-
-	pump_priv = GF_CALLOC (1, sizeof (*pump_priv),
-                            gf_afr_mt_pump_priv);
-	if (!pump_priv) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-                op_errno = ENOMEM;
-		goto out;
-	}
-
-        LOCK_INIT (&pump_priv->resume_path_lock);
-        LOCK_INIT (&pump_priv->pump_state_lock);
-
-        pump_priv->resume_path = GF_CALLOC (1, PATH_MAX,
-                                            gf_afr_mt_char);
-        if (!pump_priv->resume_path) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                ret = -1;
-                goto out;
-        }
-
-	pump_priv->env = this->ctx->env;
-        if (!pump_priv->env) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Could not create new sync-environment");
-                ret = -1;
-                goto out;
-        }
-
-        /* keep more local here as we may need them for self-heal etc */
-        this->local_pool = mem_pool_new (afr_local_t, 128);
-        if (!this->local_pool) {
-                ret = -1;
-                gf_log (this->name, GF_LOG_ERROR,
-                        "failed to create local_t's memory pool");
-                goto out;
-        }
-
-	priv->pump_private = pump_priv;
-
-        pump_change_state (this, PUMP_STATE_ABORT);
-
-	ret = 0;
-out:
-	return ret;
-}
-
-int
-fini (xlator_t *this)
-{
-        afr_private_t * priv        = NULL;
-        pump_private_t *pump_priv = NULL;
-
-        priv      = this->private;
-        this->private = NULL;
-        if (!priv)
-                goto out;
-
-        pump_priv = priv->pump_private;
-        if (!pump_priv)
-                goto afr_priv;
-
-        GF_FREE (pump_priv->resume_path);
-        LOCK_DESTROY (&pump_priv->resume_path_lock);
-        LOCK_DESTROY (&pump_priv->pump_state_lock);
-        GF_FREE (pump_priv);
-afr_priv:
-        afr_priv_destroy (priv);
-out:
-	return 0;
-}
-
-
-struct xlator_fops fops = {
-	.lookup      = pump_lookup,
-	.open        = pump_open,
-	.flush       = pump_flush,
-	.fsync       = pump_fsync,
-	.fsyncdir    = pump_fsyncdir,
-	.xattrop     = pump_xattrop,
-	.fxattrop    = pump_fxattrop,
-        .getxattr    = pump_getxattr,
-
-	/* inode write */
-	.writev      = pump_writev,
-	.truncate    = pump_truncate,
-	.ftruncate   = pump_ftruncate,
-	.setxattr    = pump_setxattr,
-        .setattr     = pump_setattr,
-	.fsetattr    = pump_fsetattr,
-	.removexattr = pump_removexattr,
-
-	/* dir read */
-	.opendir     = pump_opendir,
-	.readdir     = pump_readdir,
-	.readdirp    = pump_readdirp,
-
-	/* dir write */
-	.create      = pump_create,
-	.mknod       = pump_mknod,
-	.mkdir       = pump_mkdir,
-	.unlink      = pump_unlink,
-	.rmdir       = pump_rmdir,
-	.link        = pump_link,
-	.symlink     = pump_symlink,
-	.rename      = pump_rename,
-};
-
-struct xlator_dumpops dumpops = {
-        .priv       = afr_priv_dump,
-};
-
-
-struct xlator_cbks cbks = {
-	.release     = pump_release,
-	.releasedir  = pump_releasedir,
-        .forget      = pump_forget,
-};
-
-struct volume_options options[] = {
-	{ .key  = {NULL} },
-};
diff --git a/xlators/cluster/afr/src/pump.h b/xlators/cluster/afr/src/pump.h
deleted file mode 100644
index bc4c31a78a5..00000000000
--- a/xlators/cluster/afr/src/pump.h
+++ /dev/null
@@ -1,78 +0,0 @@
-/*
-  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
-  This file is part of GlusterFS.
-
-  This file is licensed to you under your choice of the GNU Lesser
-  General Public License, version 3 or any later version (LGPLv3 or
-  later), or the GNU General Public License, version 2 (GPLv2), in all
-  cases as published by the Free Software Foundation.
-*/
-
-#ifndef __PUMP_H__
-#define __PUMP_H__
-
-#include "syncop.h"
-
-/* FIXME: Needs to be defined in a common file */
-#define CLIENT_CMD_CONNECT "trusted.glusterfs.client-connect"
-#define CLIENT_CMD_DISCONNECT "trusted.glusterfs.client-disconnect"
-
-#define PUMP_SOURCE_COMPLETE "trusted.glusterfs.pump-source-complete"
-#define PUMP_SINK_COMPLETE "trusted.glusterfs.pump-sink-complete"
-
-#define PUMP_PATH "trusted.glusterfs.pump-path"
-
-#define PUMP_SOURCE_CHILD(xl) (xl->children->xlator)
-#define PUMP_SINK_CHILD(xl) (xl->children->next->xlator)
-
-typedef enum {
-        PUMP_STATE_RUNNING,             /* Pump is running and migrating files */
-        PUMP_STATE_RESUME,              /* Pump is resuming from a previous pause */
-        PUMP_STATE_PAUSE,               /* Pump is paused */
-        PUMP_STATE_ABORT,               /* Pump is aborted */
-        PUMP_STATE_COMMIT,              /* Pump is commited */
-} pump_state_t;
-
-typedef struct _pump_private {
-	struct syncenv *env;            /* The env pointer to the pump synctask */
-        char *resume_path;              /* path to resume from the last pause */
-        gf_lock_t resume_path_lock;     /* Synchronize resume_path changes */
-        gf_lock_t pump_state_lock;      /* Synchronize pump_state changes */
-        pump_state_t pump_state;        /* State of pump */
-        char current_file[PATH_MAX];    /* Current file being pumped */
-        uint64_t number_files_pumped;   /* Number of files pumped */
-        gf_boolean_t pump_finished;     /* Boolean to indicate pump termination */
-        char pump_start_pending;        /* Boolean to mark start pending until
-                                           CHILD_UP */
-        call_stub_t *cleaner;
-} pump_private_t;
-
-void
-build_root_loc (inode_t *inode, loc_t *loc);
-int pump_start (call_frame_t *frame, xlator_t *this);
-
-gf_boolean_t
-pump_command_start (xlator_t *this, dict_t *dict);
-
-int
-pump_execute_start (call_frame_t *frame, xlator_t *this);
-
-gf_boolean_t
-pump_command_pause (xlator_t *this, dict_t *dict);
-
-int
-pump_execute_pause (call_frame_t *frame, xlator_t *this);
-
-gf_boolean_t
-pump_command_abort (xlator_t *this, dict_t *dict);
-
-int
-pump_execute_abort (call_frame_t *frame, xlator_t *this);
-
-gf_boolean_t
-pump_command_status (xlator_t *this, dict_t *dict);
-
-int
-pump_execute_status (call_frame_t *frame, xlator_t *this);
-
-#endif /* __PUMP_H__ */
diff --git a/xlators/cluster/dht/src/Makefile.am b/xlators/cluster/dht/src/Makefile.am
index 174bea84110..56f1f2ad7c8 100644
--- a/xlators/cluster/dht/src/Makefile.am
+++ b/xlators/cluster/dht/src/Makefile.am
@@ -1,33 +1,37 @@
-
 xlator_LTLIBRARIES = dht.la nufa.la switch.la
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
 xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/cluster
 
 dht_common_source = dht-layout.c dht-helper.c dht-linkfile.c dht-rebalance.c \
 	dht-selfheal.c dht-rename.c dht-hashfn.c dht-diskusage.c \
 	dht-common.c dht-inode-write.c dht-inode-read.c dht-shared.c \
-	$(top_builddir)/xlators/lib/src/libxlator.c
+	dht-lock.c $(top_builddir)/xlators/lib/src/libxlator.c
 
 dht_la_SOURCES = $(dht_common_source) dht.c
 
 nufa_la_SOURCES = $(dht_common_source) nufa.c
 switch_la_SOURCES = $(dht_common_source) switch.c
 
-dht_la_LDFLAGS = -module -avoid-version
+dht_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
 dht_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
-nufa_la_LDFLAGS = -module -avoid-version
+nufa_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
 nufa_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
-switch_la_LDFLAGS = -module -avoid-version
+switch_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
 switch_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
-noinst_HEADERS = dht-common.h dht-mem-types.h \
-	$(top_builddir)/xlators/lib/src/libxlator.h
+noinst_HEADERS = dht-common.h dht-mem-types.h dht-messages.h \
+	dht-lock.h $(top_builddir)/xlators/lib/src/libxlator.h
 
 AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
-	-I$(top_srcdir)/xlators/lib/src
-
-AM_CFLAGS = -Wall $(GF_CFLAGS)
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src \
+	-I$(top_srcdir)/rpc/rpc-lib/src \
+	-I$(top_srcdir)/xlators/lib/src \
+	-DDATADIR=\"$(localstatedir)\" \
+	-DLIBDIR=\"$(libdir)\"
 
 CLEANFILES =
 
@@ -36,3 +40,9 @@ uninstall-local:
 
 install-data-hook:
 	ln -sf dht.so $(DESTDIR)$(xlatordir)/distribute.so
+
+if UNITTEST
+CLEANFILES += *.gcda *.gcno *_xunit.xml
+noinst_PROGRAMS =
+TESTS =
+endif
diff --git a/xlators/cluster/dht/src/dht-common.c b/xlators/cluster/dht/src/dht-common.c
index 6cb0226995f..8ba0cc4c732 100644
--- a/xlators/cluster/dht/src/dht-common.c
+++ b/xlators/cluster/dht/src/dht-common.c
@@ -8,96 +8,439 @@
   cases as published by the Free Software Foundation.
 */
 
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
 /* TODO: add NS locking */
 
-#include "glusterfs.h"
-#include "xlator.h"
 #include "libxlator.h"
 #include "dht-common.h"
-#include "defaults.h"
-#include "byte-order.h"
-#include "glusterfs-acl.h"
+#include "dht-lock.h"
+#include <glusterfs/byte-order.h>
+#include <glusterfs/quota-common-utils.h>
+#include <glusterfs/upcall-utils.h>
+#include "glusterfs/compat-errno.h"  // for ENODATA on BSD
+#include <glusterfs/common-utils.h>
 
 #include <sys/time.h>
 #include <libgen.h>
+#include <signal.h>
 
-int
-dht_aggregate (dict_t *this, char *key, data_t *value, void *data)
+static int
+dht_rmdir_readdirp_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno, gf_dirent_t *entries,
+                       dict_t *xdata);
+
+static int
+dht_link2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret);
+
+static int
+dht_set_dir_xattr_req(xlator_t *this, loc_t *loc, dict_t *xattr_req);
+
+static int
+dht_lookup_everywhere_done(call_frame_t *frame, xlator_t *this);
+
+static int
+dht_common_mark_mdsxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                             int op_ret, int op_errno, dict_t *xdata);
+
+static int
+dht_rmdir_unlock(call_frame_t *frame, xlator_t *this);
+
+static const char *dht_dbg_vxattrs[] = {DHT_DBG_HASHED_SUBVOL_PATTERN, NULL};
+
+/* Check the xdata to make sure EBADF has been set by client xlator */
+int32_t
+dht_check_remote_fd_failed_error(dht_local_t *local, int op_ret, int op_errno)
 {
-        dict_t  *dst  = NULL;
-        int64_t *ptr  = 0, *size = NULL;
-        int32_t  ret  = -1;
-        data_t  *dict_data = NULL;
+    if (op_ret == -1 && (op_errno == EBADF || op_errno == EBADFD) &&
+        !(local->fd_checked)) {
+        return 1;
+    }
+    return 0;
+}
 
-        dst = data;
+/* Sets the blocks and size values to fixed values. This is to be called
+ * only for dirs. The caller is responsible for checking the type
+ */
+int32_t
+dht_set_fixed_dir_stat(struct iatt *stat)
+{
+    if (stat) {
+        stat->ia_blocks = DHT_DIR_STAT_BLOCKS;
+        stat->ia_size = DHT_DIR_STAT_SIZE;
+        return 0;
+    }
+    return -1;
+}
 
-        if (strcmp (key, GF_XATTR_QUOTA_SIZE_KEY) == 0) {
-                ret = dict_get_bin (dst, key, (void **)&size);
-                if (ret < 0) {
-                        size = GF_CALLOC (1, sizeof (int64_t),
-                                          gf_common_mt_char);
-                        if (size == NULL) {
-                                gf_log ("dht", GF_LOG_WARNING,
-                                        "memory allocation failed");
-                                return -1;
-                        }
-                        ret = dict_set_bin (dst, key, size, sizeof (int64_t));
-                        if (ret < 0) {
-                                gf_log ("dht", GF_LOG_WARNING,
-                                        "dht aggregate dict set failed");
-                                GF_FREE (size);
-                                return -1;
-                        }
-                }
+/* Return true if key exists in array
+ */
+static gf_boolean_t
+dht_match_xattr(const char *key)
+{
+    char **xattrs_to_heal = get_xattrs_to_heal();
 
-                ptr = data_to_bin (value);
-                if (ptr == NULL) {
-                        gf_log ("dht", GF_LOG_WARNING, "data to bin failed");
-                        return -1;
-                }
+    return gf_get_index_by_elem(xattrs_to_heal, (char *)key) >= 0;
+}
+
+static int
+dht_aggregate_quota_xattr(dict_t *dst, char *key, data_t *value)
+{
+    int ret = -1;
+    quota_meta_t *meta_dst = NULL;
+    quota_meta_t *meta_src = NULL;
+    int64_t *size = NULL;
+    int64_t dst_dir_count = 0;
+    int64_t src_dir_count = 0;
+
+    if (value == NULL) {
+        gf_msg("dht", GF_LOG_WARNING, 0, DHT_MSG_DATA_NULL,
+               "data value is NULL");
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_get_bin(dst, key, (void **)&meta_dst);
+    if (ret < 0) {
+        meta_dst = GF_CALLOC(1, sizeof(quota_meta_t), gf_common_quota_meta_t);
+        if (meta_dst == NULL) {
+            gf_msg("dht", GF_LOG_WARNING, ENOMEM, DHT_MSG_NO_MEMORY,
+                   "Memory allocation failed");
+            ret = -1;
+            goto out;
+        }
+        ret = dict_set_bin(dst, key, meta_dst, sizeof(quota_meta_t));
+        if (ret < 0) {
+            gf_msg("dht", GF_LOG_WARNING, EINVAL, DHT_MSG_DICT_SET_FAILED,
+                   "dht aggregate dict set failed");
+            GF_FREE(meta_dst);
+            ret = -1;
+            goto out;
+        }
+    }
+
+    if (value->len > sizeof(int64_t)) {
+        meta_src = data_to_bin(value);
+
+        meta_dst->size = hton64(ntoh64(meta_dst->size) +
+                                ntoh64(meta_src->size));
+        meta_dst->file_count = hton64(ntoh64(meta_dst->file_count) +
+                                      ntoh64(meta_src->file_count));
 
-                *size = hton64 (ntoh64 (*size) + ntoh64 (*ptr));
+        if (value->len > (2 * sizeof(int64_t))) {
+            dst_dir_count = ntoh64(meta_dst->dir_count);
+            src_dir_count = ntoh64(meta_src->dir_count);
 
-        } else if (fnmatch (GF_XATTR_STIME_PATTERN, key, FNM_NOESCAPE) == 0) {
-                ret = gf_get_min_stime (THIS, dst, key, value);
-                if (ret < 0)
-                        return ret;
+            if (src_dir_count > dst_dir_count)
+                meta_dst->dir_count = meta_src->dir_count;
         } else {
-                /* compare user xattrs only */
-                if (!strncmp (key, "user.", strlen ("user."))) {
-                        ret = dict_lookup (dst, key, &dict_data);
-                        if (!ret && dict_data && value) {
-                                ret = is_data_equal (dict_data, value);
-                                if (!ret)
-                                        gf_log ("dht", GF_LOG_DEBUG,
-                                                "xattr mismatch for %s", key);
-                        }
-                }
-                ret = dict_set (dst, key, value);
-                if (ret)
-                        gf_log ("dht", GF_LOG_WARNING, "xattr dict set failed");
+            meta_dst->dir_count = 0;
         }
+    } else {
+        size = data_to_bin(value);
+        meta_dst->size = hton64(ntoh64(meta_dst->size) + ntoh64(*size));
+    }
 
-        return 0;
+    ret = 0;
+out:
+    return ret;
 }
 
+static int
+add_opt(char **optsp, const char *opt)
+{
+    char *newopts = NULL;
+    unsigned oldsize = 0;
+    unsigned newsize = 0;
+
+    if (*optsp == NULL)
+        newopts = gf_strdup(opt);
+    else {
+        oldsize = strlen(*optsp);
+        newsize = oldsize + 1 + strlen(opt) + 1;
+        newopts = GF_REALLOC(*optsp, newsize);
+        if (newopts)
+            sprintf(newopts + oldsize, ",%s", opt);
+    }
+    if (newopts == NULL) {
+        gf_msg("dht", GF_LOG_WARNING, 0, DHT_MSG_NO_MEMORY,
+               "Error to add choices in buffer in add_opt");
+        return -1;
+    }
+    *optsp = newopts;
+    return 0;
+}
 
-void
-dht_aggregate_xattr (dict_t *dst, dict_t *src)
+/* Return Choice list from Split brain status */
+static char *
+getChoices(const char *value)
 {
-        if ((dst == NULL) || (src == NULL)) {
+    int i = 0;
+    char *ptr = NULL;
+    char *tok = NULL;
+    char *result = NULL;
+    char *newval = NULL;
+
+    ptr = strstr(value, "Choices:");
+    if (!ptr) {
+        result = ptr;
+        goto out;
+    }
+
+    newval = gf_strdup(ptr);
+    if (!newval) {
+        result = newval;
+        goto out;
+    }
+
+    tok = strtok(newval, ":");
+    if (!tok) {
+        result = tok;
+        goto out;
+    }
+
+    while (tok) {
+        i++;
+        if (i == 2)
+            break;
+        tok = strtok(NULL, ":");
+    }
+
+    result = gf_strdup(tok);
+
+out:
+    if (newval)
+        GF_FREE(newval);
+
+    return result;
+}
+
+/* This function prepare a list of choices for key
+   (replica.split-brain-status) in   case of metadata split brain
+   only on the basis of key-value passed to this function.
+   After prepare the list of choices it update the same key in dict
+   with this value to reflect the same in
+   replica.split-brain-status attr for file.
+
+*/
+
+static int
+dht_aggregate_split_brain_xattr(dict_t *dst, char *key, data_t *value)
+{
+    int ret = 0;
+    char *oldvalue = NULL;
+    char *old_choice = NULL;
+    char *new_choice = NULL;
+    char *full_choice = NULL;
+    char *status = NULL;
+
+    if (value == NULL) {
+        gf_msg("dht", GF_LOG_WARNING, 0, DHT_MSG_DATA_NULL,
+               "GF_AFR_SBRAIN_STATUS value is NULL");
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_get_str(dst, key, &oldvalue);
+    if (ret)
+        goto out;
+
+    /* skip code that is irrelevant if !oldvalue */
+    if (!oldvalue)
+        goto out;
+
+    if (strstr(oldvalue, "not")) {
+        gf_msg_debug("dht", 0, "Need to update split-brain status in dict");
+        ret = -1;
+        goto out;
+    }
+    if (strstr(oldvalue, "metadata-split-brain:yes") &&
+        (strstr(oldvalue, "data-split-brain:no"))) {
+        if (strstr(value->data, "not")) {
+            gf_msg_debug("dht", 0, "No need to update split-brain status");
+            ret = 0;
+            goto out;
+        }
+        if (strstr(value->data, "yes") &&
+            (strncmp(oldvalue, value->data, strlen(oldvalue)))) {
+            old_choice = getChoices(oldvalue);
+            if (!old_choice) {
+                gf_msg("dht", GF_LOG_WARNING, 0, DHT_MSG_NO_MEMORY,
+                       "Error to get choices");
+                ret = -1;
+                goto out;
+            }
+
+            ret = add_opt(&full_choice, old_choice);
+            if (ret) {
+                gf_msg("dht", GF_LOG_WARNING, 0, DHT_MSG_NO_MEMORY,
+                       "Error to add choices");
+                ret = -1;
+                goto out;
+            }
+
+            new_choice = getChoices(value->data);
+            if (!new_choice) {
+                gf_msg("dht", GF_LOG_WARNING, 0, DHT_MSG_NO_MEMORY,
+                       "Error to get choices");
+                ret = -1;
                 goto out;
+            }
+
+            ret = add_opt(&full_choice, new_choice);
+            if (ret) {
+                gf_msg("dht", GF_LOG_WARNING, 0, DHT_MSG_NO_MEMORY,
+                       "Error to add choices ");
+                ret = -1;
+                goto out;
+            }
+            ret = gf_asprintf(&status,
+                              "data-split-brain:%s    "
+                              "metadata-split-brain:%s   Choices:%s",
+                              "no", "yes", full_choice);
+
+            if (-1 == ret) {
+                gf_msg("dht", GF_LOG_WARNING, 0, DHT_MSG_NO_MEMORY,
+                       "Error to prepare status ");
+                goto out;
+            }
+            ret = dict_set_dynstr(dst, key, status);
+            if (ret) {
+                gf_msg("dht", GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+                       "Failed to set full choice");
+            }
         }
+    }
 
-        dict_foreach (src, dht_aggregate, dst);
 out:
-        return;
+    if (old_choice)
+        GF_FREE(old_choice);
+    if (new_choice)
+        GF_FREE(new_choice);
+    if (full_choice)
+        GF_FREE(full_choice);
+
+    return ret;
+}
+
+static int
+dht_aggregate(dict_t *this, char *key, data_t *value, void *data)
+{
+    dict_t *dst = NULL;
+    int32_t ret = -1;
+    data_t *dict_data = NULL;
+
+    dst = data;
+
+    /* compare split brain xattr only */
+    if (strcmp(key, GF_AFR_SBRAIN_STATUS) == 0) {
+        ret = dht_aggregate_split_brain_xattr(dst, key, value);
+        if (!ret)
+            goto out;
+    } else if (strcmp(key, QUOTA_SIZE_KEY) == 0) {
+        ret = dht_aggregate_quota_xattr(dst, key, value);
+        if (ret) {
+            gf_msg("dht", GF_LOG_WARNING, 0,
+                   DHT_MSG_AGGREGATE_QUOTA_XATTR_FAILED,
+                   "Failed to aggregate quota xattr");
+        }
+        goto out;
+    } else if (fnmatch(GF_XATTR_STIME_PATTERN, key, FNM_NOESCAPE) == 0) {
+        ret = gf_get_min_stime(THIS, dst, key, value);
+        goto out;
+    } else {
+        /* compare user xattrs only */
+        if (!strncmp(key, "user.", SLEN("user."))) {
+            ret = dict_lookup(dst, key, &dict_data);
+            if (!ret && dict_data && value) {
+                ret = is_data_equal(dict_data, value);
+                if (!ret)
+                    gf_msg_debug("dht", 0, "xattr mismatch for %s", key);
+            }
+        }
+    }
+
+    ret = dict_set(dst, key, value);
+    if (ret) {
+        gf_msg("dht", GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+               "Failed to set dictionary value: key = %s", key);
+    }
+
+out:
+    return ret;
+}
+
+static void
+dht_aggregate_xattr(dict_t *dst, dict_t *src)
+{
+    if ((dst == NULL) || (src == NULL)) {
+        goto out;
+    }
+
+    dict_foreach(src, dht_aggregate, dst);
+out:
+    return;
+}
+
+/* Code to save hashed subvol on inode ctx as a mds subvol
+ */
+int
+dht_inode_ctx_mdsvol_set(inode_t *inode, xlator_t *this, xlator_t *mds_subvol)
+{
+    dht_inode_ctx_t *ctx = NULL;
+    int ret = -1;
+    uint64_t ctx_int = 0;
+    gf_boolean_t ctx_free = _gf_false;
+
+    LOCK(&inode->lock);
+    {
+        ret = __inode_ctx_get(inode, this, &ctx_int);
+        if (ctx_int) {
+            ctx = (dht_inode_ctx_t *)(uintptr_t)ctx_int;
+            ctx->mds_subvol = mds_subvol;
+        } else {
+            ctx = GF_CALLOC(1, sizeof(*ctx), gf_dht_mt_inode_ctx_t);
+            if (!ctx)
+                goto unlock;
+            ctx->mds_subvol = mds_subvol;
+            ctx_free = _gf_true;
+            ctx_int = (long)ctx;
+            ret = __inode_ctx_set(inode, this, &ctx_int);
+        }
+    }
+unlock:
+    UNLOCK(&inode->lock);
+    if (ret && ctx_free)
+        GF_FREE(ctx);
+    return ret;
+}
+
+/*Code to get mds subvol from inode ctx */
+
+int
+dht_inode_ctx_mdsvol_get(inode_t *inode, xlator_t *this, xlator_t **mdsvol)
+{
+    dht_inode_ctx_t *ctx = NULL;
+    int ret = -1;
+
+    if (!mdsvol)
+        return ret;
+
+    if (__is_root_gfid(inode->gfid)) {
+        (*mdsvol) = FIRST_CHILD(this);
+        return 0;
+    }
+
+    ret = dht_inode_ctx_get(inode, this, &ctx);
+
+    if (!ret && ctx) {
+        if (ctx->mds_subvol) {
+            *mdsvol = ctx->mds_subvol;
+            ret = 0;
+        } else {
+            ret = -1;
+        }
+    }
+
+    return ret;
 }
 
 /* TODO:
@@ -107,5106 +450,10942 @@ out:
    - complete linkfile selfheal
 */
 
-
-int
-dht_lookup_selfheal_cbk (call_frame_t *frame, void *cookie,
-                         xlator_t *this,
-                         int op_ret, int op_errno, dict_t *xdata)
+static int
+dht_lookup_selfheal_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
-        dht_layout_t *layout = NULL;
-	int           ret = -1;
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
 
-        GF_VALIDATE_OR_GOTO ("dht", frame, out);
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
 
-        local = frame->local;
-        ret = op_ret;
+    local = frame->local;
+    conf = this->private;
+    ret = op_ret;
 
-        FRAME_SU_UNDO (frame, dht_local_t);
+    FRAME_SU_UNDO(frame, dht_local_t);
 
-        if (ret == 0) {
-                layout = local->selfheal.layout;
-                ret = dht_layout_set (this, local->inode, layout);
-        }
+    if (ret == 0) {
+        layout = local->selfheal.layout;
+        ret = dht_layout_set(this, local->inode, layout);
+    }
 
-        if (local->loc.parent) {
-                dht_inode_ctx_time_update (local->loc.parent, this,
-                                           &local->postparent, 1);
-        }
+    dht_inode_ctx_time_update(local->inode, this, &local->stbuf, 1);
+    if (local->loc.parent) {
+        dht_inode_ctx_time_update(local->loc.parent, this, &local->postparent,
+                                  1);
+    }
 
-        DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
+    DHT_STRIP_PHASE1_FLAGS(&local->stbuf);
+    dht_set_fixed_dir_stat(&local->postparent);
+    /* Delete mds xattr at the time of STACK UNWIND */
+    GF_REMOVE_INTERNAL_XATTR(conf->mds_xattr_key, local->xattr);
 
-        DHT_STACK_UNWIND (lookup, frame, ret, local->op_errno, local->inode,
-                          &local->stbuf, local->xattr, &local->postparent);
+    DHT_STACK_UNWIND(lookup, frame, ret, local->op_errno, local->inode,
+                     &local->stbuf, local->xattr, &local->postparent);
 
 out:
-        return ret;
+    return ret;
 }
 
-
-int
-dht_discover_complete (xlator_t *this, call_frame_t *discover_frame)
+static int
+dht_discover_complete(xlator_t *this, call_frame_t *discover_frame)
 {
-        dht_local_t     *local = NULL;
-        call_frame_t    *main_frame = NULL;
-        int              op_errno = 0;
-        int              ret = -1;
-        dht_layout_t    *layout = NULL;
-        dht_conf_t      *conf = NULL;
+    dht_local_t *local = NULL;
+    dht_local_t *heal_local = NULL;
+    call_frame_t *main_frame = NULL;
+    call_frame_t *heal_frame = NULL;
+    int op_errno = 0;
+    int ret = -1;
+    dht_layout_t *layout = NULL;
+    dht_conf_t *conf = NULL;
+    uint32_t vol_commit_hash = 0;
+    xlator_t *source = NULL;
+    int heal_path = 0;
+    int error_while_marking_mds = 0;
+    int i = 0;
+    loc_t loc = {0};
+    int8_t is_read_only = 0, layout_anomalies = 0;
+    char gfid_local[GF_UUID_BUF_SIZE] = {0};
+
+    local = discover_frame->local;
+    layout = local->layout;
+    conf = this->private;
+    gf_uuid_unparse(local->gfid, gfid_local);
+
+    LOCK(&discover_frame->lock);
+    {
+        main_frame = local->main_frame;
+        local->main_frame = NULL;
+    }
+    UNLOCK(&discover_frame->lock);
 
-        local = discover_frame->local;
-        layout = local->layout;
-        conf = this->private;
+    if (!main_frame)
+        return 0;
 
-        LOCK(&discover_frame->lock);
-        {
-                main_frame = local->main_frame;
-                local->main_frame = NULL;
+    /* Code to update all extended attributed from
+       subvol to local->xattr on that internal xattr has found
+    */
+    if (conf->subvolume_cnt == 1)
+        local->need_xattr_heal = 0;
+    if (local->need_xattr_heal && (local->mds_xattr)) {
+        dht_dir_set_heal_xattr(this, local, local->xattr, local->mds_xattr,
+                               NULL, NULL);
+        dict_unref(local->mds_xattr);
+        local->mds_xattr = NULL;
+    }
+
+    ret = dict_get_int8(local->xattr_req, QUOTA_READ_ONLY_KEY, &is_read_only);
+    if (ret < 0)
+        gf_msg_debug(this->name, 0, "key = %s not present in dict",
+                     QUOTA_READ_ONLY_KEY);
+
+    if (local->file_count && local->dir_count) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_FILE_TYPE_MISMATCH,
+               "path %s exists as a file on one subvolume "
+               "and directory on another. "
+               "Please fix it manually",
+               local->loc.path);
+        op_errno = EIO;
+        goto out;
+    }
+
+    if (local->cached_subvol) {
+        ret = dht_layout_preset(this, local->cached_subvol, local->inode);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LAYOUT_SET_FAILED,
+                   "failed to set layout for subvolume %s",
+                   local->cached_subvol ? local->cached_subvol->name : "<nil>");
+            op_errno = EINVAL;
+            goto out;
+        }
+    } else {
+        ret = dht_layout_normalize(this, &local->loc, layout);
+        if ((ret < 0) || ((ret > 0) && (local->op_ret != 0))) {
+            /* either the layout is incorrect or the directory is
+             * not found even in one subvolume.
+             */
+            gf_msg_debug(this->name, 0,
+                         "normalizing failed on %s "
+                         "(overlaps/holes present: %s, "
+                         "ENOENT errors: %d)",
+                         local->loc.path, (ret < 0) ? "yes" : "no",
+                         (ret > 0) ? ret : 0);
+            layout_anomalies = 1;
+        } else if (local->inode) {
+            dht_layout_set(this, local->inode, layout);
+        }
+    }
+
+    if (!conf->vch_forced) {
+        ret = dict_get_uint32(local->xattr, conf->commithash_xattr_name,
+                              &vol_commit_hash);
+        if (ret == 0) {
+            conf->vol_commit_hash = vol_commit_hash;
         }
-        UNLOCK(&discover_frame->lock);
+    }
 
-        if (!main_frame)
+    if (IA_ISDIR(local->stbuf.ia_type) && !is_read_only) {
+        for (i = 0; i < layout->cnt; i++) {
+            if (!source && !layout->list[i].err)
+                source = layout->list[i].xlator;
+            if (layout->list[i].err == ENOENT ||
+                layout->list[i].err == ESTALE) {
+                heal_path = 1;
+            }
+
+            if (source && heal_path)
+                break;
+        }
+    }
+
+    if (IA_ISDIR(local->stbuf.ia_type)) {
+        /* Call function to save hashed subvol on inode ctx if
+           internal mds xattr is not present and all subvols are up
+        */
+        if (!local->op_ret && !__is_root_gfid(local->stbuf.ia_gfid))
+            (void)dht_common_mark_mdsxattr(discover_frame,
+                                           &error_while_marking_mds, 1);
+
+        if (local->need_xattr_heal && !heal_path) {
+            local->need_xattr_heal = 0;
+            ret = dht_dir_xattr_heal(this, local, &op_errno);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                       DHT_MSG_DIR_XATTR_HEAL_FAILED,
+                       "xattr heal failed for "
+                       "directory  gfid is %s ",
+                       gfid_local);
+            }
+        }
+    }
+
+    if (source && (heal_path || layout_anomalies || error_while_marking_mds)) {
+        gf_uuid_copy(loc.gfid, local->gfid);
+        if (gf_uuid_is_null(loc.gfid)) {
+            goto done;
+        }
+
+        if (local->inode)
+            loc.inode = inode_ref(local->inode);
+        else
+            goto done;
+
+        heal_frame = create_frame(this, this->ctx->pool);
+        if (heal_frame) {
+            heal_local = dht_local_init(heal_frame, &loc, NULL, 0);
+            if (!heal_local)
+                goto cleanup;
+
+            gf_uuid_copy(heal_local->gfid, local->gfid);
+            heal_frame->cookie = source;
+            heal_local->xattr = dict_ref(local->xattr);
+            heal_local->stbuf = local->stbuf;
+            heal_local->postparent = local->postparent;
+            heal_local->inode = inode_ref(loc.inode);
+            heal_local->main_frame = main_frame;
+            FRAME_SU_DO(heal_frame, dht_local_t);
+            ret = synctask_new(this->ctx->env, dht_heal_full_path,
+                               dht_heal_full_path_done, heal_frame, heal_frame);
+            if (!ret) {
+                loc_wipe(&loc);
                 return 0;
+            }
+            /*
+             * Failed to spawn the synctask. Returning
+             * with out doing heal.
+             */
+        cleanup:
+            loc_wipe(&loc);
+            DHT_STACK_DESTROY(heal_frame);
+        }
+    }
+done:
+    dht_set_fixed_dir_stat(&local->postparent);
+    /* Delete mds xattr at the time of STACK UNWIND */
+    if (local->xattr)
+        GF_REMOVE_INTERNAL_XATTR(conf->mds_xattr_key, local->xattr);
 
-        if (local->file_count && local->dir_count) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "path %s exists as a file on one subvolume "
-                        "and directory on another. "
-                        "Please fix it manually",
-                        local->loc.path);
-                op_errno = EIO;
-                goto out;
-        }
+    DHT_STACK_UNWIND(lookup, main_frame, local->op_ret, local->op_errno,
+                     local->inode, &local->stbuf, local->xattr,
+                     &local->postparent);
+    return 0;
 
-        if (local->cached_subvol) {
-                ret = dht_layout_preset (this, local->cached_subvol,
-                                         local->inode);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "failed to set layout for subvolume %s",
-                                local->cached_subvol ? local->cached_subvol->name : "<nil>");
-                        op_errno = EINVAL;
-                        goto out;
-                }
-        } else {
-                ret = dht_layout_normalize (this, &local->loc, layout);
-                if ((ret < 0) || ((ret > 0) && (local->op_ret != 0))) {
-                        /* either the layout is incorrect or the directory is
-                         * not found even in one subvolume.
-                         */
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "normalizing failed on %s "
-                                "(overlaps/holes present: %s, "
-                                "ENOENT errors: %d)", local->loc.path,
-                                (ret < 0) ? "yes" : "no", (ret > 0) ? ret : 0);
-                        if ((ret > 0) && (ret == conf->subvolume_cnt)) {
-                                op_errno = ESTALE;
-                                goto out;
-                        }
-                }
+out:
+    DHT_STACK_UNWIND(lookup, main_frame, -1, op_errno, NULL, NULL, NULL, NULL);
 
-                if (local->inode)
-                        dht_layout_set (this, local->inode, layout);
-        }
+    return ret;
+}
 
-        DHT_STACK_UNWIND (lookup, main_frame, local->op_ret, local->op_errno,
-                          local->inode, &local->stbuf, local->xattr,
-                          &local->postparent);
-        return 0;
+static int
+dht_common_mark_mdsxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                             int op_ret, int op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *prev = cookie;
+    int ret = -1;
+    dht_conf_t *conf = 0;
+    dht_layout_t *layout = NULL;
+    int32_t mds_heal_fresh_lookup = 0;
+
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+
+    local = frame->local;
+    conf = this->private;
+    layout = local->selfheal.layout;
+    mds_heal_fresh_lookup = local->mds_heal_fresh_lookup;
+
+    if (op_ret) {
+        gf_msg_debug(this->name, op_ret,
+                     "Failed to set %s on the MDS %s for path %s. ",
+                     conf->mds_xattr_key, prev->name, local->loc.path);
+    } else {
+        /* Save mds subvol on inode ctx */
+        ret = dht_inode_ctx_mdsvol_set(local->inode, this, prev);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SET_INODE_CTX_FAILED,
+                   "Failed to set mds subvol on inode ctx"
+                   " %s for %s ",
+                   prev->name, local->loc.path);
+        }
+    }
+    if (!local->mds_heal_fresh_lookup && layout) {
+        dht_selfheal_dir_setattr(frame, &local->loc, &local->stbuf, 0xffffffff,
+                                 layout);
+    }
 out:
-        DHT_STACK_UNWIND (lookup, main_frame, -1, op_errno, NULL, NULL, NULL,
-                          NULL);
-
-        return ret;
+    if (mds_heal_fresh_lookup)
+        DHT_STACK_DESTROY(frame);
+    return 0;
 }
 
+static xlator_t *
+dht_inode_get_hashed_subvol(inode_t *inode, xlator_t *this, loc_t *loc)
+{
+    char *path = NULL;
+    loc_t populate_loc = {
+        0,
+    };
+    char *name = NULL;
+    xlator_t *hash_subvol = NULL;
+
+    if (!inode)
+        return hash_subvol;
+
+    if (loc && loc->parent && loc->path) {
+        if (!loc->name) {
+            name = strrchr(loc->path, '/');
+            if (name) {
+                loc->name = name + 1;
+            } else {
+                goto out;
+            }
+        }
+        hash_subvol = dht_subvol_get_hashed(this, loc);
+        goto out;
+    }
 
-int
-dht_discover_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int op_ret, int op_errno,
-                  inode_t *inode, struct iatt *stbuf, dict_t *xattr,
-                  struct iatt *postparent)
-{
-        dht_local_t  *local                   = NULL;
-        int           this_call_cnt           = 0;
-        call_frame_t *prev                    = NULL;
-        dht_layout_t *layout                  = NULL;
-        int           ret                     = -1;
-        int           is_dir                  = 0;
-        int           is_linkfile             = 0;
-        int           attempt_unwind          = 0;
-        dht_conf_t   *conf                    = 0;
-
-        GF_VALIDATE_OR_GOTO ("dht", frame, out);
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
-        GF_VALIDATE_OR_GOTO ("dht", this->private, out);
-        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
+    if (!gf_uuid_is_null(inode->gfid)) {
+        populate_loc.inode = inode_ref(inode);
+        populate_loc.parent = inode_parent(populate_loc.inode, NULL, NULL);
+        inode_path(populate_loc.inode, NULL, &path);
 
-        local = frame->local;
-        prev  = cookie;
-        conf = this->private;
+        if (!path)
+            goto out;
 
-        layout = local->layout;
+        populate_loc.path = path;
+        if (!populate_loc.name && populate_loc.path) {
+            name = strrchr(populate_loc.path, '/');
+            if (name) {
+                populate_loc.name = name + 1;
 
-        /* Check if the gfid is different for file from other node */
-        if (!op_ret && uuid_compare (local->gfid, stbuf->ia_gfid)) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: gfid different on %s",
-                        local->loc.path, prev->this->name);
+            } else {
+                goto out;
+            }
         }
+        hash_subvol = dht_subvol_get_hashed(this, &populate_loc);
+    }
+out:
+    if (populate_loc.inode)
+        loc_wipe(&populate_loc);
+    return hash_subvol;
+}
 
+/* Common function call by revalidate/selfheal code path to populate
+   internal xattr if it is not present, mark_during_fresh_lookup value
+   determines either function is call by revalidate_cbk(discover_complete)
+   or call by selfheal code path while fresh lookup.
+   Here we do wind a call serially in case of fresh lookup and
+   for other lookup code path we do wind a call parallel.The reason
+   to wind a call serially is at the time of fresh lookup directory is not
+   discovered and at the time of revalidate_lookup directory is
+   already discovered. So, revalidate codepath can race with setxattr
+   codepath and can get into spurious heals because of an ongoing setxattr.
+   This can slow down revalidates, if healing happens in foreground.
+   However, if healing happens in background, there is no direct performance
+   penalty.
+*/
+int
+dht_common_mark_mdsxattr(call_frame_t *frame, int *errst,
+                         int mark_during_fresh_lookup)
+{
+    dht_local_t *local = NULL;
+    xlator_t *this = NULL;
+    xlator_t *hashed_subvol = NULL;
+    int ret = 0;
+    int i = 0;
+    dict_t *xattrs = NULL;
+    char gfid_local[GF_UUID_BUF_SIZE] = {
+        0,
+    };
+    int32_t zero[1] = {0};
+    dht_conf_t *conf = 0;
+    dht_layout_t *layout = NULL;
+    dht_local_t *copy_local = NULL;
+    call_frame_t *xattr_frame = NULL;
+    gf_boolean_t vol_down = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    this = frame->this;
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    local = frame->local;
+    conf = this->private;
+    layout = local->selfheal.layout;
+    local->mds_heal_fresh_lookup = mark_during_fresh_lookup;
+
+    gf_uuid_unparse(local->gfid, gfid_local);
+
+    /* Code to update hashed subvol consider as a mds subvol
+       and wind a setxattr call on hashed subvol to update
+       internal xattr
+    */
+    if (!local->xattr || !dict_get(local->xattr, conf->mds_xattr_key)) {
+        /* It means no internal MDS xattr has been set yet
+         */
+        /* Check the status of all subvol are up while call
+           this function call by lookup code path
+        */
+        if (mark_during_fresh_lookup) {
+            for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (!conf->subvolume_status[i]) {
+                    vol_down = _gf_true;
+                    break;
+                }
+            }
+            if (vol_down) {
+                gf_msg_debug(this->name, 0,
+                             "subvol %s is down. Unable to "
+                             " save mds subvol on inode for "
+                             " path %s gfid is %s ",
+                             conf->subvolumes[i]->name, local->loc.path,
+                             gfid_local);
+                goto out;
+            }
+        }
 
-        LOCK (&frame->lock);
-        {
-                /* TODO: assert equal mode on stbuf->st_mode and
-                   local->stbuf->st_mode
+        /* Calculate hashed subvol based on inode and parent node
+         */
+        hashed_subvol = dht_inode_get_hashed_subvol(local->inode, this,
+                                                    &local->loc);
+        if (!hashed_subvol) {
+            gf_msg(this->name, GF_LOG_DEBUG, 0,
+                   DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+                   "Failed to get hashed subvol for path %s"
+                   "gfid is %s ",
+                   local->loc.path, gfid_local);
+            if (errst)
+                (*errst) = 1;
+            ret = -1;
+            goto out;
+        }
+        xattrs = dict_new();
+        if (!xattrs) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+                   "dict_new failed");
+            ret = -1;
+            goto out;
+        }
+        /* Add internal MDS xattr on disk for hashed subvol
+         */
+        ret = dht_dict_set_array(xattrs, conf->mds_xattr_key, zero, 1);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_WARNING, ENOMEM, DHT_MSG_DICT_SET_FAILED,
+                   "Failed to set dictionary"
+                   "  value:key = %s for "
+                   "path %s",
+                   conf->mds_xattr_key, local->loc.path);
+            ret = -1;
+            goto out;
+        }
+        /* Create a new frame to wind a call only while
+           this function call by revalidate_cbk code path
+           To wind a call parallel need to create a new frame
+        */
+        if (mark_during_fresh_lookup) {
+            xattr_frame = create_frame(this, this->ctx->pool);
+            if (!xattr_frame) {
+                ret = -1;
+                goto out;
+            }
+            copy_local = dht_local_init(xattr_frame, &(local->loc), NULL, 0);
+            if (!copy_local) {
+                ret = -1;
+                DHT_STACK_DESTROY(xattr_frame);
+                goto out;
+            }
+            copy_local->stbuf = local->stbuf;
+            copy_local->mds_heal_fresh_lookup = mark_during_fresh_lookup;
+            if (!copy_local->inode)
+                copy_local->inode = inode_ref(local->inode);
+            gf_uuid_copy(copy_local->loc.gfid, local->gfid);
+            FRAME_SU_DO(xattr_frame, dht_local_t);
+            STACK_WIND_COOKIE(xattr_frame, dht_common_mark_mdsxattr_cbk,
+                              hashed_subvol, hashed_subvol,
+                              hashed_subvol->fops->setxattr, &local->loc,
+                              xattrs, 0, NULL);
+        } else {
+            STACK_WIND_COOKIE(frame, dht_common_mark_mdsxattr_cbk,
+                              (void *)hashed_subvol, hashed_subvol,
+                              hashed_subvol->fops->setxattr, &local->loc,
+                              xattrs, 0, NULL);
+        }
+    } else {
+        gf_msg_debug(this->name, 0,
+                     "internal xattr %s is present on subvol"
+                     "on path %s gfid is %s ",
+                     conf->mds_xattr_key, local->loc.path, gfid_local);
+        if (!mark_during_fresh_lookup)
+            dht_selfheal_dir_setattr(frame, &local->loc, &local->stbuf,
+                                     0xffffffff, layout);
+    }
 
-                   else mkdir/chmod/chown and fix
-                */
-                ret = dht_layout_merge (this, layout, prev->this,
-                                        op_ret, op_errno, xattr);
-                if (ret)
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "%s: failed to merge layouts", local->loc.path);
+out:
+    if (xattrs)
+        dict_unref(xattrs);
+    return ret;
+}
 
-                if (op_ret == -1) {
-                        local->op_errno = op_errno;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "lookup of %s on %s returned error (%s)",
-                                local->loc.path, prev->this->name,
-                                strerror (op_errno));
+/* Get the value of key from dict in the bytewise and save in array after
+   convert from network byte order to host byte order
+*/
+static int32_t
+dht_dict_get_array(dict_t *dict, char *key, int32_t value[], int32_t size,
+                   int *errst)
+{
+    void *ptr = NULL;
+    int32_t len = -1;
+    int32_t vindex = -1;
+    int32_t err = -1;
+    int ret = 0;
+
+    if (dict == NULL) {
+        (*errst) = -1;
+        return -EINVAL;
+    }
+    err = dict_get_ptr_and_len(dict, key, &ptr, &len);
+    if (err != 0) {
+        (*errst) = -1;
+        return err;
+    }
+
+    if (len != (size * sizeof(int32_t))) {
+        (*errst) = -1;
+        return -EINVAL;
+    }
+
+    for (vindex = 0; vindex < size; vindex++) {
+        value[vindex] = ntoh32(*((int32_t *)ptr + vindex));
+        if (value[vindex] < 0)
+            ret = -1;
+    }
+
+    return ret;
+}
 
-                        goto unlock;
-                }
+static int
+dht_discover_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, inode_t *inode, struct iatt *stbuf,
+                 dict_t *xattr, struct iatt *postparent)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+    dht_layout_t *layout = NULL;
+    int ret = -1;
+    int is_dir = 0;
+    int32_t check_mds = 0;
+    int is_linkfile = 0;
+    int attempt_unwind = 0;
+    dht_conf_t *conf = 0;
+    char gfid_local[GF_UUID_BUF_SIZE] = {0};
+    char gfid_node[GF_UUID_BUF_SIZE] = {0};
+    int32_t mds_xattr_val[1] = {0};
+    int errst = 0;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", this->private, out);
+    GF_VALIDATE_OR_GOTO("dht", cookie, out);
+
+    local = frame->local;
+    prev = cookie;
+    conf = this->private;
+
+    layout = local->layout;
+
+    /* Check if the gfid is different for file from other node */
+    if (!op_ret && gf_uuid_compare(local->gfid, stbuf->ia_gfid)) {
+        gf_uuid_unparse(stbuf->ia_gfid, gfid_node);
+        gf_uuid_unparse(local->gfid, gfid_local);
+
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_GFID_MISMATCH,
+               "%s: gfid different on %s, gfid local = %s"
+               "gfid other = %s",
+               local->loc.path, prev->name, gfid_local, gfid_node);
+    }
+
+    LOCK(&frame->lock);
+    {
+        /* TODO: assert equal mode on stbuf->st_mode and
+           local->stbuf->st_mode
+
+           else mkdir/chmod/chown and fix
+        */
+
+        ret = dht_layout_merge(this, layout, prev, op_ret, op_errno, xattr);
+        if (ret)
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LAYOUT_MERGE_FAILED,
+                   "%s: failed to merge layouts for subvol %s", local->loc.path,
+                   prev->name);
 
-                is_linkfile = check_is_linkfile (inode, stbuf, xattr,
-                                                 conf->link_xattr_name);
-                is_dir = check_is_dir (inode, stbuf, xattr);
+        if (op_ret == -1) {
+            local->op_errno = op_errno;
+            gf_msg_debug(this->name, op_errno,
+                         "lookup of %s on %s returned error", local->loc.path,
+                         prev->name);
 
-                if (is_dir) {
-                        local->dir_count ++;
-                } else {
-                        local->file_count ++;
-
-                        if (!is_linkfile) {
-                                /* real file */
-                                local->cached_subvol = prev->this;
-                                attempt_unwind = 1;
-                        } else {
-                                goto unlock;
-                        }
-                }
+            goto unlock;
+        }
 
-                local->op_ret = 0;
+        is_linkfile = check_is_linkfile(inode, stbuf, xattr,
+                                        conf->link_xattr_name);
+        is_dir = check_is_dir(inode, stbuf, xattr);
 
-                if (local->xattr == NULL) {
-                        local->xattr = dict_ref (xattr);
-                } else {
-                        dht_aggregate_xattr (local->xattr, xattr);
-                }
+        if (is_dir) {
+            local->dir_count++;
+        } else {
+            local->file_count++;
+
+            if (!is_linkfile && !local->cached_subvol) {
+                /* real file */
+                /* Ok, we somehow managed to find a file on
+                 * more than one subvol. ignore this or we
+                 * will end up overwriting information while a
+                 * a thread is potentially unwinding from
+                 * dht_discover_complete
+                 */
+                local->cached_subvol = prev;
+                attempt_unwind = 1;
+            } else {
+                goto unlock;
+            }
+        }
 
-                if (local->inode == NULL)
-                        local->inode = inode_ref (inode);
+        local->op_ret = 0;
 
-                dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
-                dht_iatt_merge (this, &local->postparent, postparent,
-                                prev->this);
+        if (local->xattr == NULL) {
+            local->xattr = dict_ref(xattr);
+        } else {
+            /* Don't aggregate for files. See BZ#1484709 */
+            if (is_dir)
+                dht_aggregate_xattr(local->xattr, xattr);
         }
+
+        if (local->inode == NULL)
+            local->inode = inode_ref(inode);
+
+        dht_iatt_merge(this, &local->stbuf, stbuf);
+        dht_iatt_merge(this, &local->postparent, postparent);
+
+        if (!dict_get(xattr, conf->mds_xattr_key)) {
+            goto unlock;
+        } else {
+            gf_msg_debug(this->name, 0,
+                         "internal xattr %s is present on subvol"
+                         "on path %s gfid is %s ",
+                         conf->mds_xattr_key, local->loc.path, gfid_local);
+        }
+        check_mds = dht_dict_get_array(xattr, conf->mds_xattr_key,
+                                       mds_xattr_val, 1, &errst);
+        /* save mds subvol on inode ctx */
+        ret = dht_inode_ctx_mdsvol_set(local->inode, this, prev);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SET_INODE_CTX_FAILED,
+                   "Failed to set hashed subvol for %s vol is %s",
+                   local->loc.path, prev->name);
+        }
+
+        if ((check_mds < 0) && !errst) {
+            local->mds_xattr = dict_ref(xattr);
+            gf_msg_debug(this->name, 0,
+                         "Value of %s is not zero on mds subvol"
+                         "so xattr needs to be healed on non mds"
+                         " path is %s and vol name is %s "
+                         " gfid is %s",
+                         conf->mds_xattr_key, local->loc.path, prev->name,
+                         gfid_local);
+            local->need_xattr_heal = 1;
+            local->mds_subvol = prev;
+        }
+    }
 unlock:
-        UNLOCK (&frame->lock);
+    UNLOCK(&frame->lock);
 out:
-        this_call_cnt = dht_frame_return (frame);
+    /* Make sure, the thread executing dht_discover_complete is the one
+     * which calls STACK_DESTROY (frame). In the case of "attempt_unwind",
+     * this makes sure that the thread don't call dht_frame_return, till
+     * call to dht_discover_complete is done.
+     */
+    if (attempt_unwind) {
+        dht_discover_complete(this, frame);
+    }
 
-        if (is_last_call (this_call_cnt) || attempt_unwind) {
-                dht_discover_complete (this, frame);
-        }
+    this_call_cnt = dht_frame_return(frame);
 
-        if (is_last_call (this_call_cnt))
-                DHT_STACK_DESTROY (frame);
+    if (is_last_call(this_call_cnt) && !attempt_unwind) {
+        dht_discover_complete(this, frame);
+    }
 
-        return 0;
+    if (is_last_call(this_call_cnt))
+        DHT_STACK_DESTROY(frame);
+
+    return 0;
 }
 
+static int
+dht_set_file_xattr_req(xlator_t *this, loc_t *loc, dict_t *xattr_req)
+{
+    int ret = -EINVAL;
+    dht_conf_t *conf = NULL;
+
+    conf = this->private;
+    if (!conf) {
+        goto err;
+    }
+
+    if (!xattr_req) {
+        goto err;
+    }
+
+    /* Used to check whether this is a linkto file.
+     */
+    ret = dict_set_uint32(xattr_req, conf->link_xattr_name, 256);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, DHT_MSG_DICT_SET_FAILED,
+               "Failed to set dictionary value:key = %s for "
+               "path %s",
+               conf->link_xattr_name, loc->path);
+        goto err;
+    }
+
+    /* This is used to make sure we don't unlink linkto files
+     * which are the target of an ongoing file migration.
+     */
+    ret = dict_set_uint32(xattr_req, GLUSTERFS_OPEN_FD_COUNT, 4);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, DHT_MSG_DICT_SET_FAILED,
+               "Failed to set dictionary value:key = %s for "
+               "path %s",
+               GLUSTERFS_OPEN_FD_COUNT, loc->path);
+        goto err;
+    }
+
+    ret = 0;
+err:
+    return ret;
+}
 
-int
-dht_discover (call_frame_t *frame, xlator_t *this, loc_t *loc)
+/* This is a gfid based nameless lookup. Without a name, the hashed subvol
+ * cannot be calculated so a lookup is sent to all subvols.
+ */
+static int
+dht_do_discover(call_frame_t *frame, xlator_t *this, loc_t *loc)
 {
-        int          ret;
-        dht_local_t *local = NULL;
-        dht_conf_t  *conf = NULL;
-        int          call_cnt = 0;
-        int          op_errno = EINVAL;
-        int          i = 0;
-        call_frame_t *discover_frame = NULL;
+    int ret;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int call_cnt = 0;
+    int op_errno = EINVAL;
+    int i = 0;
+    call_frame_t *discover_frame = NULL;
+
+    conf = this->private;
+    local = frame->local;
+
+    /* As we do not know if this is a file or directory, request
+     * both file and directory xattrs
+     */
+    ret = dht_set_file_xattr_req(this, loc, local->xattr_req);
+    if (ret) {
+        goto err;
+    }
+
+    ret = dht_set_dir_xattr_req(this, loc, local->xattr_req);
+    if (ret) {
+        goto err;
+    }
+
+    if (loc_is_root(loc)) {
+        /* Request the DHT commit hash xattr (trusted.glusterfs.dht.commithash)
+         * set on the brick root.
+         */
+        ret = dict_set_uint32(local->xattr_req, conf->commithash_xattr_name,
+                              sizeof(uint32_t));
+    }
 
-        conf = this->private;
-        local = frame->local;
+    call_cnt = conf->subvolume_cnt;
+    local->call_cnt = call_cnt;
 
-        ret = dict_set_uint32 (local->xattr_req, conf->xattr_name, 4 * 4);
-        if (ret)
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to set '%s' key",
-                        loc->path, conf->xattr_name);
+    local->layout = dht_layout_new(this, conf->subvolume_cnt);
 
-        ret = dict_set_uint32 (local->xattr_req, conf->link_xattr_name, 256);
-        if (ret)
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to set '%s' key",
-                        loc->path, conf->link_xattr_name);
+    if (!local->layout) {
+        op_errno = ENOMEM;
+        goto err;
+    }
 
-        call_cnt        = conf->subvolume_cnt;
-        local->call_cnt = call_cnt;
+    gf_uuid_copy(local->gfid, loc->gfid);
 
-        local->layout = dht_layout_new (this, conf->subvolume_cnt);
+    discover_frame = copy_frame(frame);
+    if (!discover_frame) {
+        op_errno = ENOMEM;
+        goto err;
+    }
 
-        if (!local->layout) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+    discover_frame->local = local;
+    frame->local = NULL;
+    local->main_frame = frame;
 
-        uuid_copy (local->gfid, loc->gfid);
+    for (i = 0; i < call_cnt; i++) {
+        STACK_WIND_COOKIE(discover_frame, dht_discover_cbk, conf->subvolumes[i],
+                          conf->subvolumes[i],
+                          conf->subvolumes[i]->fops->lookup, &local->loc,
+                          local->xattr_req);
+    }
 
-        discover_frame = copy_frame (frame);
-        if (!discover_frame) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+    return 0;
 
-        discover_frame->local = local;
-        frame->local = NULL;
-        local->main_frame = frame;
+err:
+    DHT_STACK_UNWIND(lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
 
-        for (i = 0; i < call_cnt; i++) {
-                STACK_WIND (discover_frame, dht_discover_cbk,
-                            conf->subvolumes[i],
-                            conf->subvolumes[i]->fops->lookup,
-                            &local->loc, local->xattr_req);
-        }
+    return 0;
+}
 
-        return 0;
+/* Code to call syntask to heal custom xattr from hashed subvol
+   to non hashed subvol
+*/
+int
+dht_dir_xattr_heal(xlator_t *this, dht_local_t *local, int *op_errno)
+{
+    dht_local_t *copy_local = NULL;
+    call_frame_t *copy = NULL;
+    int ret = -1;
+    char gfid_local[GF_UUID_BUF_SIZE] = {0};
+
+    if (gf_uuid_is_null(local->gfid)) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DIR_XATTR_HEAL_FAILED,
+               "No gfid exists for path %s "
+               "so healing xattr is not possible",
+               local->loc.path);
+        *op_errno = EIO;
+        goto out;
+    }
+
+    gf_uuid_unparse(local->gfid, gfid_local);
+    copy = create_frame(this, this->ctx->pool);
+    if (copy) {
+        copy_local = dht_local_init(copy, &(local->loc), NULL, 0);
+        if (!copy_local) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM,
+                   DHT_MSG_DIR_XATTR_HEAL_FAILED,
+                   "Memory allocation failed "
+                   "for path %s gfid %s ",
+                   local->loc.path, gfid_local);
+            *op_errno = ENOMEM;
+            DHT_STACK_DESTROY(copy);
+        } else {
+            copy_local->stbuf = local->stbuf;
+            gf_uuid_copy(copy_local->loc.gfid, local->gfid);
+            copy_local->mds_subvol = local->mds_subvol;
+            FRAME_SU_DO(copy, dht_local_t);
+            ret = synctask_new(this->ctx->env, dht_dir_heal_xattrs,
+                               dht_dir_heal_xattrs_done, copy, copy);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, ENOMEM,
+                       DHT_MSG_DIR_XATTR_HEAL_FAILED,
+                       "Synctask creation failed to heal xattr "
+                       "for path %s gfid %s ",
+                       local->loc.path, gfid_local);
+                *op_errno = ENOMEM;
+                DHT_STACK_DESTROY(copy);
+            }
+        }
+    }
+out:
+    return ret;
+}
 
-err:
-        DHT_STACK_UNWIND (lookup, frame, -1, op_errno, NULL, NULL, NULL,
-                          NULL);
+static int
+dht_needs_selfheal(call_frame_t *frame, xlator_t *this)
+{
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
+    int needs_selfheal = 0;
+    int ret = 0;
+
+    local = frame->local;
+    layout = local->layout;
+
+    if (local->need_attrheal || local->need_xattr_heal ||
+        local->need_selfheal) {
+        needs_selfheal = 1;
+    }
+
+    ret = dht_layout_normalize(this, &local->loc, layout);
+
+    if (ret != 0) {
+        gf_msg_debug(this->name, 0, "fixing assignment on %s", local->loc.path);
+        needs_selfheal = 1;
+    }
+    return needs_selfheal;
+}
 
+static int
+is_permission_different(ia_prot_t *prot1, ia_prot_t *prot2)
+{
+    if ((prot1->owner.read != prot2->owner.read) ||
+        (prot1->owner.write != prot2->owner.write) ||
+        (prot1->owner.exec != prot2->owner.exec) ||
+        (prot1->group.read != prot2->group.read) ||
+        (prot1->group.write != prot2->group.write) ||
+        (prot1->group.exec != prot2->group.exec) ||
+        (prot1->other.read != prot2->other.read) ||
+        (prot1->other.write != prot2->other.write) ||
+        (prot1->other.exec != prot2->other.exec) ||
+        (prot1->suid != prot2->suid) || (prot1->sgid != prot2->sgid) ||
+        (prot1->sticky != prot2->sticky)) {
+        return 1;
+    } else {
         return 0;
+    }
 }
 
-
 int
-dht_lookup_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int op_ret, int op_errno,
-                    inode_t *inode, struct iatt *stbuf, dict_t *xattr,
-                    struct iatt *postparent)
-{
-        dht_local_t  *local                   = NULL;
-        int           this_call_cnt           = 0;
-        call_frame_t *prev                    = NULL;
-        dht_layout_t *layout                  = NULL;
-        int           ret                     = -1;
-        int           is_dir                  = 0;
-
-        GF_VALIDATE_OR_GOTO ("dht", frame, out);
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
-        GF_VALIDATE_OR_GOTO ("dht", this->private, out);
-        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
+dht_lookup_dir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int op_ret, int op_errno, inode_t *inode, struct iatt *stbuf,
+                   dict_t *xattr, struct iatt *postparent)
+{
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+    dht_layout_t *layout = NULL;
+    int ret = -1;
+    int is_dir = 0;
+    int32_t check_mds = 0;
+    int errst = 0;
+    char gfid_local[GF_UUID_BUF_SIZE] = {0};
+    char gfid_node[GF_UUID_BUF_SIZE] = {0};
+    int32_t mds_xattr_val[1] = {0};
+
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", this->private, out);
+    GF_VALIDATE_OR_GOTO("dht", cookie, out);
+
+    local = frame->local;
+    prev = cookie;
+    conf = this->private;
+
+    layout = local->layout;
+    gf_msg_debug(this->name, op_errno,
+                 "%s: lookup on %s returned with op_ret = %d, op_errno = %d",
+                 local->loc.path, prev->name, op_ret, op_errno);
+
+    /* The first successful lookup*/
+    if (!op_ret && gf_uuid_is_null(local->gfid)) {
+        memcpy(local->gfid, stbuf->ia_gfid, 16);
+    }
+    if (!gf_uuid_is_null(local->gfid)) {
+        gf_uuid_unparse(local->gfid, gfid_local);
+    }
+
+    /* Check if the gfid is different for file from other node */
+    if (!op_ret && gf_uuid_compare(local->gfid, stbuf->ia_gfid)) {
+        gf_uuid_unparse(stbuf->ia_gfid, gfid_node);
+
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_GFID_MISMATCH,
+               "%s: gfid different on %s."
+               " gfid local = %s, gfid subvol = %s",
+               local->loc.path, prev->name, gfid_local, gfid_node);
+    }
+
+    LOCK(&frame->lock);
+    {
+        /* TODO: assert equal mode on stbuf->st_mode and
+           local->stbuf->st_mode
+           else mkdir/chmod/chown and fix
+        */
+        ret = dht_layout_merge(this, layout, prev, op_ret, op_errno, xattr);
 
-        local = frame->local;
-        prev  = cookie;
+        if (op_ret == -1) {
+            local->op_errno = op_errno;
 
-        layout = local->layout;
+            /* The GFID is missing on this subvol. Force a heal. */
+            if (op_errno == ENODATA) {
+                local->need_lookup_everywhere = 1;
+            }
+            goto unlock;
+        }
 
-        if (!op_ret && uuid_is_null (local->gfid))
-                memcpy (local->gfid, stbuf->ia_gfid, 16);
+        is_dir = check_is_dir(inode, stbuf, xattr);
+        if (!is_dir) {
+            gf_msg_debug(this->name, 0,
+                         "%s: lookup on %s returned non dir 0%o"
+                         "calling lookup_everywhere",
+                         local->loc.path, prev->name, stbuf->ia_type);
 
-        /* Check if the gfid is different for file from other node */
-        if (!op_ret && uuid_compare (local->gfid, stbuf->ia_gfid)) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: gfid different on %s",
-                        local->loc.path, prev->this->name);
+            local->need_lookup_everywhere = 1;
+            goto unlock;
         }
 
-        LOCK (&frame->lock);
-        {
-                /* TODO: assert equal mode on stbuf->st_mode and
-                   local->stbuf->st_mode
+        local->op_ret = 0;
+        if (local->xattr == NULL) {
+            local->xattr = dict_ref(xattr);
+        } else {
+            dht_aggregate_xattr(local->xattr, xattr);
+        }
+
+        if (__is_root_gfid(stbuf->ia_gfid)) {
+            ret = dht_dir_has_layout(xattr, conf->xattr_name);
+            if (ret >= 0) {
+                if (is_greater_time(local->prebuf.ia_ctime,
+                                    local->prebuf.ia_ctime_nsec,
+                                    stbuf->ia_ctime, stbuf->ia_ctime_nsec)) {
+                    /* Choose source */
+                    local->prebuf.ia_gid = stbuf->ia_gid;
+                    local->prebuf.ia_uid = stbuf->ia_uid;
+
+                    local->prebuf.ia_ctime = stbuf->ia_ctime;
+                    local->prebuf.ia_ctime_nsec = stbuf->ia_ctime_nsec;
+                    local->prebuf.ia_prot = stbuf->ia_prot;
+                }
+            }
+        }
 
-                   else mkdir/chmod/chown and fix
-                */
-                ret = dht_layout_merge (this, layout, prev->this,
-                                        op_ret, op_errno, xattr);
+        if (local->stbuf.ia_type != IA_INVAL) {
+            /* This is not the first subvol to respond
+             * Compare values to see if attrs need to be healed
+             */
+            if ((local->stbuf.ia_gid != stbuf->ia_gid) ||
+                (local->stbuf.ia_uid != stbuf->ia_uid) ||
+                (is_permission_different(&local->stbuf.ia_prot,
+                                         &stbuf->ia_prot))) {
+                local->need_attrheal = 1;
+            }
+        }
 
-                if (op_ret == -1) {
-                        local->op_errno = ENOENT;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "lookup of %s on %s returned error (%s)",
-                                local->loc.path, prev->this->name,
-                                strerror (op_errno));
+        if (local->inode == NULL)
+            local->inode = inode_ref(inode);
 
-                        goto unlock;
-                }
+        dht_iatt_merge(this, &local->stbuf, stbuf);
+        dht_iatt_merge(this, &local->postparent, postparent);
 
-                is_dir = check_is_dir (inode, stbuf, xattr);
-                if (!is_dir) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "lookup of %s on %s returned non dir 0%o",
-                                local->loc.path, prev->this->name,
-                                stbuf->ia_type);
-                        local->need_selfheal = 1;
-                        goto unlock;
-                }
+        if (!dict_get(xattr, conf->mds_xattr_key)) {
+            gf_msg_debug(this->name, 0,
+                         "%s: mds xattr %s is not present "
+                         "on %s(gfid = %s)",
+                         local->loc.path, conf->mds_xattr_key, prev->name,
+                         gfid_local);
+            goto unlock;
+        }
 
-                local->op_ret = 0;
-                if (local->xattr == NULL) {
-                        local->xattr = dict_ref (xattr);
-                } else {
-                        dht_aggregate_xattr (local->xattr, xattr);
-                }
+        /* Save the mds subvol info and stbuf. This is the value that will
+         * be used for healing
+         */
+        local->mds_subvol = prev;
+        local->mds_stbuf = *stbuf;
 
-                if (local->inode == NULL)
-                        local->inode = inode_ref (inode);
+        /* Save mds subvol on inode ctx */
+
+        ret = dht_inode_ctx_mdsvol_set(local->inode, this, prev);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SET_INODE_CTX_FAILED,
+                   "%s: Failed to set mds (%s)", local->loc.path, prev->name);
+        }
+        check_mds = dht_dict_get_array(xattr, conf->mds_xattr_key,
+                                       mds_xattr_val, 1, &errst);
+        if ((check_mds < 0) && !errst) {
+            /* Check if xattrs need to be healed on the directories */
+            local->mds_xattr = dict_ref(xattr);
+            gf_msg_debug(this->name, 0,
+                         "%s: %s is not zero on %s. Xattrs need to be healed."
+                         "(gfid = %s)",
+                         local->loc.path, conf->mds_xattr_key, prev->name,
+                         gfid_local);
+            local->need_xattr_heal = 1;
+        }
+    }
 
-                dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
-                dht_iatt_merge (this, &local->postparent, postparent,
-                                prev->this);
-        }
 unlock:
-        UNLOCK (&frame->lock);
+    UNLOCK(&frame->lock);
 
+    this_call_cnt = dht_frame_return(frame);
 
-        this_call_cnt = dht_frame_return (frame);
+    if (is_last_call(this_call_cnt)) {
+        /* If the mds subvol is not set correctly*/
+        if (!__is_root_gfid(local->gfid) &&
+            (!dict_get(local->xattr, conf->mds_xattr_key))) {
+            local->need_selfheal = 1;
+        }
 
-        if (is_last_call (this_call_cnt)) {
-                if (local->need_selfheal) {
-                        local->need_selfheal = 0;
-                        dht_lookup_everywhere (frame, this, &local->loc);
-                        return 0;
-                }
+        /* No need to call xattr heal code if volume count is 1
+         */
+        if (conf->subvolume_cnt == 1) {
+            local->need_xattr_heal = 0;
+        }
 
-                if (local->op_ret == 0) {
-                        ret = dht_layout_normalize (this, &local->loc, layout);
+        if (local->need_selfheal || local->need_lookup_everywhere) {
+            /* Set the gfid-req so posix will set the GFID*/
+            if (!gf_uuid_is_null(local->gfid)) {
+                /* Ok, this should _never_ happen */
+                ret = dict_set_static_bin(local->xattr_req, "gfid-req",
+                                          local->gfid, 16);
+            } else {
+                if (!gf_uuid_is_null(local->gfid_req))
+                    ret = dict_set_static_bin(local->xattr_req, "gfid-req",
+                                              local->gfid_req, 16);
+            }
+        }
 
-                        if (ret != 0) {
-                                gf_log (this->name, GF_LOG_DEBUG,
-                                        "fixing assignment on %s",
-                                        local->loc.path);
-                                goto selfheal;
-                        }
+        if (local->need_lookup_everywhere) {
+            local->need_lookup_everywhere = 0;
+            dht_lookup_everywhere(frame, this, &local->loc);
+            return 0;
+        }
 
-                        dht_layout_set (this, local->inode, layout);
-                }
+        if (local->op_ret == 0) {
+            if (dht_needs_selfheal(frame, this)) {
+                goto selfheal;
+            }
 
-                if (local->loc.parent) {
-                        dht_inode_ctx_time_update (local->loc.parent, this,
-                                                   &local->postparent, 1);
-                }
+            dht_layout_set(this, local->inode, layout);
+            if (local->inode) {
+                dht_inode_ctx_time_update(local->inode, this, &local->stbuf, 1);
+            }
 
-                DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
-                DHT_STACK_UNWIND (lookup, frame, local->op_ret, local->op_errno,
-                                  local->inode, &local->stbuf, local->xattr,
-                                  &local->postparent);
+            if (local->loc.parent) {
+                dht_inode_ctx_time_update(local->loc.parent, this,
+                                          &local->postparent, 1);
+            }
         }
 
-        return 0;
+        DHT_STRIP_PHASE1_FLAGS(&local->stbuf);
+        dht_set_fixed_dir_stat(&local->postparent);
+        /* Delete mds xattr at the time of STACK UNWIND */
+        if (local->xattr)
+            GF_REMOVE_INTERNAL_XATTR(conf->mds_xattr_key, local->xattr);
+
+        DHT_STACK_UNWIND(lookup, frame, local->op_ret, local->op_errno,
+                         local->inode, &local->stbuf, local->xattr,
+                         &local->postparent);
+    }
+
+    return 0;
 
 selfheal:
-        FRAME_SU_DO (frame, dht_local_t);
-        uuid_copy (local->loc.gfid, local->gfid);
-        ret = dht_selfheal_directory (frame, dht_lookup_selfheal_cbk,
-                                      &local->loc, layout);
+    FRAME_SU_DO(frame, dht_local_t);
+    ret = dht_selfheal_directory(frame, dht_lookup_selfheal_cbk, &local->loc,
+                                 layout);
 out:
-        return ret;
+    return ret;
+}
+
+static int
+dht_lookup_directory(call_frame_t *frame, xlator_t *this, loc_t *loc)
+{
+    int call_cnt = 0;
+    int i = 0;
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    int ret = 0;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    GF_VALIDATE_OR_GOTO("dht", this, unwind);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, unwind);
+    GF_VALIDATE_OR_GOTO("dht", this->private, unwind);
+    GF_VALIDATE_OR_GOTO("dht", loc, unwind);
+
+    conf = this->private;
+    local = frame->local;
+
+    call_cnt = conf->subvolume_cnt;
+    local->call_cnt = call_cnt;
+
+    local->layout = dht_layout_new(this, conf->subvolume_cnt);
+    if (!local->layout) {
+        goto unwind;
+    }
+
+    if (local->xattr != NULL) {
+        dict_unref(local->xattr);
+        local->xattr = NULL;
+    }
+
+    if (!gf_uuid_is_null(local->gfid)) {
+        /* use this gfid in order to heal any missing ones */
+        ret = dict_set_gfuuid(local->xattr_req, "gfid-req", local->gfid, true);
+        if (ret)
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+                   "%s: Failed to set dictionary value:"
+                   " key = gfid-req",
+                   local->loc.path);
+    }
+
+    for (i = 0; i < call_cnt; i++) {
+        STACK_WIND_COOKIE(
+            frame, dht_lookup_dir_cbk, conf->subvolumes[i], conf->subvolumes[i],
+            conf->subvolumes[i]->fops->lookup, &local->loc, local->xattr_req);
+    }
+    return 0;
+unwind:
+    DHT_STACK_UNWIND(lookup, frame, -1, ENOMEM, NULL, NULL, NULL, NULL);
+out:
+    return 0;
 }
 
 int
-dht_revalidate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int op_ret, int op_errno,
-                    inode_t *inode, struct iatt *stbuf, dict_t *xattr,
-                    struct iatt *postparent)
-{
-        dht_local_t  *local         = NULL;
-        int           this_call_cnt = 0;
-        call_frame_t *prev          = NULL;
-        dht_layout_t *layout        = NULL;
-        dht_conf_t   *conf          = NULL;
-        int           ret  = -1;
-        int           is_dir = 0;
-        int           is_linkfile = 0;
-        call_frame_t *copy          = NULL;
-        dht_local_t  *copy_local    = NULL;
-
-        GF_VALIDATE_OR_GOTO ("dht", frame, err);
-        GF_VALIDATE_OR_GOTO ("dht", this, err);
-        GF_VALIDATE_OR_GOTO ("dht", frame->local, err);
-        GF_VALIDATE_OR_GOTO ("dht", cookie, err);
+dht_revalidate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int op_ret, int op_errno, inode_t *inode, struct iatt *stbuf,
+                   dict_t *xattr, struct iatt *postparent)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+    dht_layout_t *layout = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
+    int is_dir = 0;
+    int is_linkfile = 0;
+    int follow_link = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    uint32_t vol_commit_hash = 0;
+    xlator_t *subvol = NULL;
+    int32_t check_mds = 0;
+    int errst = 0, i = 0;
+    int32_t mds_xattr_val[1] = {0};
+
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO("dht", this, err);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, err);
+    GF_VALIDATE_OR_GOTO("dht", cookie, err);
+    GF_VALIDATE_OR_GOTO("dht", this->private, err);
+
+    local = frame->local;
+    prev = cookie;
+    conf = this->private;
+
+    if (!conf->vch_forced) {
+        /* Update the commithash value if available
+         */
+        ret = dict_get_uint32(xattr, conf->commithash_xattr_name,
+                              &vol_commit_hash);
+        if (ret == 0) {
+            conf->vol_commit_hash = vol_commit_hash;
+        }
+    }
 
-        local = frame->local;
-        prev  = cookie;
-        conf = this->private;
-        if (!conf)
-                goto out;
+    gf_uuid_unparse(local->loc.gfid, gfid);
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        local->op_errno = op_errno;
-
-                        if ((op_errno != ENOTCONN)
-                            && (op_errno != ENOENT)
-                            && (op_errno != ESTALE)) {
-				gf_log (this->name, GF_LOG_INFO,
-					"subvolume %s for %s returned -1 (%s)",
-					prev->this->name, local->loc.path,
-                                        strerror (op_errno));
-			}
-                        if (op_errno == ESTALE) {
-                                /* propagate the ESTALE to parent.
-                                 * setting local->return_estale would send
-                                 * ESTALE to parent. */
-                                local->return_estale = 1;
-                        }
+    gf_msg_debug(this->name, op_errno,
+                 "%s: revalidate lookup on %s returned op_ret %d",
+                 local->loc.path, prev->name, op_ret);
+
+    LOCK(&frame->lock);
+    {
+        if (gf_uuid_is_null(local->gfid)) {
+            memcpy(local->gfid, local->loc.gfid, 16);
+        }
 
-                        /* if it is ENOENT, we may have to do a
-                         * 'lookup_everywhere()' to make sure
-                         * the file is not migrated */
-                        if (op_errno == ENOENT) {
-                                if (IA_ISREG (local->loc.inode->ia_type)) {
-                                        local->need_lookup_everywhere = 1;
-                                }
+        if (op_ret == -1) {
+            local->op_errno = op_errno;
+
+            if ((op_errno != ENOTCONN) && (op_errno != ENOENT) &&
+                (op_errno != ESTALE)) {
+                gf_msg(this->name, GF_LOG_INFO, op_errno,
+                       DHT_MSG_REVALIDATE_CBK_INFO,
+                       "Revalidate: subvolume %s for %s "
+                       "(gfid = %s) returned -1",
+                       prev->name, local->loc.path, gfid);
+            }
+            if (op_errno == ESTALE) {
+                /* propagate the ESTALE to parent.
+                 * setting local->return_estale would send
+                 * ESTALE to parent. */
+                local->return_estale = 1;
+            }
+
+            /* if it is ENOENT, we may have to do a
+             * 'lookup_everywhere()' to make sure
+             * the file is not migrated */
+            if (op_errno == ENOENT) {
+                if (IA_ISREG(local->loc.inode->ia_type)) {
+                    gf_msg_debug(this->name, 0,
+                                 "found ENOENT for %s. "
+                                 "Setting "
+                                 "need_lookup_everywhere"
+                                 " flag to 1",
+                                 local->loc.path);
+
+                    local->need_lookup_everywhere = 1;
+                } else if (IA_ISDIR(local->loc.inode->ia_type)) {
+                    layout = local->layout;
+                    for (i = 0; i < layout->cnt; i++) {
+                        if (layout->list[i].xlator == prev) {
+                            layout->list[i].err = op_errno;
+                            break;
                         }
-                        goto unlock;
+                    }
+
+                    local->need_selfheal = 1;
                 }
+            }
 
-                if (stbuf->ia_type != local->inode->ia_type) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "mismatching filetypes 0%o v/s 0%o for %s",
-                                (stbuf->ia_type), (local->inode->ia_type),
-                                local->loc.path);
+            /* The GFID is missing on this subvol. Lookup everywhere to force a
+             * gfid heal
+             */
+            if ((op_errno == ENODATA) &&
+                (IA_ISDIR(local->loc.inode->ia_type))) {
+                local->need_lookup_everywhere = 1;
+            }
 
-                        local->op_ret = -1;
-                        local->op_errno = EINVAL;
+            goto unlock;
+        }
 
-                        goto unlock;
-                }
+        if ((!IA_ISINVAL(local->inode->ia_type)) &&
+            stbuf->ia_type != local->inode->ia_type) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_FILE_TYPE_MISMATCH,
+                   "mismatching filetypes 0%o v/s 0%o for %s,"
+                   " gfid = %s",
+                   (stbuf->ia_type), (local->inode->ia_type), local->loc.path,
+                   gfid);
 
-                layout = local->layout;
+            local->op_ret = -1;
+            local->op_errno = EINVAL;
 
-                is_dir = check_is_dir (inode, stbuf, xattr);
-                is_linkfile = check_is_linkfile (inode, stbuf, xattr,
-                                                 conf->link_xattr_name);
+            goto unlock;
+        }
 
-                if (is_linkfile) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "linkfile found in revalidate for %s",
-                                local->loc.path);
-                        local->return_estale = 1;
+        layout = local->layout;
 
-                        goto unlock;
+        is_dir = check_is_dir(inode, stbuf, xattr);
+        is_linkfile = check_is_linkfile(inode, stbuf, xattr,
+                                        conf->link_xattr_name);
+        if (is_linkfile) {
+            follow_link = 1;
+            goto unlock;
+        }
+        if (is_dir) {
+            ret = dht_dir_has_layout(xattr, conf->xattr_name);
+            if (ret >= 0) {
+                if (is_greater_time(local->prebuf.ia_ctime,
+                                    local->prebuf.ia_ctime_nsec,
+                                    stbuf->ia_ctime, stbuf->ia_ctime_nsec)) {
+                    /* Choose source */
+                    local->prebuf.ia_gid = stbuf->ia_gid;
+                    local->prebuf.ia_uid = stbuf->ia_uid;
+
+                    local->prebuf.ia_ctime = stbuf->ia_ctime;
+                    local->prebuf.ia_ctime_nsec = stbuf->ia_ctime_nsec;
+
+                    if (__is_root_gfid(stbuf->ia_gfid))
+                        local->prebuf.ia_prot = stbuf->ia_prot;
                 }
+            }
+
+            if (local->stbuf.ia_type != IA_INVAL) {
+                if ((local->stbuf.ia_gid != stbuf->ia_gid) ||
+                    (local->stbuf.ia_uid != stbuf->ia_uid) ||
+                    is_permission_different(&local->stbuf.ia_prot,
+                                            &stbuf->ia_prot)) {
+                    local->need_attrheal = 1;
+                }
+            }
+
+            if (!dict_get(xattr, conf->mds_xattr_key)) {
+                gf_msg_debug(this->name, 0,
+                             "%s: internal xattr %s is not present"
+                             " on subvol %s(gfid is %s)",
+                             local->loc.path, conf->mds_xattr_key, prev->name,
+                             gfid);
+            } else {
+                check_mds = dht_dict_get_array(xattr, conf->mds_xattr_key,
+                                               mds_xattr_val, 1, &errst);
+                local->mds_subvol = prev;
+                local->mds_stbuf.ia_gid = stbuf->ia_gid;
+                local->mds_stbuf.ia_uid = stbuf->ia_uid;
+                local->mds_stbuf.ia_prot = stbuf->ia_prot;
+
+                /* save mds subvol on inode ctx */
+                ret = dht_inode_ctx_mdsvol_set(local->inode, this, prev);
+                if (ret) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0,
+                           DHT_MSG_SET_INODE_CTX_FAILED,
+                           "Failed to set MDS subvol for %s vol is %s",
+                           local->loc.path, prev->name);
+                }
+                if ((check_mds < 0) && !errst) {
+                    /* Check if xattrs need to be healed on the directory
+                     */
+                    local->mds_xattr = dict_ref(xattr);
+                    gf_msg_debug(this->name, 0,
+                                 "Value of %s is not zero on "
+                                 "hashed subvol so xattr needs to"
+                                 " be healed on non hashed"
+                                 " path is %s and vol name is %s "
+                                 " gfid is %s",
+                                 conf->mds_xattr_key, local->loc.path,
+                                 prev->name, gfid);
+                    local->need_xattr_heal = 1;
+                }
+            }
+            ret = dht_layout_dir_mismatch(this, layout, prev, &local->loc,
+                                          xattr);
+            if (ret != 0) {
+                /* In memory layout does not match on-disk layout.
+                 */
+                gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_LAYOUT_MISMATCH,
+                       "Mismatching layouts for %s, gfid = %s", local->loc.path,
+                       gfid);
 
-                if (is_dir) {
-                        ret = dht_dir_has_layout (xattr, conf->xattr_name);
-                        if (ret >= 0) {
-                                if (is_greater_time(local->stbuf.ia_ctime,
-                                                    local->stbuf.ia_ctime_nsec,
-                                                    stbuf->ia_ctime,
-                                                    stbuf->ia_ctime_nsec)) {
-                                        local->prebuf.ia_gid = stbuf->ia_gid;
-                                        local->prebuf.ia_uid = stbuf->ia_uid;
-                                }
-                        }
-                        if (local->stbuf.ia_type != IA_INVAL)
-                        {
-                                if ((local->stbuf.ia_gid != stbuf->ia_gid) ||
-                                    (local->stbuf.ia_uid != stbuf->ia_uid)) {
-                                        local->need_selfheal = 1;
-                                }
-                        }
-                        ret = dht_layout_dir_mismatch (this, layout,
-                                                       prev->this, &local->loc,
-                                                       xattr);
-                        if (ret != 0) {
-                                gf_log (this->name, GF_LOG_INFO,
-                                        "mismatching layouts for %s",
-                                        local->loc.path);
-
-                                local->layout_mismatch = 1;
+                local->layout_mismatch = 1;
 
-                                goto unlock;
-                        }
-                }
+                goto unlock;
+            }
+        }
 
-                dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
-                dht_iatt_merge (this, &local->postparent, postparent,
-                                prev->this);
+        gf_uuid_copy(local->stbuf.ia_gfid, stbuf->ia_gfid);
+        dht_iatt_merge(this, &local->stbuf, stbuf);
+        dht_iatt_merge(this, &local->postparent, postparent);
 
-                local->op_ret = 0;
+        local->op_ret = 0;
 
-                if (!local->xattr) {
-                        local->xattr = dict_ref (xattr);
-                } else if (is_dir) {
-                        dht_aggregate_xattr (local->xattr, xattr);
-                }
+        if (!local->xattr) {
+            local->xattr = dict_ref(xattr);
+        } else if (is_dir) {
+            dht_aggregate_xattr(local->xattr, xattr);
         }
+    }
 unlock:
-        UNLOCK (&frame->lock);
-out:
-        this_call_cnt = dht_frame_return (frame);
-
-        if (is_last_call (this_call_cnt)) {
-                if (!IA_ISDIR (local->stbuf.ia_type)
-                    && (local->hashed_subvol != local->cached_subvol)
-                    && (local->stbuf.ia_nlink == 1)
-                    && (conf && conf->unhashed_sticky_bit)) {
-                        local->stbuf.ia_prot.sticky = 1;
-                }
-                if (local->need_selfheal) {
-                        local->need_selfheal = 0;
-                        uuid_copy (local->gfid, local->stbuf.ia_gfid);
-                        local->stbuf.ia_gid = local->prebuf.ia_gid;
-                        local->stbuf.ia_uid = local->prebuf.ia_uid;
-                        copy = create_frame (this, this->ctx->pool);
-                        if (copy) {
-                                copy_local = dht_local_init (copy, &local->loc,
-                                                             NULL, 0);
-                                if (!copy_local)
-                                        goto cont;
-                                copy_local->stbuf = local->stbuf;
-                                copy->local = copy_local;
-                                FRAME_SU_DO (copy, dht_local_t);
-                                ret = synctask_new (this->ctx->env,
-                                                    dht_dir_attr_heal,
-                                                    dht_dir_attr_heal_done,
-                                                    copy, copy);
-                        }
-                }
-cont:
-		if (local->layout_mismatch) {
-                        /* Found layout mismatch in the directory, need to
-                           fix this in the inode context */
-                        dht_layout_unref (this, local->layout);
-                        local->layout = NULL;
-                        dht_lookup_directory (frame, this, &local->loc);
-                        return 0;
-                }
-
-                if (local->need_lookup_everywhere) {
-                        /* As the current layout gave ENOENT error, we would
-                           need a new layout */
-                        dht_layout_unref (this, local->layout);
-                        local->layout = NULL;
-
-                        /* We know that current cached subvol is no more
-                           valid, get the new one */
-                        local->cached_subvol = NULL;
-                        dht_lookup_everywhere (frame, this, &local->loc);
-                        return 0;
-                }
-                if (local->return_estale) {
-                        local->op_ret = -1;
-                        local->op_errno = ESTALE;
-                }
+    UNLOCK(&frame->lock);
 
-                if (local->loc.parent) {
-                        dht_inode_ctx_time_update (local->loc.parent, this,
-                                                   &local->postparent, 1);
-                }
+    if (follow_link) {
+        /* Found a linkto file. Follow it to see if the target file exists
+         */
+        gf_uuid_copy(local->gfid, stbuf->ia_gfid);
 
-                DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
-                DHT_STACK_UNWIND (lookup, frame, local->op_ret, local->op_errno,
-                                  local->inode, &local->stbuf, local->xattr,
-                                  &local->postparent);
+        subvol = dht_linkfile_subvol(this, inode, stbuf, xattr);
+        if (!subvol) {
+            op_errno = ESTALE;
+            local->op_ret = -1;
+        } else {
+            STACK_WIND_COOKIE(frame, dht_lookup_linkfile_cbk, subvol, subvol,
+                              subvol->fops->lookup, &local->loc,
+                              local->xattr_req);
+            return 0;
         }
+    }
 
-err:
-        return ret;
-}
+    this_call_cnt = dht_frame_return(frame);
 
+    if (is_last_call(this_call_cnt)) {
+        if (!IA_ISDIR(local->stbuf.ia_type) &&
+            (local->hashed_subvol != local->cached_subvol) &&
+            (local->stbuf.ia_nlink == 1) &&
+            (conf && conf->unhashed_sticky_bit)) {
+            local->stbuf.ia_prot.sticky = 1;
+        }
+        /* No need to call heal code if volume count is 1
+         */
+        if (conf->subvolume_cnt == 1)
+            local->need_xattr_heal = 0;
+
+        if (IA_ISDIR(local->stbuf.ia_type)) {
+            /* No mds xattr found. Trigger a heal to set it */
+            if (!__is_root_gfid(local->loc.inode->gfid) &&
+                (!dict_get(local->xattr, conf->mds_xattr_key)))
+                local->need_selfheal = 1;
+
+            if (dht_needs_selfheal(frame, this)) {
+                if (!__is_root_gfid(local->loc.inode->gfid)) {
+                    if (local->mds_subvol) {
+                        local->stbuf.ia_gid = local->mds_stbuf.ia_gid;
+                        local->stbuf.ia_uid = local->mds_stbuf.ia_uid;
+                        local->stbuf.ia_prot = local->mds_stbuf.ia_prot;
+                    }
+                } else {
+                    local->stbuf.ia_gid = local->prebuf.ia_gid;
+                    local->stbuf.ia_uid = local->prebuf.ia_uid;
+                    local->stbuf.ia_prot = local->prebuf.ia_prot;
+                }
 
-int
-dht_lookup_linkfile_create_cbk (call_frame_t *frame, void *cookie,
-                                xlator_t *this,
-                                int32_t op_ret, int32_t op_errno,
-                                inode_t *inode, struct iatt *stbuf,
-                                struct iatt *preparent, struct iatt *postparent,
-                                dict_t *xdata)
-{
-        dht_local_t  *local = NULL;
-        xlator_t     *cached_subvol = NULL;
-        dht_conf_t   *conf = NULL;
-        int           ret = -1;
-
-        GF_VALIDATE_OR_GOTO ("dht", frame, out);
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
-        GF_VALIDATE_OR_GOTO ("dht", this->private, out);
-        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
-
-        local = frame->local;
-        cached_subvol = local->cached_subvol;
-        conf = this->private;
+                layout = local->layout;
+                dht_selfheal_directory(frame, dht_lookup_selfheal_cbk,
+                                       &local->loc, layout);
+                return 0;
+            }
+        }
 
-        ret = dht_layout_preset (this, local->cached_subvol, inode);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "failed to set layout for subvolume %s",
-                        cached_subvol ? cached_subvol->name : "<nil>");
-                local->op_ret = -1;
-                local->op_errno = EINVAL;
-                goto unwind;
+        if (local->layout_mismatch) {
+            /* Found layout mismatch in the directory, need to
+               fix this in the inode context */
+            dht_layout_unref(this, local->layout);
+            local->layout = NULL;
+            dht_lookup_directory(frame, this, &local->loc);
+            return 0;
         }
 
-        local->op_ret = 0;
-        if ((local->stbuf.ia_nlink == 1)
-            && (conf && conf->unhashed_sticky_bit)) {
-                local->stbuf.ia_prot.sticky = 1;
+        if (local->need_lookup_everywhere) {
+            /* As the current layout gave ENOENT error, we would
+               need a new layout */
+            dht_layout_unref(this, local->layout);
+            local->layout = NULL;
+
+            /* We know that current cached subvol is no longer
+               valid, get the new one */
+            local->cached_subvol = NULL;
+            if (local->xattr_req) {
+                if (!gf_uuid_is_null(local->gfid)) {
+                    ret = dict_set_static_bin(local->xattr_req, "gfid-req",
+                                              local->gfid, 16);
+                }
+            }
+
+            dht_lookup_everywhere(frame, this, &local->loc);
+            return 0;
+        }
+        if (local->return_estale) {
+            local->op_ret = -1;
+            local->op_errno = ESTALE;
         }
 
         if (local->loc.parent) {
-                dht_inode_ctx_time_update (local->loc.parent, this,
-                                           postparent, 1);
+            dht_inode_ctx_time_update(local->loc.parent, this,
+                                      &local->postparent, 1);
+        }
+
+        DHT_STRIP_PHASE1_FLAGS(&local->stbuf);
+        dht_set_fixed_dir_stat(&local->postparent);
+
+        /* local->stbuf is updated only from subvols which have a layout
+         * The reason is to avoid choosing attr heal source from newly
+         * added bricks. In case e.g we have only one subvol and for
+         * some reason layout is not present on it, then local->stbuf
+         * will be EINVAL. This is an indication that the subvols
+         * active in the cluster do not have layouts on disk.
+         * Unwind with ESTALE to trigger a fresh lookup */
+        if (is_dir && local->stbuf.ia_type == IA_INVAL) {
+            local->op_ret = -1;
+            local->op_errno = ESTALE;
         }
+        /* Delete mds xattr at the time of STACK UNWIND */
+        if (local->xattr)
+            GF_REMOVE_INTERNAL_XATTR(conf->mds_xattr_key, local->xattr);
+
+        DHT_STACK_UNWIND(lookup, frame, local->op_ret, local->op_errno,
+                         local->inode, &local->stbuf, local->xattr,
+                         &local->postparent);
+    }
+
+err:
+    return ret;
+}
+
+static int
+dht_lookup_linkfile_create_cbk(call_frame_t *frame, void *cooie, xlator_t *this,
+                               int32_t op_ret, int32_t op_errno, inode_t *inode,
+                               struct iatt *stbuf, struct iatt *preparent,
+                               struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *cached_subvol = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", this->private, out);
+
+    local = frame->local;
+    cached_subvol = local->cached_subvol;
+    conf = this->private;
+
+    gf_uuid_unparse(local->loc.gfid, gfid);
+
+    if (local->locked)
+        dht_unlock_namespace(frame, &local->lock[0]);
+
+    ret = dht_layout_preset(this, local->cached_subvol, local->loc.inode);
+    if (ret < 0) {
+        gf_msg_debug(this->name, EINVAL,
+                     "Failed to set layout for subvolume %s, "
+                     "(gfid = %s)",
+                     cached_subvol ? cached_subvol->name : "<nil>", gfid);
+        local->op_ret = -1;
+        local->op_errno = EINVAL;
+        goto unwind;
+    }
+
+    local->op_ret = 0;
+    if ((local->stbuf.ia_nlink == 1) && (conf && conf->unhashed_sticky_bit)) {
+        local->stbuf.ia_prot.sticky = 1;
+    }
+
+    if (local->loc.parent) {
+        dht_inode_ctx_time_update(local->loc.parent, this, postparent, 1);
+    }
 
 unwind:
-        if (local->linked == _gf_true)
-                dht_linkfile_attr_heal (frame, this);
+    gf_msg_debug(this->name, 0,
+                 "creation of linkto on hashed subvol:%s, "
+                 "returned with op_ret %d and op_errno %d: %s",
+                 local->hashed_subvol->name, op_ret, op_errno,
+                 uuid_utoa(local->loc.gfid));
+
+    if (local->linked == _gf_true)
+        dht_linkfile_attr_heal(frame, this);
 
-        DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
-        DHT_STACK_UNWIND (lookup, frame, local->op_ret, local->op_errno,
-                          local->inode, &local->stbuf, local->xattr,
-                          &local->postparent);
+    dht_set_fixed_dir_stat(&local->postparent);
+
+    DHT_STRIP_PHASE1_FLAGS(&local->stbuf);
+    DHT_STACK_UNWIND(lookup, frame, local->op_ret, local->op_errno,
+                     local->inode, &local->stbuf, local->xattr,
+                     &local->postparent);
 out:
-        return ret;
+    return ret;
 }
 
-
-int
-dht_lookup_everywhere_done (call_frame_t *frame, xlator_t *this)
+static int
+dht_lookup_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, struct iatt *preparent,
+                      struct iatt *postparent, dict_t *xdata)
 {
-        int           ret = 0;
-        dht_local_t  *local = NULL;
-        xlator_t     *hashed_subvol = NULL;
-        xlator_t     *cached_subvol = NULL;
-        dht_layout_t *layout = NULL;
+    int this_call_cnt = 0;
+    dht_local_t *local = NULL;
+    const char *path = NULL;
 
-        local = frame->local;
-        hashed_subvol = local->hashed_subvol;
-        cached_subvol = local->cached_subvol;
+    local = (dht_local_t *)frame->local;
+    path = local->loc.path;
+    FRAME_SU_UNDO(frame, dht_local_t);
 
-        if (local->file_count && local->dir_count) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "path %s exists as a file on one subvolume "
-                        "and directory on another. "
-                        "Please fix it manually",
-                        local->loc.path);
-                DHT_STACK_UNWIND (lookup, frame, -1, EIO, NULL, NULL, NULL,
-                                  NULL);
-                return 0;
-        }
+    gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_UNLINK_LOOKUP_INFO,
+           "lookup_unlink returned with "
+           "op_ret -> %d and op-errno -> %d for %s",
+           op_ret, op_errno, ((path == NULL) ? "null" : path));
 
-        if (local->dir_count) {
-                dht_lookup_directory (frame, this, &local->loc);
-                return 0;
-        }
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        dht_lookup_everywhere_done(frame, this);
+    }
 
-        if (!cached_subvol) {
-                DHT_STACK_UNWIND (lookup, frame, -1, ENOENT, NULL, NULL, NULL,
-                                  NULL);
-                return 0;
-        }
+    return 0;
+}
 
-        if (local->need_lookup_everywhere) {
-                if (uuid_compare (local->gfid, local->inode->gfid)) {
-                        /* GFID different, return error */
-                        DHT_STACK_UNWIND (lookup, frame, -1, ENOENT, NULL,
-                                          NULL, NULL, NULL);
-                        return 0;
-                }
-                local->op_ret = 0;
-                local->op_errno = 0;
-                layout = dht_layout_for_subvol (this, cached_subvol);
-                if (!layout) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "%s: no pre-set layout for subvolume %s",
-                                local->loc.path, (cached_subvol ?
-                                                  cached_subvol->name :
-                                                  "<nil>"));
-                }
+static int
+dht_lookup_unlink_of_false_linkto_cbk(call_frame_t *frame, void *cookie,
+                                      xlator_t *this, int op_ret, int op_errno,
+                                      struct iatt *preparent,
+                                      struct iatt *postparent, dict_t *xdata)
+{
+    int this_call_cnt = 0;
+    dht_local_t *local = NULL;
+    const char *path = NULL;
 
-                ret = dht_layout_set (this, local->inode, layout);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "%s: failed to set layout for subvol %s",
-                                local->loc.path, (cached_subvol ?
-                                                  cached_subvol->name :
-                                                  "<nil>"));
-                }
+    local = (dht_local_t *)frame->local;
+    path = local->loc.path;
 
-                if (local->loc.parent) {
-                        dht_inode_ctx_time_update (local->loc.parent, this,
-                                                   &local->postparent, 1);
-                }
+    FRAME_SU_UNDO(frame, dht_local_t);
 
-                DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
-                DHT_STACK_UNWIND (lookup, frame, local->op_ret,
-                                  local->op_errno, local->inode,
-                                  &local->stbuf, local->xattr,
-                                  &local->postparent);
-                return 0;
-        }
+    gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_UNLINK_LOOKUP_INFO,
+           "lookup_unlink returned with "
+           "op_ret -> %d and op-errno -> %d for %s",
+           op_ret, op_errno, ((path == NULL) ? "null" : path));
 
-        if (!hashed_subvol) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "cannot create linkfile file for %s on %s: "
-                        "hashed subvolume cannot be found.",
-                        local->loc.path, cached_subvol->name);
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        if ((op_ret == 0) || ((op_errno != EBUSY) && (op_errno != ENOTCONN))) {
+            dht_lookup_everywhere_done(frame, this);
+        } else {
+            /*When dht_lookup_everywhere is performed, one cached
+             *and one hashed file was found and hashed file does
+             *not point to the above mentioned cached node. So it
+             *was considered as stale and an unlink was performed.
+             *But unlink fails. So may be rebalance is in progress.
+             *now ideally we have two data-files. One obtained during
+             *lookup_everywhere and one where unlink-failed. So
+             *at this point in time we cannot decide which one to
+             *choose because there are chances of first cached
+             *file is truncated after rebalance and if it is chosen
+             *as cached node, application will fail. So return EIO.*/
+
+            if (op_errno == EBUSY) {
+                gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                       DHT_MSG_UNLINK_FAILED,
+                       "Could not unlink the linkto file as "
+                       "either fd is open and/or linkto xattr "
+                       "is set for %s",
+                       ((path == NULL) ? "null" : path));
+            }
+            DHT_STACK_UNWIND(lookup, frame, -1, EIO, NULL, NULL, NULL, NULL);
+        }
+    }
+
+    return 0;
+}
 
-                local->op_ret = 0;
-                local->op_errno = 0;
+static int
+dht_lookup_unlink_stale_linkto_cbk(call_frame_t *frame, void *cookie,
+                                   xlator_t *this, int op_ret, int op_errno,
+                                   struct iatt *preparent,
+                                   struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    const char *path = NULL;
 
-                ret = dht_layout_preset (frame->this, cached_subvol,
-                                         local->inode);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "failed to set layout for subvol %s",
-                                cached_subvol ? cached_subvol->name :
-                                "<nil>");
-                        local->op_ret = -1;
-                        local->op_errno = EINVAL;
-                }
+    /* NOTE:
+     * If stale file unlink fails either there is an open-fd or is not an
+     * dht-linkto-file then posix_unlink returns EBUSY, which is overwritten
+     *  to ENOENT
+     */
 
-                if (local->loc.parent) {
-                        dht_inode_ctx_time_update (local->loc.parent, this,
-                                                   &local->postparent, 1);
-                }
+    local = frame->local;
 
-                DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
-                DHT_STACK_UNWIND (lookup, frame, local->op_ret,
-                                  local->op_errno, local->inode,
-                                  &local->stbuf, local->xattr,
-                                  &local->postparent);
-                return 0;
-        }
+    if (local) {
+        FRAME_SU_UNDO(frame, dht_local_t);
+        if (local->loc.path)
+            path = local->loc.path;
+    }
 
-        gf_log (this->name, GF_LOG_DEBUG,
-                "linking file %s existing on %s to %s (hash)",
-                local->loc.path, cached_subvol->name,
-                hashed_subvol->name);
+    gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_UNLINK_LOOKUP_INFO,
+           "Returned with op_ret %d and "
+           "op_errno %d for %s",
+           op_ret, op_errno, ((path == NULL) ? "null" : path));
 
-        ret = dht_linkfile_create (frame,
-                                   dht_lookup_linkfile_create_cbk, this,
-                                   cached_subvol, hashed_subvol, &local->loc);
+    DHT_STACK_UNWIND(lookup, frame, -1, ENOENT, NULL, NULL, NULL, NULL);
 
-        return ret;
+    return 0;
 }
 
-
-int
-dht_lookup_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int op_ret, int op_errno,
-                       struct iatt *preparent, struct iatt *postparent,
-                       dict_t *xdata)
+static int
+dht_fill_dict_to_avoid_unlink_of_migrating_file(dict_t *dict)
 {
-        int  this_call_cnt = 0;
+    int ret = 0;
 
-        this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt)) {
-                dht_lookup_everywhere_done (frame, this);
-        }
+    ret = dict_set_int32_sizen(dict, DHT_SKIP_NON_LINKTO_UNLINK, 1);
 
-        return 0;
-}
+    if (ret)
+        return -1;
 
+    ret = dict_set_int32_sizen(dict, DHT_SKIP_OPEN_FD_UNLINK, 1);
 
-int
-dht_lookup_everywhere_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                           int32_t op_ret, int32_t op_errno,
-                           inode_t *inode, struct iatt *buf, dict_t *xattr,
-                           struct iatt *postparent)
-{
-        dht_local_t  *local         = NULL;
-        int           this_call_cnt = 0;
-        call_frame_t *prev          = NULL;
-        int           is_linkfile   = 0;
-        int           is_dir        = 0;
-        xlator_t     *subvol        = NULL;
-        loc_t        *loc           = NULL;
-        xlator_t     *link_subvol   = NULL;
-        int           ret           = -1;
-        int32_t       fd_count      = 0;
-        dht_conf_t   *conf          = NULL;
-
-        GF_VALIDATE_OR_GOTO ("dht", frame, out);
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
-        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
-        GF_VALIDATE_OR_GOTO ("dht", this->private, out);
-
-        local  = frame->local;
-        loc    = &local->loc;
-        conf   = this->private;
-
-        prev   = cookie;
-        subvol = prev->this;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        if (op_errno != ENOENT)
-                                local->op_errno = op_errno;
-                        goto unlock;
-                }
+    if (ret)
+        return -1;
 
-                if (uuid_is_null (local->gfid))
-                        uuid_copy (local->gfid, buf->ia_gfid);
+    return 0;
+}
 
-                if (uuid_compare (local->gfid, buf->ia_gfid)) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "%s: gfid differs on subvolume %s",
-                                loc->path, prev->this->name);
-                }
+static int32_t
+dht_linkfile_create_lookup_cbk(call_frame_t *frame, void *cookie,
+                               xlator_t *this, int32_t op_ret, int32_t op_errno,
+                               inode_t *inode, struct iatt *buf, dict_t *xdata,
+                               struct iatt *postparent)
+{
+    dht_local_t *local = NULL;
+    int call_cnt = 0, ret = 0;
+    xlator_t *subvol = NULL;
+    uuid_t gfid = {
+        0,
+    };
+    char gfid_str[GF_UUID_BUF_SIZE] = {0};
+
+    subvol = cookie;
+    local = frame->local;
+
+    if (subvol == local->hashed_subvol) {
+        if ((op_ret == 0) || (op_errno != ENOENT))
+            local->dont_create_linkto = _gf_true;
+    } else {
+        if (gf_uuid_is_null(local->gfid))
+            gf_uuid_copy(gfid, local->loc.gfid);
+        else
+            gf_uuid_copy(gfid, local->gfid);
+
+        if ((op_ret == 0) && gf_uuid_compare(gfid, buf->ia_gfid)) {
+            gf_uuid_unparse(gfid, gfid_str);
+            gf_msg_debug(this->name, 0,
+                         "gfid (%s) different on cached subvol "
+                         "(%s) and looked up inode (%s), not "
+                         "creating linkto",
+                         uuid_utoa(buf->ia_gfid), subvol->name, gfid_str);
+            local->dont_create_linkto = _gf_true;
+        } else if (op_ret == -1) {
+            local->dont_create_linkto = _gf_true;
+        }
+    }
+
+    call_cnt = dht_frame_return(frame);
+    if (is_last_call(call_cnt)) {
+        if (local->dont_create_linkto)
+            goto no_linkto;
+        else {
+            gf_msg_debug(this->name, 0,
+                         "Creating linkto file on %s(hash) to "
+                         "%s on %s (gfid = %s)",
+                         local->hashed_subvol->name, local->loc.path,
+                         local->cached_subvol->name, gfid_str);
+
+            ret = dht_linkfile_create(frame, dht_lookup_linkfile_create_cbk,
+                                      this, local->cached_subvol,
+                                      local->hashed_subvol, &local->loc);
+
+            if (ret < 0)
+                goto no_linkto;
+        }
+    }
+
+    return 0;
+
+no_linkto:
+    gf_msg_debug(this->name, 0,
+                 "skipped linkto creation (path:%s) (gfid:%s) "
+                 "(hashed-subvol:%s) (cached-subvol:%s)",
+                 local->loc.path, gfid_str, local->hashed_subvol->name,
+                 local->cached_subvol->name);
+
+    dht_lookup_linkfile_create_cbk(frame, NULL, this, 0, 0, local->loc.inode,
+                                   &local->stbuf, &local->preparent,
+                                   &local->postparent, local->xattr);
+    return 0;
+}
 
-                is_linkfile = check_is_linkfile (inode, buf, xattr,
-                                                 conf->link_xattr_name);
-                is_dir = check_is_dir (inode, buf, xattr);
-
-                if (is_linkfile) {
-                        link_subvol = dht_linkfile_subvol (this, inode, buf,
-                                                           xattr);
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "found on %s linkfile %s (-> %s)",
-                                subvol->name, loc->path,
-                                link_subvol ? link_subvol->name : "''");
-                        goto unlock;
-                }
+static int32_t
+dht_call_lookup_linkfile_create(call_frame_t *frame, void *cookie,
+                                xlator_t *this, int32_t op_ret,
+                                int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    int i = 0;
+    xlator_t *subvol = NULL;
 
-                /* non linkfile GFID takes precedence */
-                uuid_copy (local->gfid, buf->ia_gfid);
+    local = frame->local;
+    if (gf_uuid_is_null(local->gfid))
+        gf_uuid_unparse(local->loc.gfid, gfid);
+    else
+        gf_uuid_unparse(local->gfid, gfid);
 
-                if (is_dir) {
-                        local->dir_count++;
+    if (op_ret < 0) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "protecting namespace failed, skipping linkto "
+               "creation (path:%s)(gfid:%s)(hashed-subvol:%s)"
+               "(cached-subvol:%s)",
+               local->loc.path, gfid, local->hashed_subvol->name,
+               local->cached_subvol->name);
+        goto err;
+    }
 
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "found on %s directory %s",
-                                subvol->name, loc->path);
-                } else {
-                        local->file_count++;
-
-                        if (!local->cached_subvol) {
-                                /* found one file */
-                                dht_iatt_merge (this, &local->stbuf, buf,
-                                                subvol);
-                                local->xattr = dict_ref (xattr);
-                                local->cached_subvol = subvol;
-                                gf_log (this->name, GF_LOG_DEBUG,
-                                        "found on %s file %s",
-                                        subvol->name, loc->path);
-
-                                dht_iatt_merge (this, &local->postparent,
-                                                postparent, subvol);
-                        } else {
-                                /* This is where we need 'rename' both entries logic */
-                                gf_log (this->name, GF_LOG_WARNING,
-                                        "multiple subvolumes (%s and %s) have "
-                                        "file %s (preferably rename the file "
-                                        "in the backend, and do a fresh lookup)",
-                                        local->cached_subvol->name,
-                                        subvol->name, local->loc.path);
-                        }
-                }
-        }
-unlock:
-        UNLOCK (&frame->lock);
+    local->locked = _gf_true;
 
-        if (is_linkfile) {
-                ret = dict_get_int32 (xattr, GLUSTERFS_OPEN_FD_COUNT, &fd_count);
-                /* Delete the linkfile only if there are no open fds on it.
-                   if there is a open-fd, it may be in migration */
-                if (!ret && (fd_count == 0)) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "deleting stale linkfile %s on %s",
-                                loc->path, subvol->name);
-                        STACK_WIND (frame, dht_lookup_unlink_cbk,
-                                    subvol, subvol->fops->unlink, loc, 0, NULL);
-                        return 0;
-                }
-        }
+    local->call_cnt = 2;
 
-        this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt)) {
-                dht_lookup_everywhere_done (frame, this);
-        }
+    for (i = 0; i < 2; i++) {
+        subvol = (subvol == NULL) ? local->hashed_subvol : local->cached_subvol;
 
-out:
-        return ret;
+        STACK_WIND_COOKIE(frame, dht_linkfile_create_lookup_cbk, subvol, subvol,
+                          subvol->fops->lookup, &local->loc, NULL);
+    }
+
+    return 0;
+
+err:
+    dht_lookup_linkfile_create_cbk(frame, NULL, this, 0, 0, local->loc.inode,
+                                   &local->stbuf, &local->preparent,
+                                   &local->postparent, local->xattr);
+    return 0;
 }
 
+/* Rebalance is performed from cached_node to hashed_node. Initial cached_node
+ * contains a non-linkto file. After migration it is converted to linkto and
+ * then unlinked. And at hashed_subvolume, first a linkto file is present,
+ * then after migration it is converted to a non-linkto file.
+ *
+ * Lets assume a file is present on cached subvolume and a new brick is added
+ * and new brick is the new_hashed subvolume. So fresh lookup on newly added
+ * hashed subvolume will fail and dht_lookup_everywhere gets called.  If just
+ * before sending the dht_lookup_everywhere request rebalance is in progress,
+ *
+ * from cached subvolume it may see: Nonlinkto or linkto or No file
+ * from hashed subvolume it may see: No file or linkto file or non-linkto file
+ *
+ * So this boils down to 9 cases:
+ *   at cached_subvol            at hashed_subvol
+ *   ----------------           -----------------
+ *
+ *a)   No file                     No file
+ *    [request reached after    [Request reached before
+ *       migration]                Migration]
+ *
+ *b)   No file                     Linkto File
+ *
+ *c)   No file                     Non-Linkto File
+ *
+ *d)   Linkto                      No-File
+ *
+ *e)   Linkto                      Linkto
+ *
+ *f)   Linkto                      Non-Linkto
+ *
+ *g)   NonLinkto                   No-File
+ *
+ *h)   NonLinkto                   Linkto
+ *
+ *i)   NonLinkto                   NonLinkto
+ *
+ * dht_lookup_everywhere_done takes decision based on any of the above case
+ */
 
-int
-dht_lookup_everywhere (call_frame_t *frame, xlator_t *this, loc_t *loc)
+static int
+dht_lookup_everywhere_done(call_frame_t *frame, xlator_t *this)
 {
-        dht_conf_t     *conf = NULL;
-        dht_local_t    *local = NULL;
-        int             i = 0;
-        int             call_cnt = 0;
+    int ret = 0;
+    dht_local_t *local = NULL;
+    xlator_t *hashed_subvol = NULL;
+    xlator_t *cached_subvol = NULL;
+    dht_layout_t *layout = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    gf_boolean_t found_non_linkto_on_hashed = _gf_false;
+
+    local = frame->local;
+    hashed_subvol = local->hashed_subvol;
+    cached_subvol = local->cached_subvol;
+
+    gf_uuid_unparse(local->loc.gfid, gfid);
+
+    if (local->file_count && local->dir_count) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_FILE_TYPE_MISMATCH,
+               "path %s (gfid = %s)exists as a file on one "
+               "subvolume and directory on another. "
+               "Please fix it manually",
+               local->loc.path, gfid);
+        DHT_STACK_UNWIND(lookup, frame, -1, EIO, NULL, NULL, NULL, NULL);
+        return 0;
+    }
+    if (local->op_ret && local->gfid_missing) {
+        if (gf_uuid_is_null(local->gfid_req)) {
+            DHT_STACK_UNWIND(lookup, frame, -1, ENODATA, NULL, NULL, NULL,
+                             NULL);
+            return 0;
+        }
+        /* A hack */
+        dht_lookup_directory(frame, this, &local->loc);
+        return 0;
+    }
 
-        GF_VALIDATE_OR_GOTO ("dht", frame, err);
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
-        GF_VALIDATE_OR_GOTO ("dht", this->private, out);
-        GF_VALIDATE_OR_GOTO ("dht", loc, out);
+    if (local->dir_count) {
+        dht_lookup_directory(frame, this, &local->loc);
+        return 0;
+    }
+
+    gf_msg_debug(this->name, 0,
+                 "STATUS: hashed_subvol %s "
+                 "cached_subvol %s",
+                 (hashed_subvol == NULL) ? "null" : hashed_subvol->name,
+                 (cached_subvol == NULL) ? "null" : cached_subvol->name);
+
+    if (!cached_subvol) {
+        if (local->skip_unlink.handle_valid_link && hashed_subvol) {
+            /*Purpose of "DHT_SKIP_NON_LINKTO_UNLINK":
+             * If this lookup is performed by rebalance and this
+             * rebalance process detected hashed file and by
+             * the time it sends the lookup request to cached node,
+             * file got migrated and now at initial hashed_node,
+             * final migrated file is present. With current logic,
+             * because this process fails to find the cached_node,
+             * it will unlink the file at initial hashed_node.
+             *
+             * So we avoid this by setting key, and checking at the
+             * posix_unlink that unlink the file only if file is a
+             * linkto file and not a migrated_file.
+             */
+
+            ret = dht_fill_dict_to_avoid_unlink_of_migrating_file(
+                local->xattr_req);
+
+            if (ret) {
+                /* If for some reason, setting key in the dict
+                 * fails, return with ENOENT, as with respect to
+                 * this process, it detected only a stale link
+                 * file.
+                 *
+                 * Next lookup will delete it.
+                 *
+                 * Performing deletion of stale link file when
+                 * setting key in dict fails, may cause the data
+                 * loss because of the above mentioned race.
+                 */
 
-        conf = this->private;
-        local = frame->local;
+                DHT_STACK_UNWIND(lookup, frame, -1, ENOENT, NULL, NULL, NULL,
+                                 NULL);
+            } else {
+                local->skip_unlink.handle_valid_link = _gf_false;
+
+                gf_msg_debug(this->name, 0,
+                             "No Cached was found and "
+                             "unlink on hashed was skipped"
+                             " so performing now: %s",
+                             local->loc.path);
+                FRAME_SU_DO(frame, dht_local_t);
+                STACK_WIND(frame, dht_lookup_unlink_stale_linkto_cbk,
+                           hashed_subvol, hashed_subvol->fops->unlink,
+                           &local->loc, 0, local->xattr_req);
+            }
 
-        call_cnt = conf->subvolume_cnt;
-        local->call_cnt = call_cnt;
+        } else {
+            gf_msg_debug(this->name, 0,
+                         "There was no cached file and  "
+                         "unlink on hashed is not skipped %s",
+                         local->loc.path);
 
-        if (!local->inode)
-                local->inode = inode_ref (loc->inode);
+            DHT_STACK_UNWIND(lookup, frame, -1, ENOENT, NULL, NULL, NULL, NULL);
+        }
+        return 0;
+    }
 
-        for (i = 0; i < call_cnt; i++) {
-                STACK_WIND (frame, dht_lookup_everywhere_cbk,
-                            conf->subvolumes[i],
-                            conf->subvolumes[i]->fops->lookup,
-                            loc, local->xattr_req);
+    /* At the time of dht_lookup, no file was found on hashed and that is
+     * why dht_lookup_everywhere is called, but by the time
+     * dht_lookup_everywhere
+     * reached to server, file might have already migrated. In that case we
+     * will find a migrated file at the hashed_node. In this case store the
+     * layout in context and return successfully.
+     */
+
+    if (hashed_subvol || local->need_lookup_everywhere) {
+        if (local->need_lookup_everywhere) {
+            found_non_linkto_on_hashed = _gf_true;
+
+        } else if ((local->file_count == 1) &&
+                   (hashed_subvol == cached_subvol)) {
+            gf_msg_debug(this->name, 0,
+                         "found cached file on hashed subvolume "
+                         "so store in context and return for %s",
+                         local->loc.path);
+
+            found_non_linkto_on_hashed = _gf_true;
         }
 
-        return 0;
-out:
-        DHT_STACK_UNWIND (lookup, frame, -1, EINVAL, NULL, NULL, NULL, NULL);
-err:
-        return -1;
-}
+        if (found_non_linkto_on_hashed)
+            goto preset_layout;
+    }
 
+    if (hashed_subvol) {
+        if (local->skip_unlink.handle_valid_link == _gf_true) {
+            if (cached_subvol == local->skip_unlink.hash_links_to) {
+                if (gf_uuid_compare(local->skip_unlink.cached_gfid,
+                                    local->skip_unlink.hashed_gfid)) {
+                    /*GFID different, return error*/
+                    DHT_STACK_UNWIND(lookup, frame, -1, ESTALE, NULL, NULL,
+                                     NULL, NULL);
 
-int
-dht_lookup_linkfile_cbk (call_frame_t *frame, void *cookie,
-                         xlator_t *this, int op_ret, int op_errno,
-                         inode_t *inode, struct iatt *stbuf, dict_t *xattr,
-                         struct iatt *postparent)
-{
-        call_frame_t *prev          = NULL;
-        dht_local_t  *local         = NULL;
-        xlator_t     *subvol        = NULL;
-        loc_t        *loc           = NULL;
-        dht_conf_t   *conf          = NULL;
-        int           ret           = 0;
-
-        GF_VALIDATE_OR_GOTO ("dht", frame, out);
-        GF_VALIDATE_OR_GOTO ("dht", this, unwind);
-        GF_VALIDATE_OR_GOTO ("dht", frame->local, unwind);
-        GF_VALIDATE_OR_GOTO ("dht", this->private, unwind);
-        GF_VALIDATE_OR_GOTO ("dht", cookie, unwind);
-
-        prev   = cookie;
-        subvol = prev->this;
-        conf   = this->private;
-        local  = frame->local;
-        loc    = &local->loc;
+                    return 0;
+                }
 
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "lookup of %s on %s (following linkfile) failed (%s)",
-                        local->loc.path, subvol->name, strerror (op_errno));
+                ret = dht_layout_preset(this, cached_subvol, local->loc.inode);
+                if (ret) {
+                    gf_msg(this->name, GF_LOG_INFO, 0,
+                           DHT_MSG_LAYOUT_PRESET_FAILED,
+                           "Could not set pre-set layout "
+                           "for subvolume %s",
+                           cached_subvol->name);
+                }
 
-                /* If cached subvol returned ENOTCONN, do not do
-                lookup_everywhere. We need to make sure linkfile does not get
-                removed, which can take away the namespace, and subvol is
-                anyways down. */
+                local->op_ret = (ret == 0) ? ret : -1;
+                local->op_errno = (ret == 0) ? ret : EINVAL;
 
-                if (op_errno != ENOTCONN)
-                        goto err;
-                else
-                        goto unwind;
-        }
+                /* Presence of local->cached_subvol validates
+                 * that lookup from cached node is successful
+                 */
 
-        if (check_is_dir (inode, stbuf, xattr)) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "lookup of %s on %s (following linkfile) reached dir",
-                        local->loc.path, subvol->name);
-                goto err;
-        }
+                if (!local->op_ret && local->loc.parent) {
+                    dht_inode_ctx_time_update(local->loc.parent, this,
+                                              &local->postparent, 1);
+                }
 
-        if (check_is_linkfile (inode, stbuf, xattr, conf->link_xattr_name)) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "lookup of %s on %s (following linkfile) reached link",
-                        local->loc.path, subvol->name);
-                goto err;
+                gf_msg_debug(this->name, 0,
+                             "Skipped unlinking linkto file "
+                             "on the hashed subvolume. "
+                             "Returning success as it is a "
+                             "valid linkto file. Path:%s",
+                             local->loc.path);
+
+                goto unwind_hashed_and_cached;
+            } else {
+                local->skip_unlink.handle_valid_link = _gf_false;
+
+                gf_msg_debug(this->name, 0,
+                             "Linkto file found on hashed "
+                             "subvol "
+                             "and data file found on cached "
+                             "subvolume. But linkto points to "
+                             "different cached subvolume (%s) "
+                             "path %s",
+                             (local->skip_unlink.hash_links_to
+                                  ? local->skip_unlink.hash_links_to->name
+                                  : " <nil>"),
+                             local->loc.path);
+
+                if (local->skip_unlink.opend_fd_count == 0) {
+                    ret = dht_fill_dict_to_avoid_unlink_of_migrating_file(
+                        local->xattr_req);
+
+                    if (ret) {
+                        DHT_STACK_UNWIND(lookup, frame, -1, EIO, NULL, NULL,
+                                         NULL, NULL);
+                    } else {
+                        local->call_cnt = 1;
+                        FRAME_SU_DO(frame, dht_local_t);
+                        STACK_WIND(frame, dht_lookup_unlink_of_false_linkto_cbk,
+                                   hashed_subvol, hashed_subvol->fops->unlink,
+                                   &local->loc, 0, local->xattr_req);
+                    }
+
+                    return 0;
+                }
+            }
         }
+    }
 
-        if (uuid_compare (local->gfid, stbuf->ia_gfid)) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: gfid different on data file on %s",
-                        local->loc.path, subvol->name);
-                goto err;
+preset_layout:
+
+    if (found_non_linkto_on_hashed) {
+        if (local->need_lookup_everywhere) {
+            if (gf_uuid_compare(local->gfid, local->inode->gfid)) {
+                /* GFID different, return error */
+                DHT_STACK_UNWIND(lookup, frame, -1, ENOENT, NULL, NULL, NULL,
+                                 NULL);
+                return 0;
+            }
         }
 
-        if ((stbuf->ia_nlink == 1)
-            && (conf && conf->unhashed_sticky_bit)) {
-                stbuf->ia_prot.sticky = 1;
+        local->op_ret = 0;
+        local->op_errno = 0;
+        layout = dht_layout_for_subvol(this, cached_subvol);
+        if (!layout) {
+            gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_INFO,
+                   "%s: no pre-set layout for subvolume %s,"
+                   " gfid = %s",
+                   local->loc.path,
+                   (cached_subvol ? cached_subvol->name : "<nil>"), gfid);
         }
 
-        ret = dht_layout_preset (this, prev->this, inode);
+        ret = dht_layout_set(this, local->inode, layout);
         if (ret < 0) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "failed to set layout for subvolume %s",
-                        prev->this->name);
-                op_ret   = -1;
-                op_errno = EINVAL;
+            gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_INFO,
+                   "%s: failed to set layout for subvol %s, "
+                   "gfid = %s",
+                   local->loc.path,
+                   (cached_subvol ? cached_subvol->name : "<nil>"), gfid);
         }
 
         if (local->loc.parent) {
-                dht_inode_ctx_time_update (local->loc.parent, this,
-                                           postparent, 1);
+            dht_inode_ctx_time_update(local->loc.parent, this,
+                                      &local->postparent, 1);
         }
 
-unwind:
-        DHT_STRIP_PHASE1_FLAGS (stbuf);
-        DHT_STACK_UNWIND (lookup, frame, op_ret, op_errno, inode, stbuf, xattr,
-                          postparent);
-
+        DHT_STRIP_PHASE1_FLAGS(&local->stbuf);
+        dht_set_fixed_dir_stat(&local->postparent);
+        DHT_STACK_UNWIND(lookup, frame, local->op_ret, local->op_errno,
+                         local->inode, &local->stbuf, local->xattr,
+                         &local->postparent);
         return 0;
+    }
 
-err:
-        dht_lookup_everywhere (frame, this, loc);
-out:
+    if (!hashed_subvol) {
+        gf_msg_debug(this->name, 0,
+                     "Cannot create linkfile for %s on %s: "
+                     "hashed subvolume cannot be found, gfid = %s.",
+                     local->loc.path, cached_subvol->name, gfid);
+
+        local->op_ret = 0;
+        local->op_errno = 0;
+
+        ret = dht_layout_preset(frame->this, cached_subvol, local->inode);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_LAYOUT_PRESET_FAILED,
+                   "Failed to set layout for subvol %s"
+                   ", gfid = %s",
+                   cached_subvol ? cached_subvol->name : "<nil>", gfid);
+            local->op_ret = -1;
+            local->op_errno = EINVAL;
+        }
+
+        if (local->loc.parent) {
+            dht_inode_ctx_time_update(local->loc.parent, this,
+                                      &local->postparent, 1);
+        }
+
+        DHT_STRIP_PHASE1_FLAGS(&local->stbuf);
+        dht_set_fixed_dir_stat(&local->postparent);
+        DHT_STACK_UNWIND(lookup, frame, local->op_ret, local->op_errno,
+                         local->inode, &local->stbuf, local->xattr,
+                         &local->postparent);
         return 0;
+    }
+
+    if (frame->root->op != GF_FOP_RENAME) {
+        local->current = &local->lock[0];
+        ret = dht_protect_namespace(frame, &local->loc, hashed_subvol,
+                                    &local->current->ns,
+                                    dht_call_lookup_linkfile_create);
+    } else {
+        gf_msg_debug(this->name, 0,
+                     "Creating linkto file on %s(hash) to %s on %s "
+                     "(gfid = %s)",
+                     hashed_subvol->name, local->loc.path, cached_subvol->name,
+                     gfid);
+
+        ret = dht_linkfile_create(frame, dht_lookup_linkfile_create_cbk, this,
+                                  cached_subvol, hashed_subvol, &local->loc);
+    }
+
+    return ret;
+
+unwind_hashed_and_cached:
+    DHT_STRIP_PHASE1_FLAGS(&local->stbuf);
+    dht_set_fixed_dir_stat(&local->postparent);
+    DHT_STACK_UNWIND(lookup, frame, local->op_ret, local->op_errno,
+                     local->inode, &local->stbuf, local->xattr,
+                     &local->postparent);
+    return 0;
 }
 
-
-int
-dht_lookup_directory (call_frame_t *frame, xlator_t *this, loc_t *loc)
+static int
+dht_lookup_everywhere_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, inode_t *inode,
+                          struct iatt *buf, dict_t *xattr,
+                          struct iatt *postparent)
 {
-        int           call_cnt = 0;
-        int           i = 0;
-        dht_conf_t   *conf = NULL;
-        dht_local_t  *local = NULL;
-        int           ret = 0;
-
-        GF_VALIDATE_OR_GOTO ("dht", frame, out);
-        GF_VALIDATE_OR_GOTO ("dht", this, unwind);
-        GF_VALIDATE_OR_GOTO ("dht", frame->local, unwind);
-        GF_VALIDATE_OR_GOTO ("dht", this->private, unwind);
-        GF_VALIDATE_OR_GOTO ("dht", loc, unwind);
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+    int is_linkfile = 0;
+    int is_dir = 0;
+    loc_t *loc = NULL;
+    xlator_t *link_subvol = NULL;
+    int ret = -1;
+    int32_t fd_count = 0;
+    dht_conf_t *conf = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    dict_t *dict_req = {0};
+
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", cookie, out);
+    GF_VALIDATE_OR_GOTO("dht", this->private, out);
+
+    local = frame->local;
+    loc = &local->loc;
+    conf = this->private;
+
+    prev = cookie;
+
+    gf_msg_debug(this->name, 0,
+                 "returned with op_ret %d and op_errno %d (%s) "
+                 "from subvol %s",
+                 op_ret, op_errno, loc->path, prev->name);
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            if (op_errno != ENOENT)
+                local->op_errno = op_errno;
+            if (op_errno == ENODATA)
+                local->gfid_missing = _gf_true;
+            goto unlock;
+        }
 
-        conf = this->private;
-        local = frame->local;
+        if (gf_uuid_is_null(local->gfid))
+            gf_uuid_copy(local->gfid, buf->ia_gfid);
 
-        call_cnt        = conf->subvolume_cnt;
-        local->call_cnt = call_cnt;
+        gf_uuid_unparse(local->gfid, gfid);
 
-        local->layout = dht_layout_new (this, conf->subvolume_cnt);
-        if (!local->layout) {
-                goto unwind;
+        if (gf_uuid_compare(local->gfid, buf->ia_gfid)) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_GFID_MISMATCH,
+                   "%s: gfid differs on subvolume %s,"
+                   " gfid local = %s, gfid node = %s",
+                   loc->path, prev->name, gfid, uuid_utoa(buf->ia_gfid));
         }
 
-        if (local->xattr != NULL) {
-                dict_unref (local->xattr);
-                local->xattr = NULL;
-        }
+        is_linkfile = check_is_linkfile(inode, buf, xattr,
+                                        conf->link_xattr_name);
 
-        if (!uuid_is_null (local->gfid)) {
-                ret = dict_set_static_bin (local->xattr_req, "gfid-req",
-                                           local->gfid, 16);
-                if (ret)
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "%s: failed to set gfid", local->loc.path);
+        if (is_linkfile) {
+            link_subvol = dht_linkfile_subvol(this, inode, buf, xattr);
+            gf_msg_debug(this->name, 0, "found on %s linkfile %s (-> %s)",
+                         prev->name, loc->path,
+                         link_subvol ? link_subvol->name : "''");
+            goto unlock;
         }
 
-        for (i = 0; i < call_cnt; i++) {
-                STACK_WIND (frame, dht_lookup_dir_cbk,
-                            conf->subvolumes[i],
-                            conf->subvolumes[i]->fops->lookup,
-                            &local->loc, local->xattr_req);
-        }
-        return 0;
-unwind:
-        DHT_STACK_UNWIND (lookup, frame, -1, ENOMEM, NULL, NULL, NULL, NULL);
-out:
-        return 0;
+        is_dir = check_is_dir(inode, buf, xattr);
 
-}
+        /* non linkfile GFID takes precedence but don't overwrite
+         gfid if we have already found a cached file*/
+        if (!local->cached_subvol)
+            gf_uuid_copy(local->gfid, buf->ia_gfid);
 
+        if (is_dir) {
+            local->dir_count++;
 
-int
-dht_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                int op_ret, int op_errno,
-                inode_t *inode, struct iatt *stbuf, dict_t *xattr,
-                struct iatt *postparent)
-{
-        char          is_linkfile   = 0;
-        char          is_dir        = 0;
-        xlator_t     *subvol        = NULL;
-        dht_conf_t   *conf          = NULL;
-        dht_local_t  *local         = NULL;
-        loc_t        *loc           = NULL;
-        call_frame_t *prev          = NULL;
-        int           ret           = 0;
-        dht_layout_t *parent_layout = NULL;
-
-        GF_VALIDATE_OR_GOTO ("dht", frame, err);
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
-        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
-        GF_VALIDATE_OR_GOTO ("dht", this->private, out);
-
-        conf  = this->private;
-
-        prev  = cookie;
-        local = frame->local;
-        loc   = &local->loc;
+            gf_msg_debug(this->name, 0, "found on %s directory %s", prev->name,
+                         loc->path);
+        } else {
+            local->file_count++;
 
-        /* This is required for handling stale linkfile deletion,
-         * or any more call which happens from this 'loc'.
-         */
-        if (!op_ret && uuid_is_null (local->gfid))
-                memcpy (local->gfid, stbuf->ia_gfid, 16);
-
-        if (ENTRY_MISSING (op_ret, op_errno)) {
-                gf_log (this->name, GF_LOG_TRACE, "Entry %s missing on subvol"
-                        " %s", loc->path, prev->this->name);
-                if (conf->search_unhashed == GF_DHT_LOOKUP_UNHASHED_ON) {
-                        local->op_errno = ENOENT;
-                        dht_lookup_everywhere (frame, this, loc);
-                        return 0;
-                }
-                if ((conf->search_unhashed == GF_DHT_LOOKUP_UNHASHED_AUTO) &&
-                    (loc->parent)) {
-                        ret = dht_inode_ctx_layout_get (loc->parent, this,
-                                                        &parent_layout);
-                        if (ret || !parent_layout)
-                                goto out;
-                        if (parent_layout->search_unhashed) {
-                                local->op_errno = ENOENT;
-                                dht_lookup_everywhere (frame, this, loc);
-                                return 0;
-                        }
-                }
-        }
+            gf_msg_debug(this->name, 0, "found cached file on %s for %s",
+                         prev->name, loc->path);
 
-        if (op_ret == 0) {
-                is_dir      = check_is_dir (inode, stbuf, xattr);
-                if (is_dir) {
-                        local->inode = inode_ref (inode);
-                        local->xattr = dict_ref (xattr);
-                }
-        }
+            if (!local->cached_subvol) {
+                /* found one file */
+                dht_iatt_merge(this, &local->stbuf, buf);
 
-        if (is_dir || (op_ret == -1 && op_errno == ENOTCONN)) {
-                dht_lookup_directory (frame, this, &local->loc);
-                return 0;
-        }
+                local->xattr = dict_ref(xattr);
+                local->cached_subvol = prev;
 
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_DEBUG, "Lookup of %s for subvolume"
-                       " %s failed with error %s", loc->path, prev->this->name,
-                       strerror (op_errno));
-                goto out;
+                gf_msg_debug(this->name, 0,
+                             "storing cached on %s file"
+                             " %s",
+                             prev->name, loc->path);
+
+                dht_iatt_merge(this, &local->postparent, postparent);
+
+                gf_uuid_copy(local->skip_unlink.cached_gfid, buf->ia_gfid);
+            } else {
+                /* This is where we need 'rename' both entries logic */
+                gf_msg(this->name, GF_LOG_WARNING, 0,
+                       DHT_MSG_FILE_ON_MULT_SUBVOL,
+                       "multiple subvolumes (%s and %s) have "
+                       "file %s (preferably rename the file "
+                       "in the backend,and do a fresh lookup)",
+                       local->cached_subvol->name, prev->name, local->loc.path);
+            }
         }
+    }
+unlock:
+    UNLOCK(&frame->lock);
 
-        is_linkfile = check_is_linkfile (inode, stbuf, xattr,
-                                         conf->link_xattr_name);
+    if (is_linkfile) {
+        ret = dict_get_int32(xattr, GLUSTERFS_OPEN_FD_COUNT, &fd_count);
 
-        if (!is_linkfile) {
-                /* non-directory and not a linkfile */
+        /*  Any linkto file found on the non-hashed subvolume should
+         *  be unlinked (performed in the "else if" block below)
+         *
+         *  But if a linkto file is found on hashed subvolume, it may be
+         *  pointing to valid cached node. So unlinking of linkto
+         *  file on hashed subvolume is skipped and inside
+         *  dht_lookup_everywhere_done, checks are performed. If this
+         *  linkto file is found as stale linkto file, it is deleted
+         *  otherwise unlink is skipped.
+         */
 
-                ret = dht_layout_preset (this, prev->this, inode);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "could not set pre-set layout for subvolume %s",
-                                prev->this->name);
-                        op_ret   = -1;
-                        op_errno = EINVAL;
-                        goto out;
-                }
-                goto out;
-        }
+        if (local->hashed_subvol && local->hashed_subvol == prev) {
+            local->skip_unlink.handle_valid_link = _gf_true;
+            local->skip_unlink.opend_fd_count = fd_count;
+            local->skip_unlink.hash_links_to = link_subvol;
+            gf_uuid_copy(local->skip_unlink.hashed_gfid, buf->ia_gfid);
+
+            gf_msg_debug(this->name, 0,
+                         "Found"
+                         " one linkto file on hashed subvol %s "
+                         "for %s: Skipping unlinking till "
+                         "everywhere_done",
+                         prev->name, loc->path);
+
+        } else if (!ret && (fd_count == 0)) {
+            dict_req = dict_new();
+
+            ret = dht_fill_dict_to_avoid_unlink_of_migrating_file(dict_req);
+
+            if (ret) {
+                /* Skip unlinking for dict_failure
+                 *File is found as a linkto file on non-hashed,
+                 *subvolume. In the current implementation,
+                 *finding a linkto-file on non-hashed does not
+                 *always implies that it is stale. So deletion
+                 *of file should be done only when both fd is
+                 *closed and linkto-xattr is set. In case of
+                 *dict_set failure, avoid skipping of file.
+                 *NOTE: dht_frame_return should get called for
+                 *      this block.
+                 */
+
+                dict_unref(dict_req);
+
+            } else {
+                gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_INFO,
+                       "attempting deletion of stale linkfile "
+                       "%s on %s (hashed subvol is %s)",
+                       loc->path, prev->name,
+                       (local->hashed_subvol ? local->hashed_subvol->name
+                                             : "<null>"));
+                /* *
+                 * These stale files may be created using root
+                 * user. Hence deletion will work only with
+                 * root.
+                 */
+                FRAME_SU_DO(frame, dht_local_t);
+                STACK_WIND(frame, dht_lookup_unlink_cbk, prev,
+                           prev->fops->unlink, loc, 0, dict_req);
+
+                dict_unref(dict_req);
 
-        subvol = dht_linkfile_subvol (this, inode, stbuf, xattr);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "linkfile not having link subvolume. path=%s",
-                        loc->path);
-                dht_lookup_everywhere (frame, this, loc);
                 return 0;
+            }
         }
+    }
 
-        STACK_WIND (frame, dht_lookup_linkfile_cbk,
-                    subvol, subvol->fops->lookup,
-                    &local->loc, local->xattr_req);
-
-        return 0;
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        dht_lookup_everywhere_done(frame, this);
+    }
 
 out:
-        /*
-         * FIXME: postparent->ia_size and postparent->st_blocks do not have
-         * correct values. since, postparent corresponds to a directory these
-         * two members should have values equal to sum of corresponding values
-         * from each of the subvolume. See dht_iatt_merge for reference.
-         */
+    return ret;
+}
 
-        if (!op_ret && local->loc.parent) {
-                dht_inode_ctx_time_update (local->loc.parent, this,
-                                           postparent, 1);
-        }
+int
+dht_lookup_everywhere(call_frame_t *frame, xlator_t *this, loc_t *loc)
+{
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    int i = 0;
+    int call_cnt = 0;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", this->private, out);
+    GF_VALIDATE_OR_GOTO("dht", loc, out);
+
+    conf = this->private;
+    local = frame->local;
+
+    call_cnt = conf->subvolume_cnt;
+    local->call_cnt = call_cnt;
+
+    if (!local->inode)
+        local->inode = inode_ref(loc->inode);
 
-        DHT_STRIP_PHASE1_FLAGS (stbuf);
-        DHT_STACK_UNWIND (lookup, frame, op_ret, op_errno, inode, stbuf, xattr,
-                          postparent);
+    gf_msg_debug(this->name, 0, "winding lookup call to %d subvols", call_cnt);
+
+    for (i = 0; i < call_cnt; i++) {
+        STACK_WIND_COOKIE(frame, dht_lookup_everywhere_cbk, conf->subvolumes[i],
+                          conf->subvolumes[i],
+                          conf->subvolumes[i]->fops->lookup, loc,
+                          local->xattr_req);
+    }
+
+    return 0;
+out:
+    DHT_STACK_UNWIND(lookup, frame, -1, EINVAL, NULL, NULL, NULL, NULL);
 err:
-        return 0;
+    return -1;
 }
 
-/* For directories, check if acl xattrs have been requested (by the acl xlator),
- * if not, request for them. These xattrs are needed for dht dir self-heal to
- * perform proper self-healing of dirs
- */
-void
-dht_check_and_set_acl_xattr_req (inode_t *inode, dict_t *xattr_req)
+int
+dht_lookup_linkfile_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, inode_t *inode,
+                        struct iatt *stbuf, dict_t *xattr,
+                        struct iatt *postparent)
 {
-        int     ret = 0;
+    xlator_t *prev = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *subvol = NULL;
+    loc_t *loc = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    GF_VALIDATE_OR_GOTO("dht", this, unwind);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, unwind);
+    GF_VALIDATE_OR_GOTO("dht", this->private, unwind);
+    GF_VALIDATE_OR_GOTO("dht", cookie, unwind);
+
+    prev = cookie;
+    subvol = prev;
+    conf = this->private;
+    local = frame->local;
+    loc = &local->loc;
+
+    gf_uuid_unparse(loc->gfid, gfid);
+
+    if (op_ret == -1) {
+        gf_msg(this->name, GF_LOG_INFO, op_errno, DHT_MSG_LINK_FILE_LOOKUP_INFO,
+               "Lookup of %s on %s (following linkfile) failed "
+               ",gfid = %s",
+               local->loc.path, subvol->name, gfid);
+
+        /* If cached subvol returned ENOTCONN, do not do
+        lookup_everywhere. We need to make sure linkfile does not get
+        removed, which can take away the namespace, and subvol is
+        anyways down. */
+
+        local->cached_subvol = NULL;
+        if (op_errno != ENOTCONN)
+            goto err;
+        else
+            goto unwind;
+    }
+
+    if (check_is_dir(inode, stbuf, xattr)) {
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_LINK_FILE_LOOKUP_INFO,
+               "Lookup of %s on %s (following linkfile) reached dir,"
+               " gfid = %s",
+               local->loc.path, subvol->name, gfid);
+        goto err;
+    }
+
+    if (check_is_linkfile(inode, stbuf, xattr, conf->link_xattr_name)) {
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_LINK_FILE_LOOKUP_INFO,
+               "lookup of %s on %s (following linkfile) reached link,"
+               "gfid = %s",
+               local->loc.path, subvol->name, gfid);
+        goto err;
+    }
+
+    if (gf_uuid_compare(local->gfid, stbuf->ia_gfid)) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_GFID_MISMATCH,
+               "%s: gfid different on data file on %s,"
+               " gfid local = %s, gfid node = %s ",
+               local->loc.path, subvol->name, gfid, uuid_utoa(stbuf->ia_gfid));
+        goto err;
+    }
+
+    if ((stbuf->ia_nlink == 1) && (conf && conf->unhashed_sticky_bit)) {
+        stbuf->ia_prot.sticky = 1;
+    }
+
+    ret = dht_layout_preset(this, prev, inode);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_LAYOUT_PRESET_FAILED,
+               "Failed to set layout for subvolume %s,"
+               "gfid = %s",
+               prev->name, gfid);
+        op_ret = -1;
+        op_errno = EINVAL;
+    }
+
+    if (local->loc.parent) {
+        dht_inode_ctx_time_update(local->loc.parent, this, postparent, 1);
+    }
 
-        GF_ASSERT (inode);
-        GF_ASSERT (xattr_req);
+unwind:
+    DHT_STRIP_PHASE1_FLAGS(stbuf);
+    dht_set_fixed_dir_stat(postparent);
+    DHT_STACK_UNWIND(lookup, frame, op_ret, op_errno, inode, stbuf, xattr,
+                     postparent);
 
-        if (inode->ia_type != IA_IFDIR)
-                return;
+    return 0;
 
-        if (!dict_get (xattr_req, POSIX_ACL_ACCESS_XATTR)) {
-                ret = dict_set_int8 (xattr_req, POSIX_ACL_ACCESS_XATTR, 0);
-                if (ret)
-                        gf_log (THIS->name, GF_LOG_WARNING,
-                                "failed to set key %s",
-                                POSIX_ACL_ACCESS_XATTR);
-        }
+err:
+    dht_lookup_everywhere(frame, this, loc);
+out:
+    return 0;
+}
 
-        if (!dict_get (xattr_req, POSIX_ACL_DEFAULT_XATTR)) {
-                ret = dict_set_int8 (xattr_req, POSIX_ACL_DEFAULT_XATTR, 0);
-                if (ret)
-                        gf_log (THIS->name, GF_LOG_WARNING,
-                                "failed to set key %s",
-                                POSIX_ACL_DEFAULT_XATTR);
-        }
+/* Code to get hashed subvol based on inode and loc
+   First it check if loc->parent and loc->path exist then it get
+   hashed subvol based on loc.
+*/
 
-        return;
+static gf_boolean_t
+dht_should_lookup_everywhere(xlator_t *this, dht_conf_t *conf, loc_t *loc)
+{
+    dht_layout_t *parent_layout = NULL;
+    int ret = 0;
+    gf_boolean_t lookup_everywhere = _gf_true;
+
+    /* lookup-optimize supersedes lookup-unhashed settings.
+     * If it is set, do not process search_unhashed
+     * If lookup-optimize if enabled, lookup everywhere if:
+     *  - this is the rebalance daemon.
+     *  - loc->parent is unavailable.
+     *  - parent_layout is unavailable
+     *  - parent_layout->commit_hash != conf->vol_commit_hash
+     */
+
+    if (conf->lookup_optimize) {
+        if (!conf->defrag && loc->parent) {
+            ret = dht_inode_ctx_layout_get(loc->parent, this, &parent_layout);
+            if (!ret && parent_layout &&
+                (parent_layout->commit_hash == conf->vol_commit_hash)) {
+                lookup_everywhere = _gf_false;
+            }
+        }
+        goto out;
+    } else {
+        if (conf->search_unhashed == GF_DHT_LOOKUP_UNHASHED_AUTO) {
+            if (loc->parent) {
+                ret = dht_inode_ctx_layout_get(loc->parent, this,
+                                               &parent_layout);
+                if (ret || !parent_layout ||
+                    (!parent_layout->search_unhashed)) {
+                    lookup_everywhere = _gf_false;
+                }
+            } else {
+                lookup_everywhere = _gf_false;
+            }
+
+            goto out;
+        }
+    }
+out:
+    return lookup_everywhere;
 }
 
 int
-dht_lookup (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, dict_t *xattr_req)
-{
-        xlator_t     *subvol = NULL;
-        xlator_t     *hashed_subvol = NULL;
-        dht_local_t  *local  = NULL;
-        dht_conf_t   *conf = NULL;
-        int           ret    = -1;
-        int           op_errno = -1;
-        dht_layout_t *layout = NULL;
-        int           i = 0;
-        int           call_cnt = 0;
-        loc_t         new_loc = {0,};
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        conf = this->private;
-        if (!conf)
-                goto err;
+dht_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, inode_t *inode, struct iatt *stbuf, dict_t *xattr,
+               struct iatt *postparent)
+{
+    char is_linkfile = 0;
+    char is_dir = 0;
+    xlator_t *subvol = NULL;
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    loc_t *loc = NULL;
+    xlator_t *prev = NULL;
+    int ret = 0;
+    uint32_t vol_commit_hash = 0;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", cookie, out);
+    GF_VALIDATE_OR_GOTO("dht", this->private, out);
+
+    conf = this->private;
+
+    prev = cookie;
+    local = frame->local;
+    loc = &local->loc;
+
+    gf_msg_debug(this->name, op_errno,
+                 "%s: fresh_lookup on %s returned with op_ret %d", loc->path,
+                 prev->name, op_ret);
+
+    if (op_ret == -1) {
+        if (ENTRY_MISSING(op_ret, op_errno)) {
+            if (1 == conf->subvolume_cnt) {
+                /* No need to lookup again */
+                goto out;
+            }
 
-        local = dht_local_init (frame, loc, NULL, GF_FOP_LOOKUP);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+            gf_msg_debug(this->name, 0, "Entry %s missing on subvol %s",
+                         loc->path, prev->name);
 
-        ret = dht_filter_loc_subvol_key (this, loc, &new_loc,
-                                         &hashed_subvol);
-        if (ret) {
-                loc_wipe (&local->loc);
-                ret = loc_dup (&new_loc, &local->loc);
-
-                /* we no more need 'new_loc' entries */
-                loc_wipe (&new_loc);
-
-                /* check if loc_dup() is successful */
-                if (ret == -1) {
-                        op_errno = errno;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "copying location failed for path=%s",
-                                loc->path);
-                        goto err;
-                }
-        }
+            if (dht_should_lookup_everywhere(this, conf, loc)) {
+                local->op_errno = ENOENT;
+                dht_lookup_everywhere(frame, this, loc);
+                return 0;
+            }
 
-        if (xattr_req) {
-                local->xattr_req = dict_ref (xattr_req);
         } else {
-                local->xattr_req = dict_new ();
-        }
-
-        if (uuid_is_null (loc->pargfid) && !uuid_is_null (loc->gfid) &&
-            !__is_root_gfid (loc->inode->gfid)) {
-                local->cached_subvol = NULL;
-                dht_discover (frame, this, loc);
+            /* posix returns ENODATA if the gfid is not set but the client and
+             * server protocol layers do not send the stbuf. We need to
+             * heal this so check if this is a directory on the other subvols.
+             */
+            if ((op_errno == ENOTCONN) || (op_errno == ENODATA)) {
+                dht_lookup_directory(frame, this, &local->loc);
                 return 0;
+            }
+        }
+        gf_msg_debug(this->name, op_errno, "%s: Lookup on subvolume %s failed",
+                     loc->path, prev->name);
+        goto out;
+    }
+
+    /* Lookup succeeded - op_ret = 0 */
+
+    /* This is required for handling stale linkfile deletion,
+     * or any more call which happens from this 'loc'.
+     */
+    if (gf_uuid_is_null(local->gfid)) {
+        /*This is set from the first successful response*/
+        memcpy(local->gfid, stbuf->ia_gfid, 16);
+    }
+
+    if (!conf->vch_forced) {
+        /* Update the commit hash in conf if it is found */
+        ret = dict_get_uint32(xattr, conf->commithash_xattr_name,
+                              &vol_commit_hash);
+        if (ret == 0) {
+            conf->vol_commit_hash = vol_commit_hash;
         }
+    }
 
-        if (!hashed_subvol)
-                hashed_subvol = dht_subvol_get_hashed (this, loc);
-        local->hashed_subvol = hashed_subvol;
+    is_dir = check_is_dir(inode, stbuf, xattr);
+    if (is_dir) {
+        /* A directory is present on all subvols, send the lookup to
+         * all subvols now */
+        local->inode = inode_ref(inode);
+        local->xattr = dict_ref(xattr);
+        dht_lookup_directory(frame, this, &local->loc);
+        return 0;
+    }
 
-        if (is_revalidate (loc)) {
-                layout = local->layout;
-                if (!layout) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "revalidate without cache. path=%s",
-                                loc->path);
-                        op_errno = EINVAL;
-                        goto err;
-                }
+    is_linkfile = check_is_linkfile(inode, stbuf, xattr, conf->link_xattr_name);
 
-                if (layout->gen && (layout->gen < conf->gen)) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "incomplete layout failure for path=%s",
-                                loc->path);
+    if (!is_linkfile) {
+        /* non-directory and not a linkto file. This is a data file
+         * Update the layout to point to the cached subvol
+         */
 
-                        dht_layout_unref (this, local->layout);
-                        local->layout = NULL;
-                        local->cached_subvol = NULL;
-                        goto do_fresh_lookup;
-                }
+        ret = dht_layout_preset(this, prev, inode);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_LAYOUT_PRESET_FAILED,
+                   "%s: could not set pre-set layout for subvolume %s",
+                   loc->path, prev->name);
+            op_ret = -1;
+            op_errno = EINVAL;
+            goto out;
+        }
+        goto out;
+    }
+
+    /* This is a linkto file. Get the value of the target subvol from the
+     * linkto xattr and lookup there to see if the file exists
+     */
+    subvol = dht_linkfile_subvol(this, inode, stbuf, xattr);
+    if (!subvol) {
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_INFO,
+               "%s: No link subvol for linkto", loc->path);
+        dht_lookup_everywhere(frame, this, loc);
+        return 0;
+    }
 
-                local->inode = inode_ref (loc->inode);
+    gf_msg_debug(this->name, 0, "%s: Calling lookup on linkto target %s",
+                 loc->path, subvol->name);
 
-                /* NOTE: we don't require 'trusted.glusterfs.dht.linkto' attribute,
-                 *       revalidates directly go to the cached-subvolume.
-                 */
-                ret = dict_set_uint32 (local->xattr_req,
-                                       conf->xattr_name, 4 * 4);
-
-                if (IA_ISDIR (local->inode->ia_type)) {
-                        local->call_cnt = call_cnt = conf->subvolume_cnt;
-                        for (i = 0; i < call_cnt; i++) {
-                                STACK_WIND (frame, dht_revalidate_cbk,
-                                            conf->subvolumes[i],
-                                            conf->subvolumes[i]->fops->lookup,
-                                            loc, local->xattr_req);
-                        }
-                        return 0;
-                }
+    STACK_WIND_COOKIE(frame, dht_lookup_linkfile_cbk, subvol, subvol,
+                      subvol->fops->lookup, &local->loc, local->xattr_req);
+
+    return 0;
 
-                call_cnt = local->call_cnt = layout->cnt;
+out:
+    /*
+     * FIXME: postparent->ia_size and postparent->st_blocks do not have
+     * correct values. since, postparent corresponds to a directory these
+     * two members should have values equal to sum of corresponding values
+     * from each of the subvolume. See dht_iatt_merge for reference.
+     */
+
+    if (!op_ret && local && local->loc.parent) {
+        dht_inode_ctx_time_update(local->loc.parent, this, postparent, 1);
+    }
+
+    DHT_STRIP_PHASE1_FLAGS(stbuf);
+    dht_set_fixed_dir_stat(postparent);
+    DHT_STACK_UNWIND(lookup, frame, op_ret, op_errno, inode, stbuf, xattr,
+                     postparent);
+err:
+    return 0;
+}
 
-                /* need it for self-healing linkfiles which is
-                   'in-migration' state */
-                ret = dict_set_uint32 (local->xattr_req,
-                                       GLUSTERFS_OPEN_FD_COUNT, 4);
+/* For directories, check if acl xattrs have been requested (by the acl
+ * xlator), if not, request for them. These xattrs are needed for dht dir
+ * self-heal to perform proper self-healing of dirs
+ */
+static void
+dht_check_and_set_acl_xattr_req(xlator_t *this, dict_t *xattr_req)
+{
+    int ret = 0;
 
-                /* need it for dir self-heal */
-                dht_check_and_set_acl_xattr_req (loc->inode, local->xattr_req);
+    GF_ASSERT(xattr_req);
 
-		for (i = 0; i < call_cnt; i++) {
-			subvol = layout->list[i].xlator;
+    if (!dict_get(xattr_req, POSIX_ACL_ACCESS_XATTR)) {
+        ret = dict_set_int8(xattr_req, POSIX_ACL_ACCESS_XATTR, 0);
+        if (ret)
+            gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_DICT_SET_FAILED,
+                   "Failed to set dictionary value:key = %s",
+                   POSIX_ACL_ACCESS_XATTR);
+    }
 
-			STACK_WIND (frame, dht_revalidate_cbk,
-				    subvol, subvol->fops->lookup,
-				    &local->loc, local->xattr_req);
+    if (!dict_get(xattr_req, POSIX_ACL_DEFAULT_XATTR)) {
+        ret = dict_set_int8(xattr_req, POSIX_ACL_DEFAULT_XATTR, 0);
+        if (ret)
+            gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_DICT_SET_FAILED,
+                   "Failed to set dictionary value:key = %s",
+                   POSIX_ACL_DEFAULT_XATTR);
+    }
 
-		}
-        } else {
-        do_fresh_lookup:
-                /* TODO: remove the hard-coding */
-                ret = dict_set_uint32 (local->xattr_req,
-                                       conf->xattr_name, 4 * 4);
-
-                ret = dict_set_uint32 (local->xattr_req,
-                                       conf->link_xattr_name, 256);
-
-                /* need it for self-healing linkfiles which is
-                   'in-migration' state */
-                ret = dict_set_uint32 (local->xattr_req,
-                                       GLUSTERFS_OPEN_FD_COUNT, 4);
-
-                /* need it for dir self-heal */
-                dht_check_and_set_acl_xattr_req (loc->inode, local->xattr_req);
-
-                if (!hashed_subvol) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "no subvolume in layout for path=%s, "
-                                "checking on all the subvols to see if "
-                                "it is a directory", loc->path);
-                        call_cnt        = conf->subvolume_cnt;
-                        local->call_cnt = call_cnt;
-
-                        local->layout = dht_layout_new (this,
-                                                        conf->subvolume_cnt);
-                        if (!local->layout) {
-                                op_errno = ENOMEM;
-                                goto err;
-                        }
+    return;
+}
 
-                        for (i = 0; i < call_cnt; i++) {
-                                STACK_WIND (frame, dht_lookup_dir_cbk,
-                                            conf->subvolumes[i],
-                                            conf->subvolumes[i]->fops->lookup,
-                                            &local->loc, local->xattr_req);
-                        }
-                        return 0;
-                }
+/* for directories, we need the following info:
+ * the layout : trusted.glusterfs.dht
+ * the mds information : trusted.glusterfs.dht.mds
+ * the acl info: See above
+ */
+static int
+dht_set_dir_xattr_req(xlator_t *this, loc_t *loc, dict_t *xattr_req)
+{
+    int ret = -EINVAL;
+    dht_conf_t *conf = NULL;
+
+    conf = this->private;
+    if (!conf) {
+        goto err;
+    }
+
+    if (!xattr_req) {
+        goto err;
+    }
+
+    /* Xattr to get the layout for a directory
+     */
+    ret = dict_set_uint32(xattr_req, conf->xattr_name, 4 * 4);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, DHT_MSG_DICT_SET_FAILED,
+               "Failed to set dictionary value:key = %s for "
+               "path %s",
+               conf->xattr_name, loc->path);
+        goto err;
+    }
+
+    /*Non-fatal failure */
+    ret = dict_set_uint32(xattr_req, conf->mds_xattr_key, 4);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, DHT_MSG_DICT_SET_FAILED,
+               "Failed to set dictionary value:key = %s for "
+               "path %s",
+               conf->mds_xattr_key, loc->path);
+    }
+
+    dht_check_and_set_acl_xattr_req(this, xattr_req);
+    ret = 0;
+err:
+    return ret;
+}
 
-                STACK_WIND (frame, dht_lookup_cbk,
-                            hashed_subvol, hashed_subvol->fops->lookup,
-                            loc, local->xattr_req);
+/* If the hashed subvol is present, send the lookup to only that subvol first.
+ * If no hashed subvol, send a lookup to all subvols and proceed based on the
+ * responses.
+ */
+static int
+dht_do_fresh_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc)
+{
+    int ret = -1;
+    dht_conf_t *conf = NULL;
+    xlator_t *hashed_subvol = NULL;
+    dht_local_t *local = NULL;
+    int op_errno = -1;
+    int call_cnt = 0;
+    int i = 0;
+
+    conf = this->private;
+    if (!conf) {
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    local = frame->local;
+    if (!local) {
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    /* Since we don't know whether this is a file or a directory,
+     * request all xattrs*/
+    ret = dht_set_file_xattr_req(this, loc, local->xattr_req);
+    if (ret) {
+        op_errno = -ret;
+        goto err;
+    }
+
+    ret = dht_set_dir_xattr_req(this, loc, local->xattr_req);
+    if (ret) {
+        op_errno = -ret;
+        goto err;
+    }
+
+    /* Fuse sets a random value in gfid-req. If the gfid is missing
+     * on one or more subvols, posix will set the gfid to this value,
+     * causing GFID mismatches for directories. Remove the value fuse
+     * has sent before sending the lookup.
+     */
+    ret = dict_get_gfuuid(local->xattr_req, "gfid-req", &local->gfid_req);
+    if (ret) {
+        gf_msg_debug(this->name, 0, "%s: No gfid-req available", loc->path);
+    } else {
+        dict_del(local->xattr_req, "gfid-req");
+    }
+    /* This should have been set in dht_lookup */
+    hashed_subvol = local->hashed_subvol;
+
+    if (!hashed_subvol) {
+        gf_msg_debug(this->name, 0,
+                     "%s: no subvolume in layout for path, "
+                     "checking on all the subvols to see if "
+                     "it is a directory",
+                     loc->path);
+
+        call_cnt = conf->subvolume_cnt;
+        local->call_cnt = call_cnt;
+
+        /* Allocate a layout. This will be populated and saved in
+         * the dht inode_ctx on successful lookup
+         */
+        local->layout = dht_layout_new(this, conf->subvolume_cnt);
+        if (!local->layout) {
+            op_errno = ENOMEM;
+            goto err;
         }
 
+        gf_msg_debug(this->name, 0,
+                     "%s: Found null hashed subvol. Calling lookup"
+                     " on all nodes.",
+                     loc->path);
+
+        for (i = 0; i < call_cnt; i++) {
+            STACK_WIND_COOKIE(frame, dht_lookup_dir_cbk, conf->subvolumes[i],
+                              conf->subvolumes[i],
+                              conf->subvolumes[i]->fops->lookup, &local->loc,
+                              local->xattr_req);
+        }
         return 0;
+    }
+
+    /* if the hashed_subvol is non-null, send the lookup there first so
+     * as to see whether we have a file or a directory */
+    gf_msg_debug(this->name, 0, "%s: Calling fresh lookup on %s", loc->path,
+                 hashed_subvol->name);
 
+    STACK_WIND_COOKIE(frame, dht_lookup_cbk, hashed_subvol, hashed_subvol,
+                      hashed_subvol->fops->lookup, loc, local->xattr_req);
+    return 0;
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (lookup, frame, -1, op_errno, NULL, NULL, NULL,
-                          NULL);
-        return 0;
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+    return 0;
 }
 
-
-int
-dht_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                int op_ret, int op_errno, struct iatt *preparent,
-                struct iatt *postparent, dict_t *xdata)
+static int
+dht_do_revalidate(call_frame_t *frame, xlator_t *this, loc_t *loc)
 {
-        dht_local_t  *local = NULL;
-        call_frame_t *prev  = NULL;
-
-        local = frame->local;
-        prev  = cookie;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        local->op_ret   = -1;
-                        local->op_errno = op_errno;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "subvolume %s returned -1 (%s)",
-                                prev->this->name, strerror (op_errno));
-                        goto unlock;
-                }
+    xlator_t *subvol = NULL;
+    xlator_t *mds_subvol = NULL;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
+    int op_errno = -1;
+    dht_layout_t *layout = NULL;
+    int i = 0;
+    int call_cnt = 0;
+    int gen = 0;
+
+    conf = this->private;
+    if (!conf) {
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    local = frame->local;
+    if (!local) {
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    layout = local->layout;
+    if (!layout) {
+        gf_msg_debug(this->name, 0,
+                     "path = %s. No layout found in the inode ctx.", loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    /* Generation number has changed. This layout may be stale. */
+    if (layout->gen && (layout->gen < conf->gen)) {
+        gen = layout->gen;
+        dht_layout_unref(this, local->layout);
+        local->layout = NULL;
+        local->cached_subvol = NULL;
+
+        gf_msg_debug(this->name, 0,
+                     "path = %s. In memory layout may be stale."
+                     "(layout->gen (%d) is less than "
+                     "conf->gen (%d)). Calling fresh lookup.",
+                     loc->path, gen, conf->gen);
+
+        dht_do_fresh_lookup(frame, this, loc);
+        return 0;
+    }
+
+    local->inode = inode_ref(loc->inode);
+
+    /* Since we don't know whether this has changed,
+     * request all xattrs*/
+    ret = dht_set_file_xattr_req(this, loc, local->xattr_req);
+    if (ret) {
+        op_errno = -ret;
+        goto err;
+    }
+
+    ret = dht_set_dir_xattr_req(this, loc, local->xattr_req);
+    if (ret) {
+        op_errno = -ret;
+        goto err;
+    }
+
+    if (IA_ISDIR(local->inode->ia_type)) {
+        ret = dht_inode_ctx_mdsvol_get(local->inode, this, &mds_subvol);
+        if (ret || !mds_subvol) {
+            gf_msg_debug(this->name, 0, "path = %s. No mds subvol in inode ctx",
+                         local->loc.path);
+        }
+        local->mds_subvol = mds_subvol;
+        local->call_cnt = conf->subvolume_cnt;
 
-                local->op_ret = 0;
+        /* local->call_cnt will change as responses are processed. Always use a
+         * local copy to loop through the STACK_WIND calls
+         */
 
-                local->postparent = *postparent;
-                local->preparent = *preparent;
+        call_cnt = local->call_cnt;
 
-                if (local->loc.parent) {
-                        dht_inode_ctx_time_update (local->loc.parent, this,
-                                                   &local->preparent, 0);
-                        dht_inode_ctx_time_update (local->loc.parent, this,
-                                                   &local->postparent, 1);
-                }
+        for (i = 0; i < call_cnt; i++) {
+            STACK_WIND_COOKIE(frame, dht_revalidate_cbk, conf->subvolumes[i],
+                              conf->subvolumes[i],
+                              conf->subvolumes[i]->fops->lookup, loc,
+                              local->xattr_req);
         }
-unlock:
-        UNLOCK (&frame->lock);
+        return 0;
+    }
 
-        DHT_STACK_UNWIND (unlink, frame, local->op_ret, local->op_errno,
-                          &local->preparent, &local->postparent, NULL);
+    /* If not a dir, this should be 1 */
+    local->call_cnt = layout->cnt;
+    call_cnt = local->call_cnt;
 
-        return 0;
+    for (i = 0; i < call_cnt; i++) {
+        subvol = layout->list[i].xlator;
+
+        gf_msg_debug(this->name, 0,
+                     "path = %s. Calling "
+                     "revalidate lookup on %s",
+                     loc->path, subvol->name);
+
+        STACK_WIND_COOKIE(frame, dht_revalidate_cbk, subvol, subvol,
+                          subvol->fops->lookup, &local->loc, local->xattr_req);
+    }
+    return 0;
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+    return 0;
 }
 
+/* Depending on the input, decide if this is a:
+ * fresh-lookup: loc->name is provided but no dht inode ctx
+ * revalidation: loc->name is provided, dht inode ctx is present
+ * discover: gfid based nameless lookup.
+ */
 
 int
-dht_unlink_linkfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int op_ret, int op_errno, struct iatt *preparent,
-                         struct iatt *postparent, dict_t *xdata)
+dht_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req)
 {
-        dht_local_t  *local = NULL;
-        call_frame_t *prev = NULL;
+    xlator_t *hashed_subvol = NULL;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
+    int op_errno = -1;
+    loc_t new_loc = {
+        0,
+    };
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+
+    conf = this->private;
+    if (!conf)
+        goto err;
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_LOOKUP);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    ret = dht_filter_loc_subvol_key(this, loc, &new_loc, &hashed_subvol);
+    if (ret) {
+        loc_wipe(&local->loc);
+        ret = loc_dup(&new_loc, &local->loc);
+
+        /* we no longer need 'new_loc' entries */
+        loc_wipe(&new_loc);
+
+        /* check if loc_dup() is successful */
+        if (ret == -1) {
+            op_errno = errno;
+            gf_msg_debug(this->name, errno,
+                         "copying location failed for path=%s", loc->path);
+            goto err;
+        }
+    }
+
+    if (xattr_req) {
+        local->xattr_req = dict_ref(xattr_req);
+    } else {
+        local->xattr_req = dict_new();
+    }
+
+    /* Nameless lookup */
+
+    /* This is usually sent by NFS. Lookups are done based on the gfid and
+     * no name information is available. Without the name, dht cannot calculate
+     * the hash and has to send a lookup to all subvols.
+     */
+    if (gf_uuid_is_null(loc->pargfid) && !gf_uuid_is_null(loc->gfid) &&
+        !__is_root_gfid(loc->inode->gfid)) {
+        local->cached_subvol = NULL;
+        dht_do_discover(frame, this, loc);
+        return 0;
+    }
 
-        xlator_t *cached_subvol = NULL;
+    if (loc_is_root(loc)) {
+        /* Request the DHT commit hash xattr (trusted.glusterfs.dht.commithash)
+         * set on the brick root.
+         */
+        ret = dict_set_uint32(local->xattr_req, conf->commithash_xattr_name,
+                              sizeof(uint32_t));
+    }
 
-        local = frame->local;
-        prev  = cookie;
-
-        LOCK (&frame->lock);
-        {
-                if ((op_ret == -1) && !((op_errno == ENOENT) ||
-                                        (op_errno == ENOTCONN))) {
-                        local->op_errno = op_errno;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "subvolume %s returned -1 (%s)",
-                                prev->this->name, strerror (op_errno));
-                        goto unlock;
-                }
+    if (!hashed_subvol)
+        hashed_subvol = dht_subvol_get_hashed(this, loc);
+    local->hashed_subvol = hashed_subvol;
 
-                local->op_ret = 0;
-        }
-unlock:
-        UNLOCK (&frame->lock);
+    if (is_revalidate(loc)) {
+        /* The entry has been looked up before and has a dht inode_ctx
+         */
+        dht_do_revalidate(frame, this, loc);
+        return 0;
+    } else {
+        /* Entry has not been looked up before
+         */
+        dht_do_fresh_lookup(frame, this, loc);
+        return 0;
+    }
 
-        if (local->op_ret == -1)
-                goto err;
+    return 0;
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+    return 0;
+}
 
-        cached_subvol = dht_subvol_get_cached (this, local->loc.inode);
-        if (!cached_subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for path=%s",
-                        local->loc.path);
-                local->op_errno = EINVAL;
-                goto err;
-        }
+static int
+dht_unlink_linkfile_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, struct iatt *preparent,
+                        struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
 
-        STACK_WIND (frame, dht_unlink_cbk,
-                    cached_subvol, cached_subvol->fops->unlink,
-                    &local->loc, local->flags, NULL);
+    local = frame->local;
+    prev = cookie;
 
-        return 0;
+    LOCK(&frame->lock);
+    {
+        if ((op_ret == -1) &&
+            !((op_errno == ENOENT) || (op_errno == ENOTCONN))) {
+            local->op_errno = op_errno;
+            UNLOCK(&frame->lock);
+            gf_msg_debug(this->name, op_errno,
+                         "Unlink link: subvolume %s returned -1", prev->name);
+            goto post_unlock;
+        }
 
-err:
-        DHT_STACK_UNWIND (unlink, frame, -1, local->op_errno,
-                          NULL, NULL, NULL);
-        return 0;
+        local->op_ret = 0;
+    }
+    UNLOCK(&frame->lock);
+post_unlock:
+    dht_set_fixed_dir_stat(&local->preparent);
+    dht_set_fixed_dir_stat(&local->postparent);
+    DHT_STACK_UNWIND(unlink, frame, local->op_ret, local->op_errno,
+                     &local->preparent, &local->postparent, xdata);
+
+    return 0;
 }
 
-int
-dht_err_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-             int op_ret, int op_errno, dict_t *xdata)
+static int
+dht_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, struct iatt *preparent, struct iatt *postparent,
+               dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
-        int           this_call_cnt = 0;
-        call_frame_t *prev = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    xlator_t *hashed_subvol = NULL;
 
-        local = frame->local;
-        prev = cookie;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        local->op_errno = op_errno;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "subvolume %s returned -1 (%s)",
-                                prev->this->name, strerror (op_errno));
-                        goto unlock;
-                }
+    local = frame->local;
+    prev = cookie;
 
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            if (op_errno != ENOENT) {
+                local->op_ret = -1;
+                local->op_errno = op_errno;
+            } else {
                 local->op_ret = 0;
+            }
+            UNLOCK(&frame->lock);
+            gf_msg_debug(this->name, op_errno,
+                         "Unlink: subvolume %s returned -1", prev->name);
+            goto post_unlock;
         }
-unlock:
-        UNLOCK (&frame->lock);
 
-        this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt)) {
-                DHT_STACK_UNWIND (setxattr, frame, local->op_ret,
-                                  local->op_errno, NULL);
-        }
+        local->op_ret = 0;
 
-        return 0;
+        local->postparent = *postparent;
+        local->preparent = *preparent;
+
+        if (local->loc.parent) {
+            dht_inode_ctx_time_update(local->loc.parent, this,
+                                      &local->preparent, 0);
+            dht_inode_ctx_time_update(local->loc.parent, this,
+                                      &local->postparent, 1);
+        }
+    }
+    UNLOCK(&frame->lock);
+post_unlock:
+    if (!local->op_ret) {
+        hashed_subvol = dht_subvol_get_hashed(this, &local->loc);
+        if (hashed_subvol && hashed_subvol != local->cached_subvol) {
+            /*
+             * If hashed and cached are different, then we need
+             * to unlink linkfile from hashed subvol if data
+             * file is deleted successfully
+             */
+            STACK_WIND_COOKIE(frame, dht_unlink_linkfile_cbk, hashed_subvol,
+                              hashed_subvol, hashed_subvol->fops->unlink,
+                              &local->loc, local->flags, xdata);
+            return 0;
+        }
+    }
+
+    dht_set_fixed_dir_stat(&local->preparent);
+    dht_set_fixed_dir_stat(&local->postparent);
+    DHT_STACK_UNWIND(unlink, frame, local->op_ret, local->op_errno,
+                     &local->preparent, &local->postparent, xdata);
+
+    return 0;
 }
 
-static void
-fill_layout_info (dht_layout_t *layout, char *buf)
+static int
+dht_common_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        int i = 0;
-        char tmp_buf[128] = {0,};
+    DHT_STACK_UNWIND(setxattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
-        for (i = 0; i < layout->cnt; i++) {
-                snprintf (tmp_buf, 128, "(%s %u %u)",
-                          layout->list[i].xlator->name,
-                          layout->list[i].start,
-                          layout->list[i].stop);
-                if (i)
-                        strcat (buf, " ");
-                strcat (buf, tmp_buf);
+static int
+dht_fix_layout_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
+
+    if (op_ret == 0) {
+        /* update the layout in the inode ctx */
+        local = frame->local;
+        layout = local->selfheal.layout;
+
+        dht_layout_set(this, local->loc.inode, layout);
+    }
+
+    DHT_STACK_UNWIND(setxattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+static int
+dht_err_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+            int op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+
+    local = frame->local;
+    prev = cookie;
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            local->op_errno = op_errno;
+            UNLOCK(&frame->lock);
+            gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                         prev->name);
+            goto post_unlock;
         }
+
+        local->op_ret = 0;
+    }
+    UNLOCK(&frame->lock);
+post_unlock:
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        if ((local->fop == GF_FOP_SETXATTR) ||
+            (local->fop == GF_FOP_FSETXATTR)) {
+            DHT_STACK_UNWIND(setxattr, frame, local->op_ret, local->op_errno,
+                             NULL);
+            /* 'local' itself may not be valid after this */
+            goto out;
+        }
+        if ((local->fop == GF_FOP_REMOVEXATTR) ||
+            (local->fop == GF_FOP_FREMOVEXATTR)) {
+            DHT_STACK_UNWIND(removexattr, frame, local->op_ret, local->op_errno,
+                             NULL);
+        }
+    }
+
+out:
+    return 0;
 }
 
-void
-dht_fill_pathinfo_xattr (xlator_t *this, dht_local_t *local,
-                         char *xattr_buf, int32_t alloc_len,
-                         int flag, char *layout_buf)
-{
-        if (flag && local->xattr_val)
-                snprintf (xattr_buf, alloc_len,
-                          "((<"DHT_PATHINFO_HEADER"%s> %s) (%s-layout %s))",
-                          this->name, local->xattr_val, this->name,
-                          layout_buf);
-        else if (local->xattr_val)
-                snprintf (xattr_buf, alloc_len,
-                          "(<"DHT_PATHINFO_HEADER"%s> %s)",
-                          this->name, local->xattr_val);
-        else if (flag)
-                snprintf (xattr_buf, alloc_len, "(%s-layout %s)",
-                          this->name, layout_buf);
+/* Set the value[] of key into dict after convert from
+   host byte order to network byte order
+*/
+int32_t
+dht_dict_set_array(dict_t *dict, char *key, int32_t value[], int32_t size)
+{
+    int ret = -1;
+    int32_t *ptr = NULL;
+    int32_t vindex;
+
+    if (value == NULL) {
+        return -EINVAL;
+    }
+
+    ptr = GF_MALLOC(sizeof(int32_t) * size, gf_common_mt_char);
+    if (ptr == NULL) {
+        return -ENOMEM;
+    }
+    for (vindex = 0; vindex < size; vindex++) {
+        ptr[vindex] = hton32(value[vindex]);
+    }
+    ret = dict_set_bin(dict, key, ptr, sizeof(int32_t) * size);
+    if (ret)
+        GF_FREE(ptr);
+    return ret;
 }
 
-int
-dht_vgetxattr_alloc_and_fill (dht_local_t *local, dict_t *xattr, xlator_t *this,
-                              int op_errno)
+static int
+dht_common_mds_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int32_t op_ret, int32_t op_errno, dict_t *dict,
+                           dict_t *xdata)
 {
-        int      ret       = -1;
-        char    *value     = NULL;
-        int32_t  plen      = 0;
+    dht_local_t *local = NULL;
+    call_frame_t *prev = cookie;
 
-        ret = dict_get_str (xattr, local->xsel, &value);
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Subvolume %s returned -1 (%s)", this->name,
-                        strerror (op_errno));
-                local->op_ret = -1;
-                local->op_errno = op_errno;
+    local = frame->local;
+
+    if (op_ret)
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->this->name);
+
+    if (local->fop == GF_FOP_SETXATTR) {
+        DHT_STACK_UNWIND(setxattr, frame, 0, op_errno, local->xdata);
+        /* 'local' itself may not be valid after this */
+        goto out;
+    }
+
+    if (local->fop == GF_FOP_FSETXATTR) {
+        DHT_STACK_UNWIND(fsetxattr, frame, 0, op_errno, local->xdata);
+        /* 'local' itself may not be valid after this */
+        goto out;
+    }
+
+    if (local->fop == GF_FOP_REMOVEXATTR) {
+        DHT_STACK_UNWIND(removexattr, frame, 0, op_errno, NULL);
+        /* 'local' itself may not be valid after this */
+        goto out;
+    }
+
+    if (local->fop == GF_FOP_FREMOVEXATTR) {
+        DHT_STACK_UNWIND(fremovexattr, frame, 0, op_errno, NULL);
+    }
+
+out:
+    return 0;
+}
+
+/* Code to wind a xattrop call to add 1 on current mds internal xattr
+   value
+*/
+static int
+dht_setxattr_non_mds_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int op_ret, int op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    int ret = 0;
+    dict_t *xattrop = NULL;
+    int32_t addone[1] = {1};
+    call_frame_t *prev = NULL;
+    dht_conf_t *conf = NULL;
+
+    local = frame->local;
+    prev = cookie;
+    conf = this->private;
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret && !local->op_ret) {
+            local->op_ret = op_ret;
+            local->op_errno = op_errno;
+            UNLOCK(&frame->lock);
+            gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                         prev->this->name);
+            goto post_unlock;
+        }
+    }
+    UNLOCK(&frame->lock);
+post_unlock:
+    this_call_cnt = dht_frame_return(frame);
+
+    if (is_last_call(this_call_cnt)) {
+        if (!local->op_ret) {
+            xattrop = dict_new();
+            if (!xattrop) {
+                gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_NO_MEMORY, 0,
+                       "dictionary creation failed");
+                ret = -1;
+                goto out;
+            }
+            ret = dht_dict_set_array(xattrop, conf->mds_xattr_key, addone, 1);
+            if (ret != 0) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                       "dictionary set array failed ");
+                ret = -1;
                 goto out;
+            }
+            if ((local->fop == GF_FOP_SETXATTR) ||
+                (local->fop == GF_FOP_REMOVEXATTR)) {
+                STACK_WIND(frame, dht_common_mds_xattrop_cbk, local->mds_subvol,
+                           local->mds_subvol->fops->xattrop, &local->loc,
+                           GF_XATTROP_ADD_ARRAY, xattrop, NULL);
+            } else {
+                STACK_WIND(frame, dht_common_mds_xattrop_cbk, local->mds_subvol,
+                           local->mds_subvol->fops->fxattrop, local->fd,
+                           GF_XATTROP_ADD_ARRAY, xattrop, NULL);
+            }
+        } else {
+            if (local->fop == GF_FOP_SETXATTR) {
+                DHT_STACK_UNWIND(setxattr, frame, 0, 0, local->xdata);
+                /* 'local' itself may not be valid after this */
+                goto just_return;
+            }
+
+            if (local->fop == GF_FOP_FSETXATTR) {
+                DHT_STACK_UNWIND(fsetxattr, frame, 0, 0, local->xdata);
+                /* 'local' itself may not be valid after this */
+                goto just_return;
+            }
+
+            if (local->fop == GF_FOP_REMOVEXATTR) {
+                DHT_STACK_UNWIND(removexattr, frame, 0, 0, NULL);
+                /* 'local' itself may not be valid after this */
+                goto just_return;
+            }
+
+            if (local->fop == GF_FOP_FREMOVEXATTR) {
+                DHT_STACK_UNWIND(fremovexattr, frame, 0, 0, NULL);
+                /* 'local' itself may not be valid after this */
+                goto just_return;
+            }
+        }
+    }
+out:
+    if (ret) {
+        if (local->fop == GF_FOP_SETXATTR) {
+            DHT_STACK_UNWIND(setxattr, frame, 0, 0, local->xdata);
+            /* 'local' itself may not be valid after this */
+            goto just_return;
         }
 
-        local->alloc_len += strlen(value);
-
-        if (!local->xattr_val) {
-                local->alloc_len += (strlen (DHT_PATHINFO_HEADER) + 10);
-                local->xattr_val = GF_CALLOC (local->alloc_len, sizeof (char),
-                                              gf_common_mt_char);
-                if (!local->xattr_val) {
-                        ret = -1;
-                        goto out;
-                }
+        if (local->fop == GF_FOP_FSETXATTR) {
+            DHT_STACK_UNWIND(fsetxattr, frame, 0, 0, local->xdata);
+            /* 'local' itself may not be valid after this */
+            goto just_return;
         }
 
-        if (local->xattr_val) {
-                plen = strlen (local->xattr_val);
-                if (plen) {
-                        /* extra byte(s) for \0 to be safe */
-                        local->alloc_len += (plen + 2);
-                        local->xattr_val = GF_REALLOC (local->xattr_val,
-                                                       local->alloc_len);
-                        if (!local->xattr_val) {
-                                ret = -1;
-                                goto out;
-                        }
-                }
+        if (local->fop == GF_FOP_REMOVEXATTR) {
+            DHT_STACK_UNWIND(removexattr, frame, 0, 0, NULL);
+            /* 'local' itself may not be valid after this */
+            goto just_return;
+        }
 
-                (void) strcat (local->xattr_val, value);
-                (void) strcat (local->xattr_val, " ");
-                local->op_ret = 0;
+        if (local->fop == GF_FOP_FREMOVEXATTR) {
+            DHT_STACK_UNWIND(fremovexattr, frame, 0, 0, NULL);
         }
+    }
+just_return:
+    if (xattrop)
+        dict_unref(xattrop);
+    return 0;
+}
 
-        ret = 0;
+static int
+dht_setxattr_mds_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    call_frame_t *prev = NULL;
+    xlator_t *mds_subvol = NULL;
+    int i = 0;
+
+    local = frame->local;
+    prev = cookie;
+    conf = this->private;
+    mds_subvol = local->mds_subvol;
+
+    if (op_ret == -1) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->this->name);
+        goto out;
+    }
+
+    local->op_ret = 0;
+    local->call_cnt = conf->subvolume_cnt - 1;
+    local->xdata = dict_ref(xdata);
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (mds_subvol && (mds_subvol == conf->subvolumes[i]))
+            continue;
+        if (local->fop == GF_FOP_SETXATTR) {
+            STACK_WIND(frame, dht_setxattr_non_mds_cbk, conf->subvolumes[i],
+                       conf->subvolumes[i]->fops->setxattr, &local->loc,
+                       local->xattr, local->flags, local->xattr_req);
+        }
+
+        if (local->fop == GF_FOP_FSETXATTR) {
+            STACK_WIND(frame, dht_setxattr_non_mds_cbk, conf->subvolumes[i],
+                       conf->subvolumes[i]->fops->fsetxattr, local->fd,
+                       local->xattr, local->flags, local->xattr_req);
+        }
+
+        if (local->fop == GF_FOP_REMOVEXATTR) {
+            STACK_WIND(frame, dht_setxattr_non_mds_cbk, conf->subvolumes[i],
+                       conf->subvolumes[i]->fops->removexattr, &local->loc,
+                       local->key, local->xattr_req);
+        }
+
+        if (local->fop == GF_FOP_FREMOVEXATTR) {
+            STACK_WIND(frame, dht_setxattr_non_mds_cbk, conf->subvolumes[i],
+                       conf->subvolumes[i]->fops->fremovexattr, local->fd,
+                       local->key, local->xattr_req);
+        }
+    }
+
+    return 0;
+out:
+    if (local->fop == GF_FOP_SETXATTR) {
+        DHT_STACK_UNWIND(setxattr, frame, local->op_ret, local->op_errno,
+                         xdata);
+        /* 'local' itself may not be valid after this */
+        goto just_return;
+    }
+
+    if (local->fop == GF_FOP_FSETXATTR) {
+        DHT_STACK_UNWIND(fsetxattr, frame, local->op_ret, local->op_errno,
+                         xdata);
+        /* 'local' itself may not be valid after this */
+        goto just_return;
+    }
+
+    if (local->fop == GF_FOP_REMOVEXATTR) {
+        DHT_STACK_UNWIND(removexattr, frame, local->op_ret, local->op_errno,
+                         NULL);
+        /* 'local' itself may not be valid after this */
+        goto just_return;
+    }
+
+    if (local->fop == GF_FOP_FREMOVEXATTR) {
+        DHT_STACK_UNWIND(fremovexattr, frame, local->op_ret, local->op_errno,
+                         NULL);
+    }
+
+just_return:
+    return 0;
+}
 
- out:
-        return ret;
+static int
+dht_xattrop_mds_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int op_ret, int op_errno, dict_t *dict, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    call_frame_t *prev = NULL;
+
+    local = frame->local;
+    prev = cookie;
+
+    if (op_ret == -1) {
+        local->op_errno = op_errno;
+        local->op_ret = op_ret;
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->this->name);
+        goto out;
+    }
+
+    if (local->fop == GF_FOP_SETXATTR) {
+        STACK_WIND(frame, dht_setxattr_mds_cbk, local->mds_subvol,
+                   local->mds_subvol->fops->setxattr, &local->loc, local->xattr,
+                   local->flags, local->xattr_req);
+    }
+
+    if (local->fop == GF_FOP_FSETXATTR) {
+        STACK_WIND(frame, dht_setxattr_mds_cbk, local->mds_subvol,
+                   local->mds_subvol->fops->fsetxattr, local->fd, local->xattr,
+                   local->flags, local->xattr_req);
+    }
+
+    if (local->fop == GF_FOP_REMOVEXATTR) {
+        STACK_WIND(frame, dht_setxattr_mds_cbk, local->mds_subvol,
+                   local->mds_subvol->fops->removexattr, &local->loc,
+                   local->key, local->xattr_req);
+    }
+
+    if (local->fop == GF_FOP_FREMOVEXATTR) {
+        STACK_WIND(frame, dht_setxattr_mds_cbk, local->mds_subvol,
+                   local->mds_subvol->fops->fremovexattr, local->fd, local->key,
+                   local->xattr_req);
+    }
+
+    return 0;
+out:
+    if (local->fop == GF_FOP_SETXATTR) {
+        DHT_STACK_UNWIND(setxattr, frame, local->op_ret, local->op_errno,
+                         xdata);
+        /* 'local' itself may not be valid after this */
+        goto just_return;
+    }
+
+    if (local->fop == GF_FOP_FSETXATTR) {
+        DHT_STACK_UNWIND(fsetxattr, frame, local->op_ret, local->op_errno,
+                         xdata);
+        /* 'local' itself may not be valid after this */
+        goto just_return;
+    }
+
+    if (local->fop == GF_FOP_REMOVEXATTR) {
+        DHT_STACK_UNWIND(removexattr, frame, local->op_ret, local->op_errno,
+                         NULL);
+        /* 'local' itself may not be valid after this */
+        goto just_return;
+    }
+
+    if (local->fop == GF_FOP_FREMOVEXATTR) {
+        DHT_STACK_UNWIND(fremovexattr, frame, local->op_ret, local->op_errno,
+                         NULL);
+    }
+
+just_return:
+    return 0;
 }
 
-int
-dht_vgetxattr_fill_and_set (dht_local_t *local, dict_t **dict, xlator_t *this,
-                            gf_boolean_t flag)
+static void
+fill_layout_info(dht_layout_t *layout, char *buf)
 {
-        int   ret             = -1;
-        char *xattr_buf       = NULL;
-        char layout_buf[8192] = {0,};
+    int i = 0;
+    char tmp_buf[128] = {
+        0,
+    };
+
+    for (i = 0; i < layout->cnt; i++) {
+        snprintf(tmp_buf, sizeof(tmp_buf), "(%s %u %u)",
+                 layout->list[i].xlator->name, layout->list[i].start,
+                 layout->list[i].stop);
+        if (i)
+            strcat(buf, " ");
+        strcat(buf, tmp_buf);
+    }
+}
 
-        if (flag)
-                fill_layout_info (local->layout, layout_buf);
+static void
+dht_fill_pathinfo_xattr(xlator_t *this, dht_local_t *local, char *xattr_buf,
+                        int32_t alloc_len, int flag, char *layout_buf)
+{
+    if (flag) {
+        if (local->xattr_val) {
+            snprintf(xattr_buf, alloc_len,
+                     "((<" DHT_PATHINFO_HEADER "%s> %s) (%s-layout %s))",
+                     this->name, local->xattr_val, this->name, layout_buf);
+        } else {
+            snprintf(xattr_buf, alloc_len, "(%s-layout %s)", this->name,
+                     layout_buf);
+        }
+    } else if (local->xattr_val) {
+        snprintf(xattr_buf, alloc_len, "(<" DHT_PATHINFO_HEADER "%s> %s)",
+                 this->name, local->xattr_val);
+    } else {
+        xattr_buf[0] = '\0';
+    }
+}
 
-        *dict = dict_new ();
-        if (!*dict)
-                goto out;
+static int
+dht_vgetxattr_alloc_and_fill(dht_local_t *local, dict_t *xattr, xlator_t *this,
+                             int op_errno)
+{
+    int ret = -1;
+    char *value = NULL;
+
+    ret = dict_get_str(xattr, local->xsel, &value);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, DHT_MSG_GET_XATTR_FAILED,
+               "Subvolume %s returned -1", this->name);
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+        goto out;
+    }
+
+    local->alloc_len += strlen(value);
+
+    if (!local->xattr_val) {
+        local->alloc_len += (SLEN(DHT_PATHINFO_HEADER) + 10);
+        local->xattr_val = GF_MALLOC(local->alloc_len, gf_common_mt_char);
+        if (!local->xattr_val) {
+            ret = -1;
+            goto out;
+        }
+        local->xattr_val[0] = '\0';
+    }
 
-        local->xattr_val[strlen (local->xattr_val) - 1] = '\0';
+    int plen = strlen(local->xattr_val);
+    if (plen) {
+        /* extra byte(s) for \0 to be safe */
+        local->alloc_len += (plen + 2);
+        local->xattr_val = GF_REALLOC(local->xattr_val, local->alloc_len);
+        if (!local->xattr_val) {
+            ret = -1;
+            goto out;
+        }
+    }
 
-        /* we would need max this many bytes to create xattr string
-         * extra 40 bytes is just an estimated amount of additional
-         * space required as we include translator name and some
-         * spaces, brackets etc. when forming the pathinfo string.
-         *
-         * For node-uuid we just don't have all the pretty formatting,
-         * but since this is a generic routine for pathinfo & node-uuid
-         * we dont have conditional space allocation and try to be
-         * generic
-         */
-        local->alloc_len += (2 * strlen (this->name))
-                + strlen (layout_buf)
-                + 40;
-        xattr_buf = GF_CALLOC (local->alloc_len, sizeof (char),
-                               gf_common_mt_char);
-        if (!xattr_buf)
-                goto out;
+    (void)strcat(local->xattr_val, value);
+    (void)strcat(local->xattr_val, " ");
+    local->op_ret = 0;
 
-        if (XATTR_IS_PATHINFO (local->xsel)) {
-                (void) dht_fill_pathinfo_xattr (this, local, xattr_buf,
-                                                local->alloc_len, flag,
-                                                layout_buf);
-        } else if (XATTR_IS_NODE_UUID (local->xsel)) {
-                (void) snprintf (xattr_buf, local->alloc_len, "%s",
-                                 local->xattr_val);
-        } else {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "Unknown local->xsel (%s)", local->xsel);
-                goto out;
-        }
+    ret = 0;
 
-        ret = dict_set_dynstr (*dict, local->xsel, xattr_buf);
-        GF_FREE (local->xattr_val);
+out:
+    return ret;
+}
 
- out:
-        return ret;
+static int
+dht_vgetxattr_fill_and_set(dht_local_t *local, dict_t **dict, xlator_t *this,
+                           gf_boolean_t flag)
+{
+    int ret = -1;
+    char *xattr_buf = NULL;
+    char layout_buf[8192] = {
+        0,
+    };
+
+    if (flag)
+        fill_layout_info(local->layout, layout_buf);
+
+    *dict = dict_new();
+    if (!*dict)
+        goto out;
+
+    local->xattr_val[strlen(local->xattr_val) - 1] = '\0';
+
+    /* we would need max this many bytes to create xattr string
+     * extra 40 bytes is just an estimated amount of additional
+     * space required as we include translator name and some
+     * spaces, brackets etc. when forming the pathinfo string.
+     *
+     * For node-uuid we just don't have all the pretty formatting,
+     * but since this is a generic routine for pathinfo & node-uuid
+     * we don't have conditional space allocation and try to be
+     * generic
+     */
+    local->alloc_len += (2 * strlen(this->name)) + strlen(layout_buf) + 40;
+    xattr_buf = GF_MALLOC(local->alloc_len, gf_common_mt_char);
+    if (!xattr_buf)
+        goto out;
+
+    if (XATTR_IS_PATHINFO(local->xsel)) {
+        (void)dht_fill_pathinfo_xattr(this, local, xattr_buf, local->alloc_len,
+                                      flag, layout_buf);
+    } else if ((XATTR_IS_NODE_UUID(local->xsel)) ||
+               (XATTR_IS_NODE_UUID_LIST(local->xsel))) {
+        (void)snprintf(xattr_buf, local->alloc_len, "%s", local->xattr_val);
+    } else {
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_GET_XATTR_FAILED,
+               "Unknown local->xsel (%s)", local->xsel);
+        GF_FREE(xattr_buf);
+        goto out;
+    }
+
+    ret = dict_set_dynstr(*dict, local->xsel, xattr_buf);
+    if (ret)
+        GF_FREE(xattr_buf);
+    GF_FREE(local->xattr_val);
+
+out:
+    return ret;
 }
 
-int
-dht_vgetxattr_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
+static int
+dht_find_local_subvol_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int op_ret, int op_errno, dict_t *xattr,
+                          dict_t *xdata)
 {
-        int          ret           = 0;
-        dht_local_t *local         = NULL;
-        int          this_call_cnt = 0;
-        dict_t      *dict          = NULL;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    xlator_t *prev = NULL;
+    int this_call_cnt = 0;
+    int ret = 0;
+    char *uuid_str = NULL;
+    char *uuid_list = NULL;
+    char *next_uuid_str = NULL;
+    char *saveptr = NULL;
+    uuid_t node_uuid = {
+        0,
+    };
+    char *uuid_list_copy = NULL;
+    int count = 0;
+    int i = 0;
+    int index = 0;
+    int found = 0;
+    nodeuuid_info_t *tmp_ptr = NULL;
+
+    VALIDATE_OR_GOTO(frame, out);
+    VALIDATE_OR_GOTO(frame->local, out);
+
+    local = frame->local;
+    prev = cookie;
+    conf = this->private;
+
+    VALIDATE_OR_GOTO(conf->defrag, out);
+
+    gf_msg_debug(this->name, 0, "subvol %s returned", prev->name);
+
+    LOCK(&frame->lock);
+    {
+        this_call_cnt = --local->call_cnt;
+        if (op_ret < 0) {
+            local->op_ret = -1;
+            local->op_errno = op_errno;
+            UNLOCK(&frame->lock);
+            if (op_errno == ENODATA)
+                gf_msg_debug(this->name, 0, "failed to get node-uuid");
+            else
+                gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                       DHT_MSG_GET_XATTR_FAILED, "failed to get node-uuid");
+            goto post_unlock;
+        }
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (frame->local, out);
+        ret = dict_get_str(xattr, local->xsel, &uuid_list);
 
-        local = frame->local;
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_GET_FAILED,
+                   "Failed to get %s", local->xsel);
+            local->op_ret = -1;
+            local->op_errno = EINVAL;
+            goto unlock;
+        }
 
-        LOCK (&frame->lock);
-        {
-                this_call_cnt = --local->call_cnt;
-                if (op_ret < 0) {
-                        if (op_errno != ENOTCONN) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "getxattr err (%s) for dir",
-                                        strerror (op_errno));
-				local->op_ret = -1;
-				local->op_errno = op_errno;
-                        }
+        /* As DHT will not know details of its child xlators
+         * we need to parse this twice to get the count first
+         * and allocate memory later.
+         */
+        count = 0;
+        index = conf->local_subvols_cnt;
 
-                        goto unlock;
-                }
+        uuid_list_copy = gf_strdup(uuid_list);
+        if (!uuid_list_copy)
+            goto unlock;
 
-                ret = dht_vgetxattr_alloc_and_fill (local, xattr, this,
-                                                    op_errno);
-                if (ret)
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "alloc or fill failure");
-        }
- unlock:
-        UNLOCK (&frame->lock);
+        for (uuid_str = strtok_r(uuid_list, " ", &saveptr); uuid_str;
+             uuid_str = next_uuid_str) {
+            next_uuid_str = strtok_r(NULL, " ", &saveptr);
+            if (gf_uuid_parse(uuid_str, node_uuid)) {
+                local->op_ret = -1;
+                local->op_errno = EINVAL;
+                UNLOCK(&frame->lock);
+                gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_UUID_PARSE_ERROR,
+                       "Failed to parse uuid for %s", prev->name);
+                goto post_unlock;
+            }
+
+            count++;
+            if (gf_uuid_compare(node_uuid, conf->defrag->node_uuid)) {
+                gf_msg_debug(this->name, 0,
+                             "subvol %s does not"
+                             "belong to this node",
+                             prev->name);
+            } else {
+                /* handle multiple bricks of the same replica
+                 * on the same node */
+                if (found)
+                    continue;
+                conf->local_subvols[(conf->local_subvols_cnt)++] = prev;
+                found = 1;
+                gf_msg_debug(this->name, 0,
+                             "subvol %s belongs to"
+                             " this node",
+                             prev->name);
+            }
+        }
+
+        if (!found) {
+            local->op_ret = 0;
+            goto unlock;
+        }
+
+        conf->local_nodeuuids[index].count = count;
+        conf->local_nodeuuids[index].elements = GF_CALLOC(
+            count, sizeof(nodeuuid_info_t), 1);
+
+        /* The node-uuids are guaranteed to be returned in the same
+         * order as the bricks
+         * A null node-uuid is returned for a brick that is down.
+         */
 
-        if (!is_last_call (this_call_cnt))
-                goto out;
+        saveptr = NULL;
+        i = 0;
 
-        /* -- last call: do patch ups -- */
+        for (uuid_str = strtok_r(uuid_list_copy, " ", &saveptr); uuid_str;
+             uuid_str = next_uuid_str) {
+            next_uuid_str = strtok_r(NULL, " ", &saveptr);
+            tmp_ptr = &(conf->local_nodeuuids[index].elements[i]);
+            gf_uuid_parse(uuid_str, tmp_ptr->uuid);
 
-        if (local->op_ret == -1) {
-                goto unwind;
+            if (!gf_uuid_compare(tmp_ptr->uuid, conf->defrag->node_uuid)) {
+                tmp_ptr->info = REBAL_NODEUUID_MINE;
+            }
+            i++;
+            tmp_ptr = NULL;
         }
+    }
 
-        ret = dht_vgetxattr_fill_and_set (local, &dict, this, _gf_true);
-        if (ret)
-                goto unwind;
+    local->op_ret = 0;
+unlock:
+    UNLOCK(&frame->lock);
+post_unlock:
+    if (!is_last_call(this_call_cnt))
+        goto out;
 
-        DHT_STACK_UNWIND (getxattr, frame, 0, 0, dict, xdata);
-        goto cleanup;
+    if (local->op_ret == -1) {
+        goto unwind;
+    }
 
- unwind:
-        DHT_STACK_UNWIND (getxattr, frame, -1, local->op_errno, NULL, NULL);
- cleanup:
-        if (dict)
-                dict_unref (dict);
- out:
-        return 0;
+    DHT_STACK_UNWIND(getxattr, frame, 0, 0, xattr, xdata);
+    goto out;
+
+unwind:
+
+    GF_FREE(conf->local_nodeuuids[index].elements);
+    conf->local_nodeuuids[index].elements = NULL;
+
+    DHT_STACK_UNWIND(getxattr, frame, -1, local->op_errno, NULL, xdata);
+out:
+    GF_FREE(uuid_list_copy);
+    return 0;
 }
 
-int
-dht_vgetxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
+static int
+dht_vgetxattr_dir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
 {
-        dht_local_t  *local         = NULL;
-        int           ret           = 0;
-        dict_t       *dict          = NULL;
-        call_frame_t *prev          = NULL;
-        gf_boolean_t  flag          = _gf_true;
+    int ret = 0;
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    dict_t *dict = NULL;
 
-        local = frame->local;
-        prev = cookie;
+    VALIDATE_OR_GOTO(frame, out);
+    VALIDATE_OR_GOTO(frame->local, out);
 
+    local = frame->local;
+
+    LOCK(&frame->lock);
+    {
+        this_call_cnt = --local->call_cnt;
         if (op_ret < 0) {
+            if (op_errno != ENOTCONN) {
                 local->op_ret = -1;
                 local->op_errno = op_errno;
-                gf_log (this->name, GF_LOG_ERROR, "Subvolume %s returned -1 "
-                        "(%s)", prev->this->name, strerror (op_errno));
-                goto unwind;
+                UNLOCK(&frame->lock);
+                gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                       DHT_MSG_GET_XATTR_FAILED, "getxattr err for dir");
+                goto post_unlock;
+            }
+
+            goto unlock;
         }
 
-        ret = dht_vgetxattr_alloc_and_fill (local, xattr, this,
-                                            op_errno);
+        ret = dht_vgetxattr_alloc_and_fill(local, xattr, this, op_errno);
         if (ret) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "alloc or fill failure");
-                goto unwind;
+            UNLOCK(&frame->lock);
+            gf_msg(this->name, GF_LOG_ERROR, op_errno, DHT_MSG_DICT_SET_FAILED,
+                   "alloc or fill failure");
+            goto post_unlock;
         }
+    }
+unlock:
+    UNLOCK(&frame->lock);
+post_unlock:
+    if (!is_last_call(this_call_cnt))
+        goto out;
 
-        flag = (local->layout->cnt > 1) ? _gf_true : _gf_false;
+    /* -- last call: do patch ups -- */
 
-        ret = dht_vgetxattr_fill_and_set (local, &dict, this, flag);
-        if (ret)
-                goto unwind;
+    if (local->op_ret == -1) {
+        goto unwind;
+    }
 
-        DHT_STACK_UNWIND (getxattr, frame, 0, 0, dict, xdata);
-        goto cleanup;
+    ret = dht_vgetxattr_fill_and_set(local, &dict, this, _gf_true);
+    if (ret)
+        goto unwind;
 
- unwind:
-        DHT_STACK_UNWIND (getxattr, frame, -1, local->op_errno,
-                          NULL, NULL);
- cleanup:
-        if (dict)
-                dict_unref (dict);
+    DHT_STACK_UNWIND(getxattr, frame, 0, 0, dict, xdata);
+    goto cleanup;
 
-        return 0;
+unwind:
+    DHT_STACK_UNWIND(getxattr, frame, -1, local->op_errno, NULL, NULL);
+cleanup:
+    if (dict)
+        dict_unref(dict);
+out:
+    return 0;
 }
 
-int
-dht_linkinfo_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                           int op_ret, int op_errno, dict_t *xattr,
-                           dict_t *xdata)
+static int
+dht_vgetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                  int op_errno, dict_t *xattr, dict_t *xdata)
 {
-        int   ret   = 0;
-        char *value = NULL;
+    dht_local_t *local = NULL;
+    int ret = 0;
+    dict_t *dict = NULL;
+    xlator_t *prev = NULL;
+    gf_boolean_t flag = _gf_true;
+
+    local = frame->local;
+    prev = cookie;
+
+    if (op_ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, DHT_MSG_GET_XATTR_FAILED,
+               "vgetxattr: Subvolume %s returned -1", prev->name);
+        goto unwind;
+    }
+
+    ret = dht_vgetxattr_alloc_and_fill(local, xattr, this, op_errno);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_NO_MEMORY,
+               "Allocation or fill failure");
+        goto unwind;
+    }
+
+    flag = (local->layout->cnt > 1) ? _gf_true : _gf_false;
+
+    ret = dht_vgetxattr_fill_and_set(local, &dict, this, flag);
+    if (ret)
+        goto unwind;
+
+    DHT_STACK_UNWIND(getxattr, frame, 0, 0, dict, xdata);
+    goto cleanup;
 
-        if (op_ret != -1) {
-                ret = dict_get_str (xattr, GF_XATTR_PATHINFO_KEY, &value);
-                if (!ret) {
-                        ret = dict_set_str (xattr, GF_XATTR_LINKINFO_KEY, value);
-                        if (!ret)
-                                gf_log (this->name, GF_LOG_TRACE,
-                                        "failed to set linkinfo");
-                }
+unwind:
+    DHT_STACK_UNWIND(getxattr, frame, -1, local->op_errno, NULL, NULL);
+cleanup:
+    if (dict)
+        dict_unref(dict);
+
+    return 0;
+}
+
+static int
+dht_linkinfo_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int op_ret, int op_errno, dict_t *xattr,
+                          dict_t *xdata)
+{
+    int ret = 0;
+    char *value = NULL;
+
+    if (op_ret != -1) {
+        ret = dict_get_str(xattr, GF_XATTR_PATHINFO_KEY, &value);
+        if (!ret) {
+            ret = dict_set_str(xattr, GF_XATTR_LINKINFO_KEY, value);
+            if (!ret)
+                gf_msg_trace(this->name, 0, "failed to set linkinfo");
         }
+    }
 
-        DHT_STACK_UNWIND (getxattr, frame, op_ret, op_errno, xattr, xdata);
+    DHT_STACK_UNWIND(getxattr, frame, op_ret, op_errno, xattr, xdata);
 
-        return 0;
+    return 0;
+}
+
+static int
+dht_mds_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(frame->local, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+    local = frame->local;
+
+    if (!xattr || (op_ret == -1)) {
+        local->op_ret = op_ret;
+        goto out;
+    }
+    dict_del(xattr, conf->xattr_name);
+    local->op_ret = 0;
+
+    if (!local->xattr) {
+        local->xattr = dict_copy_with_ref(xattr, NULL);
+    }
+
+out:
+    DHT_STACK_UNWIND(getxattr, frame, local->op_ret, op_errno, local->xattr,
+                     xdata);
+    return 0;
+err:
+    DHT_STACK_UNWIND(getxattr, frame, -1, EINVAL, NULL, NULL);
+    return 0;
 }
 
 int
-dht_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
+dht_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, dict_t *xattr, dict_t *xdata)
 {
-        int             this_call_cnt = 0;
-        dht_local_t     *local = NULL;
-        dht_conf_t      *conf = NULL;
+    int this_call_cnt = 0;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = 0;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (frame->local, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(frame->local, err);
+    VALIDATE_OR_GOTO(this->private, err);
 
-        conf = this->private;
-        local = frame->local;
+    conf = this->private;
+    local = frame->local;
+
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto err;
+        return 0;
+    }
 
-        this_call_cnt = dht_frame_return (frame);
+    LOCK(&frame->lock);
+    {
+        if (!xattr || (op_ret == -1)) {
+            local->op_ret = op_ret;
+            goto unlock;
+        }
 
-        if (!xattr || (op_ret == -1))
-                goto out;
+        dict_del(xattr, conf->xattr_name);
+        dict_del(xattr, conf->mds_xattr_key);
+
+        dict_del(xattr, conf->commithash_xattr_name);
 
-        if (dict_get (xattr, conf->xattr_name)) {
-                dict_del (xattr, conf->xattr_name);
+        if (frame->root->pid >= 0) {
+            GF_REMOVE_INTERNAL_XATTR("trusted.glusterfs.quota*", xattr);
+            GF_REMOVE_INTERNAL_XATTR("trusted.pgfid*", xattr);
         }
+
         local->op_ret = 0;
 
         if (!local->xattr) {
-                local->xattr = dict_copy_with_ref (xattr, NULL);
+            local->xattr = dict_copy_with_ref(xattr, NULL);
         } else {
-                /* first aggregate everything into xattr and then copy into
-                 * local->xattr. This is required as we want to have
-                 * 'local->xattr' as the proper final dictionary passed above
-                 * distribute xlator.
-                 */
-                dht_aggregate_xattr (xattr, local->xattr);
-                local->xattr = dict_copy (xattr, local->xattr);
+            dht_aggregate_xattr(local->xattr, xattr);
         }
-out:
-        if (is_last_call (this_call_cnt)) {
-                DHT_STACK_UNWIND (getxattr, frame, local->op_ret, op_errno,
-                                  local->xattr, NULL);
+
+        if (!local->xdata) {
+            local->xdata = dict_ref(xdata);
+        } else if ((local->inode && IA_ISDIR(local->inode->ia_type)) ||
+                   (local->fd && IA_ISDIR(local->fd->inode->ia_type))) {
+            dht_aggregate_xattr(local->xdata, xdata);
         }
-        return 0;
+    }
+unlock:
+    UNLOCK(&frame->lock);
+
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        /* If we have a valid xattr received from any one of the
+         * subvolume, let's return it */
+        if (local->xattr) {
+            local->op_ret = 0;
+        }
+
+        DHT_STACK_UNWIND(getxattr, frame, local->op_ret, op_errno, local->xattr,
+                         local->xdata);
+    }
+    return 0;
+err:
+    DHT_STACK_UNWIND(getxattr, frame, -1, EINVAL, NULL, NULL);
+    return 0;
 }
 
-int32_t
-dht_getxattr_unwind (call_frame_t *frame,
-                     int op_ret, int op_errno, dict_t *dict, dict_t *xdata)
+static int32_t
+dht_getxattr_unwind(call_frame_t *frame, int op_ret, int op_errno, dict_t *dict,
+                    dict_t *xdata)
 {
-        DHT_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict, xdata);
-        return 0;
+    DHT_STACK_UNWIND(getxattr, frame, op_ret, op_errno, dict, xdata);
+    return 0;
 }
 
-
-int
-dht_getxattr_get_real_filename_cbk (call_frame_t *frame, void *cookie,
-				    xlator_t *this, int op_ret, int op_errno,
-				    dict_t *xattr, dict_t *xdata)
+static int
+dht_getxattr_get_real_filename_cbk(call_frame_t *frame, void *cookie,
+                                   xlator_t *this, int op_ret, int op_errno,
+                                   dict_t *xattr, dict_t *xdata)
 {
-        int             this_call_cnt = 0;
-        dht_local_t     *local = NULL;
+    int this_call_cnt = 0;
+    dht_local_t *local = NULL;
 
+    local = frame->local;
 
-        local = frame->local;
+    LOCK(&frame->lock);
+    {
+        if (local->op_errno == EOPNOTSUPP) {
+            /* Nothing to do here, we have already found
+             * a subvol which does not have the get_real_filename
+             * optimization. If condition is for simple logic.
+             */
+            goto unlock;
+        }
 
-	if (op_ret != -1) {
-		if (local->xattr)
-			dict_unref (local->xattr);
-		local->xattr = dict_ref (xattr);
+        if (op_ret == -1) {
+            if (op_errno == EOPNOTSUPP) {
+                /* This subvol does not have the optimization.
+                 * Better let the user know we don't support it.
+                 * Remove previous results if any.
+                 */
 
-		if (local->xattr_req)
-			dict_unref (local->xattr_req);
-		local->xattr_req = dict_ref (xdata);
-	}
+                if (local->xattr) {
+                    dict_unref(local->xattr);
+                    local->xattr = NULL;
+                }
 
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt)) {
-		DHT_STACK_UNWIND (getxattr, frame, local->op_ret, op_errno,
-				  local->xattr, local->xattr_req);
-	}
+                if (local->xattr_req) {
+                    dict_unref(local->xattr_req);
+                    local->xattr_req = NULL;
+                }
 
-	return 0;
+                local->op_ret = op_ret;
+                local->op_errno = op_errno;
+                UNLOCK(&frame->lock);
+                gf_msg(this->name, GF_LOG_WARNING, op_errno,
+                       DHT_MSG_UPGRADE_BRICKS,
+                       "At least "
+                       "one of the bricks does not support "
+                       "this operation. Please upgrade all "
+                       "bricks.");
+                goto post_unlock;
+            }
+
+            if (op_errno == ENOATTR) {
+                /* Do nothing, our defaults are set to this.
+                 */
+                goto unlock;
+            }
+
+            /* This is a place holder for every other error
+             * case. I am not sure of how to interpret
+             * ENOTCONN etc. As of now, choosing to ignore
+             * down subvol and return a good result(if any)
+             * from other subvol.
+             */
+            UNLOCK(&frame->lock);
+            gf_msg(this->name, GF_LOG_WARNING, op_errno,
+                   DHT_MSG_GET_XATTR_FAILED, "Failed to get real filename.");
+            goto post_unlock;
+        }
+
+        /* This subvol has the required file.
+         * There could be other subvols which have returned
+         * success already, choosing to return the latest good
+         * result.
+         */
+        if (local->xattr)
+            dict_unref(local->xattr);
+        local->xattr = dict_ref(xattr);
+
+        if (local->xattr_req) {
+            dict_unref(local->xattr_req);
+            local->xattr_req = NULL;
+        }
+        if (xdata)
+            local->xattr_req = dict_ref(xdata);
+
+        local->op_ret = op_ret;
+        local->op_errno = 0;
+        UNLOCK(&frame->lock);
+        gf_msg_debug(this->name, 0, "Found a matching file.");
+        goto post_unlock;
+    }
+unlock:
+    UNLOCK(&frame->lock);
+post_unlock:
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        DHT_STACK_UNWIND(getxattr, frame, local->op_ret, local->op_errno,
+                         local->xattr, local->xattr_req);
+    }
+
+    return 0;
 }
 
+static int
+dht_getxattr_get_real_filename(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                               const char *key, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int i = 0;
+    dht_layout_t *layout = NULL;
+    int cnt = 0;
+    xlator_t *subvol = NULL;
 
-int
-dht_getxattr_get_real_filename (call_frame_t *frame, xlator_t *this,
-				loc_t *loc, const char *key, dict_t *xdata)
+    local = frame->local;
+    layout = local->layout;
+
+    cnt = local->call_cnt = layout->cnt;
+
+    local->op_ret = -1;
+    local->op_errno = ENOATTR;
+
+    for (i = 0; i < cnt; i++) {
+        subvol = layout->list[i].xlator;
+        STACK_WIND(frame, dht_getxattr_get_real_filename_cbk, subvol,
+                   subvol->fops->getxattr, loc, key, xdata);
+    }
+
+    return 0;
+}
+
+static int
+dht_marker_populate_args(call_frame_t *frame, int type, int *gauge,
+                         xlator_t **subvols)
 {
-	dht_local_t     *local = NULL;
-	int              i = 0;
-	dht_layout_t    *layout = NULL;
-	int              cnt = 0;
-	xlator_t        *subvol = NULL;
+    dht_local_t *local = NULL;
+    int i = 0;
+    dht_layout_t *layout = NULL;
 
+    local = frame->local;
+    layout = local->layout;
 
-	local = frame->local;
-	layout = local->layout;
+    for (i = 0; i < layout->cnt; i++)
+        subvols[i] = layout->list[i].xlator;
 
-	cnt = local->call_cnt = layout->cnt;
+    return layout->cnt;
+}
 
-	local->op_ret = -1;
-	local->op_errno = ENODATA;
+static int
+dht_is_debug_xattr_key(const char **array, char *key)
+{
+    int i = 0;
 
-	for (i = 0; i < cnt; i++) {
-		subvol = layout->list[i].xlator;
-		STACK_WIND (frame, dht_getxattr_get_real_filename_cbk,
-			    subvol, subvol->fops->getxattr,
-			    loc, key, xdata);
-	}
+    for (i = 0; array[i]; i++) {
+        if (fnmatch(array[i], key, FNM_NOESCAPE) == 0)
+            return i;
+    }
 
-	return 0;
+    return -1;
 }
 
+/* Note we already have frame->local initialised here*/
 
-int
-dht_getxattr (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, const char *key, dict_t *xdata)
-#define DHT_IS_DIR(layout)  (layout->cnt > 1)
-{
-
-        xlator_t     *subvol        = NULL;
-        xlator_t     *hashed_subvol = NULL;
-        xlator_t     *cached_subvol = NULL;
-        dht_conf_t   *conf          = NULL;
-        dht_local_t  *local         = NULL;
-        dht_layout_t *layout        = NULL;
-        xlator_t     **sub_volumes  = NULL;
-        int           op_errno      = -1;
-        int           i             = 0;
-        int           cnt           = 0;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (this->private, err);
-
-        conf   = this->private;
-
-        local = dht_local_init (frame, loc, NULL, GF_FOP_GETXATTR);
-        if (!local) {
-                op_errno = ENOMEM;
+static int
+dht_handle_debug_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                          const char *key)
+{
+    dht_local_t *local = NULL;
+    int ret = -1;
+    int op_errno = ENODATA;
+    char *value = NULL;
+    loc_t file_loc = {0};
+    const char *name = NULL;
+
+    local = frame->local;
+
+    if (dht_is_debug_xattr_key(dht_dbg_vxattrs, (char *)key) == -1) {
+        goto out;
+    }
+
+    local->xattr = dict_new();
+    if (!local->xattr) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    if (strncmp(key, DHT_DBG_HASHED_SUBVOL_KEY,
+                SLEN(DHT_DBG_HASHED_SUBVOL_KEY)) == 0) {
+        name = key + strlen(DHT_DBG_HASHED_SUBVOL_KEY);
+        if (strlen(name) == 0) {
+            op_errno = EINVAL;
+            goto out;
+        }
+
+        ret = dht_build_child_loc(this, &file_loc, loc, (char *)name);
+        if (ret) {
+            op_errno = ENOMEM;
+            goto out;
+        }
 
-                goto err;
+        local->hashed_subvol = dht_subvol_get_hashed(this, &file_loc);
+        if (local->hashed_subvol == NULL) {
+            op_errno = ENODATA;
+            goto out;
         }
 
-        layout = local->layout;
-        if (!layout) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "layout is NULL");
-                op_errno = ENOENT;
-                goto err;
+        value = gf_strdup(local->hashed_subvol->name);
+        if (!value) {
+            op_errno = ENOMEM;
+            goto out;
         }
 
-        if (key) {
-                local->key = gf_strdup (key);
-                if (!local->key) {
-                        op_errno = ENOMEM;
-                        goto err;
-                }
+        ret = dict_set_dynstr(local->xattr, (char *)key, value);
+        if (ret < 0) {
+            op_errno = -ret;
+            ret = -1;
+            goto out;
         }
+        ret = 0;
+        goto out;
+    }
 
-	if (key &&
-	    (strncmp (key, GF_XATTR_GET_REAL_FILENAME_KEY,
-		      strlen (GF_XATTR_GET_REAL_FILENAME_KEY)) == 0)
-	    && DHT_IS_DIR(layout)) {
-		dht_getxattr_get_real_filename (frame, this, loc, key, xdata);
-		return 0;
-	}
+out:
+    loc_wipe(&file_loc);
+    DHT_STACK_UNWIND(getxattr, frame, ret, op_errno, local->xattr, NULL);
+    return 0;
+}
 
-        /* for file use cached subvolume (obviously!): see if {}
-         * below
-         * for directory:
-         *  wind to all subvolumes and exclude subvolumes which
-         *  return ENOTCONN (in callback)
-         *
-         * NOTE: Don't trust inode here, as that may not be valid
-         *       (until inode_link() happens)
-         */
-        if (key && DHT_IS_DIR(layout) &&
-            ((strcmp (key, GF_XATTR_PATHINFO_KEY) == 0)
-             || (strcmp (key, GF_XATTR_NODE_UUID_KEY) == 0))) {
-                (void) strncpy (local->xsel, key, 256);
-                cnt = local->call_cnt = layout->cnt;
-                for (i = 0; i < cnt; i++) {
-                        subvol = layout->list[i].xlator;
-                        STACK_WIND (frame, dht_vgetxattr_dir_cbk,
-                                    subvol, subvol->fops->getxattr,
-                                    loc, key, NULL);
-                }
-                return 0;
-        }
+/* Virtual Xattr which returns 1 if all subvols are up,
+   else returns 0. Geo-rep then uses this virtual xattr
+   after a fresh mount and starts the I/O.
+*/
 
-        /* node-uuid or pathinfo for files */
-        if (key && ((strcmp (key, GF_XATTR_NODE_UUID_KEY) == 0)
-                    || (strcmp (key, GF_XATTR_PATHINFO_KEY) == 0))) {
-                cached_subvol = local->cached_subvol;
-                (void) strncpy (local->xsel, key, 256);
+enum dht_vxattr_subvol {
+    DHT_VXATTR_SUBVOLS_UP = 1,
+    DHT_VXATTR_SUBVOLS_DOWN = 0,
+};
 
-                local->call_cnt = 1;
-                STACK_WIND (frame, dht_vgetxattr_cbk, cached_subvol,
-                            cached_subvol->fops->getxattr, loc, key, NULL);
+int
+dht_vgetxattr_subvol_status(call_frame_t *frame, xlator_t *this,
+                            const char *key)
+{
+    dht_local_t *local = NULL;
+    int ret = -1;
+    int op_errno = ENODATA;
+    int value = DHT_VXATTR_SUBVOLS_UP;
+    int i = 0;
+    dht_conf_t *conf = NULL;
+
+    conf = this->private;
+    local = frame->local;
+
+    if (!key) {
+        op_errno = EINVAL;
+        goto out;
+    }
+    local->xattr = dict_new();
+    if (!local->xattr) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (!conf->subvolume_status[i]) {
+            value = DHT_VXATTR_SUBVOLS_DOWN;
+            gf_msg_debug(this->name, 0, "subvol %s is down ",
+                         conf->subvolumes[i]->name);
+            break;
+        }
+    }
+    ret = dict_set_int8(local->xattr, (char *)key, value);
+    if (ret < 0) {
+        op_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+    ret = 0;
 
-                return 0;
-        }
+out:
+    DHT_STACK_UNWIND(getxattr, frame, ret, op_errno, local->xattr, NULL);
+    return 0;
+}
 
-        if (key && (strcmp (key, GF_XATTR_LINKINFO_KEY) == 0)) {
-                hashed_subvol = dht_subvol_get_hashed (this, loc);
-                if (!hashed_subvol) {
-                        gf_log (this->name, GF_LOG_ERROR, "Failed to get"
-                                "hashed subvol for %s", loc->path);
-                        op_errno = EINVAL;
-                        goto err;
-                }
+int
+dht_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, const char *key,
+             dict_t *xdata)
+#define DHT_IS_DIR(layout) (layout->cnt > 1)
+{
+    xlator_t *subvol = NULL;
+    xlator_t *hashed_subvol = NULL;
+    xlator_t *mds_subvol = NULL;
+    xlator_t *cached_subvol = NULL;
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
+    int op_errno = -1;
+    int i = 0;
+    int cnt = 0;
+    char *node_uuid_key = NULL;
+    int ret = -1;
+
+    GF_CHECK_XATTR_KEY_AND_GOTO(key, IO_THREADS_QUEUE_SIZE_KEY, op_errno, err);
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_GETXATTR);
+    if (!local) {
+        op_errno = ENOMEM;
+
+        goto err;
+    }
+
+    layout = local->layout;
+    if (!layout) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LAYOUT_NULL,
+               "Layout is NULL");
+        op_errno = ENOENT;
+        goto err;
+    }
+
+    /* skip over code which is irrelevant without a valid key */
+    if (!key)
+        goto no_key;
+
+    local->key = gf_strdup(key);
+    if (!local->key) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    if (strncmp(key, conf->mds_xattr_key, strlen(key)) == 0) {
+        op_errno = ENOTSUP;
+        goto err;
+    }
+
+    if (strncmp(key, DHT_SUBVOL_STATUS_KEY, SLEN(DHT_SUBVOL_STATUS_KEY)) == 0) {
+        dht_vgetxattr_subvol_status(frame, this, key);
+        return 0;
+    }
 
-                cached_subvol = dht_subvol_get_cached (this, loc->inode);
-                if (!cached_subvol) {
-                        gf_log (this->name, GF_LOG_ERROR, "Failed to get"
-                                "cached subvol for %s", loc->path);
-                        op_errno = EINVAL;
-                        goto err;
-                }
+    /* skip over code which is irrelevant if !DHT_IS_DIR(layout) */
+    if (!DHT_IS_DIR(layout))
+        goto no_dht_is_dir;
 
-                if (hashed_subvol == cached_subvol) {
-                        op_errno = ENODATA;
-                        goto err;
-                }
-                if (hashed_subvol) {
-                        STACK_WIND (frame, dht_linkinfo_getxattr_cbk, hashed_subvol,
-                                    hashed_subvol->fops->getxattr, loc,
-                                    GF_XATTR_PATHINFO_KEY, NULL);
-                        return 0;
-                }
-                op_errno = ENODATA;
-                goto err;
+    if ((strncmp(key, GF_XATTR_GET_REAL_FILENAME_KEY,
+                 SLEN(GF_XATTR_GET_REAL_FILENAME_KEY)) == 0) &&
+        DHT_IS_DIR(layout)) {
+        dht_getxattr_get_real_filename(frame, this, loc, key, xdata);
+        return 0;
+    }
+
+    if (!strcmp(key, GF_REBAL_FIND_LOCAL_SUBVOL)) {
+        ret = gf_asprintf(&node_uuid_key, "%s", GF_XATTR_LIST_NODE_UUIDS_KEY);
+        if (ret == -1 || !node_uuid_key) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_NO_MEMORY,
+                   "Failed to copy node uuid key");
+            op_errno = ENOMEM;
+            goto err;
+        }
+        (void)snprintf(local->xsel, sizeof(local->xsel), "%s", node_uuid_key);
+        cnt = local->call_cnt = conf->subvolume_cnt;
+        for (i = 0; i < cnt; i++) {
+            STACK_WIND_COOKIE(frame, dht_find_local_subvol_cbk,
+                              conf->subvolumes[i], conf->subvolumes[i],
+                              conf->subvolumes[i]->fops->getxattr, loc,
+                              node_uuid_key, xdata);
+        }
+        if (node_uuid_key)
+            GF_FREE(node_uuid_key);
+        return 0;
+    }
+
+    if (!strcmp(key, GF_REBAL_OLD_FIND_LOCAL_SUBVOL)) {
+        ret = gf_asprintf(&node_uuid_key, "%s", GF_XATTR_NODE_UUID_KEY);
+        if (ret == -1 || !node_uuid_key) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_NO_MEMORY,
+                   "Failed to copy node uuid key");
+            op_errno = ENOMEM;
+            goto err;
+        }
+        (void)snprintf(local->xsel, sizeof(local->xsel), "%s", node_uuid_key);
+        cnt = local->call_cnt = conf->subvolume_cnt;
+        for (i = 0; i < cnt; i++) {
+            STACK_WIND_COOKIE(frame, dht_find_local_subvol_cbk,
+                              conf->subvolumes[i], conf->subvolumes[i],
+                              conf->subvolumes[i]->fops->getxattr, loc,
+                              node_uuid_key, xdata);
         }
+        if (node_uuid_key)
+            GF_FREE(node_uuid_key);
+        return 0;
+    }
+
+    /* for file use cached subvolume (obviously!): see if {}
+     * below
+     * for directory:
+     *  wind to all subvolumes and exclude subvolumes which
+     *  return ENOTCONN (in callback)
+     *
+     * NOTE: Don't trust inode here, as that may not be valid
+     *       (until inode_link() happens)
+     */
+
+    if (XATTR_IS_PATHINFO(key) || (strcmp(key, GF_XATTR_NODE_UUID_KEY) == 0) ||
+        (strcmp(key, GF_XATTR_LIST_NODE_UUIDS_KEY) == 0)) {
+        (void)snprintf(local->xsel, sizeof(local->xsel), "%s", key);
+        cnt = local->call_cnt = layout->cnt;
+        for (i = 0; i < cnt; i++) {
+            subvol = layout->list[i].xlator;
+            STACK_WIND(frame, dht_vgetxattr_dir_cbk, subvol,
+                       subvol->fops->getxattr, loc, key, xdata);
+        }
+        return 0;
+    }
 
-        if (key && (!strcmp (GF_XATTR_MARKER_KEY, key))
-            && (GF_CLIENT_PID_GSYNCD == frame->root->pid)) {
-                if (DHT_IS_DIR(layout)) {
-                        cnt = layout->cnt;
-                } else {
-                        cnt = 1;
-                }
+no_dht_is_dir:
+    /* node-uuid or pathinfo for files */
+    if (XATTR_IS_PATHINFO(key) || (strcmp(key, GF_XATTR_NODE_UUID_KEY) == 0)) {
+        cached_subvol = local->cached_subvol;
+        (void)snprintf(local->xsel, sizeof(local->xsel), "%s", key);
+        local->call_cnt = 1;
+        STACK_WIND_COOKIE(frame, dht_vgetxattr_cbk, cached_subvol,
+                          cached_subvol, cached_subvol->fops->getxattr, loc,
+                          key, xdata);
 
-                sub_volumes = alloca ( cnt * sizeof (xlator_t *));
-                for (i = 0; i < cnt; i++)
-                        *(sub_volumes + i) = layout->list[i].xlator;
-
-                if (cluster_getmarkerattr (frame, this, loc, key,
-                                           local, dht_getxattr_unwind,
-                                           sub_volumes, cnt,
-                                           MARKER_UUID_TYPE, marker_uuid_default_gauge,
-                                           conf->vol_uuid)) {
-                        op_errno = EINVAL;
-                        goto err;
-                }
+        return 0;
+    }
 
-                return 0;
+    if (strcmp(key, GF_XATTR_LINKINFO_KEY) == 0) {
+        hashed_subvol = dht_subvol_get_hashed(this, loc);
+        if (!hashed_subvol) {
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+                   "Failed to get hashed subvol for %s", loc->path);
+            op_errno = EINVAL;
+            goto err;
         }
 
-        if (key && *conf->vol_uuid) {
-                if ((match_uuid_local (key, conf->vol_uuid) == 0) &&
-                    (GF_CLIENT_PID_GSYNCD == frame->root->pid)) {
-                        if (DHT_IS_DIR(layout)) {
-                                cnt = layout->cnt;
-                        } else {
-                                cnt = 1;
-                        }
-                        sub_volumes = alloca ( cnt * sizeof (xlator_t *));
-                        for (i = 0; i < cnt; i++)
-                                sub_volumes[i] = layout->list[i].xlator;
-
-                        if (cluster_getmarkerattr (frame, this, loc, key,
-                                                   local, dht_getxattr_unwind,
-                                                   sub_volumes, cnt,
-                                                   MARKER_XTIME_TYPE,
-                                                   marker_xtime_default_gauge,
-                                                   conf->vol_uuid)) {
-                                op_errno = EINVAL;
-                                goto err;
-                        }
+        cached_subvol = dht_subvol_get_cached(this, loc->inode);
+        if (!cached_subvol) {
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   DHT_MSG_CACHED_SUBVOL_GET_FAILED,
+                   "Failed to get cached subvol for %s", loc->path);
+            op_errno = EINVAL;
+            goto err;
+        }
 
-                        return 0;
-                }
+        if (hashed_subvol == cached_subvol) {
+            op_errno = ENODATA;
+            goto err;
         }
 
-        if (DHT_IS_DIR(layout)) {
-                cnt = local->call_cnt = layout->cnt;
+        STACK_WIND(frame, dht_linkinfo_getxattr_cbk, hashed_subvol,
+                   hashed_subvol->fops->getxattr, loc, GF_XATTR_PATHINFO_KEY,
+                   xdata);
+        return 0;
+    }
+
+    if (dht_is_debug_xattr_key(dht_dbg_vxattrs, (char *)key) >= 0) {
+        dht_handle_debug_getxattr(frame, this, loc, key);
+        return 0;
+    }
+
+no_key:
+    if (cluster_handle_marker_getxattr(frame, loc, key, conf->vol_uuid,
+                                       dht_getxattr_unwind,
+                                       dht_marker_populate_args) == 0)
+        return 0;
+
+    if (DHT_IS_DIR(layout)) {
+        local->call_cnt = conf->subvolume_cnt;
+        cnt = conf->subvolume_cnt;
+        ret = dht_inode_ctx_mdsvol_get(loc->inode, this, &mds_subvol);
+        if (!mds_subvol) {
+            gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+                   "Cannot determine MDS, fetching xattr %s randomly"
+                   " from a subvol for path %s ",
+                   key, loc->path);
         } else {
-                cnt = local->call_cnt  = 1;
+            /* TODO need to handle it, As of now we are
+               choosing availability instead of chossing
+               consistencty, in case of mds_subvol is
+               down winding a getxattr call on other subvol
+               and return xattr
+            */
+            local->mds_subvol = mds_subvol;
+            for (i = 0; i < cnt; i++) {
+                if (conf->subvolumes[i] == mds_subvol) {
+                    if (!conf->subvolume_status[i]) {
+                        gf_msg(this->name, GF_LOG_INFO, 0,
+                               DHT_MSG_HASHED_SUBVOL_DOWN,
+                               "MDS %s is down for path"
+                               " path %s so fetching xattr "
+                               "%s randomly from a subvol ",
+                               local->mds_subvol->name, loc->path, key);
+                        ret = 1;
+                    }
+                }
+            }
         }
 
-        for (i = 0; i < cnt; i++) {
-                subvol = layout->list[i].xlator;
-                STACK_WIND (frame, dht_getxattr_cbk,
-                            subvol, subvol->fops->getxattr,
-                            loc, key, NULL);
+        if (!ret && key && local->mds_subvol && dht_match_xattr(key)) {
+            STACK_WIND(frame, dht_mds_getxattr_cbk, local->mds_subvol,
+                       local->mds_subvol->fops->getxattr, loc, key, xdata);
+
+            return 0;
         }
-        return 0;
+    } else {
+        cnt = local->call_cnt = 1;
+    }
+
+    for (i = 0; i < cnt; i++) {
+        subvol = layout->list[i].xlator;
+        STACK_WIND(frame, dht_getxattr_cbk, subvol, subvol->fops->getxattr, loc,
+                   key, xdata);
+    }
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (getxattr, frame, -1, op_errno, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(getxattr, frame, -1, op_errno, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 #undef DHT_IS_DIR
 
 int
-dht_fgetxattr (call_frame_t *frame, xlator_t *this,
-               fd_t *fd, const char *key, dict_t *xdata)
-{
-        xlator_t     *subvol        = NULL;
-        dht_local_t  *local         = NULL;
-        dht_layout_t *layout        = NULL;
-        int           op_errno      = -1;
-        int           i             = 0;
-        int           cnt           = 0;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-        VALIDATE_OR_GOTO (this->private, err);
-
-        local = dht_local_init (frame, NULL, fd, GF_FOP_FGETXATTR);
-        if (!local) {
-                op_errno = ENOMEM;
-
-                goto err;
+dht_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *key,
+              dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
+    int op_errno = -1;
+    int i = 0;
+    int cnt = 0;
+    xlator_t *mds_subvol = NULL;
+    int ret = -1;
+    dht_conf_t *conf = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+    VALIDATE_OR_GOTO(fd->inode, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_FGETXATTR);
+    if (!local) {
+        op_errno = ENOMEM;
+
+        goto err;
+    }
+
+    layout = local->layout;
+    if (!layout) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LAYOUT_NULL,
+               "Layout is NULL");
+        op_errno = ENOENT;
+        goto err;
+    }
+
+    if (key) {
+        local->key = gf_strdup(key);
+        if (!local->key) {
+            op_errno = ENOMEM;
+            goto err;
+        }
+    }
+
+    gf_uuid_unparse(fd->inode->gfid, gfid);
+
+    if ((fd->inode->ia_type == IA_IFDIR) && key &&
+        (strncmp(key, GF_XATTR_LOCKINFO_KEY, SLEN(GF_XATTR_LOCKINFO_KEY)) !=
+         0)) {
+        local->call_cnt = conf->subvolume_cnt;
+        cnt = conf->subvolume_cnt;
+        ret = dht_inode_ctx_mdsvol_get(fd->inode, this, &mds_subvol);
+
+        if (!mds_subvol) {
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+                   "cannot determine MDS, fetching xattr %s "
+                   " randomly from a subvol for gfid %s ",
+                   key, gfid);
+        } else {
+            /* TODO need to handle it, As of now we are
+               choosing availability instead of chossing
+               consistencty, in case of hashed_subvol is
+               down winding a getxattr call on other subvol
+               and return xattr
+            */
+            local->mds_subvol = mds_subvol;
+            for (i = 0; i < cnt; i++) {
+                if (conf->subvolumes[i] == mds_subvol) {
+                    if (!conf->subvolume_status[i]) {
+                        gf_msg(this->name, GF_LOG_WARNING, 0,
+                               DHT_MSG_HASHED_SUBVOL_DOWN,
+                               "MDS subvolume %s is down"
+                               " for gfid %s so fetching xattr "
+                               " %s randomly from a subvol ",
+                               local->mds_subvol->name, gfid, key);
+                        ret = 1;
+                    }
+                }
+            }
         }
 
-        layout = local->layout;
-        if (!layout) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "layout is NULL");
-                op_errno = ENOENT;
-                goto err;
-        }
+        if (!ret && key && local->mds_subvol && dht_match_xattr(key)) {
+            STACK_WIND(frame, dht_mds_getxattr_cbk, local->mds_subvol,
+                       local->mds_subvol->fops->fgetxattr, fd, key, NULL);
 
-        if (key) {
-                local->key = gf_strdup (key);
-                if (!local->key) {
-                        op_errno = ENOMEM;
-                        goto err;
-                }
+            return 0;
         }
 
-        if ((fd->inode->ia_type == IA_IFDIR)
-            && (strncmp (key, GF_XATTR_LOCKINFO_KEY,
-                         strlen (GF_XATTR_LOCKINFO_KEY) != 0))) {
-                cnt = local->call_cnt = layout->cnt;
-        } else {
-                cnt = local->call_cnt  = 1;
-        }
+    } else {
+        cnt = local->call_cnt = 1;
+    }
 
-        for (i = 0; i < cnt; i++) {
-                subvol = layout->list[i].xlator;
-                STACK_WIND (frame, dht_getxattr_cbk,
-                            subvol, subvol->fops->fgetxattr,
-                            fd, key, NULL);
-        }
-        return 0;
+    for (i = 0; i < cnt; i++) {
+        subvol = layout->list[i].xlator;
+        STACK_WIND(frame, dht_getxattr_cbk, subvol, subvol->fops->fgetxattr, fd,
+                   key, NULL);
+    }
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (fgetxattr, frame, -1, op_errno, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(fgetxattr, frame, -1, op_errno, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
-int
-dht_fsetxattr (call_frame_t *frame, xlator_t *this,
-               fd_t *fd, dict_t *xattr, int flags, dict_t *xdata)
+static int
+dht_setxattr2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
 {
-        xlator_t     *subvol   = NULL;
-        dht_local_t  *local    = NULL;
-        int           op_errno = EINVAL;
-        dht_conf_t   *conf     = NULL;
+    dht_local_t *local = NULL;
+    int op_errno = EINVAL;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-        VALIDATE_OR_GOTO (this->private, err);
+    if (!frame || !frame->local)
+        goto err;
 
-        conf = this->private;
+    local = frame->local;
+    op_errno = local->op_errno;
 
-        GF_IF_INTERNAL_XATTR_GOTO (conf->wild_xattr_name, xattr,
-                                   op_errno, err);
-
-        local = dht_local_init (frame, NULL, fd, GF_FOP_FSETXATTR);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        DHT_STACK_UNWIND(setxattr, frame, local->op_ret, local->op_errno,
+                         local->rebalance.xdata);
+        return 0;
+    }
 
-        subvol = local->cached_subvol;
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
-        }
+    if (subvol == NULL)
+        goto err;
 
-        local->call_cnt = 1;
+    local->call_cnt = 2; /* This is the second attempt */
 
-        STACK_WIND (frame, dht_err_cbk, subvol, subvol->fops->fsetxattr,
-                    fd, xattr, flags, NULL);
+    if (local->fop == GF_FOP_SETXATTR) {
+        STACK_WIND_COOKIE(frame, dht_file_setxattr_cbk, subvol, subvol,
+                          subvol->fops->setxattr, &local->loc,
+                          local->rebalance.xattr, local->rebalance.flags,
+                          local->xattr_req);
+    } else {
+        STACK_WIND_COOKIE(frame, dht_file_setxattr_cbk, subvol, subvol,
+                          subvol->fops->fsetxattr, local->fd,
+                          local->rebalance.xattr, local->rebalance.flags,
+                          local->xattr_req);
+    }
 
-        return 0;
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (fsetxattr, frame, -1, op_errno, NULL);
-
-        return 0;
+    DHT_STACK_UNWIND(setxattr, frame, (local ? local->op_ret : -1), op_errno,
+                     NULL);
+    return 0;
 }
 
-
-static int
-dht_common_setxattr_cbk (call_frame_t *frame, void *cookie,
-                         xlator_t *this, int32_t op_ret, int32_t op_errno,
-                         dict_t *xdata)
+int
+dht_file_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, dict_t *xdata)
 {
-        DHT_STACK_UNWIND (setxattr, frame, op_ret, op_errno, xdata);
+    int ret = -1;
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    struct iatt *stbuf = NULL;
+    inode_t *inode = NULL;
+    xlator_t *subvol1 = NULL, *subvol2 = NULL;
 
+    local = frame->local;
+    prev = cookie;
+
+    local->op_errno = op_errno;
+
+    if ((local->fop == GF_FOP_FSETXATTR) &&
+        dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
         return 0;
-}
+    }
 
-int
-dht_checking_pathinfo_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                           int op_ret, int op_errno, dict_t *xattr,
-                           dict_t *xdata)
-{
-        int           i     = -1;
-        int           ret   = -1;
-        char         *value = NULL;
-        dht_local_t  *local = NULL;
-        dht_conf_t   *conf  = NULL;
-        call_frame_t *prev  = NULL;
-        int           this_call_cnt = 0;
+    if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1.",
+                     prev->name);
+        goto out;
+    }
 
-        local = frame->local;
-        prev = cookie;
-        conf = this->private;
+    if (local->call_cnt != 1)
+        goto out;
 
-        if (op_ret == -1)
-                goto out;
+    ret = dict_get_bin(xdata, DHT_IATT_IN_XDATA_KEY, (void **)&stbuf);
 
+    if ((!op_ret) && !stbuf) {
+        goto out;
+    }
 
-        ret = dict_get_str (xattr, GF_XATTR_PATHINFO_KEY, &value);
-        if (ret)
-                goto out;
+    local->op_ret = op_ret;
+    local->rebalance.target_op_fn = dht_setxattr2;
+    if (xdata)
+        local->rebalance.xdata = dict_ref(xdata);
 
-        if (!strcmp (value, local->key)) {
-                for (i = 0; i < conf->subvolume_cnt; i++) {
-                        if (conf->subvolumes[i] == prev->this)
-                                conf->decommissioned_bricks[i] = prev->this;
-                }
+    /* Phase 2 of migration */
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(stbuf)) {
+        ret = dht_rebalance_complete_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+    /* Phase 1 of migration */
+    if (IS_DHT_MIGRATION_PHASE1(stbuf)) {
+        inode = (local->fd) ? local->fd->inode : local->loc.inode;
+
+        ret = dht_inode_ctx_get_mig_info(this, inode, &subvol1, &subvol2);
+        if (!dht_mig_info_is_invalid(local->cached_subvol, subvol1, subvol2)) {
+            dht_setxattr2(this, subvol2, frame, 0);
+            return 0;
         }
 
+        ret = dht_rebalance_in_progress_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
 out:
-        this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt)) {
-                DHT_STACK_UNWIND (setxattr, frame, local->op_ret, ENOTSUP, NULL);
-        }
-        return 0;
 
+    if (local->fop == GF_FOP_SETXATTR) {
+        DHT_STACK_UNWIND(setxattr, frame, op_ret, op_errno, xdata);
+    } else {
+        DHT_STACK_UNWIND(fsetxattr, frame, op_ret, op_errno, xdata);
+    }
+
+    return 0;
 }
 
-int
-dht_setxattr (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, dict_t *xattr, int flags, dict_t *xdata)
-{
-        xlator_t     *subvol   = NULL;
-        dht_local_t  *local    = NULL;
-        dht_conf_t   *conf     = NULL;
-        dht_layout_t *layout   = NULL;
-        int           i        = 0;
-        int           op_errno = EINVAL;
-        int           ret      = -1;
-        data_t       *tmp      = NULL;
-        uint32_t      dir_spread = 0;
-        char          value[4096] = {0,};
-        gf_dht_migrate_data_type_t forced_rebalance = GF_DHT_MIGRATE_DATA;
-        int           call_cnt = 0;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        conf   = this->private;
-
-        GF_IF_INTERNAL_XATTR_GOTO (conf->wild_xattr_name, xattr,
-                                   op_errno, err);
-
-        local = dht_local_init (frame, loc, NULL, GF_FOP_SETXATTR);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
+/* Function is call by dict_foreach_fnmatch if key is match with
+   user.* and set boolean flag to true
+*/
+static int
+dht_is_user_xattr(dict_t *this, char *key, data_t *value, void *data)
+{
+    gf_boolean_t *user_xattr_found = data;
+    *user_xattr_found = _gf_true;
+    return 0;
+}
+
+/* Common code to wind a (f)(set|remove)xattr call to set xattr on directory
+ */
+static int
+dht_dir_common_set_remove_xattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                                fd_t *fd, dict_t *xattr, int flags,
+                                dict_t *xdata, int *op_errno)
+
+{
+    dict_t *xattrop = NULL;
+    int32_t subone[1] = {-1};
+    gf_boolean_t uxattr_key_found = _gf_false;
+    xlator_t *mds_subvol = NULL;
+    xlator_t *travvol = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
+    int i = 0;
+    int call_cnt = 0;
+    dht_local_t *local = NULL;
+    char gfid_local[GF_UUID_BUF_SIZE] = {0};
+    char **xattrs_to_heal;
+
+    conf = this->private;
+    local = frame->local;
+    call_cnt = conf->subvolume_cnt;
+    local->flags = flags;
+    xattrs_to_heal = get_xattrs_to_heal();
+
+    if (!gf_uuid_is_null(local->gfid)) {
+        gf_uuid_unparse(local->gfid, gfid_local);
+    }
+
+    if ((local->fop == GF_FOP_SETXATTR) || (local->fop == GF_FOP_FSETXATTR)) {
+        /* Check if any user xattr present in xattr
+         */
+        dict_foreach_fnmatch(xattr, "user*", dht_is_user_xattr,
+                             &uxattr_key_found);
+
+        /* Check if any custom key xattr present in dict xattr
+           and start index from 1 because user xattr already
+           checked in previous line
+        */
+        for (i = 1; xattrs_to_heal[i]; i++)
+            if (dict_get(xattr, xattrs_to_heal[i]))
+                uxattr_key_found = _gf_true;
+    }
+
+    if ((local->fop == GF_FOP_REMOVEXATTR) ||
+        (local->fop == GF_FOP_FREMOVEXATTR)) {
+        /* Check if any custom key xattr present in local->key
+         */
+        for (i = 0; xattrs_to_heal[i]; i++)
+            if (strstr(local->key, xattrs_to_heal[i]))
+                uxattr_key_found = _gf_true;
+    }
+
+    /* If there is no custom key xattr present or gfid is root
+       or call_cnt is 1 then wind a (f)setxattr call on all subvols
+    */
+    if (!uxattr_key_found || __is_root_gfid(local->gfid) || call_cnt == 1) {
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            travvol = conf->subvolumes[i];
+            if ((local->fop == GF_FOP_SETXATTR) ||
+                (local->fop == GF_FOP_FSETXATTR)) {
+                if (fd) {
+                    STACK_WIND_COOKIE(frame, dht_err_cbk, travvol, travvol,
+                                      travvol->fops->fsetxattr, fd, xattr,
+                                      flags, xdata);
+                } else {
+                    STACK_WIND_COOKIE(frame, dht_err_cbk, travvol, travvol,
+                                      travvol->fops->setxattr, loc, xattr,
+                                      flags, xdata);
+                }
+            }
+
+            if ((local->fop == GF_FOP_REMOVEXATTR) ||
+                (local->fop == GF_FOP_FREMOVEXATTR)) {
+                if (fd) {
+                    STACK_WIND_COOKIE(frame, dht_err_cbk, travvol, travvol,
+                                      travvol->fops->fremovexattr, fd,
+                                      local->key, local->xattr_req);
+                } else {
+                    STACK_WIND_COOKIE(frame, dht_err_cbk, travvol, travvol,
+                                      travvol->fops->removexattr, loc,
+                                      local->key, local->xattr_req);
+                }
+            }
         }
 
-        subvol = local->cached_subvol;
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for path=%s", loc->path);
-                op_errno = EINVAL;
+        return 0;
+    }
+
+    /* Calculate hash subvol based on inode and parent inode
+     */
+    if (fd) {
+        ret = dht_inode_ctx_mdsvol_get(fd->inode, this, &mds_subvol);
+    } else {
+        ret = dht_inode_ctx_mdsvol_get(loc->inode, this, &mds_subvol);
+    }
+    if (ret || !mds_subvol) {
+        if (fd) {
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+                   "Failed to get mds subvol for fd %p"
+                   "gfid is %s ",
+                   fd, gfid_local);
+        } else {
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+                   "%s: Failed to get mds subvol. (gfid is %s)", loc->path,
+                   gfid_local);
+        }
+        (*op_errno) = ENOENT;
+        goto err;
+    }
+
+    local->mds_subvol = mds_subvol;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (conf->subvolumes[i] == mds_subvol) {
+            if (!conf->subvolume_status[i]) {
+                gf_msg(this->name, GF_LOG_WARNING, 0,
+                       DHT_MSG_HASHED_SUBVOL_DOWN,
+                       "MDS subvol is down for path "
+                       " %s gfid is %s Unable to set xattr ",
+                       local->loc.path, gfid_local);
+                (*op_errno) = ENOTCONN;
                 goto err;
+            }
+        }
+    }
+
+    if (uxattr_key_found) {
+        xattrop = dict_new();
+        if (!xattrop) {
+            gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_NO_MEMORY, 0,
+                   "dictionary creation failed for path %s "
+                   "for gfid is %s ",
+                   local->loc.path, gfid_local);
+            (*op_errno) = ENOMEM;
+            goto err;
+        }
+        local->xattr = dict_ref(xattr);
+        /* Subtract current MDS xattr value to -1 , value of MDS
+           xattr represents no. of times xattr modification failed
+           on non MDS subvols.
+        */
+        ret = dht_dict_set_array(xattrop, conf->mds_xattr_key, subone, 1);
+        if (ret != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                   "dictionary set array failed for path %s "
+                   "for gfid is %s ",
+                   local->loc.path, gfid_local);
+            if (xattrop)
+                dict_unref(xattrop);
+            (*op_errno) = ret;
+            goto err;
+        }
+        /* Wind a xattrop call to use ref counting approach
+           update mds xattr to -1 before update xattr on
+           hashed subvol and update mds xattr to +1 after update
+           xattr on all non hashed subvol
+        */
+        if (fd) {
+            STACK_WIND(frame, dht_xattrop_mds_cbk, local->mds_subvol,
+                       local->mds_subvol->fops->fxattrop, fd,
+                       GF_XATTROP_ADD_ARRAY, xattrop, NULL);
+        } else {
+            STACK_WIND(frame, dht_xattrop_mds_cbk, local->mds_subvol,
+                       local->mds_subvol->fops->xattrop, loc,
+                       GF_XATTROP_ADD_ARRAY, xattrop, NULL);
         }
+        if (xattrop)
+            dict_unref(xattrop);
+    }
 
-        layout = local->layout;
-        if (!layout) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no layout for path=%s", loc->path);
-                op_errno = EINVAL;
-                goto err;
+    return 0;
+err:
+    return -1;
+}
+
+int
+dht_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xattr,
+              int flags, dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    dht_local_t *local = NULL;
+    int op_errno = EINVAL;
+    dht_conf_t *conf = NULL;
+    dht_layout_t *layout = NULL;
+    int ret = -1;
+    int call_cnt = 0;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+    VALIDATE_OR_GOTO(fd->inode, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+
+    if (!conf->defrag)
+        GF_IF_INTERNAL_XATTR_GOTO(conf->wild_xattr_name, xattr, op_errno, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_FSETXATTR);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    layout = local->layout;
+    if (!layout) {
+        gf_msg_debug(this->name, 0, "no layout for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    local->xattr_req = xdata ? dict_ref(xdata) : dict_new();
+    local->call_cnt = call_cnt = layout->cnt;
+
+    if (IA_ISDIR(fd->inode->ia_type)) {
+        local->hashed_subvol = NULL;
+        ret = dht_dir_common_set_remove_xattr(frame, this, NULL, fd, xattr,
+                                              flags, xdata, &op_errno);
+        if (ret)
+            goto err;
+    } else {
+        local->call_cnt = 1;
+        local->rebalance.xattr = dict_ref(xattr);
+        local->rebalance.flags = flags;
+
+        ret = dict_set_int8(local->xattr_req, DHT_IATT_IN_XDATA_KEY, 1);
+        if (ret) {
+            gf_msg_debug(this->name, 0,
+                         "Failed to set dictionary key %s for fd=%p",
+                         DHT_IATT_IN_XDATA_KEY, fd);
         }
 
-        local->call_cnt = call_cnt = layout->cnt;
+        STACK_WIND_COOKIE(frame, dht_file_setxattr_cbk, subvol, subvol,
+                          subvol->fops->fsetxattr, fd, xattr, flags,
+                          local->xattr_req);
+    }
+    return 0;
 
-        tmp = dict_get (xattr, "distribute.migrate-data");
-        if (tmp) {
-                if (IA_ISDIR (loc->inode->ia_type)) {
-                        op_errno = ENOTSUP;
-                        goto err;
-                }
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(fsetxattr, frame, -1, op_errno, NULL);
 
-                /* TODO: need to interpret the 'value' for more meaning
-                   (ie, 'target' subvolume given there, etc) */
-                memcpy (value, tmp->data, tmp->len);
-                if (strcmp (value, "force") == 0)
-                       forced_rebalance =
-                                GF_DHT_MIGRATE_DATA_EVEN_IF_LINK_EXISTS;
-
-                if (conf->decommission_in_progress)
-                        forced_rebalance = GF_DHT_MIGRATE_HARDLINK;
-
-                local->rebalance.target_node = dht_subvol_get_hashed (this, loc);
-                if (!local->rebalance.target_node) {
-                        gf_log (this->name, GF_LOG_ERROR, "Failed to get "
-                                "hashed subvol for %s", loc->path);
-                        op_errno = EINVAL;
-                        goto err;
-                }
+    return 0;
+}
 
-                local->rebalance.from_subvol = local->cached_subvol;
+static int
+dht_checking_pathinfo_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int op_ret, int op_errno, dict_t *xattr,
+                          dict_t *xdata)
+{
+    int i = -1;
+    int ret = -1;
+    char *value = NULL;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    xlator_t *prev = NULL;
+    int this_call_cnt = 0;
+
+    local = frame->local;
+    prev = cookie;
+    conf = this->private;
+
+    if (op_ret == -1)
+        goto out;
+
+    ret = dict_get_str(xattr, GF_XATTR_PATHINFO_KEY, &value);
+    if (ret)
+        goto out;
+
+    if (!strcmp(value, local->key)) {
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            if (conf->subvolumes[i] == prev)
+                conf->decommissioned_bricks[i] = prev;
+        }
+    }
 
-                if (local->rebalance.target_node == local->rebalance.from_subvol) {
-                        op_errno = EEXIST;
-                        goto err;
-                }
-                if (local->rebalance.target_node) {
-                        local->flags = forced_rebalance;
+out:
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        DHT_STACK_UNWIND(setxattr, frame, local->op_ret, ENOTSUP, NULL);
+    }
+    return 0;
+}
 
-                        ret = dht_start_rebalance_task (this, frame);
-                        if (!ret)
-                                return 0;
+static int
+dht_nuke_dir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                 struct iatt *postparent, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(setxattr, frame, op_ret, op_errno, NULL);
+    return 0;
+}
 
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "%s: failed to create a new synctask",
-                                loc->path);
-                }
-                op_errno = EINVAL;
-                goto err;
+static int
+dht_nuke_dir(call_frame_t *frame, xlator_t *this, loc_t *loc, data_t *tmp)
+{
+    if (!IA_ISDIR(loc->inode->ia_type)) {
+        DHT_STACK_UNWIND(setxattr, frame, -1, ENOTSUP, NULL);
+        return 0;
+    }
 
-        }
+    /* Setxattr didn't need the parent, but rmdir does. */
+    loc->parent = inode_parent(loc->inode, NULL, NULL);
+    if (!loc->parent) {
+        DHT_STACK_UNWIND(setxattr, frame, -1, ENOENT, NULL);
+        return 0;
+    }
+    gf_uuid_copy(loc->pargfid, loc->parent->gfid);
+
+    if (!loc->name && loc->path) {
+        loc->name = strrchr(loc->path, '/');
+        if (loc->name) {
+            ++(loc->name);
+        }
+    }
+
+    /*
+     * We do this instead of calling dht_rmdir_do directly for two reasons.
+     * The first is that we want to reuse all of the initialization that
+     * dht_rmdir does, so if it ever changes we'll just follow along.  The
+     * second (i.e. why we don't use STACK_WIND_TAIL) is so that we don't
+     * obscure the fact that we came in via this path instead of a genuine
+     * rmdir.  That makes debugging just a tiny bit easier.
+     */
+    STACK_WIND(frame, dht_nuke_dir_cbk, this, this->fops->rmdir, loc, 1, NULL);
+
+    return 0;
+}
 
-        tmp = dict_get (xattr, "decommission-brick");
-        if (tmp) {
-                /* This operation should happen only on '/' */
-                if (!__is_root_gfid (loc->inode->gfid)) {
-                        op_errno = ENOTSUP;
-                        goto err;
-                }
+int
+dht_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr,
+             int flags, dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    dht_methods_t *methods = NULL;
+    dht_layout_t *layout = NULL;
+    int i = 0;
+    int op_errno = EINVAL;
+    int ret = -1;
+    data_t *tmp = NULL;
+    uint32_t dir_spread = 0;
+    char value[4096] = {
+        0,
+    };
+    gf_dht_migrate_data_type_t forced_rebalance = GF_DHT_MIGRATE_DATA;
+    int call_cnt = 0;
+    uint32_t new_hash = 0;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+
+    conf = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, conf, err);
+
+    methods = &(conf->methods);
+
+    /* Rebalance daemon is allowed to set internal keys */
+    if (!conf->defrag)
+        GF_IF_INTERNAL_XATTR_GOTO(conf->wild_xattr_name, xattr, op_errno, err);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_SETXATTR);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for path=%s",
+                     loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    layout = local->layout;
+    if (!layout) {
+        gf_msg_debug(this->name, 0, "no layout for path=%s", loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    local->call_cnt = call_cnt = layout->cnt;
+    tmp = dict_get(xattr, conf->mds_xattr_key);
+    if (tmp) {
+        op_errno = ENOTSUP;
+        goto err;
+    }
+
+    tmp = dict_get(xattr, GF_XATTR_FILE_MIGRATE_KEY);
+    if (tmp) {
+        if (IA_ISDIR(loc->inode->ia_type)) {
+            op_errno = ENOTSUP;
+            goto err;
+        }
+
+        /* TODO: need to interpret the 'value' for more meaning
+           (ie, 'target' subvolume given there, etc) */
+        memcpy(value, tmp->data, tmp->len);
+        if (strcmp(value, "force") == 0)
+            forced_rebalance = GF_DHT_MIGRATE_DATA_EVEN_IF_LINK_EXISTS;
+
+        if (conf->decommission_in_progress)
+            forced_rebalance = GF_DHT_MIGRATE_HARDLINK;
+
+        if (!loc->path) {
+            op_errno = EINVAL;
+            goto err;
+        }
+
+        if (!local->loc.name)
+            local->loc.name = strrchr(local->loc.path, '/') + 1;
+
+        if (!local->loc.parent)
+            local->loc.parent = inode_parent(local->loc.inode, NULL, NULL);
+
+        if ((!local->loc.name) || (!local->loc.parent)) {
+            op_errno = EINVAL;
+            goto err;
+        }
+
+        if (gf_uuid_is_null(local->loc.pargfid))
+            gf_uuid_copy(local->loc.pargfid, local->loc.parent->gfid);
+
+        methods->migration_get_dst_subvol(this, local);
+
+        if (!local->rebalance.target_node) {
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+                   "Failed to get hashed subvol for %s", loc->path);
+            op_errno = EINVAL;
+            goto err;
+        }
+
+        local->rebalance.from_subvol = local->cached_subvol;
+
+        if (local->rebalance.target_node == local->rebalance.from_subvol) {
+            op_errno = EEXIST;
+            goto err;
+        }
+        if (local->rebalance.target_node) {
+            local->flags = forced_rebalance;
 
-                memcpy (value, tmp->data, ((tmp->len < 4095) ? tmp->len : 4095));
-                local->key = gf_strdup (value);
-                local->call_cnt = conf->subvolume_cnt;
+            frame->root->pid = GF_CLIENT_PID_DEFRAG;
 
-                for (i = 0 ; i < conf->subvolume_cnt; i++) {
-                        /* Get the pathinfo, and then compare */
-                        STACK_WIND (frame, dht_checking_pathinfo_cbk,
-                                    conf->subvolumes[i],
-                                    conf->subvolumes[i]->fops->getxattr,
-                                    loc, GF_XATTR_PATHINFO_KEY, NULL);
-                }
+            ret = dht_start_rebalance_task(this, frame);
+            if (!ret)
                 return 0;
+
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_REBALANCE_START_FAILED,
+                   "%s: failed to create a new rebalance synctask", loc->path);
         }
+        op_errno = EINVAL;
+        goto err;
+    }
 
-        tmp = dict_get (xattr, GF_XATTR_FIX_LAYOUT_KEY);
-        if (tmp) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "fixing the layout of %s", loc->path);
+    tmp = dict_get(xattr, "decommission-brick");
+    if (tmp) {
+        /* This operation should happen only on '/' */
+        if (!__is_root_gfid(loc->inode->gfid)) {
+            op_errno = ENOTSUP;
+            goto err;
+        }
 
-                ret = dht_fix_directory_layout (frame, dht_common_setxattr_cbk,
-                                                layout);
-                if (ret) {
-                        op_errno = ENOTCONN;
-                        goto err;
-                }
-                return ret;
+        memcpy(value, tmp->data, min(tmp->len, 4095));
+        local->key = gf_strdup(value);
+        local->call_cnt = conf->subvolume_cnt;
+
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            /* Get the pathinfo, and then compare */
+            STACK_WIND_COOKIE(frame, dht_checking_pathinfo_cbk,
+                              conf->subvolumes[i], conf->subvolumes[i],
+                              conf->subvolumes[i]->fops->getxattr, loc,
+                              GF_XATTR_PATHINFO_KEY, NULL);
         }
+        return 0;
+    }
 
-        tmp = dict_get (xattr, "distribute.directory-spread-count");
-        if (tmp) {
-                /* Setxattr value is packed as 'binary', not string */
-                memcpy (value, tmp->data, ((tmp->len < 4095)?tmp->len:4095));
-                ret = gf_string2uint32 (value, &dir_spread);
-                if (!ret && ((dir_spread <= conf->subvolume_cnt) &&
-                             (dir_spread > 0))) {
-                        layout->spread_cnt = dir_spread;
-
-                        ret = dht_fix_directory_layout (frame,
-                                                        dht_common_setxattr_cbk,
-                                                        layout);
-                        if (ret) {
-                                op_errno = ENOTCONN;
-                                goto err;
-                        }
-                        return ret;
-                }
-                gf_log (this->name, GF_LOG_ERROR,
-                        "wrong 'directory-spread-count' value (%s)", value);
-                op_errno = ENOTSUP;
+    tmp = dict_get(xattr, GF_XATTR_FIX_LAYOUT_KEY);
+    if (tmp) {
+        ret = dict_get_uint32(xattr, "new-commit-hash", &new_hash);
+        if (ret == 0) {
+            gf_msg_debug(this->name, 0,
+                         "updating commit hash for %s from %u to %u",
+                         uuid_utoa(loc->gfid), layout->commit_hash, new_hash);
+            layout->commit_hash = new_hash;
+
+            ret = dht_update_commit_hash_for_layout(frame);
+            if (ret) {
+                op_errno = ENOTCONN;
                 goto err;
+            }
+            return ret;
         }
 
-        for (i = 0; i < call_cnt; i++) {
-                STACK_WIND (frame, dht_err_cbk,
-                            layout->list[i].xlator,
-                            layout->list[i].xlator->fops->setxattr,
-                            loc, xattr, flags, xdata);
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_FIX_LAYOUT_INFO,
+               "fixing the layout of %s", loc->path);
+
+        ret = dht_fix_directory_layout(frame, dht_fix_layout_setxattr_cbk,
+                                       layout);
+        if (ret) {
+            op_errno = ENOTCONN;
+            goto err;
         }
+        return ret;
+    }
+
+    tmp = dict_get(xattr, "distribute.directory-spread-count");
+    if (tmp) {
+        /* Setxattr value is packed as 'binary', not string */
+        memcpy(value, tmp->data, min(tmp->len, 4095));
+        ret = gf_string2uint32(value, &dir_spread);
+        if (!ret && ((dir_spread <= conf->subvolume_cnt) && (dir_spread > 0))) {
+            layout->spread_cnt = dir_spread;
+
+            ret = dht_fix_directory_layout(frame, dht_common_setxattr_cbk,
+                                           layout);
+            if (ret) {
+                op_errno = ENOTCONN;
+                goto err;
+            }
+            return ret;
+        }
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_OPERATION_NOT_SUP,
+               "wrong 'directory-spread-count' value (%s)", value);
+        op_errno = ENOTSUP;
+        goto err;
+    }
+
+    tmp = dict_get(xattr, "glusterfs.dht.nuke");
+    if (tmp) {
+        return dht_nuke_dir(frame, this, loc, tmp);
+    }
+    local->xattr_req = xdata ? dict_ref(xdata) : dict_new();
+
+    if (IA_ISDIR(loc->inode->ia_type)) {
+        local->hashed_subvol = NULL;
+        ret = dht_dir_common_set_remove_xattr(frame, this, loc, NULL, xattr,
+                                              flags, xdata, &op_errno);
+        if (ret)
+            goto err;
+    } else {
+        local->rebalance.xattr = dict_ref(xattr);
+        local->rebalance.flags = flags;
+        local->call_cnt = 1;
 
-        return 0;
+        ret = dict_set_int8(local->xattr_req, DHT_IATT_IN_XDATA_KEY, 1);
+
+        STACK_WIND_COOKIE(frame, dht_file_setxattr_cbk, subvol, subvol,
+                          subvol->fops->setxattr, loc, xattr, flags,
+                          local->xattr_req);
+    }
+
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (setxattr, frame, -1, op_errno, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(setxattr, frame, -1, op_errno, NULL);
 
-        return 0;
+    return 0;
 }
 
-
-int
-dht_removexattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int op_ret, int op_errno, dict_t *xdata)
+static int
+dht_removexattr2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
 {
-        dht_local_t  *local = NULL;
-        int           this_call_cnt = 0;
-        call_frame_t *prev = NULL;
+    dht_local_t *local = NULL;
+    int op_errno = EINVAL;
 
-        local = frame->local;
-        prev = cookie;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        local->op_errno = op_errno;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "subvolume %s returned -1 (%s)",
-                                prev->this->name, strerror (op_errno));
-                        goto unlock;
-                }
+    if (!frame || !frame->local)
+        goto err;
 
-                local->op_ret = 0;
-        }
-unlock:
-        UNLOCK (&frame->lock);
+    local = frame->local;
+    op_errno = local->op_errno;
 
-        this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt)) {
-                DHT_STACK_UNWIND (removexattr, frame, local->op_ret,
-                                  local->op_errno, NULL);
-        }
+    local->call_cnt = 2; /* This is the second attempt */
 
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        DHT_STACK_UNWIND(removexattr, frame, local->op_ret, local->op_errno,
+                         local->rebalance.xdata);
         return 0;
-}
+    }
 
+    if (subvol == NULL)
+        goto err;
+
+    if (local->fop == GF_FOP_REMOVEXATTR) {
+        STACK_WIND_COOKIE(frame, dht_file_removexattr_cbk, subvol, subvol,
+                          subvol->fops->removexattr, &local->loc, local->key,
+                          local->xattr_req);
+    } else {
+        STACK_WIND_COOKIE(frame, dht_file_removexattr_cbk, subvol, subvol,
+                          subvol->fops->fremovexattr, local->fd, local->key,
+                          local->xattr_req);
+    }
+
+    return 0;
+
+err:
+    DHT_STACK_UNWIND(removexattr, frame, -1, op_errno, NULL);
+    return 0;
+}
 
 int
-dht_removexattr (call_frame_t *frame, xlator_t *this,
-                 loc_t *loc, const char *key, dict_t *xdata)
+dht_file_removexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int op_ret, int op_errno, dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-        dht_layout_t *layout = NULL;
-        int           call_cnt = 0;
-        dht_conf_t   *conf = NULL;
+    int ret = -1;
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    struct iatt *stbuf = NULL;
+    inode_t *inode = NULL;
+    xlator_t *subvol1 = NULL, *subvol2 = NULL;
 
-        int i;
+    local = frame->local;
+    prev = cookie;
 
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (this->private, err);
+    local->op_errno = op_errno;
 
-        conf = this->private;
+    if ((local->fop == GF_FOP_FREMOVEXATTR) &&
+        dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
 
-        GF_IF_NATIVE_XATTR_GOTO (conf->wild_xattr_name, key, op_errno, err);
+    if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->name);
+        goto out;
+    }
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
+    if (local->call_cnt != 1)
+        goto out;
 
-        local = dht_local_init (frame, loc, NULL, GF_FOP_REMOVEXATTR);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+    ret = dict_get_bin(xdata, DHT_IATT_IN_XDATA_KEY, (void **)&stbuf);
 
-        subvol = local->cached_subvol;
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for path=%s", loc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
+    if ((!op_ret) && !stbuf) {
+        goto out;
+    }
 
-        layout = local->layout;
-        if (!local->layout) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no layout for path=%s", loc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
+    local->op_ret = 0;
 
-        local->call_cnt = call_cnt = layout->cnt;
-        local->key = gf_strdup (key);
+    local->rebalance.target_op_fn = dht_removexattr2;
+    if (xdata)
+        local->rebalance.xdata = dict_ref(xdata);
 
-        for (i = 0; i < call_cnt; i++) {
-                STACK_WIND (frame, dht_removexattr_cbk,
-                            layout->list[i].xlator,
-                            layout->list[i].xlator->fops->removexattr,
-                            loc, key, NULL);
-        }
+    /* Phase 2 of migration */
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(stbuf)) {
+        ret = dht_rebalance_complete_check(this, frame);
+        if (!ret)
+            return 0;
+    }
 
-        return 0;
+    /* Phase 1 of migration */
+    if (IS_DHT_MIGRATION_PHASE1(stbuf)) {
+        inode = (local->fd) ? local->fd->inode : local->loc.inode;
 
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (removexattr, frame, -1, op_errno, NULL);
+        ret = dht_inode_ctx_get_mig_info(this, inode, &subvol1, &subvol2);
+        if (!dht_mig_info_is_invalid(local->cached_subvol, subvol1, subvol2)) {
+            dht_removexattr2(this, subvol2, frame, 0);
+            return 0;
+        }
 
-        return 0;
+        ret = dht_rebalance_in_progress_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+out:
+    if (local->fop == GF_FOP_REMOVEXATTR) {
+        DHT_STACK_UNWIND(removexattr, frame, op_ret, op_errno, xdata);
+    } else {
+        DHT_STACK_UNWIND(fremovexattr, frame, op_ret, op_errno, xdata);
+    }
+    return 0;
 }
 
 int
-dht_fremovexattr (call_frame_t *frame, xlator_t *this,
-                  fd_t *fd, const char *key, dict_t *xdata)
+dht_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                const char *key, dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-        dht_layout_t *layout = NULL;
-        int           call_cnt = 0;
-        dht_conf_t   *conf = 0;
-
-        int i;
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
+    int call_cnt = 0;
+    dht_conf_t *conf = NULL;
+    int ret = 0;
+
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+
+    GF_IF_NATIVE_XATTR_GOTO(conf->wild_xattr_name, key, op_errno, err);
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_REMOVEXATTR);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for path=%s",
+                     loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    layout = local->layout;
+    if (!local->layout) {
+        gf_msg_debug(this->name, 0, "no layout for path=%s", loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+    local->xattr_req = (xdata) ? dict_ref(xdata) : dict_new();
+
+    local->call_cnt = call_cnt = layout->cnt;
+    local->key = gf_strdup(key);
+
+    if (key && (strncmp(key, conf->mds_xattr_key, strlen(key)) == 0)) {
+        op_errno = ENOTSUP;
+        goto err;
+    }
+
+    if (IA_ISDIR(loc->inode->ia_type)) {
+        local->hashed_subvol = NULL;
+        ret = dht_dir_common_set_remove_xattr(frame, this, loc, NULL, NULL, 0,
+                                              local->xattr_req, &op_errno);
+        if (ret)
+            goto err;
 
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (this->private, err);
+    } else {
+        local->call_cnt = 1;
+        ret = dict_set_int8(local->xattr_req, DHT_IATT_IN_XDATA_KEY, 1);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_DICT_SET_FAILED,
+                   "Failed to "
+                   "set dictionary key %s for %s",
+                   DHT_IATT_IN_XDATA_KEY, loc->path);
+        }
 
-        conf = this->private;
+        STACK_WIND_COOKIE(frame, dht_file_removexattr_cbk, subvol, subvol,
+                          subvol->fops->removexattr, loc, key,
+                          local->xattr_req);
+    }
 
-        GF_IF_NATIVE_XATTR_GOTO (conf->wild_xattr_name, key, op_errno, err);
+    return 0;
 
-        VALIDATE_OR_GOTO (frame, err);
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(removexattr, frame, -1, op_errno, NULL);
 
-        local = dht_local_init (frame, NULL, fd, GF_FOP_FREMOVEXATTR);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+    return 0;
+}
 
-        subvol = local->cached_subvol;
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for inode=%s",
-                        uuid_utoa (fd->inode->gfid));
-                op_errno = EINVAL;
-                goto err;
-        }
+int
+dht_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *key,
+                 dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
+    int call_cnt = 0;
+    dht_conf_t *conf = 0;
+    int ret = 0;
+
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+
+    GF_IF_NATIVE_XATTR_GOTO(conf->wild_xattr_name, key, op_errno, err);
+
+    VALIDATE_OR_GOTO(frame, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_FREMOVEXATTR);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for inode=%s",
+                     uuid_utoa(fd->inode->gfid));
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    layout = local->layout;
+    if (!local->layout) {
+        gf_msg_debug(this->name, 0, "no layout for inode=%s",
+                     uuid_utoa(fd->inode->gfid));
+        op_errno = EINVAL;
+        goto err;
+    }
+    local->xattr_req = xdata ? dict_ref(xdata) : dict_new();
+
+    local->call_cnt = call_cnt = layout->cnt;
+    local->key = gf_strdup(key);
+
+    if (IA_ISDIR(fd->inode->ia_type)) {
+        local->hashed_subvol = NULL;
+        ret = dht_dir_common_set_remove_xattr(frame, this, NULL, fd, NULL, 0,
+                                              local->xattr_req, &op_errno);
+        if (ret)
+            goto err;
 
-        layout = local->layout;
-        if (!local->layout) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no layout for inode=%s", uuid_utoa (fd->inode->gfid));
-                op_errno = EINVAL;
-                goto err;
+    } else {
+        local->call_cnt = 1;
+        ret = dict_set_int8(local->xattr_req, DHT_IATT_IN_XDATA_KEY, 1);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_DICT_SET_FAILED,
+                   "Failed to "
+                   "set dictionary key %s for fd=%p",
+                   DHT_IATT_IN_XDATA_KEY, fd);
         }
 
-        local->call_cnt = call_cnt = layout->cnt;
-        local->key = gf_strdup (key);
-
-        for (i = 0; i < call_cnt; i++) {
-                STACK_WIND (frame, dht_removexattr_cbk,
-                            layout->list[i].xlator,
-                            layout->list[i].xlator->fops->fremovexattr,
-                            fd, key, NULL);
-        }
+        STACK_WIND_COOKIE(frame, dht_file_removexattr_cbk, subvol, subvol,
+                          subvol->fops->fremovexattr, fd, key,
+                          local->xattr_req);
+    }
 
-        return 0;
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (fremovexattr, frame, -1, op_errno, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(fremovexattr, frame, -1, op_errno, NULL);
 
-        return 0;
+    return 0;
 }
 
-
 int
-dht_fd_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-            int op_ret, int op_errno, fd_t *fd, dict_t *xdata)
+dht_fd_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+           int op_errno, fd_t *fd, dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
-        int           this_call_cnt = 0;
-        call_frame_t *prev = NULL;
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
 
-        local = frame->local;
-        prev = cookie;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        local->op_errno = op_errno;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "subvolume %s returned -1 (%s)",
-                                prev->this->name, strerror (op_errno));
-                        goto unlock;
-                }
+    local = frame->local;
+    prev = cookie;
 
-                local->op_ret = 0;
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            local->op_errno = op_errno;
+            UNLOCK(&frame->lock);
+            gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                         prev->name);
+            goto post_unlock;
         }
-unlock:
-        UNLOCK (&frame->lock);
 
-        this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt))
-                DHT_STACK_UNWIND (open, frame, local->op_ret, local->op_errno,
-                                  local->fd, NULL);
-
-        return 0;
+        local->op_ret = 0;
+    }
+    UNLOCK(&frame->lock);
+post_unlock:
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt))
+        DHT_STACK_UNWIND(open, frame, local->op_ret, local->op_errno, local->fd,
+                         NULL);
+
+    return 0;
 }
 
 /*
  * dht_normalize_stats -
  */
 static void
-dht_normalize_stats (struct statvfs *buf, unsigned long bsize,
-                     unsigned long frsize)
+dht_normalize_stats(struct statvfs *buf, unsigned long bsize,
+                    unsigned long frsize)
 {
-        double factor = 0;
+    double factor = 0;
+
+    if (buf->f_bsize != bsize) {
+        buf->f_bsize = bsize;
+    }
+
+    if (buf->f_frsize != frsize) {
+        factor = ((double)buf->f_frsize) / frsize;
+        buf->f_frsize = frsize;
+        buf->f_blocks = (fsblkcnt_t)(factor * buf->f_blocks);
+        buf->f_bfree = (fsblkcnt_t)(factor * buf->f_bfree);
+        buf->f_bavail = (fsblkcnt_t)(factor * buf->f_bavail);
+    }
+}
 
-        if (buf->f_bsize != bsize) {
-                buf->f_bsize = bsize;
+static int
+dht_statfs_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, struct statvfs *statvfs, dict_t *xdata)
+{
+    gf_boolean_t event = _gf_false;
+    qdstatfs_action_t action = qdstatfs_action_OFF;
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    int bsize = 0;
+    int frsize = 0;
+    GF_UNUSED int ret = 0;
+    unsigned long new_usage = 0;
+    unsigned long cur_usage = 0;
+
+    local = frame->local;
+    GF_ASSERT(local);
+
+    if (xdata)
+        ret = dict_get_int8(xdata, "quota-deem-statfs", (int8_t *)&event);
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            local->op_errno = op_errno;
+            goto unlock;
         }
+        if (!statvfs) {
+            op_errno = EINVAL;
+            local->op_ret = -1;
+            goto unlock;
+        }
+        local->op_ret = 0;
 
-        if (buf->f_frsize != frsize) {
-                factor = ((double) buf->f_frsize) / frsize;
-                buf->f_frsize = frsize;
-                buf->f_blocks = (fsblkcnt_t) (factor * buf->f_blocks);
-                buf->f_bfree  = (fsblkcnt_t) (factor * buf->f_bfree);
-                buf->f_bavail = (fsblkcnt_t) (factor * buf->f_bavail);
-
+        if (local->quota_deem_statfs) {
+            if (event == _gf_true) {
+                action = qdstatfs_action_COMPARE;
+            } else {
+                action = qdstatfs_action_NEGLECT;
+            }
+        } else {
+            if (event == _gf_true) {
+                action = qdstatfs_action_REPLACE;
+                local->quota_deem_statfs = _gf_true;
+            }
         }
-}
 
-int
-dht_statfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                int op_ret, int op_errno, struct statvfs *statvfs, dict_t *xdata)
-{
-        dht_local_t *local         = NULL;
-        int          this_call_cnt = 0;
-        int          bsize         = 0;
-        int          frsize        = 0;
+        if (local->quota_deem_statfs) {
+            switch (action) {
+                case qdstatfs_action_NEGLECT:
+                    goto unlock;
 
+                case qdstatfs_action_REPLACE:
+                    local->statvfs = *statvfs;
+                    goto unlock;
 
-        local = frame->local;
+                case qdstatfs_action_COMPARE:
+                    new_usage = statvfs->f_blocks - statvfs->f_bfree;
+                    cur_usage = local->statvfs.f_blocks -
+                                local->statvfs.f_bfree;
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        local->op_errno = op_errno;
-                        goto unlock;
-                }
-                local->op_ret = 0;
-
-                if (local->statvfs.f_bsize != 0) {
-                        bsize = max(local->statvfs.f_bsize, statvfs->f_bsize);
-                        frsize = max(local->statvfs.f_frsize, statvfs->f_frsize);
-                        dht_normalize_stats(&local->statvfs, bsize, frsize);
-                        dht_normalize_stats(statvfs, bsize, frsize);
-                } else {
-                        local->statvfs.f_bsize    = statvfs->f_bsize;
-                        local->statvfs.f_frsize   = statvfs->f_frsize;
-                }
-
-                local->statvfs.f_blocks  += statvfs->f_blocks;
-                local->statvfs.f_bfree   += statvfs->f_bfree;
-                local->statvfs.f_bavail  += statvfs->f_bavail;
-                local->statvfs.f_files   += statvfs->f_files;
-                local->statvfs.f_ffree   += statvfs->f_ffree;
-                local->statvfs.f_favail  += statvfs->f_favail;
-                local->statvfs.f_fsid     = statvfs->f_fsid;
-                local->statvfs.f_flag     = statvfs->f_flag;
-                local->statvfs.f_namemax  = statvfs->f_namemax;
+                    /* Take the max of the usage from subvols */
+                    if (new_usage >= cur_usage)
+                        local->statvfs = *statvfs;
+                    goto unlock;
 
+                default:
+                    break;
+            }
         }
-unlock:
-        UNLOCK (&frame->lock);
 
+        if (local->statvfs.f_bsize != 0) {
+            bsize = max(local->statvfs.f_bsize, statvfs->f_bsize);
+            frsize = max(local->statvfs.f_frsize, statvfs->f_frsize);
+            dht_normalize_stats(&local->statvfs, bsize, frsize);
+            dht_normalize_stats(statvfs, bsize, frsize);
+        } else {
+            local->statvfs.f_bsize = statvfs->f_bsize;
+            local->statvfs.f_frsize = statvfs->f_frsize;
+        }
+
+        local->statvfs.f_blocks += statvfs->f_blocks;
+        local->statvfs.f_bfree += statvfs->f_bfree;
+        local->statvfs.f_bavail += statvfs->f_bavail;
+        local->statvfs.f_files += statvfs->f_files;
+        local->statvfs.f_ffree += statvfs->f_ffree;
+        local->statvfs.f_favail += statvfs->f_favail;
+        local->statvfs.f_fsid = statvfs->f_fsid;
+        local->statvfs.f_flag = statvfs->f_flag;
+        local->statvfs.f_namemax = statvfs->f_namemax;
+    }
+unlock:
+    UNLOCK(&frame->lock);
 
-        this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt))
-                DHT_STACK_UNWIND (statfs, frame, local->op_ret, local->op_errno,
-                                  &local->statvfs, xdata);
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt))
+        DHT_STACK_UNWIND(statfs, frame, local->op_ret, local->op_errno,
+                         &local->statvfs, xdata);
 
-        return 0;
+    return 0;
 }
 
-
 int
-dht_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+dht_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        dht_local_t  *local  = NULL;
-        dht_conf_t   *conf = NULL;
-        int           op_errno = -1;
-        int           i = -1;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int op_errno = -1;
+    int i = -1;
+    inode_t *inode = NULL;
+    inode_table_t *itable = NULL;
+    static uuid_t root_gfid = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1};
+    loc_t newloc = {
+        0,
+    };
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (this->private, err);
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(this->private, err);
 
-        conf = this->private;
+    conf = this->private;
 
-        local = dht_local_init (frame, NULL, NULL, GF_FOP_STATFS);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
+    local = dht_local_init(frame, NULL, NULL, GF_FOP_STATFS);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    if (loc->inode && !IA_ISDIR(loc->inode->ia_type)) {
+        itable = loc->inode->table;
+        if (!itable) {
+            op_errno = EINVAL;
+            goto err;
         }
 
-        if (IA_ISDIR (loc->inode->ia_type)) {
-                local->call_cnt = conf->subvolume_cnt;
+        loc = &local->loc2;
 
-                for (i = 0; i < conf->subvolume_cnt; i++) {
-                        STACK_WIND (frame, dht_statfs_cbk,
-                                    conf->subvolumes[i],
-                                    conf->subvolumes[i]->fops->statfs, loc,
-                                    xdata);
-                }
-                return 0;
+        inode = inode_find(itable, root_gfid);
+        if (!inode) {
+            op_errno = EINVAL;
+            goto err;
         }
 
-        subvol = dht_subvol_get_cached (this, loc->inode);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for path=%s", loc->path);
-                op_errno = EINVAL;
-                goto err;
+        dht_build_root_loc(inode, &newloc);
+        loc = &newloc;
+    }
+
+    local->call_cnt = conf->subvolume_cnt;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        STACK_WIND(frame, dht_statfs_cbk, conf->subvolumes[i],
+                   conf->subvolumes[i]->fops->statfs, loc, xdata);
+    }
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(statfs, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
+}
+
+int
+dht_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+            dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int op_errno = -1;
+    int i = -1;
+    int ret = 0;
+    gf_boolean_t new_xdata = _gf_false;
+    xlator_t **subvolumes = NULL;
+    int call_count = 0;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+
+    local = dht_local_init(frame, loc, fd, GF_FOP_OPENDIR);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+    local->first_up_subvol = dht_first_up_subvol(this);
+
+    if (!xdata) {
+        xdata = dict_new();
+        if (!xdata) {
+            op_errno = ENOMEM;
+            goto err;
+        }
+        new_xdata = _gf_true;
+    }
+
+    ret = dict_set_uint32(xdata, conf->link_xattr_name, 256);
+    if (ret)
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+               "Failed to set dictionary value : key = %s",
+               conf->link_xattr_name);
+
+    /* dht_readdirp will wind to all subvols so open has to be sent to
+     * all subvols whether or not conf->local_subvols is set */
+
+    call_count = local->call_cnt = conf->subvolume_cnt;
+    subvolumes = conf->subvolumes;
+
+    /* In case of parallel-readdir, the readdir-ahead will be loaded
+     * below dht, in this case, if we want to enable or disable SKIP_DIRs
+     * it has to be done in opendir, so that prefetching logic in
+     * readdir-ahead, honors it */
+    for (i = 0; i < call_count; i++) {
+        if (conf->readdir_optimize == _gf_true) {
+            if (subvolumes[i] != local->first_up_subvol) {
+                ret = dict_set_int32(xdata, GF_READDIR_SKIP_DIRS, 1);
+                if (ret)
+                    gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                           "Failed to set dictionary"
+                           " value :key = %s, ret:%d",
+                           GF_READDIR_SKIP_DIRS, ret);
+            }
         }
 
-        local->call_cnt = 1;
+        STACK_WIND_COOKIE(frame, dht_fd_cbk, subvolumes[i], subvolumes[i],
+                          subvolumes[i]->fops->opendir, loc, fd, xdata);
+        dict_del(xdata, GF_READDIR_SKIP_DIRS);
+    }
 
-        STACK_WIND (frame, dht_statfs_cbk,
-                    subvol, subvol->fops->statfs, loc, xdata);
+    if (new_xdata)
+        dict_unref(xdata);
 
-        return 0;
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (statfs, frame, -1, op_errno, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(opendir, frame, -1, op_errno, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
+/* dht_readdirp_cbk creates a new dentry and dentry->inode is not assigned.
+   This functions assigns an inode if all of the following conditions are
+   true:
 
-int
-dht_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
-             dict_t *xdata)
+   * DHT has only one child. In this case the entire layout is present on
+   this single child and hence we can set complete layout in inode.
+   * backend has complete layout and there are no anomalies in it and from
+   this information layout can be constructed and set in inode.
+*/
+
+static void
+dht_populate_inode_for_dentry(xlator_t *this, xlator_t *subvol,
+                              gf_dirent_t *entry, gf_dirent_t *orig_entry)
 {
-        dht_local_t  *local  = NULL;
-        dht_conf_t   *conf = NULL;
-        int           op_errno = -1;
-        int           i = -1;
+    dht_layout_t *layout = NULL;
+    int ret = 0;
+    loc_t loc = {
+        0,
+    };
+
+    if (gf_uuid_is_null(orig_entry->d_stat.ia_gfid)) {
+        /* this skips the '..' entry for the root of the volume */
+        return;
+    }
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (this->private, err);
+    gf_uuid_copy(loc.gfid, orig_entry->d_stat.ia_gfid);
+    loc.inode = inode_ref(orig_entry->inode);
 
-        conf = this->private;
+    if (is_revalidate(&loc)) {
+        goto out;
+    }
 
-        local = dht_local_init (frame, loc, fd, GF_FOP_OPENDIR);
-        if (!local) {
-                op_errno = ENOMEM;
+    layout = dht_layout_new(this, 1);
+    if (!layout)
+        goto out;
 
-                goto err;
+    ret = dht_layout_merge(this, layout, subvol, 0, 0, orig_entry->dict);
+    if (!ret) {
+        ret = dht_layout_normalize(this, &loc, layout);
+        if (ret == 0) {
+            dht_layout_set(this, orig_entry->inode, layout);
+            entry->inode = inode_ref(orig_entry->inode);
+            layout = NULL;
         }
+    }
 
-        local->call_cnt = conf->subvolume_cnt;
+    if (layout)
+        dht_layout_unref(this, layout);
 
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                STACK_WIND (frame, dht_fd_cbk,
-                            conf->subvolumes[i],
-                            conf->subvolumes[i]->fops->opendir,
-                            loc, fd, xdata);
-        }
+out:
+    loc_wipe(&loc);
+    return;
+}
 
-        return 0;
+/* Posix returns op_errno = ENOENT to indicate that there are no more
+ * entries
+ */
+static int
+dht_readdirp_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, gf_dirent_t *orig_entries, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    gf_dirent_t entries;
+    gf_dirent_t *orig_entry = NULL;
+    gf_dirent_t *entry = NULL;
+    xlator_t *prev = NULL;
+    xlator_t *next_subvol = NULL;
+    off_t next_offset = 0;
+    int count = 0;
+    dht_layout_t *layout = NULL;
+    dht_conf_t *conf = NULL;
+    dht_methods_t *methods = NULL;
+    xlator_t *subvol = 0;
+    xlator_t *hashed_subvol = 0;
+    int ret = 0;
+    int readdir_optimize = 0;
+    inode_table_t *itable = NULL;
+    inode_t *inode = NULL;
+    gf_boolean_t skip_hashed_check = _gf_false;
+
+    INIT_LIST_HEAD(&entries.list);
+
+    prev = cookie;
+    local = frame->local;
+    GF_VALIDATE_OR_GOTO(this->name, local->fd, unwind);
+
+    itable = local->fd->inode->table;
+
+    conf = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, conf, unwind);
+
+    methods = &(conf->methods);
+
+    if (op_ret <= 0) {
+        goto done;
+    }
+
+    /* Why aren't we skipping DHT entirely in case of a single subvol?
+     * Because if this was a larger volume earlier and all but one subvol
+     * was removed, there might be stale linkto files on the subvol.
+     */
+    if (conf->subvolume_cnt == 1) {
+        /* return all directory and file entries except
+         * linkto files for a single child DHT
+         */
+        skip_hashed_check = _gf_true;
+    }
 
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (opendir, frame, -1, op_errno, NULL, NULL);
+    if (!local->layout)
+        local->layout = dht_layout_get(this, local->fd->inode);
 
-        return 0;
-}
+    layout = local->layout;
 
+    /* This will skip the entries on the subvol without a layout,
+     * hence preventing the crash but rmdir might fail with
+     * "directory not empty" errors*/
 
-int
-dht_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
-                  int op_errno, gf_dirent_t *orig_entries, dict_t *xdata)
-{
-        dht_local_t  *local = NULL;
-        gf_dirent_t   entries;
-        gf_dirent_t  *orig_entry = NULL;
-        gf_dirent_t  *entry = NULL;
-        call_frame_t *prev = NULL;
-        xlator_t     *next_subvol = NULL;
-        off_t         next_offset = 0;
-        int           count = 0;
-        dht_layout_t *layout = 0;
-        dht_conf_t   *conf   = NULL;
-        xlator_t     *subvol = 0;
-        int           ret    = 0;
-
-        INIT_LIST_HEAD (&entries.list);
-        prev = cookie;
-        local = frame->local;
-        conf  = this->private;
+    if (layout == NULL)
+        goto done;
 
-        if (op_ret < 0)
-                goto done;
+    if (conf->readdir_optimize == _gf_true)
+        readdir_optimize = 1;
 
-        if (!local->layout)
-                local->layout = dht_layout_get (this, local->fd->inode);
+    gf_msg_debug(this->name, 0, "Processing entries from %s", prev->name);
 
-        layout = local->layout;
+    list_for_each_entry(orig_entry, (&orig_entries->list), list)
+    {
+        next_offset = orig_entry->d_off;
 
-        list_for_each_entry (orig_entry, (&orig_entries->list), list) {
-                next_offset = orig_entry->d_off;
-                if (check_is_dir (NULL, (&orig_entry->d_stat), NULL) &&
-                    (prev->this != local->first_up_subvol)) {
-                        continue;
-                }
-                if (check_is_linkfile (NULL, (&orig_entry->d_stat),
-                                       orig_entry->dict,
-                                       conf->link_xattr_name)) {
-                        continue;
-                }
+        gf_msg_debug(this->name, 0, "%s: entry = %s, type = %d", prev->name,
+                     orig_entry->d_name, orig_entry->d_type);
 
-                entry = gf_dirent_for_name (orig_entry->d_name);
-                if (!entry) {
+        if (IA_ISINVAL(orig_entry->d_stat.ia_type)) {
+            /*stat failed somewhere- display this entry but the data may
+             * be inaccurate.
+             */
+            gf_msg_debug(this->name, EINVAL, "Invalid stat for %s (gfid %s)",
+                         orig_entry->d_name,
+                         uuid_utoa(orig_entry->d_stat.ia_gfid));
+        }
 
-                        goto unwind;
-                }
+        if (check_is_linkfile(NULL, (&orig_entry->d_stat), orig_entry->dict,
+                              conf->link_xattr_name)) {
+            gf_msg_debug(this->name, 0, "%s: %s is a linkto file", prev->name,
+                         orig_entry->d_name);
+            continue;
+        }
 
-                /* Do this if conf->search_unhashed is set to "auto" */
-                if (conf->search_unhashed == GF_DHT_LOOKUP_UNHASHED_AUTO) {
-                        subvol = dht_layout_search (this, layout,
-                                                    orig_entry->d_name);
-                        if (!subvol || (subvol != prev->this)) {
-                                /* TODO: Count the number of entries which need
-                                   linkfile to prove its existence in fs */
-                                layout->search_unhashed++;
-                        }
-                }
+        if (skip_hashed_check) {
+            goto list;
+        }
 
-                dht_itransform (this, prev->this, orig_entry->d_off,
-                                &entry->d_off);
-
-                entry->d_stat = orig_entry->d_stat;
-                entry->d_ino  = orig_entry->d_ino;
-                entry->d_type = orig_entry->d_type;
-                entry->d_len  = orig_entry->d_len;
-
-                if (orig_entry->dict)
-                        entry->dict = dict_ref (orig_entry->dict);
-
-                /* making sure we set the inode ctx right with layout,
-                   currently possible only for non-directories, so for
-                   directories don't set entry inodes */
-                if (!IA_ISDIR(entry->d_stat.ia_type)) {
-                        ret = dht_layout_preset (this, prev->this,
-                                                 orig_entry->inode);
-                        if (ret)
-                                gf_log (this->name, GF_LOG_WARNING,
-                                        "failed to link the layout in inode");
-                        entry->inode = inode_ref (orig_entry->inode);
-                } else if (orig_entry->inode) {
-                        dht_inode_ctx_time_update (orig_entry->inode, this,
-                                                   &entry->d_stat, 1);
+        if (check_is_dir(NULL, (&orig_entry->d_stat), NULL)) {
+            /*Directory entries filtering :
+             * a) If rebalance is running, pick from first_up_subvol
+             * b) (rebalance not running)hashed subvolume is NULL or
+             * down then filter in first_up_subvolume. Other wise the
+             * corresponding hashed subvolume will take care of the
+             * directory entry.
+             */
+            if (readdir_optimize) {
+                if (prev == local->first_up_subvol)
+                    goto list;
+                else
+                    continue;
+            }
+
+            hashed_subvol = methods->layout_search(this, layout,
+                                                   orig_entry->d_name);
+
+            if (prev == hashed_subvol)
+                goto list;
+            if ((hashed_subvol && dht_subvol_status(conf, hashed_subvol)) ||
+                (prev != local->first_up_subvol))
+                continue;
+
+            goto list;
+        }
+
+    list:
+        entry = gf_dirent_for_name(orig_entry->d_name);
+        if (!entry) {
+            goto unwind;
+        }
+
+        /* Do this if conf->search_unhashed is set to "auto" */
+        if (conf->search_unhashed == GF_DHT_LOOKUP_UNHASHED_AUTO) {
+            subvol = methods->layout_search(this, layout, orig_entry->d_name);
+            if (!subvol || (subvol != prev)) {
+                /* TODO: Count the number of entries which need
+                   linkfile to prove its existence in fs */
+                layout->search_unhashed++;
+            }
+        }
+
+        entry->d_off = orig_entry->d_off;
+        entry->d_stat = orig_entry->d_stat;
+        entry->d_ino = orig_entry->d_ino;
+        entry->d_type = orig_entry->d_type;
+        entry->d_len = orig_entry->d_len;
+
+        if (orig_entry->dict)
+            entry->dict = dict_ref(orig_entry->dict);
+
+        /* making sure we set the inode ctx right with layout,
+           currently possible only for non-directories, so for
+           directories don't set entry inodes */
+        if (IA_ISDIR(entry->d_stat.ia_type)) {
+            entry->d_stat.ia_blocks = DHT_DIR_STAT_BLOCKS;
+            entry->d_stat.ia_size = DHT_DIR_STAT_SIZE;
+            if (orig_entry->inode) {
+                dht_inode_ctx_time_update(orig_entry->inode, this,
+                                          &entry->d_stat, 1);
+
+                if (conf->subvolume_cnt == 1) {
+                    dht_populate_inode_for_dentry(this, prev, entry,
+                                                  orig_entry);
                 }
-
-                list_add_tail (&entry->list, &entries.list);
-                count++;
+            }
+        } else {
+            if (orig_entry->dict &&
+                dict_get(orig_entry->dict, conf->link_xattr_name)) {
+                /* Strip out the S and T flags set by rebalance*/
+                DHT_STRIP_PHASE1_FLAGS(&entry->d_stat);
+            }
+
+            if (orig_entry->inode) {
+                ret = dht_layout_preset(this, prev, orig_entry->inode);
+                if (ret)
+                    gf_msg(this->name, GF_LOG_WARNING, 0,
+                           DHT_MSG_LAYOUT_SET_FAILED,
+                           "failed to link the layout "
+                           "in inode for %s",
+                           orig_entry->d_name);
+
+                entry->inode = inode_ref(orig_entry->inode);
+            } else if (itable) {
+                /*
+                 * orig_entry->inode might be null if any upper
+                 * layer xlators below client set to null, to
+                 * force a lookup on the inode even if the inode
+                 * is present in the inode table. In that case
+                 * we just update the ctx to make sure we didn't
+                 * missed anything.
+                 */
+                inode = inode_find(itable, orig_entry->d_stat.ia_gfid);
+                if (inode) {
+                    ret = dht_layout_preset(this, prev, inode);
+                    if (ret)
+                        gf_msg(this->name, GF_LOG_WARNING, 0,
+                               DHT_MSG_LAYOUT_SET_FAILED,
+                               "failed to link the layout"
+                               " in inode for %s",
+                               orig_entry->d_name);
+                    inode_unref(inode);
+                    inode = NULL;
+                }
+            }
         }
-        op_ret = count;
-        /* We need to ensure that only the last subvolume's end-of-directory
-         * notification is respected so that directory reading does not stop
-         * before all subvolumes have been read. That could happen because the
-         * posix for each subvolume sends a ENOENT on end-of-directory but in
-         * distribute we're not concerned only with a posix's view of the
-         * directory but the aggregated namespace' view of the directory.
-         */
-        if (prev->this != dht_last_up_subvol (this))
-                op_errno = 0;
 
-done:
-        if (count == 0) {
-                /* non-zero next_offset means that
-                   EOF is not yet hit on the current subvol
-                */
-                if (next_offset == 0) {
-                        next_subvol = dht_subvol_next (this, prev->this);
-                } else {
-                        next_subvol = prev->this;
-                }
+        gf_msg_debug(this->name, 0, "%s: Adding entry = %s", prev->name,
+                     entry->d_name);
 
-                if (!next_subvol) {
-                        goto unwind;
-                }
+        list_add_tail(&entry->list, &entries.list);
+        count++;
+    }
 
-		if (conf->readdir_optimize == _gf_true) {
-                        if (next_subvol != local->first_up_subvol) {
-                                ret = dict_set_int32 (local->xattr,
-                                                      GF_READDIR_SKIP_DIRS, 1);
-                                if (ret)
-                                        gf_log (this->name, GF_LOG_ERROR,
-					         "dict set failed");
-		        } else {
-                                 dict_del (local->xattr,
-                                           GF_READDIR_SKIP_DIRS);
-                        }
-                }
+done:
 
-                STACK_WIND (frame, dht_readdirp_cbk,
-                            next_subvol, next_subvol->fops->readdirp,
-                            local->fd, local->size, next_offset,
-                            local->xattr);
-                return 0;
+    /* We need to ensure that only the last subvolume's end-of-directory
+     * notification is respected so that directory reading does not stop
+     * before all subvolumes have been read. That could happen because the
+     * posix for each subvolume sends a ENOENT on end-of-directory but in
+     * distribute we're not concerned only with a posix's view of the
+     * directory but the aggregated namespace' view of the directory.
+     * Possible values:
+     * op_ret == 0 and op_errno != 0
+     *   if op_errno != ENOENT : Error.Unwind.
+     *   if op_errno == ENOENT : There are no more entries on this subvol.
+     *                           Move to the next one.
+     * op_ret > 0 and count == 0 :
+     *    The subvol returned entries to dht but all were stripped out.
+     *    For example, if they were linkto files or dirs where
+     *    hashed_subvol != prev. Try to get some entries by winding
+     *    to the next subvol. This can be dangerous if parallel readdir
+     *    is enabled as it grows the stack.
+     *
+     * op_ret > 0 and count > 0:
+     *   We found some entries. Unwind even if the buffer is not full.
+     *
+     */
+
+    op_ret = count;
+    if (count == 0) {
+        /* non-zero next_offset means that
+         * EOF is not yet hit on the current subvol
+         */
+        if ((next_offset == 0) || (op_errno == ENOENT)) {
+            next_offset = 0;
+            next_subvol = dht_subvol_next(this, prev);
+        } else {
+            next_subvol = prev;
+        }
+
+        if (!next_subvol) {
+            goto unwind;
         }
 
+        if (conf->readdir_optimize == _gf_true) {
+            if (next_subvol != local->first_up_subvol) {
+                ret = dict_set_int32(local->xattr, GF_READDIR_SKIP_DIRS, 1);
+                if (ret)
+                    gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                           "Failed to set dictionary value"
+                           ":key = %s",
+                           GF_READDIR_SKIP_DIRS);
+            } else {
+                dict_del(local->xattr, GF_READDIR_SKIP_DIRS);
+            }
+        }
+
+        STACK_WIND_COOKIE(frame, dht_readdirp_cbk, next_subvol, next_subvol,
+                          next_subvol->fops->readdirp, local->fd, local->size,
+                          next_offset, local->xattr);
+        return 0;
+    }
+
 unwind:
-        if (op_ret < 0)
-                op_ret = 0;
+    /* We need to ensure that only the last subvolume's end-of-directory
+     * notification is respected so that directory reading does not stop
+     * before all subvolumes have been read. That could happen because the
+     * posix for each subvolume sends a ENOENT on end-of-directory but in
+     * distribute we're not concerned only with a posix's view of the
+     * directory but the aggregated namespace' view of the directory.
+     */
+    if (op_ret < 0)
+        op_ret = 0;
+
+    if (prev != dht_last_up_subvol(this))
+        op_errno = 0;
+
+    DHT_STACK_UNWIND(readdirp, frame, op_ret, op_errno, &entries, NULL);
+
+    gf_dirent_free(&entries);
+    return 0;
+}
 
-        DHT_STACK_UNWIND (readdirp, frame, op_ret, op_errno, &entries, NULL);
+static int
+dht_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                int op_errno, gf_dirent_t *orig_entries, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    gf_dirent_t entries;
+    gf_dirent_t *orig_entry = NULL;
+    gf_dirent_t *entry = NULL;
+    xlator_t *prev = NULL;
+    xlator_t *next_subvol = NULL;
+    off_t next_offset = 0;
+    int count = 0;
+    dht_layout_t *layout = 0;
+    xlator_t *subvol = 0;
+    dht_conf_t *conf = NULL;
+    dht_methods_t *methods = NULL;
+    gf_boolean_t skip_hashed_check = _gf_false;
 
-        gf_dirent_free (&entries);
+    INIT_LIST_HEAD(&entries.list);
 
-        return 0;
-}
+    prev = cookie;
+    local = frame->local;
 
+    conf = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, conf, done);
 
+    methods = &(conf->methods);
 
-int
-dht_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int op_ret, int op_errno, gf_dirent_t *orig_entries,
-                 dict_t *xdata)
-{
-        dht_local_t  *local = NULL;
-        gf_dirent_t   entries;
-        gf_dirent_t  *orig_entry = NULL;
-        gf_dirent_t  *entry = NULL;
-        call_frame_t *prev = NULL;
-        xlator_t     *next_subvol = NULL;
-        off_t         next_offset = 0;
-        int           count = 0;
-        dht_layout_t *layout = 0;
-        xlator_t     *subvol = 0;
-
-        INIT_LIST_HEAD (&entries.list);
-        prev = cookie;
-        local = frame->local;
+    if (op_ret <= 0)
+        goto done;
 
-        if (op_ret < 0)
-                goto done;
+    if (!local->layout)
+        local->layout = dht_layout_get(this, local->fd->inode);
 
-        if (!local->layout)
-                local->layout = dht_layout_get (this, local->fd->inode);
+    layout = local->layout;
 
-        layout = local->layout;
+    gf_msg_debug(this->name, 0, "Processing entries from %s", prev->name);
 
-        list_for_each_entry (orig_entry, (&orig_entries->list), list) {
-                next_offset = orig_entry->d_off;
+    if (conf->subvolume_cnt == 1) {
+        /*return everything*/
+        skip_hashed_check = _gf_true;
+        count = op_ret;
+        goto done;
+    }
 
-                subvol = dht_layout_search (this, layout, orig_entry->d_name);
+    list_for_each_entry(orig_entry, (&orig_entries->list), list)
+    {
+        next_offset = orig_entry->d_off;
 
-                if (!subvol || (subvol == prev->this)) {
-                        entry = gf_dirent_for_name (orig_entry->d_name);
-                        if (!entry) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "memory allocation failed :(");
-                                goto unwind;
-                        }
+        gf_msg_debug(this->name, 0, "%s: entry = %s, type = %d", prev->name,
+                     orig_entry->d_name, orig_entry->d_type);
+
+        subvol = methods->layout_search(this, layout, orig_entry->d_name);
 
-                        dht_itransform (this, prev->this, orig_entry->d_off,
-                                        &entry->d_off);
+        if (!subvol || (subvol == prev)) {
+            entry = gf_dirent_for_name(orig_entry->d_name);
+            if (!entry) {
+                gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+                       "Memory allocation failed ");
+                goto unwind;
+            }
 
-                        entry->d_ino  = orig_entry->d_ino;
-                        entry->d_type = orig_entry->d_type;
-                        entry->d_len  = orig_entry->d_len;
+            entry->d_off = orig_entry->d_off;
+            entry->d_ino = orig_entry->d_ino;
+            entry->d_type = orig_entry->d_type;
+            entry->d_len = orig_entry->d_len;
 
-                        list_add_tail (&entry->list, &entries.list);
-                        count++;
-                }
-        }
-        op_ret = count;
-        /* We need to ensure that only the last subvolume's end-of-directory
-         * notification is respected so that directory reading does not stop
-         * before all subvolumes have been read. That could happen because the
-         * posix for each subvolume sends a ENOENT on end-of-directory but in
-         * distribute we're not concerned only with a posix's view of the
-         * directory but the aggregated namespace' view of the directory.
-         */
-        if (prev->this != dht_last_up_subvol (this))
-                op_errno = 0;
+            gf_msg_debug(this->name, 0, "%s: Adding = entry %s", prev->name,
+                         entry->d_name);
 
+            list_add_tail(&entry->list, &entries.list);
+            count++;
+        }
+    }
 done:
-        if (count == 0) {
-                /* non-zero next_offset means that
-                   EOF is not yet hit on the current subvol
-                */
-                if (next_offset == 0) {
-                        next_subvol = dht_subvol_next (this, prev->this);
+    op_ret = count;
+    /* We need to ensure that only the last subvolume's end-of-directory
+     * notification is respected so that directory reading does not stop
+     * before all subvolumes have been read. That could happen because the
+     * posix for each subvolume sends a ENOENT on end-of-directory but in
+     * distribute we're not concerned only with a posix's view of the
+     * directory but the aggregated namespace' view of the directory.
+     */
+    if (count == 0) {
+        if ((next_offset == 0) || (op_errno == ENOENT)) {
+            next_offset = 0;
+            next_subvol = dht_subvol_next(this, prev);
+        } else {
+            next_subvol = prev;
+        }
+
+        if (!next_subvol) {
+            goto unwind;
+        }
+
+        STACK_WIND_COOKIE(frame, dht_readdir_cbk, next_subvol, next_subvol,
+                          next_subvol->fops->readdir, local->fd, local->size,
+                          next_offset, NULL);
+        return 0;
+    }
+
+unwind:
+    /* We need to ensure that only the last subvolume's end-of-directory
+     * notification is respected so that directory reading does not stop
+     * before all subvolumes have been read. That could happen because the
+     * posix for each subvolume sends a ENOENT on end-of-directory but in
+     * distribute we're not concerned only with a posix's view of the
+     * directory but the aggregated namespace' view of the directory.
+     */
+
+    if (prev != dht_last_up_subvol(this))
+        op_errno = 0;
+
+    if (!skip_hashed_check) {
+        DHT_STACK_UNWIND(readdir, frame, op_ret, op_errno, &entries, NULL);
+        gf_dirent_free(&entries);
+
+    } else {
+        DHT_STACK_UNWIND(readdir, frame, op_ret, op_errno, orig_entries, NULL);
+    }
+    return 0;
+}
+
+static int
+dht_do_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+               off_t yoff, int whichop, dict_t *dict)
+{
+    dht_local_t *local = NULL;
+    int op_errno = -1;
+    xlator_t *xvol = NULL;
+    int ret = 0;
+    dht_conf_t *conf = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+
+    local = dht_local_init(frame, NULL, NULL, whichop);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->fd = fd_ref(fd);
+    local->size = size;
+    local->xattr_req = (dict) ? dict_ref(dict) : NULL;
+    local->first_up_subvol = dht_first_up_subvol(this);
+    local->op_ret = -1;
+
+    dht_deitransform(this, yoff, &xvol);
+
+    /* TODO: do proper readdir */
+    if (whichop == GF_FOP_READDIRP) {
+        if (dict)
+            local->xattr = dict_ref(dict);
+        else
+            local->xattr = dict_new();
+
+        if (local->xattr) {
+            ret = dict_set_uint32(local->xattr, conf->link_xattr_name, 256);
+            if (ret)
+                gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+                       "Failed to set dictionary value"
+                       " : key = %s",
+                       conf->link_xattr_name);
+
+            if (conf->readdir_optimize == _gf_true) {
+                if (xvol != local->first_up_subvol) {
+                    ret = dict_set_int32(local->xattr, GF_READDIR_SKIP_DIRS, 1);
+                    if (ret)
+                        gf_msg(this->name, GF_LOG_ERROR, 0,
+                               DHT_MSG_DICT_SET_FAILED,
+                               "Failed to set "
+                               "dictionary value: "
+                               "key = %s",
+                               GF_READDIR_SKIP_DIRS);
                 } else {
-                        next_subvol = prev->this;
+                    dict_del(local->xattr, GF_READDIR_SKIP_DIRS);
                 }
+            }
 
-                if (!next_subvol) {
-                        goto unwind;
+            if (conf->subvolume_cnt == 1) {
+                ret = dict_set_uint32(local->xattr, conf->xattr_name, 4 * 4);
+                if (ret) {
+                    gf_msg(this->name, GF_LOG_WARNING, ENOMEM,
+                           DHT_MSG_DICT_SET_FAILED,
+                           "Failed to set dictionary "
+                           "value:key = %s ",
+                           conf->xattr_name);
                 }
-
-                STACK_WIND (frame, dht_readdir_cbk,
-                            next_subvol, next_subvol->fops->readdir,
-                            local->fd, local->size, next_offset, NULL);
-                return 0;
+            }
         }
 
-unwind:
-        if (op_ret < 0)
-                op_ret = 0;
+        STACK_WIND_COOKIE(frame, dht_readdirp_cbk, xvol, xvol,
+                          xvol->fops->readdirp, fd, size, yoff, local->xattr);
+    } else {
+        STACK_WIND_COOKIE(frame, dht_readdir_cbk, xvol, xvol,
+                          xvol->fops->readdir, fd, size, yoff, local->xattr);
+    }
 
-        DHT_STACK_UNWIND (readdir, frame, op_ret, op_errno, &entries, NULL);
+    return 0;
 
-        gf_dirent_free (&entries);
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(readdir, frame, -1, op_errno, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
-
 int
-dht_do_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-                off_t yoff, int whichop, dict_t *dict)
-{
-        dht_local_t  *local  = NULL;
-        int           op_errno = -1;
-        xlator_t     *xvol = NULL;
-        off_t         xoff = 0;
-        int           ret = 0;
-        dht_conf_t   *conf = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (this->private, err);
-
-        conf = this->private;
-
-        local = dht_local_init (frame, NULL, NULL, whichop);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+dht_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t yoff, dict_t *xdata)
+{
+    int op = GF_FOP_READDIR;
+    dht_conf_t *conf = NULL;
+    int i = 0;
 
-        local->fd = fd_ref (fd);
-        local->size = size;
-        local->xattr_req = (dict)? dict_ref (dict) : NULL;
-        local->first_up_subvol = dht_first_up_subvol (this);
+    conf = this->private;
+    if (!conf)
+        goto out;
 
-        dht_deitransform (this, yoff, &xvol, (uint64_t *)&xoff);
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (!conf->subvolume_status[i]) {
+            op = GF_FOP_READDIRP;
+            break;
+        }
+    }
 
-        /* TODO: do proper readdir */
-        if (whichop == GF_FOP_READDIRP) {
-                if (dict)
-                        local->xattr = dict_ref (dict);
-                else
-                        local->xattr = dict_new ();
+    if (conf->use_readdirp)
+        op = GF_FOP_READDIRP;
 
-                if (local->xattr) {
-                        ret = dict_set_uint32 (local->xattr,
-                                               conf->link_xattr_name, 256);
-                        if (ret)
-                                gf_log (this->name, GF_LOG_WARNING,
-                                        "failed to set '%s' key",
-                                        conf->link_xattr_name);
-			if (conf->readdir_optimize == _gf_true) {
-                                if (xvol != local->first_up_subvol) {
-				        ret = dict_set_int32 (local->xattr,
-			                               GF_READDIR_SKIP_DIRS, 1);
-				        if (ret)
-					        gf_log (this->name,
-                                                        GF_LOG_ERROR,
-						        "Dict set failed");
-                                } else {
-                                        dict_del (local->xattr,
-                                                  GF_READDIR_SKIP_DIRS);
-                                }
-			}
-                }
+out:
+    dht_do_readdir(frame, this, fd, size, yoff, op, 0);
+    return 0;
+}
 
-                STACK_WIND (frame, dht_readdirp_cbk, xvol, xvol->fops->readdirp,
-                            fd, size, xoff, local->xattr);
-        } else {
-                STACK_WIND (frame, dht_readdir_cbk, xvol, xvol->fops->readdir,
-                            fd, size, xoff, local->xattr);
-        }
+int
+dht_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+             off_t yoff, dict_t *dict)
+{
+    dht_do_readdir(frame, this, fd, size, yoff, GF_FOP_READDIRP, dict);
+    return 0;
+}
 
-        return 0;
+static int
+dht_fsyncdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
 
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (readdir, frame, -1, op_errno, NULL, NULL);
+    local = frame->local;
 
-        return 0;
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1)
+            local->op_errno = op_errno;
+        else if (op_ret == 0)
+            local->op_ret = 0;
+    }
+    UNLOCK(&frame->lock);
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt))
+        DHT_STACK_UNWIND(fsyncdir, frame, local->op_ret, local->op_errno,
+                         xdata);
+
+    return 0;
 }
 
-
 int
-dht_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-             off_t yoff, dict_t *xdata)
+dht_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd, int datasync,
+             dict_t *xdata)
 {
-        int          op = GF_FOP_READDIR;
-        dht_conf_t  *conf = NULL;
-        int          i = 0;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int op_errno = -1;
+    int i = -1;
 
-        conf = this->private;
-        if (!conf)
-                goto out;
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+    VALIDATE_OR_GOTO(this->private, err);
 
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                if (!conf->subvolume_status[i]) {
-                        op = GF_FOP_READDIRP;
-                        break;
-                }
-        }
+    conf = this->private;
 
-        if (conf->use_readdirp)
-                op = GF_FOP_READDIRP;
+    local = dht_local_init(frame, NULL, NULL, GF_FOP_FSYNCDIR);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
 
-out:
-        dht_do_readdir (frame, this, fd, size, yoff, op, 0);
-        return 0;
+    local->fd = fd_ref(fd);
+    local->call_cnt = conf->subvolume_cnt;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        STACK_WIND(frame, dht_fsyncdir_cbk, conf->subvolumes[i],
+                   conf->subvolumes[i]->fops->fsyncdir, fd, datasync, xdata);
+    }
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(fsyncdir, frame, -1, op_errno, NULL);
+
+    return 0;
 }
 
 int
-dht_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-              off_t yoff, dict_t *dict)
+dht_newfile_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                int op_errno, inode_t *inode, struct iatt *stbuf,
+                struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
 {
-        dht_do_readdir (frame, this, fd, size, yoff, GF_FOP_READDIRP, dict);
-        return 0;
+    xlator_t *prev = NULL;
+    int ret = -1;
+    dht_local_t *local = NULL;
+
+    if (op_ret == -1)
+        goto out;
+
+    local = frame->local;
+    if (!local) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    prev = cookie;
+
+    if (local->loc.parent) {
+        dht_inode_ctx_time_update(local->loc.parent, this, preparent, 0);
+        dht_inode_ctx_time_update(local->loc.parent, this, postparent, 1);
+    }
+
+    ret = dht_layout_preset(this, prev, inode);
+    if (ret < 0) {
+        gf_msg_debug(this->name, EINVAL,
+                     "could not set pre-set layout for subvolume %s",
+                     prev ? prev->name : NULL);
+        op_ret = -1;
+        op_errno = EINVAL;
+        goto out;
+    }
+    if (local->linked == _gf_true)
+        dht_linkfile_attr_heal(frame, this);
+out:
+    /*
+     * FIXME: ia_size and st_blocks of preparent and postparent do not have
+     * correct values. since, preparent and postparent buffers correspond
+     * to a directory these two members should have values equal to sum of
+     * corresponding values from each of the subvolume.
+     * See dht_iatt_merge for reference.
+     */
+    DHT_STRIP_PHASE1_FLAGS(stbuf);
+    dht_set_fixed_dir_stat(postparent);
+    dht_set_fixed_dir_stat(preparent);
+
+    if (local && local->lock[0].layout.parent_layout.locks) {
+        /* store op_errno for failure case*/
+        local->op_errno = op_errno;
+        local->refresh_layout_unlock(frame, this, op_ret, 1);
+
+        if (op_ret == 0) {
+            DHT_STACK_UNWIND(mknod, frame, op_ret, op_errno, inode, stbuf,
+                             preparent, postparent, xdata);
+        }
+    } else {
+        DHT_STACK_UNWIND(mknod, frame, op_ret, op_errno, inode, stbuf,
+                         preparent, postparent, xdata);
+    }
+
+    return 0;
 }
 
+static int
+dht_mknod_linkfile_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                              int32_t op_ret, int32_t op_errno, inode_t *inode,
+                              struct iatt *stbuf, struct iatt *preparent,
+                              struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *cached_subvol = NULL;
+    dht_conf_t *conf = NULL;
 
+    local = frame->local;
 
-int
-dht_fsyncdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int op_ret, int op_errno, dict_t *xdata)
+    if (!local || !local->cached_subvol) {
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (op_ret == -1) {
+        local->op_errno = op_errno;
+        goto err;
+    }
+
+    conf = this->private;
+    if (!conf) {
+        local->op_errno = EINVAL;
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    cached_subvol = local->cached_subvol;
+
+    if (local->params) {
+        dict_del(local->params, conf->link_xattr_name);
+        dict_del(local->params, GLUSTERFS_INTERNAL_FOP_KEY);
+    }
+
+    STACK_WIND_COOKIE(frame, dht_newfile_cbk, (void *)cached_subvol,
+                      cached_subvol, cached_subvol->fops->mknod, &local->loc,
+                      local->mode, local->rdev, local->umask, local->params);
+
+    return 0;
+err:
+    if (local && local->lock[0].layout.parent_layout.locks) {
+        local->refresh_layout_unlock(frame, this, -1, 1);
+    } else {
+        DHT_STACK_UNWIND(mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                         NULL);
+    }
+    return 0;
+}
+
+static int
+dht_mknod_wind_to_avail_subvol(call_frame_t *frame, xlator_t *this,
+                               xlator_t *subvol, loc_t *loc, dev_t rdev,
+                               mode_t mode, mode_t umask, dict_t *params)
 {
-        dht_local_t  *local = NULL;
-        int           this_call_cnt = 0;
+    dht_local_t *local = NULL;
+    xlator_t *avail_subvol = NULL;
 
+    local = frame->local;
 
-        local = frame->local;
+    if (!dht_is_subvol_filled(this, subvol)) {
+        gf_msg_debug(this->name, 0, "creating %s on %s", loc->path,
+                     subvol->name);
+
+        STACK_WIND_COOKIE(frame, dht_newfile_cbk, (void *)subvol, subvol,
+                          subvol->fops->mknod, loc, mode, rdev, umask, params);
+    } else {
+        avail_subvol = dht_free_disk_available_subvol(this, subvol, local);
+
+        if (avail_subvol != subvol) {
+            local->params = dict_ref(params);
+            local->rdev = rdev;
+            local->mode = mode;
+            local->umask = umask;
+            local->cached_subvol = avail_subvol;
+            local->hashed_subvol = subvol;
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1)
-                        local->op_errno = op_errno;
+            gf_msg_debug(this->name, 0, "creating %s on %s (link at %s)",
+                         loc->path, avail_subvol->name, subvol->name);
 
-                if (op_ret == 0)
-                        local->op_ret = 0;
+            dht_linkfile_create(frame, dht_mknod_linkfile_create_cbk, this,
+                                avail_subvol, subvol, loc);
+
+            goto out;
         }
-        UNLOCK (&frame->lock);
 
-        this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt))
-                DHT_STACK_UNWIND (fsyncdir, frame, local->op_ret,
-                                  local->op_errno, xdata);
+        gf_msg_debug(this->name, 0, "creating %s on %s", loc->path,
+                     subvol->name);
+
+        STACK_WIND_COOKIE(frame, dht_newfile_cbk, (void *)subvol, subvol,
+                          subvol->fops->mknod, loc, mode, rdev, umask, params);
+    }
+out:
+    return 0;
+}
+
+static int32_t
+dht_mknod_do(call_frame_t *frame)
+{
+    dht_local_t *local = NULL;
+    dht_layout_t *refreshed = NULL;
+    xlator_t *subvol = NULL;
+    xlator_t *this = NULL;
+    dht_conf_t *conf = NULL;
+    dht_methods_t *methods = NULL;
+
+    local = frame->local;
+
+    this = THIS;
+
+    conf = this->private;
+
+    GF_VALIDATE_OR_GOTO(this->name, conf, err);
+
+    methods = &(conf->methods);
+
+    /* We don't need parent_loc anymore */
+    loc_wipe(&local->loc);
+
+    loc_copy(&local->loc, &local->loc2);
+
+    loc_wipe(&local->loc2);
+
+    refreshed = local->selfheal.refreshed_layout;
+
+    subvol = methods->layout_search(this, refreshed, local->loc.name);
+
+    if (!subvol) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+               "no subvolume in "
+               "layout for path=%s",
+               local->loc.path);
+        local->op_errno = ENOENT;
+        goto err;
+    }
+
+    dht_mknod_wind_to_avail_subvol(frame, this, subvol, &local->loc,
+                                   local->rdev, local->mode, local->umask,
+                                   local->params);
+    return 0;
+err:
+    local->refresh_layout_unlock(frame, this, -1, 1);
+
+    return 0;
+}
+
+static int32_t
+dht_mknod_unlock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    DHT_STACK_DESTROY(frame);
+    return 0;
+}
+
+static int32_t
+dht_mknod_finish(call_frame_t *frame, xlator_t *this, int op_ret,
+                 int invoke_cbk)
+{
+    dht_local_t *local = NULL, *lock_local = NULL;
+    call_frame_t *lock_frame = NULL;
+    int lock_count = 0;
+
+    local = frame->local;
+    lock_count = dht_lock_count(local->lock[0].layout.parent_layout.locks,
+                                local->lock[0].layout.parent_layout.lk_count);
+    if (lock_count == 0)
+        goto done;
+
+    lock_frame = copy_frame(frame);
+    if (lock_frame == NULL) {
+        goto done;
+    }
+
+    lock_local = dht_local_init(lock_frame, &local->loc, NULL,
+                                lock_frame->root->op);
+    if (lock_local == NULL) {
+        goto done;
+    }
+
+    lock_local->lock[0]
+        .layout.parent_layout.locks = local->lock[0].layout.parent_layout.locks;
+    lock_local->lock[0].layout.parent_layout.lk_count =
+        local->lock[0].layout.parent_layout.lk_count;
+
+    local->lock[0].layout.parent_layout.locks = NULL;
+    local->lock[0].layout.parent_layout.lk_count = 0;
+
+    dht_unlock_inodelk(lock_frame,
+                       lock_local->lock[0].layout.parent_layout.locks,
+                       lock_local->lock[0].layout.parent_layout.lk_count,
+                       dht_mknod_unlock_cbk);
+    lock_frame = NULL;
 
+done:
+    if (lock_frame != NULL) {
+        DHT_STACK_DESTROY(lock_frame);
+    }
+
+    if (op_ret == 0)
         return 0;
+
+    DHT_STACK_UNWIND(mknod, frame, op_ret, local->op_errno, NULL, NULL, NULL,
+                     NULL, NULL);
+    return 0;
 }
 
+static int32_t
+dht_mknod_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+
+    local = frame->local;
 
-int
-dht_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd,
-              int datasync, dict_t *xdata)
+    if (!local) {
+        goto err;
+    }
+
+    if (op_ret < 0) {
+        gf_msg("DHT", GF_LOG_ERROR, 0, DHT_MSG_INODE_LK_ERROR,
+               "mknod lock failed for file: %s", local->loc2.name);
+
+        local->op_errno = op_errno;
+
+        goto err;
+    }
+
+    local->refresh_layout_unlock = dht_mknod_finish;
+
+    local->refresh_layout_done = dht_mknod_do;
+
+    dht_refresh_layout(frame);
+
+    return 0;
+err:
+    if (local)
+        dht_mknod_finish(frame, this, -1, 0);
+    else
+        DHT_STACK_UNWIND(mknod, frame, -1, EINVAL, NULL, NULL, NULL, NULL,
+                         NULL);
+    return 0;
+}
+
+static int32_t
+dht_mknod_lock(call_frame_t *frame, xlator_t *subvol)
 {
-        dht_local_t  *local  = NULL;
-        dht_conf_t   *conf = NULL;
-        int           op_errno = -1;
-        int           i = -1;
+    dht_local_t *local = NULL;
+    int count = 1, ret = -1;
+    dht_lock_t **lk_array = NULL;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (this->private, err);
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO(frame->this->name, frame->local, err);
 
-        conf = this->private;
+    local = frame->local;
 
-        local = dht_local_init (frame, NULL, NULL, GF_FOP_FSYNCDIR);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+    lk_array = GF_CALLOC(count, sizeof(*lk_array), gf_common_mt_pointer);
 
-        local->fd = fd_ref (fd);
-        local->call_cnt = conf->subvolume_cnt;
+    if (lk_array == NULL)
+        goto err;
 
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                STACK_WIND (frame, dht_fsyncdir_cbk,
-                            conf->subvolumes[i],
-                            conf->subvolumes[i]->fops->fsyncdir,
-                            fd, datasync, xdata);
-        }
+    lk_array[0] = dht_lock_new(frame->this, subvol, &local->loc, F_RDLCK,
+                               DHT_LAYOUT_HEAL_DOMAIN, NULL,
+                               IGNORE_ENOENT_ESTALE);
 
-        return 0;
+    if (lk_array[0] == NULL)
+        goto err;
 
+    local->lock[0].layout.parent_layout.locks = lk_array;
+    local->lock[0].layout.parent_layout.lk_count = count;
+
+    ret = dht_blocking_inodelk(frame, lk_array, count, dht_mknod_lock_cbk);
+
+    if (ret < 0) {
+        local->lock[0].layout.parent_layout.locks = NULL;
+        local->lock[0].layout.parent_layout.lk_count = 0;
+        goto err;
+    }
+
+    return 0;
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (fsyncdir, frame, -1, op_errno, NULL);
+    if (lk_array != NULL) {
+        dht_lock_array_free(lk_array, count);
+        GF_FREE(lk_array);
+    }
 
-        return 0;
+    return -1;
 }
 
+static int
+dht_refresh_parent_layout_resume(call_frame_t *frame, xlator_t *this, int ret,
+                                 int invoke_cbk)
+{
+    dht_local_t *local = NULL, *parent_local = NULL;
+    call_stub_t *stub = NULL;
+    call_frame_t *parent_frame = NULL;
 
-int
-dht_newfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int op_ret, int op_errno,
-                 inode_t *inode, struct iatt *stbuf, struct iatt *preparent,
-                 struct iatt *postparent, dict_t *xdata)
+    local = frame->local;
+
+    stub = local->stub;
+    local->stub = NULL;
+
+    parent_frame = stub->frame;
+    parent_local = parent_frame->local;
+
+    if (ret < 0) {
+        parent_local->op_ret = -1;
+        parent_local->op_errno = local->op_errno ? local->op_errno : EIO;
+    } else {
+        parent_local->op_ret = 0;
+    }
+
+    call_resume(stub);
+
+    DHT_STACK_DESTROY(frame);
+
+    return 0;
+}
+
+static int
+dht_refresh_parent_layout_done(call_frame_t *frame)
 {
-        xlator_t     *prev = NULL;
-        int           ret = -1;
-        dht_local_t  *local = NULL;
+    dht_local_t *local = NULL;
+    int ret = 0;
 
+    local = frame->local;
 
-        if (op_ret == -1)
-                goto out;
+    if (local->op_ret < 0) {
+        ret = -1;
+        goto resume;
+    }
 
-        local = frame->local;
-        if (!local) {
-                op_ret = -1;
-                op_errno = EINVAL;
-                goto out;
-        }
+    dht_layout_set(frame->this, local->loc.inode,
+                   local->selfheal.refreshed_layout);
 
-        prev = cookie;
+resume:
+    dht_refresh_parent_layout_resume(frame, frame->this, ret, 1);
+    return 0;
+}
 
-        if (local->loc.parent) {
+static int
+dht_handle_parent_layout_change(xlator_t *this, call_stub_t *stub)
+{
+    call_frame_t *refresh_frame = NULL, *frame = NULL;
+    dht_local_t *refresh_local = NULL, *local = NULL;
 
-                dht_inode_ctx_time_update (local->loc.parent, this,
-                                           preparent, 0);
-                dht_inode_ctx_time_update (local->loc.parent, this,
-                                           postparent, 1);
-        }
+    frame = stub->frame;
+    local = frame->local;
 
-        ret = dht_layout_preset (this, prev, inode);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "could not set pre-set layout for subvolume %s",
-                        prev? prev->name: NULL);
-                op_ret   = -1;
-                op_errno = EINVAL;
-                goto out;
-        }
-        if (local->linked == _gf_true)
-                dht_linkfile_attr_heal (frame, this);
-out:
-        /*
-         * FIXME: ia_size and st_blocks of preparent and postparent do not have
-         * correct values. since, preparent and postparent buffers correspond
-         * to a directory these two members should have values equal to sum of
-         * corresponding values from each of the subvolume.
-         * See dht_iatt_merge for reference.
-         */
-        DHT_STRIP_PHASE1_FLAGS (stbuf);
-        DHT_STACK_UNWIND (mknod, frame, op_ret, op_errno, inode, stbuf,
-                          preparent, postparent, xdata);
-        return 0;
+    refresh_frame = copy_frame(frame);
+    if (!refresh_frame) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+               "mem allocation failed for refresh_frame");
+        return -1;
+    }
+
+    refresh_local = dht_local_init(refresh_frame, NULL, NULL, stub->fop);
+    if (!refresh_local) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+               "mem allocation failed for refresh_local");
+        return -1;
+    }
+
+    refresh_local->loc.inode = inode_ref(local->loc.parent);
+    gf_uuid_copy(refresh_local->loc.gfid, local->loc.parent->gfid);
+
+    refresh_local->stub = stub;
+
+    refresh_local->refresh_layout_unlock = dht_refresh_parent_layout_resume;
+    refresh_local->refresh_layout_done = dht_refresh_parent_layout_done;
+
+    dht_refresh_layout(refresh_frame);
+    return 0;
 }
 
-int
-dht_mknod_linkfile_create_cbk (call_frame_t *frame, void *cookie,
-                               xlator_t *this,
-                               int32_t op_ret, int32_t op_errno,
-                               inode_t *inode, struct iatt *stbuf,
-                               struct iatt *preparent, struct iatt *postparent,
-                               dict_t *xdata)
+static int32_t
+dht_call_mkdir_stub(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
-        xlator_t     *cached_subvol = NULL;
+    dht_local_t *local = NULL;
+    call_stub_t *stub = NULL;
 
-        if (op_ret == -1)
-                goto err;
+    local = frame->local;
+    stub = local->stub;
+    local->stub = NULL;
 
-        local = frame->local;
-        if (!local || !local->cached_subvol) {
-                op_errno = EINVAL;
-                goto err;
-        }
+    if (op_ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+    } else {
+        local->op_ret = 0;
+    }
 
-        cached_subvol = local->cached_subvol;
+    call_resume(stub);
 
-        STACK_WIND_COOKIE (frame, dht_newfile_cbk, (void *)cached_subvol,
-                           cached_subvol, cached_subvol->fops->mknod,
-                           &local->loc, local->mode, local->rdev, local->umask,
-                           local->params);
+    return 0;
+}
 
-        return 0;
+static int32_t
+dht_guard_parent_layout_and_namespace(xlator_t *subvol, call_stub_t *stub)
+{
+    dht_local_t *local = NULL;
+    int ret = -1;
+    loc_t *loc = NULL;
+    xlator_t *hashed_subvol = NULL, *this = NULL;
+    ;
+    call_frame_t *frame = NULL;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    int32_t *parent_disk_layout = NULL;
+    dht_layout_t *parent_layout = NULL;
+    dht_conf_t *conf = NULL;
+
+    GF_VALIDATE_OR_GOTO("dht", stub, err);
+
+    frame = stub->frame;
+    this = frame->this;
+
+    conf = this->private;
+
+    local = frame->local;
+
+    local->stub = stub;
+
+    /* TODO: recheck whether we should lock on src or dst if we do similar
+     * stale layout checks for rename.
+     */
+    loc = &stub->args.loc;
+
+    gf_uuid_unparse(loc->parent->gfid, pgfid);
+
+    if (local->params == NULL) {
+        local->params = dict_new();
+        if (local->params == NULL) {
+            local->op_errno = ENOMEM;
+            gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+                   DHT_MSG_PARENT_LAYOUT_CHANGED,
+                   "%s (%s/%s) (path: %s): "
+                   "dict allocation failed",
+                   gf_fop_list[stub->fop], pgfid, loc->name, loc->path);
+            goto err;
+        }
+    }
+
+    hashed_subvol = dht_subvol_get_hashed(this, loc);
+    if (hashed_subvol == NULL) {
+        local->op_errno = EINVAL;
+
+        gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "%s (%s/%s) (path: %s): "
+               "hashed subvolume not found",
+               gf_fop_list[stub->fop], pgfid, loc->name, loc->path);
+        goto err;
+    }
+
+    parent_layout = dht_layout_get(this, loc->parent);
+
+    ret = dht_disk_layout_extract_for_subvol(this, parent_layout, hashed_subvol,
+                                             &parent_disk_layout);
+    if (ret == -1) {
+        local->op_errno = EINVAL;
+        gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "%s (%s/%s) (path: %s): "
+               "extracting in-memory layout of parent failed. ",
+               gf_fop_list[stub->fop], pgfid, loc->name, loc->path);
+        goto err;
+    }
+
+    memcpy((void *)local->parent_disk_layout, (void *)parent_disk_layout,
+           sizeof(local->parent_disk_layout));
+
+    dht_layout_unref(this, parent_layout);
+    parent_layout = NULL;
+
+    ret = dict_set_str(local->params, GF_PREOP_PARENT_KEY, conf->xattr_name);
+    if (ret < 0) {
+        local->op_errno = -ret;
+        gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "%s (%s/%s) (path: %s): "
+               "setting %s key in params dictionary failed. ",
+               gf_fop_list[stub->fop], pgfid, loc->name, loc->path,
+               GF_PREOP_PARENT_KEY);
+        goto err;
+    }
+
+    ret = dict_set_bin(local->params, conf->xattr_name, parent_disk_layout,
+                       4 * 4);
+    if (ret < 0) {
+        local->op_errno = -ret;
+        gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "%s (%s/%s) (path: %s): "
+               "setting parent-layout in params dictionary failed. ",
+               gf_fop_list[stub->fop], pgfid, loc->name, loc->path);
+        goto err;
+    }
+
+    parent_disk_layout = NULL;
+    local->hashed_subvol = hashed_subvol;
+
+    local->current = &local->lock[0];
+    ret = dht_protect_namespace(frame, loc, hashed_subvol, &local->current->ns,
+                                dht_call_mkdir_stub);
+    if (ret < 0)
+        goto err;
+
+    return 0;
 err:
-        DHT_STACK_UNWIND (mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL,
-                          NULL);
-        return 0;
+
+    if (parent_disk_layout != NULL)
+        GF_FREE(parent_disk_layout);
+
+    if (parent_layout != NULL)
+        dht_layout_unref(this, parent_layout);
+
+    return -1;
 }
 
 int
-dht_mknod (call_frame_t *frame, xlator_t *this,
-           loc_t *loc, mode_t mode, dev_t rdev, mode_t umask, dict_t *params)
+dht_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+          dev_t rdev, mode_t umask, dict_t *params)
 {
-        xlator_t    *subvol = NULL;
-        int          op_errno = -1;
-        xlator_t    *avail_subvol = NULL;
-        dht_local_t *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    int i = 0;
+    int ret = 0;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+
+    conf = this->private;
+
+    dht_get_du_info(frame, this, loc);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_MKNOD);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = dht_subvol_get_hashed(this, loc);
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no subvolume in layout for path=%s",
+                     loc->path);
+        op_errno = EIO;
+        goto err;
+    }
+
+    /* Post remove-brick, the client layout may not be in sync with
+     * disk layout because of lack of lookup. Hence,a mknod call
+     * may fall on the decommissioned brick.  Hence, if the
+     * hashed_subvol is part of decommissioned bricks  list, do a
+     * lookup on parent dir. If a fix-layout is already done by the
+     * remove-brick process, the parent directory layout will be in
+     * sync with that of the disk. If fix-layout is still ending
+     * on the parent directory, we can let the file get created on
+     * the decommissioned brick which will be eventually migrated to
+     * non-decommissioned brick based on the new layout.
+     */
+
+    if (conf->decommission_subvols_cnt) {
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            if (conf->decommissioned_bricks[i] &&
+                conf->decommissioned_bricks[i] == subvol) {
+                gf_msg_debug(this->name, 0,
+                             "hashed subvol:%s is "
+                             "part of decommission brick list for "
+                             "file: %s",
+                             subvol->name, loc->path);
+
+                /* dht_refresh_layout needs directory info in
+                 * local->loc. Hence, storing the parent_loc in
+                 * local->loc and storing the create context in
+                 * local->loc2. We will restore this information
+                 * in dht_creation do */
+
+                ret = loc_copy(&local->loc2, &local->loc);
+                if (ret) {
+                    gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+                           "loc_copy failed %s", loc->path);
 
-        dht_get_du_info (frame, this, loc);
+                    goto err;
+                }
 
-        local = dht_local_init (frame, loc, NULL, GF_FOP_MKNOD);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+                local->params = dict_ref(params);
+                local->rdev = rdev;
+                local->mode = mode;
+                local->umask = umask;
 
-        subvol = dht_subvol_get_hashed (this, loc);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no subvolume in layout for path=%s",
-                        loc->path);
-                op_errno = ENOENT;
-                goto err;
-        }
+                loc_wipe(&local->loc);
 
-        if (!dht_is_subvol_filled (this, subvol)) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "creating %s on %s", loc->path, subvol->name);
+                ret = dht_build_parent_loc(this, &local->loc, loc, &op_errno);
 
-                STACK_WIND_COOKIE (frame, dht_newfile_cbk, (void *)subvol,
-                                   subvol, subvol->fops->mknod, loc, mode,
-                                   rdev, umask, params);
-        } else {
+                if (ret) {
+                    gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_LOC_FAILED,
+                           "parent loc build failed");
+                    goto err;
+                }
 
-                avail_subvol = dht_free_disk_available_subvol (this, subvol,
-                                                               local);
-                if (avail_subvol != subvol) {
-                        /* Choose the minimum filled volume, and create the
-                           files there */
-
-                        local->params = dict_ref (params);
-                        local->cached_subvol = avail_subvol;
-                        local->mode = mode;
-                        local->rdev = rdev;
-                        local->umask = umask;
-                        dht_linkfile_create (frame,
-                                             dht_mknod_linkfile_create_cbk,
-                                             this, avail_subvol, subvol, loc);
-                } else {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "creating %s on %s", loc->path, subvol->name);
+                ret = dht_mknod_lock(frame, subvol);
 
-                        STACK_WIND_COOKIE (frame, dht_newfile_cbk,
-                                           (void *)subvol, subvol,
-                                           subvol->fops->mknod, loc, mode,
-                                           rdev, umask, params);
+                if (ret < 0) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INODE_LK_ERROR,
+                           "locking parent failed");
+                    goto err;
                 }
+
+                goto done;
+            }
         }
+    }
 
-        return 0;
+    dht_mknod_wind_to_avail_subvol(frame, this, subvol, loc, rdev, mode, umask,
+                                   params);
+
+done:
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (mknod, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
+int
+dht_symlink(call_frame_t *frame, xlator_t *this, const char *linkname,
+            loc_t *loc, mode_t umask, dict_t *params)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_SYMLINK);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = dht_subvol_get_hashed(this, loc);
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no subvolume in layout for path=%s",
+                     loc->path);
+        op_errno = EIO;
+        goto err;
+    }
+
+    gf_msg_trace(this->name, 0, "creating %s on %s", loc->path, subvol->name);
+
+    STACK_WIND_COOKIE(frame, dht_newfile_cbk, (void *)subvol, subvol,
+                      subvol->fops->symlink, linkname, loc, umask, params);
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(link, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+
+    return 0;
+}
 
 int
-dht_symlink (call_frame_t *frame, xlator_t *this,
-             const char *linkname, loc_t *loc, mode_t umask, dict_t *params)
+dht_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+           dict_t *xdata)
 {
-        xlator_t    *subvol = NULL;
-        int          op_errno = -1;
-        dht_local_t *local = NULL;
+    xlator_t *cached_subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_UNLINK);
+    if (!local) {
+        op_errno = ENOMEM;
+
+        goto err;
+    }
+
+    cached_subvol = local->cached_subvol;
+    if (!cached_subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for path=%s",
+                     loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    local->flags = xflag;
+    STACK_WIND_COOKIE(frame, dht_unlink_cbk, cached_subvol, cached_subvol,
+                      cached_subvol->fops->unlink, loc, xflag, xdata);
+
+    return 0;
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(unlink, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
+static int
+dht_remove_stale_linkto_cbk(int ret, call_frame_t *sync_frame, void *data)
+{
+    DHT_STACK_DESTROY(sync_frame);
+    return 0;
+}
 
-        local = dht_local_init (frame, loc, NULL, GF_FOP_SYMLINK);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+static int
+dht_remove_stale_linkto(void *data)
+{
+    call_frame_t *frame = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *this = NULL;
+    dict_t *xdata_in = NULL;
+    int ret = 0;
+
+    GF_VALIDATE_OR_GOTO("dht", data, out);
+
+    frame = data;
+    local = frame->local;
+    this = frame->this;
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", local, out);
+    GF_VALIDATE_OR_GOTO("dht", local->link_subvol, out);
+
+    xdata_in = dict_new();
+    if (!xdata_in)
+        goto out;
+
+    ret = dht_fill_dict_to_avoid_unlink_of_migrating_file(xdata_in);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, 0,
+               "Failed to set keys for stale linkto"
+               "deletion on path %s",
+               local->loc.path);
+        goto out;
+    }
+
+    ret = syncop_unlink(local->link_subvol, &local->loc, xdata_in, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, 0,
+               "Removal of linkto failed"
+               " on path %s at subvol %s",
+               local->loc.path, local->link_subvol->name);
+    }
+out:
+    if (xdata_in)
+        dict_unref(xdata_in);
+    return ret;
+}
 
-        subvol = dht_subvol_get_hashed (this, loc);
+static int
+dht_link_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+             int op_errno, inode_t *inode, struct iatt *stbuf,
+             struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int ret = -1;
+    gf_boolean_t stbuf_merged = _gf_false;
+    xlator_t *subvol = NULL;
+    call_frame_t *cleanup_frame = NULL;
+    dht_local_t *cleanup_local = NULL;
+
+    local = frame->local;
+
+    if (op_ret == -1) {
+        /* Remove the linkto if exists */
+        if (local->linked) {
+            cleanup_frame = create_frame(this, this->ctx->pool);
+            if (cleanup_frame) {
+                cleanup_local = dht_local_init(cleanup_frame, &local->loc2,
+                                               NULL, 0);
+                if (!cleanup_local || !local->link_subvol) {
+                    DHT_STACK_DESTROY(cleanup_frame);
+                    goto out;
+                }
+                cleanup_local->link_subvol = local->link_subvol;
+                FRAME_SU_DO(cleanup_frame, dht_local_t);
+                ret = synctask_new(this->ctx->env, dht_remove_stale_linkto,
+                                   dht_remove_stale_linkto_cbk, cleanup_frame,
+                                   cleanup_frame);
+            }
+        }
+        /* No continuation on DHT inode missing errors, as we should
+         * then have a good stbuf that states P2 happened. We would
+         * get inode missing if, the file completed migrated between
+         * the lookup and the link call */
+        goto out;
+    }
+
+    /* Update parent on success, even if P1/2 checks are positive.
+     * The second call on success will further update the parent */
+    if (local->loc.parent) {
+        dht_inode_ctx_time_update(local->loc.parent, this, preparent, 0);
+        dht_inode_ctx_time_update(local->loc.parent, this, postparent, 1);
+    }
+
+    /* Update linkto attrs, if this is the first call and non-P2,
+     * if we detect P2 then we need to trust the attrs from the
+     * second call, not the first */
+    if (local->linked == _gf_true &&
+        ((local->call_cnt == 1 && !IS_DHT_MIGRATION_PHASE2(stbuf)) ||
+         (local->call_cnt != 1 && IS_DHT_MIGRATION_PHASE2(&local->stbuf)))) {
+        dht_iatt_merge(this, &local->stbuf, stbuf);
+        stbuf_merged = _gf_true;
+        dht_linkfile_attr_heal(frame, this);
+    }
+
+    /* No further P1/2 checks if we are in the second iteration of
+     * the call */
+    if (local->call_cnt != 1) {
+        goto out;
+    } else {
+        /* Preserve the return values, in case the migration decides
+         * to recreate the link on the same subvol that the current
+         * hased for the link was created on. */
+        dht_iatt_merge(this, &local->preparent, preparent);
+        dht_iatt_merge(this, &local->postparent, postparent);
+        if (!stbuf_merged) {
+            dht_iatt_merge(this, &local->stbuf, stbuf);
+            stbuf_merged = _gf_true;
+        }
+
+        local->inode = inode_ref(inode);
+    }
+
+    local->op_ret = op_ret;
+    local->op_errno = op_errno;
+    local->rebalance.target_op_fn = dht_link2;
+    dht_set_local_rebalance(this, local, stbuf, preparent, postparent, xdata);
+
+    /* Check if the rebalance phase2 is true */
+    if (IS_DHT_MIGRATION_PHASE2(stbuf)) {
+        ret = dht_inode_ctx_get_mig_info(this, local->loc.inode, NULL, &subvol);
         if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no subvolume in layout for path=%s",
-                        loc->path);
-                op_errno = ENOENT;
-                goto err;
-        }
+            /* Phase 2 of migration */
+            ret = dht_rebalance_complete_check(this, frame);
+            if (!ret)
+                return 0;
+        } else {
+            dht_link2(this, subvol, frame, 0);
+            return 0;
+        }
+    }
+
+    /* Check if the rebalance phase1 is true */
+    if (IS_DHT_MIGRATION_PHASE1(stbuf)) {
+        ret = dht_inode_ctx_get_mig_info(this, local->loc.inode, NULL, &subvol);
+        if (subvol) {
+            dht_link2(this, subvol, frame, 0);
+            return 0;
+        }
+        ret = dht_rebalance_in_progress_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+out:
+    DHT_STRIP_PHASE1_FLAGS(stbuf);
+
+    dht_set_fixed_dir_stat(preparent);
+    dht_set_fixed_dir_stat(postparent);
+    DHT_STACK_UNWIND(link, frame, op_ret, op_errno, inode, stbuf, preparent,
+                     postparent, NULL);
+
+    return 0;
+}
+
+static int
+dht_link2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
+{
+    dht_local_t *local = NULL;
+    int op_errno = EINVAL;
+
+    local = frame->local;
+    if (!local)
+        goto err;
+
+    op_errno = local->op_errno;
 
-        gf_log (this->name, GF_LOG_TRACE,
-                "creating %s on %s", loc->path, subvol->name);
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        dht_set_fixed_dir_stat(&local->preparent);
+        dht_set_fixed_dir_stat(&local->postparent);
 
-        STACK_WIND_COOKIE (frame, dht_newfile_cbk, (void *)subvol, subvol,
-                           subvol->fops->symlink, linkname, loc, umask,
-                           params);
+        DHT_STACK_UNWIND(link, frame, local->op_ret, op_errno, local->inode,
+                         &local->stbuf, &local->preparent, &local->postparent,
+                         NULL);
+        return 0;
+    }
+
+    if (subvol == NULL) {
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    /* Second call to create link file could result in EEXIST as the
+     * first call created the linkto in the currently
+     * migrating subvol, which could be the new hashed subvol */
+    if (local->link_subvol == subvol) {
+        DHT_STRIP_PHASE1_FLAGS(&local->stbuf);
+        dht_set_fixed_dir_stat(&local->preparent);
+        dht_set_fixed_dir_stat(&local->postparent);
+        DHT_STACK_UNWIND(link, frame, 0, 0, local->inode, &local->stbuf,
+                         &local->preparent, &local->postparent, NULL);
 
         return 0;
+    }
+
+    local->call_cnt = 2;
 
+    STACK_WIND(frame, dht_link_cbk, subvol, subvol->fops->link, &local->loc,
+               &local->loc2, local->xattr_req);
+
+    return 0;
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (link, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL, NULL);
+    DHT_STACK_UNWIND(link, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
-
-int
-dht_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
-            dict_t *xdata)
+static int
+dht_link_linkfile_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, inode_t *inode,
+                      struct iatt *stbuf, struct iatt *preparent,
+                      struct iatt *postparent, dict_t *xdata)
 {
-        xlator_t    *cached_subvol = NULL;
-        xlator_t    *hashed_subvol = NULL;
-        int          op_errno = -1;
-        dht_local_t *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-
-        if (dht_filter_loc_subvol_key (this, loc, &local->loc,
-                                       &cached_subvol)) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "unlinking %s on %s (given path %s)",
-                        local->loc.path, cached_subvol->name, loc->path);
-                STACK_WIND (frame, dht_unlink_cbk,
-                            cached_subvol, cached_subvol->fops->unlink,
-                            &local->loc, xflag, xdata);
-                goto done;
-        }
+    dht_local_t *local = NULL;
+    xlator_t *srcvol = NULL;
 
-        local = dht_local_init (frame, loc, NULL, GF_FOP_UNLINK);
-        if (!local) {
-                op_errno = ENOMEM;
+    if (op_ret == -1)
+        goto err;
 
-                goto err;
-        }
+    local = frame->local;
+    srcvol = local->linkfile.srcvol;
 
-        hashed_subvol = dht_subvol_get_hashed (this, loc);
-        if (!hashed_subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no subvolume in layout for path=%s",
-                        loc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
+    STACK_WIND(frame, dht_link_cbk, srcvol, srcvol->fops->link, &local->loc,
+               &local->loc2, local->xattr_req);
 
-        cached_subvol = local->cached_subvol;
-        if (!cached_subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for path=%s", loc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
+    return 0;
 
-        local->flags = xflag;
-        if (hashed_subvol != cached_subvol) {
-                STACK_WIND (frame, dht_unlink_linkfile_cbk,
-                            hashed_subvol, hashed_subvol->fops->unlink, loc,
-                            xflag, xdata);
-        } else {
-                STACK_WIND (frame, dht_unlink_cbk,
-                            cached_subvol, cached_subvol->fops->unlink, loc,
-                            xflag, xdata);
-        }
-done:
-        return 0;
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (unlink, frame, -1, op_errno, NULL, NULL, NULL);
+    DHT_STRIP_PHASE1_FLAGS(stbuf);
+    dht_set_fixed_dir_stat(preparent);
+    dht_set_fixed_dir_stat(postparent);
+    DHT_STACK_UNWIND(link, frame, op_ret, op_errno, inode, stbuf, preparent,
+                     postparent, xdata);
 
-        return 0;
+    return 0;
 }
 
+int
+dht_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+         dict_t *xdata)
+{
+    xlator_t *cached_subvol = NULL;
+    xlator_t *hashed_subvol = NULL;
+    int op_errno = -1;
+    int ret = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(oldloc, err);
+    VALIDATE_OR_GOTO(newloc, err);
+
+    local = dht_local_init(frame, oldloc, NULL, GF_FOP_LINK);
+    if (!local) {
+        op_errno = ENOMEM;
+
+        goto err;
+    }
+    local->call_cnt = 1;
+
+    cached_subvol = local->cached_subvol;
+    if (!cached_subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for path=%s",
+                     oldloc->path);
+        op_errno = ENOENT;
+        goto err;
+    }
+
+    hashed_subvol = dht_subvol_get_hashed(this, newloc);
+    if (!hashed_subvol) {
+        gf_msg_debug(this->name, 0, "no subvolume in layout for path=%s",
+                     newloc->path);
+        op_errno = EIO;
+        goto err;
+    }
+
+    ret = loc_copy(&local->loc2, newloc);
+    if (ret == -1) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    if (hashed_subvol != cached_subvol) {
+        gf_uuid_copy(local->gfid, oldloc->inode->gfid);
+        dht_linkfile_create(frame, dht_link_linkfile_cbk, this, cached_subvol,
+                            hashed_subvol, newloc);
+    } else {
+        STACK_WIND(frame, dht_link_cbk, cached_subvol,
+                   cached_subvol->fops->link, oldloc, newloc, xdata);
+    }
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(link, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+
+    return 0;
+}
 
 int
-dht_link_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-              int op_ret, int op_errno,
-              inode_t *inode, struct iatt *stbuf, struct iatt *preparent,
-              struct iatt *postparent, dict_t *xdata)
+dht_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, fd_t *fd, inode_t *inode, struct iatt *stbuf,
+               struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
 {
-        call_frame_t *prev = NULL;
-        dht_layout_t *layout = NULL;
-        dht_local_t  *local = NULL;
+    xlator_t *prev = NULL;
+    int ret = -1;
+    dht_local_t *local = NULL;
+    gf_boolean_t parent_layout_changed = _gf_false;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    xlator_t *subvol = NULL;
+
+    local = frame->local;
+
+    local = frame->local;
+    if (!local) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    if (op_ret == -1) {
+        local->op_errno = op_errno;
+        parent_layout_changed = (xdata &&
+                                 dict_get(xdata, GF_PREOP_CHECK_FAILED))
+                                    ? _gf_true
+                                    : _gf_false;
+
+        if (parent_layout_changed) {
+            if (local && local->lock[0].layout.parent_layout.locks) {
+                /* Returning failure as the layout could not be fixed even under
+                 * the lock */
+                goto out;
+            }
 
-        prev = cookie;
+            gf_uuid_unparse(local->loc.parent->gfid, pgfid);
+            gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_PARENT_LAYOUT_CHANGED,
+                   "create (%s/%s) (path: %s): parent layout "
+                   "changed. Attempting a layout refresh and then a "
+                   "retry",
+                   pgfid, local->loc.name, local->loc.path);
 
-        local = frame->local;
+            /*
+              dht_refresh_layout needs directory info in local->loc.Hence,
+              storing the parent_loc in local->loc and storing the create
+              context in local->loc2. We will restore this information in
+              dht_creation_do.
+             */
+
+            loc_wipe(&local->loc2);
+
+            ret = loc_copy(&local->loc2, &local->loc);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+                       "loc_copy failed %s", local->loc.path);
 
-        if (op_ret == -1)
                 goto out;
+            }
 
-        layout = dht_layout_for_subvol (this, prev->this);
-        if (!layout) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no pre-set layout for subvolume %s",
-                        prev->this->name);
-                op_ret   = -1;
-                op_errno = EINVAL;
+            loc_wipe(&local->loc);
+
+            ret = dht_build_parent_loc(this, &local->loc, &local->loc2,
+                                       &op_errno);
+
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_LOC_FAILED,
+                       "parent loc build failed");
                 goto out;
-        }
+            }
 
-        if (local->loc.parent) {
-                dht_inode_ctx_time_update (local->loc.parent, this,
-                                           preparent, 0);
-                dht_inode_ctx_time_update (local->loc.parent, this,
-                                           postparent, 1);
-        }
-        if (local->linked == _gf_true) {
-                local->stbuf = *stbuf;
-                dht_linkfile_attr_heal (frame, this);
+            subvol = dht_subvol_get_hashed(this, &local->loc2);
+
+            ret = dht_create_lock(frame, subvol);
+            if (ret < 0) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INODE_LK_ERROR,
+                       "locking parent failed");
+                goto out;
+            }
+
+            return 0;
         }
-out:
-        DHT_STRIP_PHASE1_FLAGS (stbuf);
-        DHT_STACK_UNWIND (link, frame, op_ret, op_errno, inode, stbuf, preparent,
-                          postparent, NULL);
 
-        return 0;
-}
+        goto out;
+    }
 
+    prev = cookie;
 
-int
-dht_link_linkfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int op_ret, int op_errno,
-                       inode_t *inode, struct iatt *stbuf,
-                       struct iatt *preparent, struct iatt *postparent,
-                       dict_t *xdata)
-{
-        dht_local_t  *local = NULL;
-        xlator_t     *srcvol = NULL;
+    if (local->loc.parent) {
+        dht_inode_ctx_time_update(local->loc.parent, this, preparent, 0);
 
-        if (op_ret == -1)
-                goto err;
+        dht_inode_ctx_time_update(local->loc.parent, this, postparent, 1);
+    }
 
-        local = frame->local;
-        srcvol = local->linkfile.srcvol;
+    ret = dht_fd_ctx_set(this, fd, prev);
+    if (ret != 0) {
+        gf_msg_debug(this->name, 0,
+                     "Possible fd leak. "
+                     "Could not set fd ctx for subvol %s",
+                     prev->name);
+    }
 
-        STACK_WIND (frame, dht_link_cbk, srcvol, srcvol->fops->link,
-                    &local->loc, &local->loc2, xdata);
+    ret = dht_layout_preset(this, prev, inode);
+    if (ret != 0) {
+        gf_msg_debug(this->name, 0, "could not set preset layout for subvol %s",
+                     prev->name);
+        op_ret = -1;
+        op_errno = EINVAL;
+        goto out;
+    }
 
-        return 0;
+    local->op_errno = op_errno;
 
-err:
-        DHT_STRIP_PHASE1_FLAGS (stbuf);
-        DHT_STACK_UNWIND (link, frame, op_ret, op_errno, inode, stbuf, preparent,
-                          postparent, NULL);
+    if (local->linked == _gf_true) {
+        local->stbuf = *stbuf;
+        dht_linkfile_attr_heal(frame, this);
+    }
+out:
 
-        return 0;
+    DHT_STRIP_PHASE1_FLAGS(stbuf);
+    dht_set_fixed_dir_stat(preparent);
+    dht_set_fixed_dir_stat(postparent);
+
+    if (local && local->lock[0].layout.parent_layout.locks) {
+        /* store op_errno for failure case*/
+        local->op_errno = op_errno;
+        local->refresh_layout_unlock(frame, this, op_ret, 1);
+
+        if (op_ret == 0) {
+            DHT_STACK_UNWIND(create, frame, op_ret, op_errno, fd, inode, stbuf,
+                             preparent, postparent, xdata);
+        }
+    } else {
+        DHT_STACK_UNWIND(create, frame, op_ret, op_errno, fd, inode, stbuf,
+                         preparent, postparent, xdata);
+    }
+    return 0;
 }
 
+static int
+dht_create_linkfile_create_cbk(call_frame_t *frame, void *cookie,
+                               xlator_t *this, int32_t op_ret, int32_t op_errno,
+                               inode_t *inode, struct iatt *stbuf,
+                               struct iatt *preparent, struct iatt *postparent,
+                               dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *cached_subvol = NULL;
+    dht_conf_t *conf = NULL;
+
+    local = frame->local;
+    if (!local) {
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (op_ret == -1) {
+        local->op_errno = op_errno;
+        goto err;
+    }
+
+    conf = this->private;
+    if (!conf) {
+        local->op_errno = EINVAL;
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    cached_subvol = local->cached_subvol;
+
+    if (local->params) {
+        dict_del(local->params, conf->link_xattr_name);
+        dict_del(local->params, GLUSTERFS_INTERNAL_FOP_KEY);
+    }
+
+    STACK_WIND_COOKIE(frame, dht_create_cbk, cached_subvol, cached_subvol,
+                      cached_subvol->fops->create, &local->loc, local->flags,
+                      local->mode, local->umask, local->fd, local->params);
+
+    return 0;
+err:
+    if (local && local->lock[0].layout.parent_layout.locks) {
+        local->refresh_layout_unlock(frame, this, -1, 1);
+    } else {
+        DHT_STACK_UNWIND(create, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                         NULL, NULL);
+    }
+    return 0;
+}
 
-int
-dht_link (call_frame_t *frame, xlator_t *this,
-          loc_t *oldloc, loc_t *newloc, dict_t *xdata)
+static int
+dht_create_wind_to_avail_subvol(call_frame_t *frame, xlator_t *this,
+                                xlator_t *subvol, loc_t *loc, int32_t flags,
+                                mode_t mode, mode_t umask, fd_t *fd,
+                                dict_t *params)
 {
-        xlator_t    *cached_subvol = NULL;
-        xlator_t    *hashed_subvol = NULL;
-        int          op_errno = -1;
-        int          ret = -1;
-        dht_local_t *local = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *avail_subvol = NULL;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (oldloc, err);
-        VALIDATE_OR_GOTO (newloc, err);
+    local = frame->local;
 
-        local = dht_local_init (frame, oldloc, NULL, GF_FOP_LINK);
-        if (!local) {
-                op_errno = ENOMEM;
+    if (!dht_is_subvol_filled(this, subvol)) {
+        gf_msg_debug(this->name, 0, "creating %s on %s", loc->path,
+                     subvol->name);
 
-                goto err;
-        }
+        dht_set_parent_layout_in_dict(loc, this, local);
 
-        cached_subvol = local->cached_subvol;
-        if (!cached_subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for path=%s", oldloc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
+        STACK_WIND_COOKIE(frame, dht_create_cbk, subvol, subvol,
+                          subvol->fops->create, loc, flags, mode, umask, fd,
+                          params);
 
-        hashed_subvol = dht_subvol_get_hashed (this, newloc);
-        if (!hashed_subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no subvolume in layout for path=%s",
-                        newloc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
+    } else {
+        avail_subvol = dht_free_disk_available_subvol(this, subvol, local);
 
-        ret = loc_copy (&local->loc2, newloc);
-        if (ret == -1) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+        if (avail_subvol != subvol) {
+            local->cached_subvol = avail_subvol;
+            local->hashed_subvol = subvol;
 
-        if (hashed_subvol != cached_subvol) {
-                uuid_copy (local->gfid, oldloc->inode->gfid);
-                dht_linkfile_create (frame, dht_link_linkfile_cbk, this,
-                                     cached_subvol, hashed_subvol, newloc);
-        } else {
-                STACK_WIND (frame, dht_link_cbk,
-                            cached_subvol, cached_subvol->fops->link,
-                            oldloc, newloc, xdata);
+            gf_msg_debug(this->name, 0, "creating %s on %s (link at %s)",
+                         loc->path, avail_subvol->name, subvol->name);
+
+            dht_linkfile_create(frame, dht_create_linkfile_create_cbk, this,
+                                avail_subvol, subvol, loc);
+
+            goto out;
         }
 
-        return 0;
+        gf_msg_debug(this->name, 0, "creating %s on %s", loc->path,
+                     subvol->name);
 
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (link, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+        dht_set_parent_layout_in_dict(loc, this, local);
 
-        return 0;
+        STACK_WIND_COOKIE(frame, dht_create_cbk, subvol, subvol,
+                          subvol->fops->create, loc, flags, mode, umask, fd,
+                          params);
+    }
+out:
+    return 0;
 }
 
-
 int
-dht_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                int op_ret, int op_errno,
-                fd_t *fd, inode_t *inode, struct iatt *stbuf,
-                struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+dht_build_parent_loc(xlator_t *this, loc_t *parent, loc_t *child,
+                     int32_t *op_errno)
 {
-        call_frame_t *prev = NULL;
-        int           ret = -1;
-        dht_local_t  *local = NULL;
+    inode_table_t *table = NULL;
+    int ret = -1;
 
-        if (op_ret == -1)
-                goto out;
+    if (!parent || !child) {
+        if (op_errno)
+            *op_errno = EINVAL;
+        goto out;
+    }
 
-        local = frame->local;
-        if (!local) {
-                op_ret = -1;
-                op_errno = EINVAL;
-                goto out;
+    if (child->parent) {
+        parent->inode = inode_ref(child->parent);
+        if (!parent->inode) {
+            if (op_errno)
+                *op_errno = EINVAL;
+            goto out;
         }
 
-        prev = cookie;
+        gf_uuid_copy(parent->gfid, child->pargfid);
 
-        if (local->loc.parent) {
-                dht_inode_ctx_time_update (local->loc.parent, this,
-                                           preparent, 0);
+        ret = 0;
 
-                dht_inode_ctx_time_update (local->loc.parent, this,
-                                           postparent, 1);
+        goto out;
+    } else {
+        if (gf_uuid_is_null(child->pargfid)) {
+            if (op_errno)
+                *op_errno = EINVAL;
+            goto out;
         }
 
-        ret = dht_layout_preset (this, prev->this, inode);
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "could not set preset layout for subvol %s",
-                        prev->this->name);
-                op_ret   = -1;
-                op_errno = EINVAL;
+        table = this->itable;
+
+        if (!table) {
+            if (op_errno) {
+                *op_errno = EINVAL;
                 goto out;
+            }
         }
-        if (local->linked == _gf_true) {
-                local->stbuf = *stbuf;
-                dht_linkfile_attr_heal (frame, this);
+
+        parent->inode = inode_find(table, child->pargfid);
+
+        if (!parent->inode) {
+            if (op_errno) {
+                *op_errno = EINVAL;
+                goto out;
+            }
         }
+
+        gf_uuid_copy(parent->gfid, child->pargfid);
+
+        ret = 0;
+    }
+
 out:
-        DHT_STRIP_PHASE1_FLAGS (stbuf);
-        DHT_STACK_UNWIND (create, frame, op_ret, op_errno, fd, inode, stbuf, preparent,
-                          postparent, NULL);
-        return 0;
+    return ret;
 }
 
-
-int
-dht_create_linkfile_create_cbk (call_frame_t *frame, void *cookie,
-                                xlator_t *this,
-                                int32_t op_ret, int32_t op_errno,
-                                inode_t *inode, struct iatt *stbuf,
-                                struct iatt *preparent, struct iatt *postparent,
-                                dict_t *xdata)
+static int32_t
+dht_create_do(call_frame_t *frame)
 {
-        dht_local_t  *local = NULL;
-        xlator_t     *cached_subvol = NULL;
+    dht_local_t *local = NULL;
+    dht_layout_t *refreshed = NULL;
+    xlator_t *subvol = NULL;
+    xlator_t *this = NULL;
+    dht_conf_t *conf = NULL;
+    dht_methods_t *methods = NULL;
 
-        if (op_ret == -1)
-                goto err;
+    local = frame->local;
 
-        local = frame->local;
-        cached_subvol = local->cached_subvol;
+    this = THIS;
 
-        STACK_WIND (frame, dht_create_cbk,
-                    cached_subvol, cached_subvol->fops->create,
-                    &local->loc, local->flags, local->mode,
-                    local->umask, local->fd, local->params);
+    conf = this->private;
 
-        return 0;
+    GF_VALIDATE_OR_GOTO(this->name, conf, err);
+
+    methods = &(conf->methods);
+
+    /* We don't need parent_loc anymore */
+    loc_wipe(&local->loc);
+
+    loc_copy(&local->loc, &local->loc2);
+
+    loc_wipe(&local->loc2);
+
+    refreshed = local->selfheal.refreshed_layout;
+
+    subvol = methods->layout_search(this, refreshed, local->loc.name);
+
+    if (!subvol) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+               "no subvolume in "
+               "layout for path=%s",
+               local->loc.path);
+        local->op_errno = ENOENT;
+        goto err;
+    }
+
+    dht_create_wind_to_avail_subvol(frame, this, subvol, &local->loc,
+                                    local->flags, local->mode, local->umask,
+                                    local->fd, local->params);
+    return 0;
 err:
-        DHT_STACK_UNWIND (create, frame, -1, op_errno, NULL, NULL, NULL,
-                          NULL, NULL, NULL);
+    local->refresh_layout_unlock(frame, this, -1, 1);
+
+    return 0;
+}
+
+static int32_t
+dht_create_unlock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    DHT_STACK_DESTROY(frame);
+    return 0;
+}
+
+static int32_t
+dht_create_finish(call_frame_t *frame, xlator_t *this, int op_ret,
+                  int invoke_cbk)
+{
+    dht_local_t *local = NULL, *lock_local = NULL;
+    call_frame_t *lock_frame = NULL;
+    int lock_count = 0;
+
+    local = frame->local;
+    lock_count = dht_lock_count(local->lock[0].layout.parent_layout.locks,
+                                local->lock[0].layout.parent_layout.lk_count);
+    if (lock_count == 0)
+        goto done;
+
+    lock_frame = copy_frame(frame);
+    if (lock_frame == NULL) {
+        goto done;
+    }
+
+    lock_local = dht_local_init(lock_frame, &local->loc, NULL,
+                                lock_frame->root->op);
+    if (lock_local == NULL) {
+        goto done;
+    }
+
+    lock_local->lock[0]
+        .layout.parent_layout.locks = local->lock[0].layout.parent_layout.locks;
+    lock_local->lock[0].layout.parent_layout.lk_count =
+        local->lock[0].layout.parent_layout.lk_count;
+
+    local->lock[0].layout.parent_layout.locks = NULL;
+    local->lock[0].layout.parent_layout.lk_count = 0;
+
+    dht_unlock_inodelk(lock_frame,
+                       lock_local->lock[0].layout.parent_layout.locks,
+                       lock_local->lock[0].layout.parent_layout.lk_count,
+                       dht_create_unlock_cbk);
+    lock_frame = NULL;
+
+done:
+    if (lock_frame != NULL) {
+        DHT_STACK_DESTROY(lock_frame);
+    }
+
+    if (op_ret == 0)
         return 0;
+
+    DHT_STACK_UNWIND(create, frame, op_ret, local->op_errno, NULL, NULL, NULL,
+                     NULL, NULL, NULL);
+    return 0;
 }
 
-int
-dht_create (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, int32_t flags, mode_t mode,
-            mode_t umask, fd_t *fd, dict_t *params)
+static int32_t
+dht_create_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        int          op_errno = -1;
-        xlator_t    *subvol = NULL;
-        dht_local_t *local = NULL;
-        xlator_t    *avail_subvol = NULL;
+    dht_local_t *local = NULL;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
+    local = frame->local;
 
-        dht_get_du_info (frame, this, loc);
+    if (!local) {
+        goto err;
+    }
 
-        local = dht_local_init (frame, loc, fd, GF_FOP_CREATE);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+    if (op_ret < 0) {
+        gf_msg("DHT", GF_LOG_ERROR, 0, DHT_MSG_INODE_LK_ERROR,
+               "Create lock failed for file: %s", local->loc2.name);
 
-        if (dht_filter_loc_subvol_key (this, loc, &local->loc,
-                                       &subvol)) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "creating %s on %s (got create on %s)",
-                        local->loc.path, subvol->name, loc->path);
-                STACK_WIND (frame, dht_create_cbk,
-                            subvol, subvol->fops->create,
-                            &local->loc, flags, mode, umask, fd, params);
-                goto done;
-        }
+        local->op_errno = op_errno;
 
-        subvol = dht_subvol_get_hashed (this, loc);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no subvolume in layout for path=%s",
-                        loc->path);
-                op_errno = ENOENT;
-                goto err;
-        }
+        goto err;
+    }
 
-        if (!dht_is_subvol_filled (this, subvol)) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "creating %s on %s", loc->path, subvol->name);
-                STACK_WIND (frame, dht_create_cbk,
-                            subvol, subvol->fops->create,
-                            loc, flags, mode, umask, fd, params);
-                goto done;
-        }
-        /* Choose the minimum filled volume, and create the
-           files there */
-        avail_subvol = dht_free_disk_available_subvol (this, subvol, local);
-        if (avail_subvol != subvol) {
-                local->params = dict_ref (params);
-                local->flags = flags;
-                local->mode = mode;
-                local->umask = umask;
-                local->cached_subvol = avail_subvol;
-                local->hashed_subvol = subvol;
-                gf_log (this->name, GF_LOG_TRACE,
-                        "creating %s on %s (link at %s)", loc->path,
-                        avail_subvol->name, subvol->name);
-                dht_linkfile_create (frame, dht_create_linkfile_create_cbk,
-                                     this, avail_subvol, subvol, loc);
-                goto done;
-        }
-        gf_log (this->name, GF_LOG_TRACE,
-                "creating %s on %s", loc->path, subvol->name);
-        STACK_WIND (frame, dht_create_cbk,
-                    subvol, subvol->fops->create,
-                    loc, flags, mode, umask, fd, params);
-done:
-        return 0;
+    local->refresh_layout_unlock = dht_create_finish;
 
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (create, frame, -1, op_errno, NULL, NULL, NULL,
-                          NULL, NULL, NULL);
+    local->refresh_layout_done = dht_create_do;
 
-        return 0;
-}
+    dht_refresh_layout(frame);
 
+    return 0;
+err:
+    if (local)
+        dht_create_finish(frame, this, -1, 0);
+    else
+        DHT_STACK_UNWIND(create, frame, -1, EINVAL, NULL, NULL, NULL, NULL,
+                         NULL, NULL);
+    return 0;
+}
 
-int
-dht_mkdir_selfheal_cbk (call_frame_t *frame, void *cookie,
-                        xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+int32_t
+dht_create_lock(call_frame_t *frame, xlator_t *subvol)
 {
-        dht_local_t   *local = NULL;
-        dht_layout_t  *layout = NULL;
+    dht_local_t *local = NULL;
+    int count = 1, ret = -1;
+    dht_lock_t **lk_array = NULL;
 
-        local = frame->local;
-        layout = local->selfheal.layout;
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO(frame->this->name, frame->local, err);
 
-        if (op_ret == 0) {
-                dht_layout_set (this, local->inode, layout);
-                if (local->loc.parent) {
-                        dht_inode_ctx_time_update (local->loc.parent, this,
-                                                   &local->preparent, 0);
+    local = frame->local;
 
-                        dht_inode_ctx_time_update (local->loc.parent, this,
-                                                   &local->postparent, 1);
-                }
-        }
+    lk_array = GF_CALLOC(count, sizeof(*lk_array), gf_common_mt_pointer);
 
-        DHT_STACK_UNWIND (mkdir, frame, op_ret, op_errno,
-                          local->inode, &local->stbuf, &local->preparent,
-                          &local->postparent, NULL);
+    if (lk_array == NULL)
+        goto err;
 
-        return 0;
+    lk_array[0] = dht_lock_new(frame->this, subvol, &local->loc, F_RDLCK,
+                               DHT_LAYOUT_HEAL_DOMAIN, NULL,
+                               IGNORE_ENOENT_ESTALE);
+
+    if (lk_array[0] == NULL)
+        goto err;
+
+    local->lock[0].layout.parent_layout.locks = lk_array;
+    local->lock[0].layout.parent_layout.lk_count = count;
+
+    ret = dht_blocking_inodelk(frame, lk_array, count, dht_create_lock_cbk);
+
+    if (ret < 0) {
+        local->lock[0].layout.parent_layout.locks = NULL;
+        local->lock[0].layout.parent_layout.lk_count = 0;
+        goto err;
+    }
+
+    return 0;
+err:
+    if (lk_array != NULL) {
+        dht_lock_array_free(lk_array, count);
+        GF_FREE(lk_array);
+    }
+
+    return -1;
 }
 
 int
-dht_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-               int op_ret, int op_errno, inode_t *inode, struct iatt *stbuf,
-               struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+dht_set_parent_layout_in_dict(loc_t *loc, xlator_t *this, dht_local_t *local)
 {
-        dht_local_t  *local = NULL;
-        int           this_call_cnt = 0;
-        int           ret = -1;
-        gf_boolean_t subvol_filled = _gf_false;
-        call_frame_t *prev = NULL;
-        dht_layout_t *layout = NULL;
+    dht_conf_t *conf = this->private;
+    dht_layout_t *parent_layout = NULL;
+    int *parent_disk_layout = NULL;
+    xlator_t *hashed_subvol = NULL;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    int ret = 0;
+
+    gf_uuid_unparse(loc->parent->gfid, pgfid);
+
+    parent_layout = dht_layout_get(this, loc->parent);
+    hashed_subvol = dht_subvol_get_hashed(this, loc);
+
+    ret = dht_disk_layout_extract_for_subvol(this, parent_layout, hashed_subvol,
+                                             &parent_disk_layout);
+    if (ret == -1) {
+        gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "%s (%s/%s) (path: %s): "
+               "extracting in-memory layout of parent failed. ",
+               gf_fop_list[local->fop], pgfid, loc->name, loc->path);
+        goto err;
+    }
+
+    ret = dict_set_str_sizen(local->params, GF_PREOP_PARENT_KEY,
+                             conf->xattr_name);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "%s (%s/%s) (path: %s): "
+               "setting %s key in params dictionary failed. ",
+               gf_fop_list[local->fop], pgfid, loc->name, loc->path,
+               GF_PREOP_PARENT_KEY);
+        goto err;
+    }
+
+    ret = dict_set_bin(local->params, conf->xattr_name, parent_disk_layout,
+                       4 * 4);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "%s (%s/%s) (path: %s): "
+               "setting parent-layout in params dictionary failed. ",
+               gf_fop_list[local->fop], pgfid, loc->name, loc->path);
+        goto err;
+    }
 
-        local = frame->local;
-        prev  = cookie;
-        layout = local->layout;
+err:
+    dht_layout_unref(this, parent_layout);
+    return ret;
+}
 
-        subvol_filled = dht_is_subvol_filled (this, prev->this);
+int
+dht_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+           mode_t mode, mode_t umask, fd_t *fd, dict_t *params)
+{
+    int op_errno = -1;
+    xlator_t *subvol = NULL;
+    xlator_t *hashed_subvol = NULL;
+    dht_local_t *local = NULL;
+    int i = 0;
+    dht_conf_t *conf = NULL;
+    int ret = 0;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+
+    conf = this->private;
+
+    dht_get_du_info(frame, this, loc);
+
+    local = dht_local_init(frame, loc, fd, GF_FOP_CREATE);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->params = dict_ref(params);
+    local->flags = flags;
+    local->mode = mode;
+    local->umask = umask;
+
+    if (dht_filter_loc_subvol_key(this, loc, &local->loc, &subvol)) {
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_INFO,
+               "creating %s on %s (got create on %s)", local->loc.path,
+               subvol->name, loc->path);
+
+        /* Since lookup-optimize is enabled by default, we need
+         * to create the linkto file if required.
+         * Note this does not check for decommisioned bricks
+         * and min-free-disk limits as this is a debugging tool
+         * and not expected to be used in production.
+         */
+        hashed_subvol = dht_subvol_get_hashed(this, &local->loc);
 
-        LOCK (&frame->lock);
-        {
-                if (subvol_filled && (op_ret != -1)) {
-                        ret = dht_layout_merge (this, layout, prev->this,
-                                                -1, ENOSPC, NULL);
-                } else {
-			if (op_ret == -1 && op_errno == EEXIST)
-				/* Very likely just a race between mkdir and
-				   self-heal (from lookup of a concurrent mkdir
-				   attempt).
-				   Ignore error for now. layout setting will
-				   anyways fail if this was a different (old)
-				   pre-existing different directory.
-				*/
-				op_ret = 0;
-                        ret = dht_layout_merge (this, layout, prev->this,
-                                                op_ret, op_errno, NULL);
+        if (hashed_subvol && (hashed_subvol != subvol)) {
+            /* Create the linkto file and then the data file */
+            local->cached_subvol = subvol;
+            local->hashed_subvol = hashed_subvol;
+
+            dht_linkfile_create(frame, dht_create_linkfile_create_cbk, this,
+                                subvol, hashed_subvol, &local->loc);
+            goto done;
+        }
+        /* We either don't have a hashed subvol or the hashed subvol is
+         * the same as the one specified. No need to create the linkto
+         * file as we expect a lookup everywhere if there are problems
+         * with the parent layout
+         */
+
+        dht_set_parent_layout_in_dict(loc, this, local);
+
+        STACK_WIND_COOKIE(frame, dht_create_cbk, subvol, subvol,
+                          subvol->fops->create, &local->loc, flags, mode, umask,
+                          fd, params);
+        goto done;
+    }
+
+    subvol = dht_subvol_get_hashed(this, loc);
+    if (!subvol) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+               "no subvolume in layout for path=%s", loc->path);
+
+        op_errno = EIO;
+        goto err;
+    }
+
+    /* Post remove-brick, the client layout may not be in sync with
+     * disk layout because of lack of lookup. Hence,a create call
+     * may fall on the decommissioned brick.  Hence, if the
+     * hashed_subvol is part of decommissioned bricks  list, do a
+     * lookup on parent dir. If a fix-layout is already done by the
+     * remove-brick process, the parent directory layout will be in
+     * sync with that of the disk. If fix-layout is still ending
+     * on the parent directory, we can let the file get created on
+     * the decommissioned brick which will be eventually migrated to
+     * non-decommissioned brick based on the new layout.
+     */
+
+    if (conf->decommission_subvols_cnt) {
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            if (conf->decommissioned_bricks[i] &&
+                conf->decommissioned_bricks[i] == subvol) {
+                gf_msg_debug(this->name, 0,
+                             "hashed subvol:%s is "
+                             "part of decommission brick list for "
+                             "file: %s",
+                             subvol->name, loc->path);
+
+                /* dht_refresh_layout needs directory info in
+                 * local->loc. Hence, storing the parent_loc in
+                 * local->loc and storing the create context in
+                 * local->loc2. We will restore this information
+                 * in dht_creation do */
+
+                ret = loc_copy(&local->loc2, &local->loc);
+                if (ret) {
+                    gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+                           "loc_copy failed %s", loc->path);
+
+                    goto err;
                 }
-                if (ret)
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "%s: failed to merge layouts", local->loc.path);
 
-                if (op_ret == -1) {
-                        local->op_errno = op_errno;
-                        goto unlock;
+                loc_wipe(&local->loc);
+
+                ret = dht_build_parent_loc(this, &local->loc, loc, &op_errno);
+
+                if (ret) {
+                    gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_LOC_FAILED,
+                           "parent loc build failed");
+                    goto err;
+                }
+
+                ret = dht_create_lock(frame, subvol);
+
+                if (ret < 0) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INODE_LK_ERROR,
+                           "locking parent failed");
+                    goto err;
                 }
-                dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
-                dht_iatt_merge (this, &local->preparent, preparent, prev->this);
-                dht_iatt_merge (this, &local->postparent, postparent,
-                                prev->this);
-        }
-unlock:
-        UNLOCK (&frame->lock);
 
-        this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt)) {
-                dht_selfheal_new_directory (frame, dht_mkdir_selfheal_cbk,
-                                            layout);
+                goto done;
+            }
         }
+    }
 
-        return 0;
+    dht_create_wind_to_avail_subvol(frame, this, subvol, loc, flags, mode,
+                                    umask, fd, params);
+done:
+    return 0;
+
+err:
+
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(create, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL,
+                     NULL);
+
+    return 0;
 }
 
-int
-dht_mkdir_hashed_cbk (call_frame_t *frame, void *cookie,
-                      xlator_t *this, int op_ret, int op_errno,
-                      inode_t *inode, struct iatt *stbuf,
-                      struct iatt *preparent, struct iatt *postparent,
-                      dict_t *xdata)
-{
-        dht_local_t  *local = NULL;
-        int           ret = -1;
-        call_frame_t *prev = NULL;
-        dht_layout_t *layout = NULL;
-        dht_conf_t   *conf = NULL;
-        int           i = 0;
-        xlator_t     *hashed_subvol = NULL;
-
-        VALIDATE_OR_GOTO (this->private, err);
+static int
+dht_mkdir_selfheal_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
 
-        local = frame->local;
-        prev  = cookie;
-        layout = local->layout;
-        conf = this->private;
-        hashed_subvol = local->hashed_subvol;
+    local = frame->local;
+    layout = local->selfheal.layout;
 
-        if (uuid_is_null (local->loc.gfid) && !op_ret)
-                uuid_copy (local->loc.gfid, stbuf->ia_gfid);
+    FRAME_SU_UNDO(frame, dht_local_t);
+    dht_set_fixed_dir_stat(&local->preparent);
+    dht_set_fixed_dir_stat(&local->postparent);
 
-        if (dht_is_subvol_filled (this, hashed_subvol))
-                ret = dht_layout_merge (this, layout, prev->this,
-                                        -1, ENOSPC, NULL);
-        else
-                ret = dht_layout_merge (this, layout, prev->this,
-                                        op_ret, op_errno, NULL);
+    if (op_ret == 0) {
+        dht_layout_set(this, local->inode, layout);
 
-        /* TODO: we may have to return from the function
-           if layout merge fails. For now, lets just log an error */
-        if (ret)
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to merge layouts", local->loc.path);
+        dht_inode_ctx_time_update(local->inode, this, &local->stbuf, 1);
+        if (local->loc.parent) {
+            dht_inode_ctx_time_update(local->loc.parent, this,
+                                      &local->preparent, 0);
 
-        if (op_ret == -1) {
-                local->op_errno = op_errno;
-                goto err;
+            dht_inode_ctx_time_update(local->loc.parent, this,
+                                      &local->postparent, 1);
         }
-        local->op_ret = 0;
+    }
 
-        dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
-        dht_iatt_merge (this, &local->preparent, preparent, prev->this);
-        dht_iatt_merge (this, &local->postparent, postparent, prev->this);
+    DHT_STACK_UNWIND(mkdir, frame, op_ret, op_errno, local->inode,
+                     &local->stbuf, &local->preparent, &local->postparent,
+                     NULL);
 
-        local->call_cnt = conf->subvolume_cnt - 1;
+    return 0;
+}
 
-        if (uuid_is_null (local->loc.gfid))
-                uuid_copy (local->loc.gfid, stbuf->ia_gfid);
-        if (local->call_cnt == 0) {
-                dht_selfheal_directory (frame, dht_mkdir_selfheal_cbk,
-                                        &local->loc, layout);
+static int
+dht_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, inode_t *inode, struct iatt *stbuf,
+              struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    int ret = -1;
+    gf_boolean_t subvol_filled = _gf_false;
+    gf_boolean_t dir_exists = _gf_false;
+    xlator_t *prev = NULL;
+    dht_layout_t *layout = NULL;
+
+    local = frame->local;
+    prev = cookie;
+    layout = local->layout;
+
+    subvol_filled = dht_is_subvol_filled(this, prev);
+
+    LOCK(&frame->lock);
+    {
+        if (subvol_filled && (op_ret != -1)) {
+            ret = dht_layout_merge(this, layout, prev, -1, ENOSPC, NULL);
+        } else {
+            if (op_ret == -1 && op_errno == EEXIST) {
+                /* Very likely just a race between mkdir and
+                   self-heal (from lookup of a concurrent mkdir
+                   attempt).
+                   Ignore error for now. layout setting will
+                   anyways fail if this was a different (old)
+                   pre-existing different directory.
+                */
+                op_ret = 0;
+                dir_exists = _gf_true;
+            }
+            ret = dht_layout_merge(this, layout, prev, op_ret, op_errno, NULL);
         }
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                if (conf->subvolumes[i] == hashed_subvol)
-                        continue;
-                STACK_WIND (frame, dht_mkdir_cbk,
-                            conf->subvolumes[i],
-                            conf->subvolumes[i]->fops->mkdir, &local->loc,
-                            local->mode, local->umask, local->params);
+        if (ret)
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LAYOUT_MERGE_FAILED,
+                   "%s: failed to merge layouts for subvol %s", local->loc.path,
+                   prev->name);
+
+        if (op_ret == -1) {
+            local->op_errno = op_errno;
+            goto unlock;
         }
-        return 0;
-err:
-        DHT_STACK_UNWIND (mkdir, frame, -1, op_errno, NULL, NULL, NULL,
-                          NULL, NULL);
-        return 0;
+
+        if (dir_exists)
+            goto unlock;
+
+        dht_iatt_merge(this, &local->stbuf, stbuf);
+        dht_iatt_merge(this, &local->preparent, preparent);
+        dht_iatt_merge(this, &local->postparent, postparent);
+    }
+unlock:
+    UNLOCK(&frame->lock);
+
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        /*Unlock entrylk and inodelk once mkdir is done on all subvols*/
+        dht_unlock_namespace(frame, &local->lock[0]);
+        FRAME_SU_DO(frame, dht_local_t);
+        dht_selfheal_new_directory(frame, dht_mkdir_selfheal_cbk, layout);
+    }
+
+    return 0;
 }
 
+static int
+dht_mkdir_hashed_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, inode_t *inode,
+                     struct iatt *stbuf, struct iatt *preparent,
+                     struct iatt *postparent, dict_t *xdata);
 
- int
-dht_mkdir (call_frame_t *frame, xlator_t *this,
-           loc_t *loc, mode_t mode, mode_t umask, dict_t *params)
+static int
+dht_mkdir_helper(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+                 mode_t umask, dict_t *params)
 {
-        dht_local_t  *local  = NULL;
-        dht_conf_t   *conf = NULL;
-        int           op_errno = -1;
-        xlator_t     *hashed_subvol = NULL;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int op_errno = -1, ret = -1;
+    xlator_t *hashed_subvol = NULL;
+    int32_t *parent_disk_layout = NULL;
+    dht_layout_t *parent_layout = NULL;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+    VALIDATE_OR_GOTO(loc->path, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    gf_uuid_unparse(loc->parent->gfid, pgfid);
+
+    conf = this->private;
+    local = frame->local;
+
+    if (local->op_ret == -1) {
+        gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "mkdir (%s/%s) (path: %s): refreshing parent layout "
+               "failed.",
+               pgfid, loc->name, loc->path);
+
+        op_errno = local->op_errno;
+        goto err;
+    }
+
+    local->op_ret = -1;
+
+    hashed_subvol = dht_subvol_get_hashed(this, loc);
+    if (hashed_subvol == NULL) {
+        gf_msg_debug(this->name, 0,
+                     "mkdir (%s/%s) (path: %s): hashed subvol not "
+                     "found",
+                     pgfid, loc->name, loc->path);
+        op_errno = ENOENT;
+        goto err;
+    }
+
+    local->hashed_subvol = hashed_subvol;
+
+    parent_layout = dht_layout_get(this, loc->parent);
+
+    ret = dht_disk_layout_extract_for_subvol(this, parent_layout, hashed_subvol,
+                                             &parent_disk_layout);
+    if (ret == -1) {
+        gf_msg(this->name, GF_LOG_WARNING, EIO, DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "mkdir (%s/%s) (path: %s): "
+               "extracting in-memory layout of parent failed. ",
+               pgfid, loc->name, loc->path);
+        goto err;
+    }
+
+    if (memcmp(local->parent_disk_layout, parent_disk_layout,
+               sizeof(local->parent_disk_layout)) == 0) {
+        gf_msg(this->name, GF_LOG_WARNING, EIO, DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "mkdir (%s/%s) (path: %s): loop detected. "
+               "parent layout didn't change even though "
+               "previous attempt of mkdir failed because of "
+               "in-memory layout not matching with that on disk.",
+               pgfid, loc->name, loc->path);
+        op_errno = EIO;
+        goto err;
+    }
+
+    memcpy((void *)local->parent_disk_layout, (void *)parent_disk_layout,
+           sizeof(local->parent_disk_layout));
+
+    dht_layout_unref(this, parent_layout);
+    parent_layout = NULL;
+
+    ret = dict_set_str(params, GF_PREOP_PARENT_KEY, conf->xattr_name);
+    if (ret < 0) {
+        local->op_errno = -ret;
+        gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "mkdir (%s/%s) (path: %s): "
+               "setting %s key in params dictionary failed. ",
+               pgfid, loc->name, loc->path, GF_PREOP_PARENT_KEY);
+        goto err;
+    }
+
+    ret = dict_set_bin(params, conf->xattr_name, parent_disk_layout, 4 * 4);
+    if (ret < 0) {
+        local->op_errno = -ret;
+        gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "setting parent-layout in params dictionary failed. "
+               "mkdir (%s/%s) (path: %s)",
+               pgfid, loc->name, loc->path);
+        goto err;
+    }
+
+    parent_disk_layout = NULL;
+
+    STACK_WIND_COOKIE(frame, dht_mkdir_hashed_cbk, hashed_subvol, hashed_subvol,
+                      hashed_subvol->fops->mkdir, loc, mode, umask, params);
+
+    return 0;
 
+err:
+    dht_unlock_namespace(frame, &local->lock[0]);
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (this->private, err);
+    op_errno = local ? local->op_errno : op_errno;
+    DHT_STACK_UNWIND(mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
 
-        conf = this->private;
+    if (parent_disk_layout != NULL)
+        GF_FREE(parent_disk_layout);
 
-        dht_get_du_info (frame, this, loc);
+    if (parent_layout != NULL)
+        dht_layout_unref(this, parent_layout);
 
-        local = dht_local_init (frame, loc, NULL, GF_FOP_MKDIR);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+    return 0;
+}
 
-        hashed_subvol = dht_subvol_get_hashed (this, loc);
-        if (hashed_subvol == NULL) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "hashed subvol not found for %s",
-                        loc->path);
-                op_errno = EINVAL;
+static int
+dht_mkdir_hashed_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, inode_t *inode,
+                     struct iatt *stbuf, struct iatt *preparent,
+                     struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int ret = -1;
+    xlator_t *prev = NULL;
+    dht_layout_t *layout = NULL;
+    dht_conf_t *conf = NULL;
+    int i = 0;
+    xlator_t *hashed_subvol = NULL;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    gf_boolean_t parent_layout_changed = _gf_false;
+    call_stub_t *stub = NULL;
+
+    local = frame->local;
+    prev = cookie;
+    layout = local->layout;
+    conf = this->private;
+    hashed_subvol = local->hashed_subvol;
+
+    gf_uuid_unparse(local->loc.parent->gfid, pgfid);
+
+    if (gf_uuid_is_null(local->loc.gfid) && !op_ret)
+        gf_uuid_copy(local->loc.gfid, stbuf->ia_gfid);
+
+    if (op_ret == -1) {
+        local->op_errno = op_errno;
+
+        parent_layout_changed = (xdata &&
+                                 dict_get(xdata, GF_PREOP_CHECK_FAILED))
+                                    ? 1
+                                    : 0;
+        if (parent_layout_changed) {
+            gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_PARENT_LAYOUT_CHANGED,
+                   "mkdir (%s/%s) (path: %s): parent layout "
+                   "changed. Attempting a refresh and then a "
+                   "retry",
+                   pgfid, local->loc.name, local->loc.path);
+
+            stub = fop_mkdir_stub(frame, dht_mkdir_helper, &local->loc,
+                                  local->mode, local->umask, local->params);
+            if (stub == NULL) {
                 goto err;
-        }
-
-        local->hashed_subvol = hashed_subvol;
-        local->mode = mode;
-        local->umask = umask;
-        local->params = dict_ref (params);
-        local->inode  = inode_ref (loc->inode);
+            }
 
-        local->layout = dht_layout_new (this, conf->subvolume_cnt);
-        if (!local->layout) {
-                op_errno = ENOMEM;
+            ret = dht_handle_parent_layout_change(this, stub);
+            if (ret) {
                 goto err;
-        }
+            }
+
+            stub = NULL;
+
+            return 0;
+        }
+
+        goto err;
+    }
+
+    dict_del(local->params, GF_PREOP_PARENT_KEY);
+    dict_del(local->params, conf->xattr_name);
+
+    if (dht_is_subvol_filled(this, hashed_subvol))
+        ret = dht_layout_merge(this, layout, prev, -1, ENOSPC, NULL);
+    else
+        ret = dht_layout_merge(this, layout, prev, op_ret, op_errno, NULL);
+
+    /* TODO: we may have to return from the function
+       if layout merge fails. For now, lets just log an error */
+    if (ret)
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LAYOUT_MERGE_FAILED,
+               "%s: failed to merge layouts for subvol %s", local->loc.path,
+               prev->name);
+
+    local->op_ret = 0;
+
+    dht_iatt_merge(this, &local->stbuf, stbuf);
+    dht_iatt_merge(this, &local->preparent, preparent);
+    dht_iatt_merge(this, &local->postparent, postparent);
+
+    local->call_cnt = conf->subvolume_cnt - 1;
+    /* Delete internal mds xattr from params dict to avoid store
+      internal mds xattr on other subvols
+    */
+    dict_del(local->params, conf->mds_xattr_key);
+
+    if (gf_uuid_is_null(local->loc.gfid))
+        gf_uuid_copy(local->loc.gfid, stbuf->ia_gfid);
+
+    /* Set hashed subvol as a mds subvol on inode ctx */
+    /*if (!local->inode)
+            local->inode  = inode_ref (inode);
+    */
+    ret = dht_inode_ctx_mdsvol_set(local->inode, this, hashed_subvol);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SET_INODE_CTX_FAILED,
+               "Failed to set hashed subvol for %s on inode vol is %s",
+               local->loc.path, hashed_subvol->name);
+    }
+
+    if (local->call_cnt == 0) {
+        /*Unlock namespace lock once mkdir is done on all subvols*/
+        dht_unlock_namespace(frame, &local->lock[0]);
+        FRAME_SU_DO(frame, dht_local_t);
+        dht_selfheal_directory(frame, dht_mkdir_selfheal_cbk, &local->loc,
+                               layout);
+        return 0;
+    }
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (conf->subvolumes[i] == hashed_subvol)
+            continue;
+        STACK_WIND_COOKIE(frame, dht_mkdir_cbk, conf->subvolumes[i],
+                          conf->subvolumes[i], conf->subvolumes[i]->fops->mkdir,
+                          &local->loc, local->mode, local->umask,
+                          local->params);
+    }
+
+    return 0;
+err:
+    if (local->op_ret != 0) {
+        dht_unlock_namespace(frame, &local->lock[0]);
+    }
 
-        STACK_WIND (frame, dht_mkdir_hashed_cbk,
-                    hashed_subvol,
-                    hashed_subvol->fops->mkdir,
-                    loc, mode, umask, params);
+    DHT_STACK_UNWIND(mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
 
-        return 0;
+    return 0;
+}
 
+static int
+dht_mkdir_guard_parent_layout_cbk(call_frame_t *frame, xlator_t *this,
+                                  loc_t *loc, mode_t mode, mode_t umask,
+                                  dict_t *params)
+{
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = 0;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    int ret = -1;
+    int32_t zero[1] = {0};
+
+    local = frame->local;
+    conf = this->private;
+
+    gf_uuid_unparse(loc->parent->gfid, pgfid);
+
+    if (local->op_ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "mkdir (%s/%s) (path: %s): "
+               "Acquiring lock on parent to guard against "
+               "layout-change failed.",
+               pgfid, loc->name, loc->path);
+        goto err;
+    }
+
+    local->op_ret = -1;
+    /* Add internal MDS xattr on disk for hashed subvol
+     */
+    ret = dht_dict_set_array(params, conf->mds_xattr_key, zero, 1);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, DHT_MSG_DICT_SET_FAILED,
+               "Failed to set dictionary value:key = %s for "
+               "path %s",
+               conf->mds_xattr_key, loc->path);
+    }
+
+    STACK_WIND_COOKIE(frame, dht_mkdir_hashed_cbk, local->hashed_subvol,
+                      local->hashed_subvol, local->hashed_subvol->fops->mkdir,
+                      loc, mode, umask, params);
+
+    return 0;
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (mkdir, frame, -1, op_errno, NULL, NULL, NULL,
-                          NULL, NULL);
+    DHT_STACK_UNWIND(mkdir, frame, -1, local->op_errno, NULL, NULL, NULL, NULL,
+                     NULL);
 
-        return 0;
+    return 0;
 }
 
-
 int
-dht_rmdir_selfheal_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int op_ret, int op_errno, dict_t *xdata)
+dht_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+          mode_t umask, dict_t *params)
 {
-        dht_local_t  *local = NULL;
-
-        local = frame->local;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int op_errno = EINVAL, ret = -1;
+    xlator_t *hashed_subvol = NULL;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    call_stub_t *stub = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+    VALIDATE_OR_GOTO(loc->path, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    gf_uuid_unparse(loc->parent->gfid, pgfid);
+
+    conf = this->private;
+
+    if (!params || !dict_get(params, "gfid-req")) {
+        op_errno = EPERM;
+        gf_msg_callingfn(this->name, GF_LOG_WARNING, op_errno,
+                         DHT_MSG_GFID_NULL,
+                         "mkdir: %s is received "
+                         "without gfid-req %p",
+                         loc->path, params);
+        goto err;
+    }
+
+    dht_get_du_info(frame, this, loc);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_MKDIR);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    hashed_subvol = dht_subvol_get_hashed(this, loc);
+    if (hashed_subvol == NULL) {
+        gf_msg_debug(this->name, 0, "hashed subvol not found for %s",
+                     loc->path);
+        local->op_errno = EIO;
+        goto err;
+    }
+
+    local->hashed_subvol = hashed_subvol;
+    local->mode = mode;
+    local->umask = umask;
+    if (params)
+        local->params = dict_ref(params);
+
+    local->inode = inode_ref(loc->inode);
+
+    local->layout = dht_layout_new(this, conf->subvolume_cnt);
+    if (!local->layout) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    /* set the newly created directory hash to the commit hash
+     * if the configuration option is set. If configuration option
+     * is not set, the older clients may still be connecting to the
+     * volume and hence we need to preserve the 1 in disk[0] part of the
+     * layout xattr */
+    if (conf->lookup_optimize)
+        local->layout->commit_hash = conf->vol_commit_hash;
+    else
+        local->layout->commit_hash = DHT_LAYOUT_HASH_INVALID;
+
+    stub = fop_mkdir_stub(frame, dht_mkdir_guard_parent_layout_cbk, loc, mode,
+                          umask, params);
+    if (stub == NULL) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "mkdir (%s/%s) (path: %s): "
+               "creating stub failed.",
+               pgfid, loc->name, loc->path);
+        local->op_errno = ENOMEM;
+        goto err;
+    }
+
+    ret = dht_guard_parent_layout_and_namespace(this, stub);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "mkdir (%s/%s) (path: %s) cannot wind lock request to "
+               "guard parent layout",
+               pgfid, loc->name, loc->path);
+        goto err;
+    }
+
+    return 0;
 
-        DHT_STACK_UNWIND (rmdir, frame, local->op_ret, local->op_errno,
-                          &local->preparent, &local->postparent, NULL);
+err:
+    op_errno = local ? local->op_errno : op_errno;
+    DHT_STACK_UNWIND(mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
-
-int
-dht_rmdir_hashed_subvol_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-               int op_ret, int op_errno, struct iatt *preparent,
-               struct iatt *postparent, dict_t *xdata)
+static int
+dht_rmdir_selfheal_cbk(call_frame_t *heal_frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno, dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
-        int           this_call_cnt = 0;
-        call_frame_t *prev = NULL;
+    dht_local_t *local = NULL;
+    dht_local_t *heal_local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        local = frame->local;
-        prev  = cookie;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        local->op_errno = op_errno;
-                        local->op_ret   = -1;
-                        if (op_errno != ENOENT && op_errno != EACCES) {
-                                local->need_selfheal = 1;
-                        }
+    heal_local = heal_frame->local;
+    main_frame = heal_local->main_frame;
+    local = main_frame->local;
 
+    DHT_STACK_DESTROY(heal_frame);
+    dht_set_fixed_dir_stat(&local->preparent);
+    dht_set_fixed_dir_stat(&local->postparent);
 
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "rmdir on %s for %s failed (%s)",
-                                prev->this->name, local->loc.path,
-                                strerror (op_errno));
-                        goto unlock;
-                }
+    DHT_STACK_UNWIND(rmdir, main_frame, local->op_ret, local->op_errno,
+                     &local->preparent, &local->postparent, NULL);
 
-                dht_iatt_merge (this, &local->preparent, preparent, prev->this);
-                dht_iatt_merge (this, &local->postparent, postparent,
-                                prev->this);
+    return 0;
+}
 
+static int
+dht_rmdir_hashed_subvol_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int op_ret, int op_errno, struct iatt *preparent,
+                            struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    dht_local_t *heal_local = NULL;
+    call_frame_t *heal_frame = NULL;
+    dht_conf_t *conf = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    prev = cookie;
+    conf = this->private;
+
+    gf_uuid_unparse(local->loc.gfid, gfid);
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            local->op_errno = op_errno;
+            local->op_ret = -1;
+            if (conf->subvolume_cnt != 1) {
+                if (op_errno != ENOENT && op_errno != EACCES &&
+                    op_errno != ESTALE) {
+                    local->need_selfheal = 1;
+                }
+            }
+
+            gf_msg_debug(this->name, op_errno,
+                         "rmdir on %s for %s failed "
+                         "(gfid = %s)",
+                         prev->name, local->loc.path, gfid);
+            goto unlock;
         }
+
+        dht_iatt_merge(this, &local->preparent, preparent);
+        dht_iatt_merge(this, &local->postparent, postparent);
+    }
 unlock:
-        UNLOCK (&frame->lock);
-
-        this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt)) {
-               if (local->need_selfheal) {
-                        local->layout =
-                                dht_layout_get (this, local->loc.inode);
-
-                        /* TODO: neater interface needed below */
-                        local->stbuf.ia_type = local->loc.inode->ia_type;
-
-                        uuid_copy (local->gfid, local->loc.inode->gfid);
-                        dht_selfheal_restore (frame, dht_rmdir_selfheal_cbk,
-                                              &local->loc, local->layout);
-               } else {
-
-                        if (local->loc.parent) {
-                                dht_inode_ctx_time_update (local->loc.parent,
-                                                           this,
-                                                           &local->preparent,
-                                                           0);
-
-                                dht_inode_ctx_time_update (local->loc.parent,
-                                                           this,
-                                                           &local->postparent,
-                                                           1);
-                        }
+    UNLOCK(&frame->lock);
 
-                        DHT_STACK_UNWIND (rmdir, frame, local->op_ret,
-                                          local->op_errno, &local->preparent,
-                                          &local->postparent, NULL);
-               }
-        }
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        if (local->need_selfheal) {
+            dht_rmdir_unlock(frame, this);
+            local->layout = dht_layout_get(this, local->loc.inode);
 
-        return 0;
-}
+            /* TODO: neater interface needed below */
+            local->stbuf.ia_type = local->loc.inode->ia_type;
 
+            gf_uuid_copy(local->gfid, local->loc.inode->gfid);
 
-int
-dht_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-               int op_ret, int op_errno, struct iatt *preparent,
-               struct iatt *postparent, dict_t *xdata)
-{
-        dht_local_t  *local = NULL;
-        int           this_call_cnt = 0;
-        call_frame_t *prev = NULL;
-        int           done = 0;
+            /* Use a different frame or else the rmdir op_ret is
+             * overwritten by that of the selfheal */
 
-        local = frame->local;
-        prev  = cookie;
+            heal_frame = copy_frame(frame);
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        local->op_errno = op_errno;
-                        local->op_ret   = -1;
+            if (heal_frame == NULL) {
+                goto err;
+            }
 
-                        if (op_errno != ENOENT && op_errno != EACCES) {
-                                local->need_selfheal = 1;
-                        }
+            heal_local = dht_local_init(heal_frame, &local->loc, NULL, 0);
+            if (!heal_local) {
+                DHT_STACK_DESTROY(heal_frame);
+                goto err;
+            }
 
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "rmdir on %s for %s failed (%s)",
-                                prev->this->name, local->loc.path,
-                                strerror (op_errno));
-                        goto unlock;
-                }
+            heal_local->inode = inode_ref(local->loc.inode);
+            heal_local->main_frame = frame;
+            gf_uuid_copy(heal_local->gfid, local->loc.inode->gfid);
 
-                /* Track if rmdir succeeded on atleast one subvol*/
-                local->fop_succeeded = 1;
-                dht_iatt_merge (this, &local->preparent, preparent, prev->this);
-                dht_iatt_merge (this, &local->postparent, postparent,
-                                prev->this);
-        }
-unlock:
-        UNLOCK (&frame->lock);
+            dht_selfheal_restore(heal_frame, dht_rmdir_selfheal_cbk,
+                                 &heal_local->loc, heal_local->layout);
+            return 0;
+        } else {
+            if (local->loc.parent) {
+                dht_inode_ctx_time_update(local->loc.parent, this,
+                                          &local->preparent, 0);
 
+                dht_inode_ctx_time_update(local->loc.parent, this,
+                                          &local->postparent, 1);
+            }
 
-        this_call_cnt = dht_frame_return (frame);
+            dht_set_fixed_dir_stat(&local->preparent);
+            dht_set_fixed_dir_stat(&local->postparent);
 
-        /* if local->hashed_subvol, we are yet to wind to hashed_subvol. */
-        if (local->hashed_subvol && (this_call_cnt == 1)) {
-                done = 1;
-        } else if (!local->hashed_subvol && !this_call_cnt) {
-                done = 1;
+            dht_rmdir_unlock(frame, this);
+            DHT_STACK_UNWIND(rmdir, frame, local->op_ret, local->op_errno,
+                             &local->preparent, &local->postparent, NULL);
         }
+    }
 
+    return 0;
 
-        if (done) {
-                if (local->need_selfheal && local->fop_succeeded) {
-                        local->layout =
-                                dht_layout_get (this, local->loc.inode);
+err:
+    DHT_STACK_UNWIND(rmdir, frame, local->op_ret, local->op_errno, NULL, NULL,
+                     NULL);
+    return 0;
+}
 
-                        /* TODO: neater interface needed below */
-                        local->stbuf.ia_type = local->loc.inode->ia_type;
+static int
+dht_rmdir_unlock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    DHT_STACK_DESTROY(frame);
+    return 0;
+}
 
-                        uuid_copy (local->gfid, local->loc.inode->gfid);
-                        dht_selfheal_restore (frame, dht_rmdir_selfheal_cbk,
-                                              &local->loc, local->layout);
-                } else if (this_call_cnt) {
-                        /* If non-hashed subvol's have responded, proceed */
+static int
+dht_rmdir_unlock(call_frame_t *frame, xlator_t *this)
+{
+    dht_local_t *local = NULL, *lock_local = NULL;
+    call_frame_t *lock_frame = NULL;
+    int lock_count = 0;
 
-                        local->need_selfheal = 0;
-                        STACK_WIND (frame, dht_rmdir_hashed_subvol_cbk,
-                                    local->hashed_subvol,
-                                    local->hashed_subvol->fops->rmdir,
-                                    &local->loc, local->flags, NULL);
-                } else if (!this_call_cnt) {
-                        /* All subvol's have responded, proceed */
+    local = frame->local;
 
-                        if (local->loc.parent) {
+    /* Unlock entrylk */
+    dht_unlock_entrylk_wrapper(frame, &local->lock[0].ns.directory_ns);
 
-                                dht_inode_ctx_time_update (local->loc.parent,
-                                                           this,
-                                                           &local->preparent,
-                                                           0);
+    /* Unlock inodelk */
+    lock_count = dht_lock_count(local->lock[0].ns.parent_layout.locks,
+                                local->lock[0].ns.parent_layout.lk_count);
 
-                                dht_inode_ctx_time_update (local->loc.parent,
-                                                           this,
-                                                           &local->postparent,
-                                                           1);
+    if (lock_count == 0)
+        goto done;
 
-                        }
+    lock_frame = copy_frame(frame);
+    if (lock_frame == NULL)
+        goto done;
 
-                        DHT_STACK_UNWIND (rmdir, frame, local->op_ret,
-                                          local->op_errno, &local->preparent,
-                                          &local->postparent, NULL);
-                }
-        }
+    lock_local = dht_local_init(lock_frame, &local->loc, NULL,
+                                lock_frame->root->op);
+    if (lock_local == NULL)
+        goto done;
 
-        return 0;
-}
+    lock_local->lock[0].ns.parent_layout.locks = local->lock[0]
+                                                     .ns.parent_layout.locks;
+    lock_local->lock[0]
+        .ns.parent_layout.lk_count = local->lock[0].ns.parent_layout.lk_count;
 
+    local->lock[0].ns.parent_layout.locks = NULL;
+    local->lock[0].ns.parent_layout.lk_count = 0;
+    dht_unlock_inodelk(lock_frame, lock_local->lock[0].ns.parent_layout.locks,
+                       lock_local->lock[0].ns.parent_layout.lk_count,
+                       dht_rmdir_unlock_cbk);
+    lock_frame = NULL;
 
-int
-dht_rmdir_do (call_frame_t *frame, xlator_t *this)
+done:
+    if (lock_frame != NULL) {
+        DHT_STACK_DESTROY(lock_frame);
+    }
+
+    return 0;
+}
+
+static int
+dht_rmdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, struct iatt *preparent, struct iatt *postparent,
+              dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
-        dht_conf_t   *conf = NULL;
-        int           i = 0;
-        xlator_t     *hashed_subvol = NULL;
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+    int done = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    dht_local_t *heal_local = NULL;
+    call_frame_t *heal_frame = NULL;
+    int ret = -1;
+
+    local = frame->local;
+    prev = cookie;
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            if ((op_errno != ENOENT) && (op_errno != ESTALE)) {
+                local->op_errno = op_errno;
+                local->op_ret = -1;
 
-        VALIDATE_OR_GOTO (this->private, err);
+                if (op_errno != EACCES)
+                    local->need_selfheal = 1;
+            }
 
-        conf = this->private;
-        local = frame->local;
+            gf_uuid_unparse(local->loc.gfid, gfid);
 
-        if (local->op_ret == -1)
-                goto err;
+            gf_msg_debug(this->name, op_errno,
+                         "rmdir on %s for %s failed."
+                         "(gfid = %s)",
+                         prev->name, local->loc.path, gfid);
+            goto unlock;
+        }
 
-        local->call_cnt = conf->subvolume_cnt;
+        /* Track if rmdir succeeded on at least one subvol*/
+        local->fop_succeeded = 1;
+        dht_iatt_merge(this, &local->preparent, preparent);
+        dht_iatt_merge(this, &local->postparent, postparent);
+    }
+unlock:
+    UNLOCK(&frame->lock);
 
-        /* first remove from non-hashed_subvol */
-        hashed_subvol = dht_subvol_get_hashed (this, &local->loc);
+    this_call_cnt = dht_frame_return(frame);
 
-        if (!hashed_subvol) {
-                gf_log (this->name, GF_LOG_WARNING, "failed to get hashed "
-                        "subvol for %s",local->loc.path);
-        } else {
-                local->hashed_subvol = hashed_subvol;
-        }
+    /* if local->hashed_subvol, we are yet to wind to hashed_subvol. */
+    if (local->hashed_subvol && (this_call_cnt == 1)) {
+        done = 1;
+    } else if (!local->hashed_subvol && !this_call_cnt) {
+        done = 1;
+    }
+
+    if (done) {
+        if (local->need_selfheal && local->fop_succeeded) {
+            dht_rmdir_unlock(frame, this);
+            local->layout = dht_layout_get(this, local->loc.inode);
+
+            /* TODO: neater interface needed below */
+            local->stbuf.ia_type = local->loc.inode->ia_type;
+
+            gf_uuid_copy(local->gfid, local->loc.inode->gfid);
+            heal_frame = copy_frame(frame);
+            if (heal_frame == NULL) {
+                goto err;
+            }
+
+            heal_local = dht_local_init(heal_frame, &local->loc, NULL, 0);
+            if (!heal_local) {
+                DHT_STACK_DESTROY(heal_frame);
+                goto err;
+            }
+
+            heal_local->inode = inode_ref(local->loc.inode);
+            heal_local->main_frame = frame;
+            gf_uuid_copy(heal_local->gfid, local->loc.inode->gfid);
+            ret = dht_selfheal_restore(heal_frame, dht_rmdir_selfheal_cbk,
+                                       &heal_local->loc, heal_local->layout);
+            if (ret) {
+                DHT_STACK_DESTROY(heal_frame);
+                goto err;
+            }
+
+        } else if (this_call_cnt) {
+            /* If non-hashed subvol's have responded, proceed */
+            if (local->op_ret == 0) {
+                /* Delete the dir from the hashed subvol if:
+                 * The fop succeeded on at least one subvol
+                 *  and did not fail on any
+                 *  or
+                 *  The fop failed with ENOENT/ESTALE on
+                 *  all subvols */
+
+                STACK_WIND_COOKIE(frame, dht_rmdir_hashed_subvol_cbk,
+                                  local->hashed_subvol, local->hashed_subvol,
+                                  local->hashed_subvol->fops->rmdir,
+                                  &local->loc, local->flags, NULL);
+            } else {
+                /* hashed-subvol was non-NULL and rmdir failed on
+                 * all non hashed-subvols. Unwind rmdir with
+                 * local->op_ret and local->op_errno. */
+                dht_rmdir_unlock(frame, this);
+                DHT_STACK_UNWIND(rmdir, frame, local->op_ret, local->op_errno,
+                                 &local->preparent, &local->postparent, NULL);
 
-        /* When DHT has only 1 child */
-        if (conf->subvolume_cnt == 1) {
-                STACK_WIND (frame, dht_rmdir_hashed_subvol_cbk,
-                            conf->subvolumes[0],
-                            conf->subvolumes[0]->fops->rmdir,
-                            &local->loc, local->flags, NULL);
                 return 0;
-        }
+            }
+        } else if (!this_call_cnt) {
+            /* All subvol's have responded, proceed */
 
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                if (hashed_subvol &&
-                    (hashed_subvol == conf->subvolumes[i]))
-                        continue;
+            if (local->loc.parent) {
+                dht_inode_ctx_time_update(local->loc.parent, this,
+                                          &local->preparent, 0);
+
+                dht_inode_ctx_time_update(local->loc.parent, this,
+                                          &local->postparent, 1);
+            }
+
+            dht_set_fixed_dir_stat(&local->preparent);
+            dht_set_fixed_dir_stat(&local->postparent);
 
-                STACK_WIND (frame, dht_rmdir_cbk,
-                            conf->subvolumes[i],
-                            conf->subvolumes[i]->fops->rmdir,
-                            &local->loc, local->flags, NULL);
+            dht_rmdir_unlock(frame, this);
+            DHT_STACK_UNWIND(rmdir, frame, local->op_ret, local->op_errno,
+                             &local->preparent, &local->postparent, NULL);
         }
+    }
 
-        return 0;
+    return 0;
 
 err:
-        DHT_STACK_UNWIND (rmdir, frame, local->op_ret, local->op_errno,
-                          &local->preparent, &local->postparent, NULL);
-        return 0;
+    DHT_STACK_UNWIND(rmdir, frame, -1, local->op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
+static int
+dht_rmdir_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int i = 0;
+    xlator_t *hashed_subvol;
+
+    conf = this->private;
+    local = frame->local;
 
-int
-dht_rmdir_linkfile_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                               int op_ret, int op_errno, struct iatt *preparent,
-                               struct iatt *postparent, dict_t *xdata)
+    if (op_ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_INODE_LK_ERROR,
+               "acquiring entrylk after inodelk failed rmdir for %s)",
+               local->loc.path);
+
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+        goto err;
+    }
+
+    hashed_subvol = local->hashed_subvol;
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (hashed_subvol && (hashed_subvol == conf->subvolumes[i]))
+            continue;
+
+        STACK_WIND_COOKIE(frame, dht_rmdir_cbk, conf->subvolumes[i],
+                          conf->subvolumes[i], conf->subvolumes[i]->fops->rmdir,
+                          &local->loc, local->flags, NULL);
+    }
+
+    return 0;
+
+err:
+    DHT_STACK_UNWIND(rmdir, frame, local->op_ret, local->op_errno,
+                     &local->preparent, &local->postparent, NULL);
+
+    return 0;
+}
+
+static int
+dht_rmdir_do(call_frame_t *frame, xlator_t *this)
 {
-        dht_local_t    *local = NULL;
-        call_frame_t   *prev = NULL;
-        xlator_t       *src = NULL;
-        call_frame_t   *main_frame = NULL;
-        dht_local_t    *main_local = NULL;
-        int             this_call_cnt = 0;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
+    xlator_t *hashed_subvol = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
 
-        local  = frame->local;
-        prev   = cookie;
-        src    = prev->this;
+    VALIDATE_OR_GOTO(frame->local, err);
+    local = frame->local;
+    VALIDATE_OR_GOTO(this->private, out);
+    conf = this->private;
 
-        main_frame = local->main_frame;
-        main_local = main_frame->local;
+    if (local->op_ret == -1)
+        goto out;
 
-        if (op_ret == 0) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "unlinked linkfile %s on %s",
-                        local->loc.path, src->name);
-        } else {
-                main_local->op_ret   = -1;
-                main_local->op_errno = op_errno;
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "unlink of %s on %s failed (%s)",
-                        local->loc.path, src->name, strerror (op_errno));
-        }
+    local->call_cnt = conf->subvolume_cnt;
+
+    /* first remove from non-hashed_subvol */
+    hashed_subvol = dht_subvol_get_hashed(this, &local->loc);
 
-        this_call_cnt = dht_frame_return (main_frame);
-        if (is_last_call (this_call_cnt))
-                dht_rmdir_do (main_frame, this);
+    if (!hashed_subvol) {
+        gf_uuid_unparse(local->loc.gfid, gfid);
 
-        DHT_STACK_DESTROY (frame);
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+               "Failed to get hashed subvol for %s (gfid = %s)",
+               local->loc.path, gfid);
+    } else {
+        local->hashed_subvol = hashed_subvol;
+    }
+
+    /* When DHT has only 1 child */
+    if (conf->subvolume_cnt == 1) {
+        STACK_WIND_COOKIE(frame, dht_rmdir_hashed_subvol_cbk,
+                          conf->subvolumes[0], conf->subvolumes[0],
+                          conf->subvolumes[0]->fops->rmdir, &local->loc,
+                          local->flags, NULL);
         return 0;
+    }
+
+    local->current = &local->lock[0];
+    ret = dht_protect_namespace(frame, &local->loc, local->hashed_subvol,
+                                &local->current->ns, dht_rmdir_lock_cbk);
+    if (ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = errno ? errno : EINVAL;
+        goto out;
+    }
+
+    return 0;
+
+out:
+    dht_set_fixed_dir_stat(&local->preparent);
+    dht_set_fixed_dir_stat(&local->postparent);
+
+    DHT_STACK_UNWIND(rmdir, frame, local->op_ret, local->op_errno,
+                     &local->preparent, &local->postparent, NULL);
+    return 0;
+err:
+    DHT_STACK_UNWIND(rmdir, frame, -1, EINVAL, NULL, NULL, NULL);
+    return 0;
 }
 
+static void
+dht_rmdir_readdirp_done(call_frame_t *readdirp_frame, xlator_t *this)
+{
+    call_frame_t *main_frame = NULL;
+    dht_local_t *main_local = NULL;
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+
+    local = readdirp_frame->local;
+    main_frame = local->main_frame;
+    main_local = main_frame->local;
+
+    /* At least one readdirp failed.
+     * This is a bit hit or miss - if readdirp failed on more than
+     * one subvol, we don't know which error is returned.
+     */
+    if (local->op_ret == -1) {
+        main_local->op_ret = local->op_ret;
+        main_local->op_errno = local->op_errno;
+    }
+
+    this_call_cnt = dht_frame_return(main_frame);
+
+    if (is_last_call(this_call_cnt))
+        dht_rmdir_do(main_frame, this);
+
+    DHT_STACK_DESTROY(readdirp_frame);
+}
 
-int
-dht_rmdir_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int op_ret, int op_errno, inode_t *inode,
-                      struct iatt *stbuf, dict_t *xattr, struct iatt *parent)
+/* Keep sending readdirp on the subvol until it returns no more entries
+ * It is possible that not all entries will fit in a single readdirp in
+ * which case the rmdir will keep failing with ENOTEMPTY
+ */
+
+static int
+dht_rmdir_readdirp_do(call_frame_t *readdirp_frame, xlator_t *this)
 {
-        dht_local_t    *local = NULL;
-        call_frame_t   *prev = NULL;
-        xlator_t       *src = NULL;
-        call_frame_t   *main_frame = NULL;
-        dht_local_t    *main_local = NULL;
-        int             this_call_cnt = 0;
-        dht_conf_t     *conf = this->private;
+    dht_local_t *local = NULL;
 
-        local = frame->local;
-        prev  = cookie;
-        src   = prev->this;
+    local = readdirp_frame->local;
 
-        main_frame = local->main_frame;
-        main_local = main_frame->local;
+    if (local->op_ret == -1) {
+        /* there is no point doing another readdirp on this
+         * subvol . */
+        dht_rmdir_readdirp_done(readdirp_frame, this);
+        return 0;
+    }
 
-        if (op_ret != 0)
-                goto err;
+    STACK_WIND_COOKIE(readdirp_frame, dht_rmdir_readdirp_cbk,
+                      local->hashed_subvol, local->hashed_subvol,
+                      local->hashed_subvol->fops->readdirp, local->fd, 4096, 0,
+                      local->xattr);
 
-        if (!check_is_linkfile (inode, stbuf, xattr, conf->link_xattr_name)) {
-                main_local->op_ret  = -1;
-                main_local->op_errno = ENOTEMPTY;
+    return 0;
+}
 
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s on %s found to be not a linkfile (type=0%o)",
-                        local->loc.path, src->name, stbuf->ia_type);
-                goto err;
-        }
+static int
+dht_rmdir_linkfile_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                              int op_ret, int op_errno, struct iatt *preparent,
+                              struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    xlator_t *src = NULL;
+    call_frame_t *readdirp_frame = NULL;
+    dht_local_t *readdirp_local = NULL;
+    int this_call_cnt = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    prev = cookie;
+    src = prev;
+
+    readdirp_frame = local->main_frame;
+    readdirp_local = readdirp_frame->local;
+
+    gf_uuid_unparse(local->loc.gfid, gfid);
+
+    if (op_ret == 0) {
+        gf_msg_trace(this->name, 0, "Unlinked linkfile %s on %s, gfid = %s",
+                     local->loc.path, src->name, gfid);
+    } else {
+        if (op_errno != ENOENT) {
+            readdirp_local->op_ret = -1;
+            readdirp_local->op_errno = op_errno;
+        }
+        gf_msg_debug(this->name, op_errno,
+                     "Unlink of %s on %s failed. (gfid = %s)", local->loc.path,
+                     src->name, gfid);
+    }
+
+    this_call_cnt = dht_frame_return(readdirp_frame);
+
+    if (is_last_call(this_call_cnt))
+        dht_rmdir_readdirp_do(readdirp_frame, this);
+
+    DHT_STACK_DESTROY(frame);
+    return 0;
+}
 
-        STACK_WIND (frame, dht_rmdir_linkfile_unlink_cbk,
-                    src, src->fops->unlink, &local->loc, 0, NULL);
-        return 0;
+static int
+dht_rmdir_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, inode_t *inode,
+                     struct iatt *stbuf, dict_t *xattr, struct iatt *parent)
+{
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    xlator_t *src = NULL;
+    call_frame_t *readdirp_frame = NULL;
+    dht_local_t *readdirp_local = NULL;
+    int this_call_cnt = 0;
+    dht_conf_t *conf = this->private;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    prev = cookie;
+    src = prev;
+
+    gf_msg_debug(this->name, 0, "dht_rmdir_lookup_cbk %s", local->loc.path);
+
+    readdirp_frame = local->main_frame;
+    readdirp_local = readdirp_frame->local;
+
+    if (op_ret != 0) {
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_FILE_LOOKUP_FAILED,
+               "lookup failed for %s on %s", local->loc.path, src->name);
+        goto err;
+    }
+
+    if (!check_is_linkfile(inode, stbuf, xattr, conf->link_xattr_name)) {
+        readdirp_local->op_ret = -1;
+        readdirp_local->op_errno = ENOTEMPTY;
+
+        gf_uuid_unparse(local->loc.gfid, gfid);
+
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_NOT_LINK_FILE_ERROR,
+               "%s on %s is not a linkfile (type=0%o, gfid = %s)",
+               local->loc.path, src->name, stbuf->ia_type, gfid);
+        goto err;
+    }
+
+    STACK_WIND_COOKIE(frame, dht_rmdir_linkfile_unlink_cbk, src, src,
+                      src->fops->unlink, &local->loc, 0, NULL);
+    return 0;
 err:
 
-        this_call_cnt = dht_frame_return (main_frame);
-        if (is_last_call (this_call_cnt))
-                dht_rmdir_do (main_frame, this);
+    this_call_cnt = dht_frame_return(readdirp_frame);
+    if (is_last_call(this_call_cnt)) {
+        dht_rmdir_readdirp_do(readdirp_frame, this);
+    }
 
-        DHT_STACK_DESTROY (frame);
-        return 0;
+    DHT_STACK_DESTROY(frame);
+    return 0;
 }
 
+static int
+dht_rmdir_cached_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int op_ret, int op_errno, inode_t *inode,
+                            struct iatt *stbuf, dict_t *xattr,
+                            struct iatt *parent)
+{
+    dht_local_t *local = NULL;
+    xlator_t *src = NULL;
+    call_frame_t *readdirp_frame = NULL;
+    dht_local_t *readdirp_local = NULL;
+    int this_call_cnt = 0;
+    dht_conf_t *conf = this->private;
+    dict_t *xattrs = NULL;
+    int ret = 0;
+
+    local = frame->local;
+    src = local->hashed_subvol;
+
+    /* main_frame here is the readdirp_frame */
+
+    readdirp_frame = local->main_frame;
+    readdirp_local = readdirp_frame->local;
+
+    gf_msg_debug(this->name, 0, "returning for %s ", local->loc.path);
+
+    if (op_ret == 0) {
+        readdirp_local->op_ret = -1;
+        readdirp_local->op_errno = ENOTEMPTY;
+
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_SUBVOL_ERROR,
+               "%s found on cached subvol %s", local->loc.path, src->name);
+        goto err;
+    } else if (op_errno != ENOENT) {
+        readdirp_local->op_ret = -1;
+        readdirp_local->op_errno = op_errno;
+
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_SUBVOL_ERROR,
+               "%s not found on cached subvol %s", local->loc.path, src->name);
+        goto err;
+    }
+
+    xattrs = dict_new();
+    if (!xattrs) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+               "dict_new failed");
+        goto err;
+    }
+
+    ret = dict_set_uint32(xattrs, conf->link_xattr_name, 256);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+               "Failed to set dictionary value: key = %s",
+               conf->link_xattr_name);
+        if (xattrs)
+            dict_unref(xattrs);
+        goto err;
+    }
+    STACK_WIND_COOKIE(frame, dht_rmdir_lookup_cbk, src, src, src->fops->lookup,
+                      &local->loc, xattrs);
+    if (xattrs)
+        dict_unref(xattrs);
+
+    return 0;
+err:
 
-int
-dht_rmdir_is_subvol_empty (call_frame_t *frame, xlator_t *this,
-                           gf_dirent_t *entries, xlator_t *src)
-{
-        int                 ret = 0;
-        int                 build_ret = 0;
-        gf_dirent_t        *trav = NULL;
-        call_frame_t       *lookup_frame = NULL;
-        dht_local_t        *lookup_local = NULL;
-        dht_local_t        *local = NULL;
-        dict_t             *xattrs = NULL;
-        dht_conf_t         *conf = this->private;
+    this_call_cnt = dht_frame_return(readdirp_frame);
 
-        local = frame->local;
+    /* Once all the lookups/unlinks etc have returned, proceed to wind
+     * readdirp on the subvol again until no entries are returned.
+     * This is required if there are more entries than can be returned
+     * in a single readdirp call.
+     */
 
-        list_for_each_entry (trav, &entries->list, list) {
-                if (strcmp (trav->d_name, ".") == 0)
-                        continue;
-                if (strcmp (trav->d_name, "..") == 0)
-                        continue;
-                if (check_is_linkfile (NULL, (&trav->d_stat), trav->dict,
-                                              conf->link_xattr_name)) {
-                        ret++;
-                        continue;
-                }
+    if (is_last_call(this_call_cnt))
+        dht_rmdir_readdirp_do(readdirp_frame, this);
 
-                /* this entry is either a directory which is neither "." nor "..",
-                   or a non directory which is not a linkfile. the directory is to
-                   be treated as non-empty
-                */
-                return 0;
+    DHT_STACK_DESTROY(frame);
+    return 0;
+}
+
+static int
+dht_rmdir_is_subvol_empty(call_frame_t *frame, xlator_t *this,
+                          gf_dirent_t *entries, xlator_t *src)
+{
+    int ret = 0;
+    int build_ret = 0;
+    gf_dirent_t *trav = NULL;
+    call_frame_t *lookup_frame = NULL;
+    dht_local_t *lookup_local = NULL;
+    dht_local_t *local = NULL;
+    dict_t *xattrs = NULL;
+    dht_conf_t *conf = this->private;
+    xlator_t *subvol = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    int count = 0;
+    gf_boolean_t unwind = _gf_false;
+
+    local = frame->local;
+
+    list_for_each_entry(trav, &entries->list, list)
+    {
+        if (strcmp(trav->d_name, ".") == 0)
+            continue;
+        if (strcmp(trav->d_name, "..") == 0)
+            continue;
+        if (check_is_linkfile(NULL, (&trav->d_stat), trav->dict,
+                              conf->link_xattr_name)) {
+            count++;
+            continue;
+        }
+
+        /* this entry is either a directory which is neither "." nor "..",
+           or a non directory which is not a linkfile. the directory is to
+           be treated as non-empty
+        */
+        return 0;
+    }
+
+    xattrs = dict_new();
+    if (!xattrs) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+               "dict_new failed");
+        return -1;
+    }
+
+    ret = dict_set_uint32(xattrs, conf->link_xattr_name, 256);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+               "Failed to set dictionary value: key = %s",
+               conf->link_xattr_name);
+
+        if (xattrs)
+            dict_unref(xattrs);
+        return -1;
+    }
+
+    local->call_cnt = count;
+    ret = 0;
+
+    list_for_each_entry(trav, &entries->list, list)
+    {
+        if (strcmp(trav->d_name, ".") == 0)
+            continue;
+        if (strcmp(trav->d_name, "..") == 0)
+            continue;
+
+        lookup_frame = copy_frame(frame);
+
+        if (!lookup_frame) {
+            /* out of memory, let the rmdir fail
+               (as non-empty, unfortunately) */
+            goto err;
         }
 
-        xattrs = dict_new ();
-        if (!xattrs) {
-                gf_log (this->name, GF_LOG_ERROR, "dict_new failed");
-                return -1;
+        lookup_local = dht_local_init(lookup_frame, NULL, NULL, GF_FOP_LOOKUP);
+        if (!lookup_local) {
+            goto err;
         }
 
-        ret = dict_set_uint32 (xattrs, conf->link_xattr_name, 256);
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "failed to set linkto key"
-                        " in dict");
-                if (xattrs)
-                        dict_unref (xattrs);
-                return -1;
-        }
-
-        list_for_each_entry (trav, &entries->list, list) {
-                if (strcmp (trav->d_name, ".") == 0)
-                        continue;
-                if (strcmp (trav->d_name, "..") == 0)
-                        continue;
-
-                lookup_frame = NULL;
-                lookup_local = NULL;
-
-                lookup_frame = copy_frame (frame);
-                if (!lookup_frame) {
-                        /* out of memory, let the rmdir fail
-                           (as non-empty, unfortunately) */
-                        goto err;
-                }
+        lookup_frame->local = lookup_local;
+        lookup_local->main_frame = frame;
+        lookup_local->hashed_subvol = src;
 
-                lookup_local = mem_get0 (this->local_pool);
-                if (!lookup_local) {
-                        goto err;
-                }
+        build_ret = dht_build_child_loc(this, &lookup_local->loc, &local->loc,
+                                        trav->d_name);
+        if (build_ret != 0)
+            goto err;
 
-                lookup_frame->local = lookup_local;
-                lookup_local->main_frame = frame;
+        gf_uuid_copy(lookup_local->loc.gfid, trav->d_stat.ia_gfid);
 
-                build_ret = dht_build_child_loc (this, &lookup_local->loc,
-                                                 &local->loc, trav->d_name);
-                if (build_ret != 0)
-                        goto err;
+        gf_uuid_unparse(lookup_local->loc.gfid, gfid);
 
-                uuid_copy (lookup_local->loc.gfid, trav->d_stat.ia_gfid);
+        gf_msg_trace(this->name, 0, "looking up %s on subvolume %s, gfid = %s",
+                     lookup_local->loc.path, src->name, gfid);
 
-                gf_log (this->name, GF_LOG_TRACE,
-                        "looking up %s on %s",
-                        lookup_local->loc.path, src->name);
+        subvol = dht_linkfile_subvol(this, NULL, &trav->d_stat, trav->dict);
+        if (!subvol || (subvol == src)) {
+            /* we need to delete the linkto file if it does not have a
+             * valid subvol or it points to itself.
+             */
+            gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_INVALID_LINKFILE,
+                   "Linkfile does not have link subvolume. "
+                   "path = %s, gfid = %s",
+                   lookup_local->loc.path, gfid);
 
-                LOCK (&frame->lock);
-                {
-                        local->call_cnt++;
-                }
-                UNLOCK (&frame->lock);
+            gf_msg_debug(this->name, 0, "looking up %s on subvol %s, gfid = %s",
+                         lookup_local->loc.path, src->name, gfid);
+
+            STACK_WIND_COOKIE(lookup_frame, dht_rmdir_lookup_cbk, src, src,
+                              src->fops->lookup, &lookup_local->loc, xattrs);
+        } else {
+            gf_msg_debug(this->name, 0,
+                         "Looking up linkfile target %s on "
+                         " subvol %s, gfid = %s",
+                         lookup_local->loc.path, subvol->name, gfid);
 
-                STACK_WIND (lookup_frame, dht_rmdir_lookup_cbk,
-                            src, src->fops->lookup,
-                            &lookup_local->loc, xattrs);
-                ret++;
+            STACK_WIND(lookup_frame, dht_rmdir_cached_lookup_cbk, subvol,
+                       subvol->fops->lookup, &lookup_local->loc, xattrs);
         }
+        ret++;
 
-        if (xattrs)
-                dict_unref (xattrs);
+        lookup_frame = NULL;
+        lookup_local = NULL;
+    }
 
-        return ret;
+    if (xattrs)
+        dict_unref(xattrs);
+
+    return ret;
 err:
-        if (xattrs)
-                dict_unref (xattrs);
+    if (xattrs)
+        dict_unref(xattrs);
 
-        DHT_STACK_DESTROY (lookup_frame);
-        return 0;
+    if (lookup_frame)
+        DHT_STACK_DESTROY(lookup_frame);
+
+    /* Handle the case where the wound calls have unwound before the
+     * loop processing is done
+     */
+
+    LOCK(&frame->lock);
+    {
+        local->op_ret = -1;
+        local->op_errno = ENOTEMPTY;
+
+        local->call_cnt -= (count - ret);
+        if (!local->call_cnt)
+            unwind = _gf_true;
+    }
+    UNLOCK(&frame->lock);
+
+    if (!unwind) {
+        return ret;
+    }
+    return 0;
 }
 
+/*
+ * No more entries on this subvol. Proceed to the actual rmdir operation.
+ */
 
-int
-dht_rmdir_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int op_ret, int op_errno, gf_dirent_t *entries,
-                        dict_t *xdata)
+static int
+dht_rmdir_readdirp_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno, gf_dirent_t *entries,
+                       dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
-        int           this_call_cnt = -1;
-        call_frame_t *prev = NULL;
-        xlator_t     *src = NULL;
-        int           ret = 0;
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    xlator_t *src = NULL;
+    int ret = 0;
+    char *path = NULL;
+
+    local = frame->local;
+    prev = cookie;
+    src = prev;
+
+    if (op_ret > 2) {
+        /* dht_rmdir_is_subvol_empty() may free the frame,
+         * copy path for logging.
+         */
+        path = gf_strdup(local->loc.path);
 
-        local = frame->local;
-        prev  = cookie;
-        src   = prev->this;
-
-        if (op_ret > 2) {
-                ret = dht_rmdir_is_subvol_empty (frame, this, entries, src);
-
-                switch (ret) {
-                case 0: /* non linkfiles exist */
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "readdir on %s for %s returned %d entries",
-                                prev->this->name, local->loc.path, op_ret);
-                        local->op_ret = -1;
-                        local->op_errno = ENOTEMPTY;
-                        break;
-                default:
-                        /* @ret number of linkfiles are getting unlinked */
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "readdir on %s for %s found %d linkfiles",
-                                prev->this->name, local->loc.path, ret);
-                        break;
-                }
+        ret = dht_rmdir_is_subvol_empty(frame, this, entries, src);
+
+        switch (ret) {
+            case 0: /* non linkfiles exist */
+                gf_msg_trace(this->name, 0,
+                             "readdir on %s for %s returned %d "
+                             "entries",
+                             prev->name, local->loc.path, op_ret);
+                local->op_ret = -1;
+                local->op_errno = ENOTEMPTY;
+                break;
+            default:
+                /* @ret number of linkfiles are getting unlinked */
+                gf_msg_trace(this->name, 0,
+                             "readdir on %s for %s found %d "
+                             "linkfiles",
+                             prev->name, path, ret);
+                break;
         }
+    }
 
-        this_call_cnt = dht_frame_return (frame);
+    /* readdirp failed or no linkto files were found on this subvol */
+    if (!ret)
+        dht_rmdir_readdirp_done(frame, this);
 
-        if (is_last_call (this_call_cnt)) {
-                dht_rmdir_do (frame, this);
-        }
+    GF_FREE(path);
+    return 0;
+}
 
+static int
+dht_rmdir_opendir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, fd_t *fd, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = -1;
+    xlator_t *prev = NULL;
+    int ret = 0;
+    dht_conf_t *conf = this->private;
+    dict_t *dict = NULL;
+    int i = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    dht_local_t *readdirp_local = NULL;
+    call_frame_t *readdirp_frame = NULL;
+    int cnt = 0;
+
+    local = frame->local;
+    prev = cookie;
+
+    this_call_cnt = dht_frame_return(frame);
+    if (op_ret == -1) {
+        gf_uuid_unparse(local->loc.gfid, gfid);
+
+        gf_msg_debug(this->name, op_errno,
+                     "opendir on %s for %s failed, "
+                     "gfid = %s,",
+                     prev->name, local->loc.path, gfid);
+        if ((op_errno != ENOENT) && (op_errno != ESTALE)) {
+            local->op_ret = -1;
+            local->op_errno = op_errno;
+        }
+        goto err;
+    }
+
+    if (!is_last_call(this_call_cnt))
         return 0;
-}
 
+    if (local->op_ret == -1)
+        goto err;
 
-int
-dht_rmdir_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int op_ret, int op_errno, fd_t *fd, dict_t *xdata)
-{
-        dht_local_t  *local = NULL;
-        int           this_call_cnt = -1;
-        call_frame_t *prev = NULL;
-        dict_t       *dict = NULL;
-        int           ret = 0;
-        dht_conf_t   *conf = this->private;
+    fd_bind(fd);
 
-        local = frame->local;
-        prev  = cookie;
+    dict = dict_new();
+    if (!dict) {
+        local->op_ret = -1;
+        local->op_errno = ENOMEM;
+        goto err;
+    }
 
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "opendir on %s for %s failed (%s)",
-                        prev->this->name, local->loc.path,
-                        strerror (op_errno));
-                if (op_errno != ENOENT) {
-                        local->op_ret = -1;
-                        local->op_errno = op_errno;
-                }
-                goto err;
+    ret = dict_set_uint32(dict, conf->link_xattr_name, 256);
+    if (ret)
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+               "%s: Failed to set dictionary value:key = %s", local->loc.path,
+               conf->link_xattr_name);
+
+    cnt = local->call_cnt = conf->subvolume_cnt;
+
+    /* Create a separate frame per subvol as we might need
+     * to resend readdirp multiple times to get all the
+     * entries.
+     */
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        readdirp_frame = copy_frame(frame);
+
+        if (!readdirp_frame) {
+            cnt--;
+            /* Reduce the local->call_cnt as well */
+            (void)dht_frame_return(frame);
+            continue;
         }
 
-        dict = dict_new ();
-        if (!dict) {
-                local->op_ret = -1;
-                local->op_errno = ENOMEM;
-                goto err;
+        readdirp_local = dht_local_init(readdirp_frame, &local->loc, local->fd,
+                                        0);
+
+        if (!readdirp_local) {
+            DHT_STACK_DESTROY(readdirp_frame);
+            cnt--;
+            /* Reduce the local->call_cnt as well */
+            dht_frame_return(frame);
+            continue;
         }
+        readdirp_local->main_frame = frame;
+        readdirp_local->op_ret = 0;
+        readdirp_local->xattr = dict_ref(dict);
+        /* overload this field to save the subvol info */
+        readdirp_local->hashed_subvol = conf->subvolumes[i];
 
-        ret = dict_set_uint32 (dict, conf->link_xattr_name, 256);
-        if (ret)
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to set '%s' key",
-                        local->loc.path, conf->link_xattr_name);
+        STACK_WIND_COOKIE(readdirp_frame, dht_rmdir_readdirp_cbk,
+                          conf->subvolumes[i], conf->subvolumes[i],
+                          conf->subvolumes[i]->fops->readdirp,
+                          readdirp_local->fd, 4096, 0, readdirp_local->xattr);
+    }
 
-        STACK_WIND (frame, dht_rmdir_readdirp_cbk,
-                    prev->this, prev->this->fops->readdirp,
-                    local->fd, 4096, 0, dict);
+    if (dict)
+        dict_unref(dict);
 
-        if (dict)
-                dict_unref (dict);
+    /* Could not wind readdirp to any subvol */
 
-        return 0;
+    if (!cnt)
+        goto err;
 
-err:
-        this_call_cnt = dht_frame_return (frame);
+    return 0;
 
-        if (is_last_call (this_call_cnt)) {
-                dht_rmdir_do (frame, this);
-        }
+err:
+    if (is_last_call(this_call_cnt)) {
+        dht_rmdir_do(frame, this);
+    }
 
-        return 0;
+    return 0;
 }
 
-
 int
-dht_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
-           dict_t *xdata)
+dht_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+          dict_t *xdata)
 {
-        dht_local_t  *local  = NULL;
-        dht_conf_t   *conf = NULL;
-        int           op_errno = -1;
-        int           i = -1;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int op_errno = -1;
+    int i = -1;
+    int ret = -1;
+    dict_t *xattr_req = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+    VALIDATE_OR_GOTO(loc->path, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_RMDIR);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->call_cnt = conf->subvolume_cnt;
+    local->op_ret = 0;
+    local->fop_succeeded = 0;
+
+    local->flags = flags;
+
+    local->fd = fd_create(local->loc.inode, frame->root->pid);
+    if (!local->fd) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    if (flags) {
+        return dht_rmdir_do(frame, this);
+    }
+    if (xdata) {
+        xattr_req = dict_ref(xdata);
+    } else {
+        xattr_req = dict_new();
+    }
+    if (xattr_req) {
+        ret = dict_set_uint32(xattr_req, conf->link_xattr_name, 256);
+        /* If parallel-readdir is enabled, this is required
+         * to handle stale linkto files in the directory
+         * being deleted. If this fails, log an error but
+         * do not prevent the operation.
+         */
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, 0, "%s: failed to set key %s",
+                   loc->path, conf->link_xattr_name);
+        }
+    } else {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "%s: failed to set key %s",
+               loc->path, conf->link_xattr_name);
+    }
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        STACK_WIND_COOKIE(frame, dht_rmdir_opendir_cbk, conf->subvolumes[i],
+                          conf->subvolumes[i],
+                          conf->subvolumes[i]->fops->opendir, loc, local->fd,
+                          xattr_req);
+    }
+
+    if (xattr_req) {
+        dict_unref(xattr_req);
+    }
+    return 0;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (this->private, err);
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(rmdir, frame, -1, op_errno, NULL, NULL, NULL);
 
-        conf = this->private;
+    return 0;
+}
 
-        local = dht_local_init (frame, loc, NULL, GF_FOP_RMDIR);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+static int
+dht_entrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *xdata)
 
-        local->call_cnt = conf->subvolume_cnt;
-        local->op_ret   = 0;
-        local->fop_succeeded = 0;
+{
+    DHT_STACK_UNWIND(entrylk, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
-        local->flags = flags;
+/* TODO
+ * Sending entrylk to cached subvol can result in stale lock
+ * as described in the bug 1311002.
+ */
+int
+dht_entrylk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+            const char *basename, entrylk_cmd cmd, entrylk_type type,
+            dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
 
-        local->fd = fd_create (local->loc.inode, frame->root->pid);
-        if (!local->fd) {
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
 
-                op_errno = ENOMEM;
-                goto err;
-        }
+    local = dht_local_init(frame, loc, NULL, GF_FOP_ENTRYLK);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
 
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                STACK_WIND (frame, dht_rmdir_opendir_cbk,
-                            conf->subvolumes[i],
-                            conf->subvolumes[i]->fops->opendir,
-                            loc, local->fd, NULL);
-        }
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_uuid_unparse(loc->gfid, gfid);
 
-        return 0;
+        gf_msg_debug(this->name, 0,
+                     "no cached subvolume for path=%s, "
+                     "gfid = %s",
+                     loc->path, gfid);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    local->call_cnt = 1;
+
+    STACK_WIND(frame, dht_entrylk_cbk, subvol, subvol->fops->entrylk, volume,
+               loc, basename, cmd, type, xdata);
+
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (rmdir, frame, -1, op_errno,
-                          NULL, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(entrylk, frame, -1, op_errno, NULL);
 
-        return 0;
+    return 0;
 }
 
-int
-dht_entrylk_cbk (call_frame_t *frame, void *cookie,
-                 xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata)
+static int
+dht_fentrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
 
 {
-        DHT_STACK_UNWIND (entrylk, frame, op_ret, op_errno, xdata);
-        return 0;
+    DHT_STACK_UNWIND(fentrylk, frame, op_ret, op_errno, NULL);
+    return 0;
 }
 
-
 int
-dht_entrylk (call_frame_t *frame, xlator_t *this,
-             const char *volume, loc_t *loc, const char *basename,
-             entrylk_cmd cmd, entrylk_type type, dict_t *xdata)
-{
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        local = dht_local_init (frame, loc, NULL, GF_FOP_ENTRYLK);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+dht_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+             const char *basename, entrylk_cmd cmd, entrylk_type type,
+             dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
 
-        subvol = local->cached_subvol;
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for path=%s", loc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+    VALIDATE_OR_GOTO(fd->inode, err);
 
-        local->call_cnt = 1;
+    gf_uuid_unparse(fd->inode->gfid, gfid);
 
-        STACK_WIND (frame, dht_entrylk_cbk,
-                    subvol, subvol->fops->entrylk,
-                    volume, loc, basename, cmd, type, xdata);
+    subvol = dht_subvol_get_cached(this, fd->inode);
+    if (!subvol) {
+        gf_msg_debug(this->name, 0,
+                     "No cached subvolume for fd=%p,"
+                     " gfid = %s",
+                     fd, gfid);
+        op_errno = EINVAL;
+        goto err;
+    }
 
-        return 0;
+    STACK_WIND(frame, dht_fentrylk_cbk, subvol, subvol->fops->fentrylk, volume,
+               fd, basename, cmd, type, xdata);
+
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (entrylk, frame, -1, op_errno, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(fentrylk, frame, -1, op_errno, NULL);
 
-        return 0;
+    return 0;
 }
 
+static int32_t
+dht_ipc_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
 
-int
-dht_fentrylk_cbk (call_frame_t *frame, void *cookie,
-                  xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata)
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
 
-{
-        DHT_STACK_UNWIND (fentrylk, frame, op_ret, op_errno, NULL);
-        return 0;
-}
+    local = frame->local;
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret < 0 && op_errno != ENOTCONN) {
+            local->op_errno = op_errno;
+            goto unlock;
+        }
+        local->op_ret = 0;
+    }
+unlock:
+    UNLOCK(&frame->lock);
 
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        DHT_STACK_UNWIND(ipc, frame, local->op_ret, local->op_errno, NULL);
+    }
 
-int
-dht_fentrylk (call_frame_t *frame, xlator_t *this,
-              const char *volume, fd_t *fd, const char *basename,
-              entrylk_cmd cmd, entrylk_type type, dict_t *xdata)
+out:
+    return 0;
+}
+
+int32_t
+dht_ipc(call_frame_t *frame, xlator_t *this, int32_t op, dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
+    dht_local_t *local = NULL;
+    int op_errno = EINVAL;
+    dht_conf_t *conf = NULL;
+    int call_cnt = 0;
+    int i = 0;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
 
-        subvol = dht_subvol_get_cached (this, fd->inode);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
-        }
+    if (op != GF_IPC_TARGET_UPCALL)
+        goto wind_default;
 
-        STACK_WIND (frame, dht_fentrylk_cbk,
-                    subvol, subvol->fops->fentrylk,
-                    volume, fd, basename, cmd, type, xdata);
+    VALIDATE_OR_GOTO(this->private, err);
+    conf = this->private;
 
-        return 0;
+    local = dht_local_init(frame, NULL, NULL, GF_FOP_IPC);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    call_cnt = conf->subvolume_cnt;
+    local->call_cnt = call_cnt;
+
+    if (xdata) {
+        if (dict_set_int8(xdata, conf->xattr_name, 0) < 0)
+            goto err;
+    }
+
+    for (i = 0; i < call_cnt; i++) {
+        STACK_WIND(frame, dht_ipc_cbk, conf->subvolumes[i],
+                   conf->subvolumes[i]->fops->ipc, op, xdata);
+    }
+
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (fentrylk, frame, -1, op_errno, NULL);
+    DHT_STACK_UNWIND(ipc, frame, -1, op_errno, NULL);
 
-        return 0;
-}
+    return 0;
 
+wind_default:
+    STACK_WIND(frame, default_ipc_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ipc, op, xdata);
+    return 0;
+}
 
 int
-dht_forget (xlator_t *this, inode_t *inode)
+dht_forget(xlator_t *this, inode_t *inode)
 {
-        uint64_t        ctx_int = 0;
-        dht_inode_ctx_t *ctx    = NULL;
-        dht_layout_t *layout = NULL;
+    uint64_t ctx_int = 0;
+    dht_inode_ctx_t *ctx = NULL;
+    dht_layout_t *layout = NULL;
 
-        inode_ctx_del (inode, this, &ctx_int);
+    inode_ctx_del(inode, this, &ctx_int);
 
-        if (!ctx_int)
-                return 0;
+    if (!ctx_int)
+        return 0;
 
-        ctx = (dht_inode_ctx_t *) (long) ctx_int;
+    ctx = (dht_inode_ctx_t *)(long)ctx_int;
 
-        layout = ctx->layout;
-        ctx->layout = NULL;
-        dht_layout_unref (this, layout);
-        GF_FREE (ctx);
+    layout = ctx->layout;
+    ctx->layout = NULL;
+    dht_layout_unref(this, layout);
+    GF_FREE(ctx);
 
-        return 0;
+    return 0;
 }
 
-
 int
-dht_notify (xlator_t *this, int event, void *data, ...)
-{
-        xlator_t                *subvol = NULL;
-        int                      cnt    = -1;
-        int                      i      = -1;
-        dht_conf_t              *conf   = NULL;
-        int                      ret    = -1;
-        int                      propagate = 0;
-
-        int                      had_heard_from_all = 0;
-        int                      have_heard_from_all = 0;
-        struct timeval           time = {0,};
-        gf_defrag_info_t        *defrag = NULL;
-        dict_t                  *dict   = NULL;
-        gf_defrag_type           cmd    = 0;
-        dict_t                  *output = NULL;
-        va_list                  ap;
-
-
-        conf = this->private;
-        if (!conf)
-                return ret;
-
-        /* had all subvolumes reported status once till now? */
-        had_heard_from_all = 1;
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                if (!conf->last_event[i]) {
-                        had_heard_from_all = 0;
-                }
-        }
-
-        switch (event) {
+dht_notify(xlator_t *this, int event, void *data, ...)
+{
+    xlator_t *subvol = NULL;
+    int cnt = -1;
+    int i = -1;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
+    int propagate = 0;
+
+    int had_heard_from_all = 0;
+    int have_heard_from_all = 0;
+    gf_defrag_info_t *defrag = NULL;
+    dict_t *dict = NULL;
+    gf_defrag_type cmd = 0;
+    dict_t *output = NULL;
+    va_list ap;
+    struct gf_upcall *up_data = NULL;
+    struct gf_upcall_cache_invalidation *up_ci = NULL;
+
+    conf = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, conf, out);
+
+    /* had all subvolumes reported status once till now? */
+    had_heard_from_all = 1;
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (!conf->last_event[i]) {
+            had_heard_from_all = 0;
+        }
+    }
+
+    switch (event) {
         case GF_EVENT_CHILD_UP:
-                subvol = data;
+            subvol = data;
 
-                conf->gen++;
+            conf->gen++;
 
-                for (i = 0; i < conf->subvolume_cnt; i++) {
-                        if (subvol == conf->subvolumes[i]) {
-                                cnt = i;
-                                break;
-                        }
+            for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (subvol == conf->subvolumes[i]) {
+                    cnt = i;
+                    break;
                 }
+            }
 
-                if (cnt == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "got GF_EVENT_CHILD_UP bad subvolume %s",
-                                subvol->name);
-                        break;
-                }
+            if (cnt == -1) {
+                gf_msg_debug(this->name, 0,
+                             "got GF_EVENT_CHILD_UP bad "
+                             "subvolume %s",
+                             subvol->name);
+                break;
+            }
 
-                gettimeofday (&time, NULL);
-                LOCK (&conf->subvolume_lock);
-                {
-                        conf->subvolume_status[cnt] = 1;
-                        conf->last_event[cnt] = event;
-                        conf->subvol_up_time[cnt] = time.tv_sec;
-                }
-                UNLOCK (&conf->subvolume_lock);
+            LOCK(&conf->subvolume_lock);
+            {
+                conf->subvolume_status[cnt] = 1;
+                conf->last_event[cnt] = event;
+                conf->subvol_up_time[cnt] = gf_time();
+            }
+            UNLOCK(&conf->subvolume_lock);
 
-                /* one of the node came back up, do a stat update */
-                dht_get_du_info_for_subvol (this, cnt);
+            /* one of the node came back up, do a stat update */
+            dht_get_du_info_for_subvol(this, cnt);
 
-                break;
+            break;
 
-        case GF_EVENT_CHILD_MODIFIED:
-                subvol = data;
+        case GF_EVENT_SOME_DESCENDENT_UP:
+            subvol = data;
+            conf->gen++;
+            propagate = 1;
 
-                conf->gen++;
-                propagate = 1;
+            break;
 
-                break;
+        case GF_EVENT_SOME_DESCENDENT_DOWN:
+            subvol = data;
+            propagate = 1;
 
-        case GF_EVENT_CHILD_DOWN:
-                subvol = data;
-
-                if (conf->assert_no_child_down) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Received CHILD_DOWN. Exiting");
-                        if (conf->defrag) {
-                                gf_defrag_stop (conf->defrag, NULL);
-                        } else {
-                                kill (getpid(), SIGTERM);
-                        }
-                }
+            break;
 
-                for (i = 0; i < conf->subvolume_cnt; i++) {
-                        if (subvol == conf->subvolumes[i]) {
-                                cnt = i;
-                                break;
-                        }
-                }
+        case GF_EVENT_CHILD_DOWN:
+            subvol = data;
 
-                if (cnt == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "got GF_EVENT_CHILD_DOWN bad subvolume %s",
-                                subvol->name);
-                        break;
+            if (conf->assert_no_child_down) {
+                gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_CHILD_DOWN,
+                       "Received CHILD_DOWN. Exiting");
+                if (conf->defrag) {
+                    gf_defrag_stop(conf, GF_DEFRAG_STATUS_FAILED, NULL);
+                } else {
+                    kill(getpid(), SIGTERM);
                 }
+            }
 
-                LOCK (&conf->subvolume_lock);
-                {
-                        conf->subvolume_status[cnt] = 0;
-                        conf->last_event[cnt] = event;
-                        conf->subvol_up_time[cnt] = 0;
+            for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (subvol == conf->subvolumes[i]) {
+                    cnt = i;
+                    break;
                 }
-                UNLOCK (&conf->subvolume_lock);
+            }
 
+            if (cnt == -1) {
+                gf_msg_debug(this->name, 0,
+                             "got GF_EVENT_CHILD_DOWN bad "
+                             "subvolume %s",
+                             subvol->name);
                 break;
+            }
 
-        case GF_EVENT_CHILD_CONNECTING:
-                subvol = data;
+            LOCK(&conf->subvolume_lock);
+            {
+                conf->subvolume_status[cnt] = 0;
+                conf->last_event[cnt] = event;
+                conf->subvol_up_time[cnt] = 0;
+            }
+            UNLOCK(&conf->subvolume_lock);
 
-                for (i = 0; i < conf->subvolume_cnt; i++) {
-                        if (subvol == conf->subvolumes[i]) {
-                                cnt = i;
-                                break;
-                        }
-                }
+            for (i = 0; i < conf->subvolume_cnt; i++)
+                if (conf->last_event[i] != event)
+                    event = GF_EVENT_SOME_DESCENDENT_DOWN;
+            break;
 
-                if (cnt == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "got GF_EVENT_CHILD_CONNECTING bad subvolume %s",
-                                subvol->name);
-                        break;
-                }
+        case GF_EVENT_CHILD_CONNECTING:
+            subvol = data;
 
-                LOCK (&conf->subvolume_lock);
-                {
-                        conf->last_event[cnt] = event;
+            for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (subvol == conf->subvolumes[i]) {
+                    cnt = i;
+                    break;
                 }
-                UNLOCK (&conf->subvolume_lock);
+            }
 
+            if (cnt == -1) {
+                gf_msg_debug(this->name, 0,
+                             "got GF_EVENT_CHILD_CONNECTING"
+                             " bad subvolume %s",
+                             subvol->name);
                 break;
-        case GF_EVENT_VOLUME_DEFRAG:
-        {
-                if (!conf->defrag) {
-                        return ret;
-                }
-                defrag = conf->defrag;
+            }
 
-                dict = data;
-                va_start (ap, data);
-                output = va_arg (ap, dict_t*);
+            LOCK(&conf->subvolume_lock);
+            {
+                conf->last_event[cnt] = event;
+            }
+            UNLOCK(&conf->subvolume_lock);
 
-                ret = dict_get_int32 (dict, "rebalance-command",
-                                      (int32_t*)&cmd);
-                if (ret)
-                        return ret;
-                LOCK (&defrag->lock);
-                {
-                        if (defrag->is_exiting)
-                                goto unlock;
-                        if (cmd == GF_DEFRAG_CMD_STATUS)
-                                gf_defrag_status_get (defrag, output);
-                        else if (cmd == GF_DEFRAG_CMD_STOP)
-                                gf_defrag_stop (defrag, output);
-                }
-unlock:
-                UNLOCK (&defrag->lock);
-                return 0;
-                break;
-        }
+            break;
+        case GF_EVENT_VOLUME_DEFRAG: {
+            if (!conf->defrag) {
+                return ret;
+            }
+            defrag = conf->defrag;
 
+            dict = data;
+            va_start(ap, data);
+            output = va_arg(ap, dict_t *);
+
+            ret = dict_get_int32(dict, "rebalance-command", (int32_t *)&cmd);
+            if (ret) {
+                va_end(ap);
+                return ret;
+            }
+            LOCK(&defrag->lock);
+            {
+                if (defrag->is_exiting)
+                    goto unlock;
+                if ((cmd == GF_DEFRAG_CMD_STATUS) ||
+                    (cmd == GF_DEFRAG_CMD_DETACH_STATUS))
+                    gf_defrag_status_get(conf, output);
+                else if (cmd == GF_DEFRAG_CMD_DETACH_START)
+                    defrag->cmd = GF_DEFRAG_CMD_DETACH_START;
+                else if (cmd == GF_DEFRAG_CMD_STOP ||
+                         cmd == GF_DEFRAG_CMD_DETACH_STOP)
+                    gf_defrag_stop(conf, GF_DEFRAG_STATUS_STOPPED, output);
+            }
+        unlock:
+            UNLOCK(&defrag->lock);
+            va_end(ap);
+            return ret;
+            break;
+        }
+        case GF_EVENT_UPCALL:
+            up_data = (struct gf_upcall *)data;
+            if (up_data->event_type != GF_UPCALL_CACHE_INVALIDATION)
+                break;
+            up_ci = (struct gf_upcall_cache_invalidation *)up_data->data;
+
+            /* Since md-cache will be aggressively filtering lookups,
+             * the stale layout issue will be more pronounced. Hence
+             * when a layout xattr is changed by the rebalance process
+             * notify all the md-cache clients to invalidate the existing
+             * stat cache and send the lookup next time*/
+            if (up_ci->dict && dict_get(up_ci->dict, conf->xattr_name))
+                up_ci->flags |= UP_EXPLICIT_LOOKUP;
+
+            /* TODO: Instead of invalidating iatt, update the new
+             * hashed/cached subvolume in dht inode_ctx */
+            if (IS_DHT_LINKFILE_MODE(&up_ci->stat))
+                up_ci->flags |= UP_EXPLICIT_LOOKUP;
+
+            propagate = 1;
+            break;
         default:
-                propagate = 1;
+            propagate = 1;
+            break;
+    }
+
+    /* have all subvolumes reported status once by now? */
+    have_heard_from_all = 1;
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (!conf->last_event[i])
+            have_heard_from_all = 0;
+    }
+
+    /* if all subvols have reported status, no need to hide anything
+       or wait for anything else. Just propagate blindly */
+    if (have_heard_from_all) {
+        propagate = 1;
+    }
+
+    if (!had_heard_from_all && have_heard_from_all) {
+        static int run_defrag = 0;
+        /* This is the first event which completes aggregation
+           of events from all subvolumes. If at least one subvol
+           had come up, propagate CHILD_UP, but only this time
+        */
+        event = GF_EVENT_CHILD_DOWN;
+
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            if (conf->last_event[i] == GF_EVENT_CHILD_UP) {
+                event = GF_EVENT_CHILD_UP;
                 break;
+            }
+
+            if (conf->last_event[i] == GF_EVENT_CHILD_CONNECTING) {
+                event = GF_EVENT_CHILD_CONNECTING;
+                /* continue to check other events for CHILD_UP */
+            }
         }
 
+        /* Rebalance is started with assert_no_child_down. So we do
+         * not need to handle CHILD_DOWN event here.
+         *
+         * If there is a graph switch, we should not restart the
+         * rebalance daemon. Use 'run_defrag' to indicate if the
+         * thread has already started.
+         */
+        if (conf->defrag && !run_defrag) {
+            run_defrag = 1;
+            ret = gf_thread_create(&conf->defrag->th, NULL, gf_defrag_start,
+                                   this, "dhtdg");
+            if (ret) {
+                GF_FREE(conf->defrag);
+                conf->defrag = NULL;
+                kill(getpid(), SIGTERM);
+            }
+        }
+    }
+
+    ret = 0;
+    if (propagate)
+        ret = default_notify(this, event, data);
+out:
+    return ret;
+}
 
-        /* have all subvolumes reported status once by now? */
-        have_heard_from_all = 1;
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                if (!conf->last_event[i])
-                        have_heard_from_all = 0;
-        }
+int
+dht_inode_ctx_layout_get(inode_t *inode, xlator_t *this, dht_layout_t **layout)
+{
+    dht_inode_ctx_t *ctx = NULL;
+    int ret = -1;
 
-        /* if all subvols have reported status, no need to hide anything
-           or wait for anything else. Just propagate blindly */
-        if (have_heard_from_all) {
-                propagate = 1;
+    ret = dht_inode_ctx_get(inode, this, &ctx);
 
+    if (!ret && ctx) {
+        if (ctx->layout) {
+            if (layout)
+                *layout = ctx->layout;
+            ret = 0;
+        } else {
+            ret = -1;
         }
+    }
 
+    return ret;
+}
 
-        if (!had_heard_from_all && have_heard_from_all) {
-                /* This is the first event which completes aggregation
-                   of events from all subvolumes. If at least one subvol
-                   had come up, propagate CHILD_UP, but only this time
-                */
-                event = GF_EVENT_CHILD_DOWN;
+void
+dht_log_new_layout_for_dir_selfheal(xlator_t *this, loc_t *loc,
+                                    dht_layout_t *layout)
+{
+    char string[2048] = {0};
+    char *output_string = NULL;
+    int len = 0;
+    int off = 0;
+    int i = 0;
+    gf_loglevel_t log_level = gf_log_get_loglevel();
+    int ret = 0;
+
+    if (log_level < GF_LOG_INFO)
+        return;
 
-                for (i = 0; i < conf->subvolume_cnt; i++) {
-                        if (conf->last_event[i] == GF_EVENT_CHILD_UP) {
-                                event = GF_EVENT_CHILD_UP;
-                                break;
-                        }
+    if (!layout)
+        return;
 
-                        if (conf->last_event[i] == GF_EVENT_CHILD_CONNECTING) {
-                                event = GF_EVENT_CHILD_CONNECTING;
-                                /* continue to check other events for CHILD_UP */
-                        }
-                }
+    if (!layout->cnt)
+        return;
 
-                /* rebalance is started with assert_no_child_down. So we do
-                 * not need to handle CHILD_DOWN event here.
-                 */
-                if (conf->defrag) {
-                        ret = gf_thread_create (&conf->defrag->th, NULL,
-						gf_defrag_start, this);
-                        if (ret) {
-                                conf->defrag = NULL;
-                                GF_FREE (conf->defrag);
-                                kill (getpid(), SIGTERM);
-                        }
-                }
-        }
+    if (!loc)
+        return;
+
+    if (!loc->path)
+        return;
+
+    ret = snprintf(string, sizeof(string), "Setting layout of %s with ",
+                   loc->path);
+
+    if (ret < 0)
+        return;
+
+    len += ret;
+
+    /* Calculation  of total length of the string required to calloc
+     * output_string. Log includes subvolume-name, start-range, end-range
+     * and err value.
+     *
+     * This log will help to debug cases where:
+     * a) Different processes set different layout of a directory.
+     * b) Error captured in lookup, which will be filled in layout->err
+     * (like ENOENT, ESTALE etc)
+     */
+
+    for (i = 0; i < layout->cnt; i++) {
+        ret = snprintf(string, sizeof(string),
+                       "[Subvol_name: %s, Err: %d , Start: "
+                       "0x%x, Stop: 0x%x, Hash: 0x%x], ",
+                       layout->list[i].xlator->name, layout->list[i].err,
+                       layout->list[i].start, layout->list[i].stop,
+                       layout->list[i].commit_hash);
+
+        if (ret < 0)
+            return;
+
+        len += ret;
+    }
+
+    len++;
+
+    output_string = GF_MALLOC(len + 1, gf_common_mt_char);
+
+    if (!output_string)
+        return;
+
+    ret = snprintf(output_string, len + 1, "Setting layout of %s with ",
+                   loc->path);
+
+    if (ret < 0)
+        goto err;
+
+    off += ret;
+
+    for (i = 0; i < layout->cnt; i++) {
+        ret = snprintf(output_string + off, len - off,
+                       "[Subvol_name: %s, Err: %d , Start: "
+                       "0x%x, Stop: 0x%x, Hash: 0x%x], ",
+                       layout->list[i].xlator->name, layout->list[i].err,
+                       layout->list[i].start, layout->list[i].stop,
+                       layout->list[i].commit_hash);
+
+        if (ret < 0)
+            goto err;
+
+        off += ret;
+    }
+
+    gf_msg(this->name, GF_LOG_DEBUG, 0, DHT_MSG_LOG_FIXED_LAYOUT, "%s",
+           output_string);
 
+err:
+    GF_FREE(output_string);
+}
+
+int32_t
+dht_migration_get_dst_subvol(xlator_t *this, dht_local_t *local)
+{
+    int ret = -1;
+
+    if (!local)
+        goto out;
+
+    local->rebalance.target_node = dht_subvol_get_hashed(this, &local->loc);
+
+    if (local->rebalance.target_node)
         ret = 0;
-        if (propagate)
-                ret = default_notify (this, event, data);
 
-        return ret;
+out:
+    return ret;
+}
+
+/*
+This function should not be called more then once during a FOP
+handling path. It is valid only for for ops on files
+*/
+int32_t
+dht_set_local_rebalance(xlator_t *this, dht_local_t *local, struct iatt *stbuf,
+                        struct iatt *prebuf, struct iatt *postbuf,
+                        dict_t *xdata)
+{
+    if (!local)
+        return -1;
+
+    if (local->rebalance.set) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_REBAL_STRUCT_SET,
+               "local->rebalance already set");
+    }
+
+    if (stbuf)
+        memcpy(&local->rebalance.stbuf, stbuf, sizeof(struct iatt));
+
+    if (prebuf)
+        memcpy(&local->rebalance.prebuf, prebuf, sizeof(struct iatt));
+
+    if (postbuf)
+        memcpy(&local->rebalance.postbuf, postbuf, sizeof(struct iatt));
+
+    if (xdata)
+        local->rebalance.xdata = dict_ref(xdata);
+
+    local->rebalance.set = 1;
+
+    return 0;
+}
+
+int32_t
+dht_release(xlator_t *this, fd_t *fd)
+{
+    return dht_fd_ctx_destroy(this, fd);
+}
+
+static int
+dht_pt_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, inode_t *inode, struct iatt *stbuf,
+                 struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (!op_ret) {
+        dht_layout_set(this, inode, local->layout);
+    }
+
+    DHT_STACK_UNWIND(mkdir, frame, op_ret, op_errno, inode, stbuf, preparent,
+                     postparent, NULL);
+
+    return 0;
+}
+
+int32_t
+dht_pt_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+             mode_t umask, dict_t *xdata)
+{
+    dht_layout_t *layout = NULL;
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    bool free_xdata = false;
+    int ret = 0;
+    int op_errno = 0;
+    int32_t *disk_layout_p = NULL;
+
+    conf = this->private;
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_MKDIR);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    layout = dht_layout_new(this, conf->subvolume_cnt);
+    if (!layout)
+        goto wind;
+
+    local->layout = layout;
+
+    if (!xdata) {
+        xdata = dict_new();
+        if (!xdata)
+            goto wind;
+        free_xdata = true;
+    }
+
+    /*Set the xlator or the following will crash*/
+    layout->list[0].xlator = conf->subvolumes[0];
+
+    dht_selfheal_layout_new_directory(frame, loc, layout);
+
+    dht_disk_layout_extract(this, layout, 0, &disk_layout_p);
+
+    ret = dict_set_bin(xdata, conf->xattr_name, disk_layout_p, 4 * 4);
+    if (ret) {
+        gf_msg("dht", GF_LOG_DEBUG, EINVAL, DHT_MSG_DICT_SET_FAILED,
+               "dht layout dict set failed");
+    }
+wind:
+    STACK_WIND(frame, dht_pt_mkdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mkdir, loc, mode, umask, xdata);
+    if (free_xdata)
+        dict_unref(xdata);
+    return 0;
+
+err:
+    op_errno = local ? local->op_errno : op_errno;
+    DHT_STACK_UNWIND(mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+
+    return 0;
+}
+
+static int
+dht_pt_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
+{
+    dht_conf_t *conf = NULL;
+
+    conf = this->private;
+    dict_del(xattr, conf->xattr_name);
+    dict_del(xattr, conf->mds_xattr_key);
+    dict_del(xattr, conf->commithash_xattr_name);
+
+    if (frame->root->pid >= 0) {
+        GF_REMOVE_INTERNAL_XATTR("trusted.glusterfs.quota*", xattr);
+        GF_REMOVE_INTERNAL_XATTR("trusted.pgfid*", xattr);
+    }
+
+    DHT_STACK_UNWIND(getxattr, frame, op_ret, op_errno, xattr, xdata);
+    return 0;
+}
+
+int
+dht_pt_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                const char *key, dict_t *xdata)
+{
+    STACK_WIND(frame, dht_pt_getxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->getxattr, loc, key, xdata);
+    return 0;
+}
+
+static int
+dht_pt_fgetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
+{
+    dht_conf_t *conf = NULL;
+
+    conf = this->private;
+    dict_del(xattr, conf->xattr_name);
+
+    if (frame->root->pid >= 0) {
+        GF_REMOVE_INTERNAL_XATTR("trusted.glusterfs.quota*", xattr);
+        GF_REMOVE_INTERNAL_XATTR("trusted.pgfid*", xattr);
+    }
+
+    DHT_STACK_UNWIND(fgetxattr, frame, op_ret, op_errno, xattr, xdata);
+    return 0;
 }
 
 int
-dht_inode_ctx_layout_get (inode_t *inode, xlator_t *this, dht_layout_t **layout)
+dht_pt_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *key,
+                 dict_t *xdata)
 {
-        dht_inode_ctx_t         *ctx            = NULL;
-        int                      ret            = -1;
+    STACK_WIND(frame, dht_pt_fgetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fgetxattr, fd, key, xdata);
+    return 0;
+}
 
-        ret = dht_inode_ctx_get (inode, this, &ctx);
+/* The job of this function is to check if all the xlators have updated
+ * error in the layout. */
+int
+dht_dir_layout_error_check(xlator_t *this, inode_t *inode)
+{
+    dht_layout_t *layout = NULL;
+    int i = 0;
 
-        if (!ret && ctx) {
-                if (ctx->layout) {
-                        if (layout)
-                                *layout = ctx->layout;
-                        ret = 0;
-                } else {
-                        ret = -1;
-                }
+    layout = dht_layout_get(this, inode);
+    for (i = 0; i < layout->cnt; i++) {
+        if (layout->list[i].err == 0) {
+            return 0;
         }
+    }
 
-        return ret;
+    /* Returning the first xlator error as all xlators have errors */
+    return layout->list[0].err;
 }
diff --git a/xlators/cluster/dht/src/dht-common.h b/xlators/cluster/dht/src/dht-common.h
index da8923e9b24..fe0dc3db34a 100644
--- a/xlators/cluster/dht/src/dht-common.h
+++ b/xlators/cluster/dht/src/dht-common.h
@@ -8,775 +8,1377 @@
   cases as published by the Free Software Foundation.
 */
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
 #include <regex.h>
 
 #include "dht-mem-types.h"
+#include "dht-messages.h"
+#include <glusterfs/call-stub.h>
 #include "libxlator.h"
-#include "syncop.h"
+#include <glusterfs/syncop.h>
+#include <glusterfs/refcount.h>
+#include <glusterfs/timer.h>
+#include "protocol-common.h"
+#include <glusterfs/glusterfs-acl.h>
 
 #ifndef _DHT_H
 #define _DHT_H
 
-#define GF_XATTR_FIX_LAYOUT_KEY     "distribute.fix.layout"
-#define GF_DHT_LOOKUP_UNHASHED_ON   1
+#define GF_XATTR_FIX_LAYOUT_KEY "distribute.fix.layout"
+#define GF_XATTR_FILE_MIGRATE_KEY "trusted.distribute.migrate-data"
+#define DHT_MDS_STR "mds"
+#define GF_DHT_LOOKUP_UNHASHED_OFF 0
+#define GF_DHT_LOOKUP_UNHASHED_ON 1
 #define GF_DHT_LOOKUP_UNHASHED_AUTO 2
-#define DHT_PATHINFO_HEADER         "DISTRIBUTE:"
+#define DHT_PATHINFO_HEADER "DISTRIBUTE:"
+#define DHT_FILE_MIGRATE_DOMAIN "dht.file.migrate"
+/* Layout synchronization */
+#define DHT_LAYOUT_HEAL_DOMAIN "dht.layout.heal"
+/* Namespace synchronization */
+#define DHT_ENTRY_SYNC_DOMAIN "dht.entry.sync"
+#define DHT_LAYOUT_HASH_INVALID 1
+#define MAX_REBAL_THREADS sysconf(_SC_NPROCESSORS_ONLN)
+
+#define DHT_DIR_STAT_BLOCKS 8
+#define DHT_DIR_STAT_SIZE 4096
+
+/* Virtual xattr for subvols status */
+
+#define DHT_SUBVOL_STATUS_KEY "dht.subvol.status"
+
+/* Virtual xattrs for debugging */
+
+#define DHT_DBG_HASHED_SUBVOL_PATTERN "dht.file.hashed-subvol.*"
+#define DHT_DBG_HASHED_SUBVOL_KEY "dht.file.hashed-subvol."
 
-#include <fnmatch.h>
+/* Rebalance nodeuuid flags */
+#define REBAL_NODEUUID_MINE 0x01
 
-typedef int (*dht_selfheal_dir_cbk_t) (call_frame_t *frame, void *cookie,
-                                       xlator_t     *this,
-                                       int32_t       op_ret, int32_t op_errno,
-                                       dict_t *xdata);
-typedef int (*dht_defrag_cbk_fn_t) (xlator_t        *this, call_frame_t *frame,
-                                    int              ret);
+typedef int (*dht_selfheal_dir_cbk_t)(call_frame_t *frame, void *cookie,
+                                      xlator_t *this, int32_t op_ret,
+                                      int32_t op_errno, dict_t *xdata);
+typedef int (*dht_defrag_cbk_fn_t)(xlator_t *this, xlator_t *dst_node,
+                                   call_frame_t *frame, int ret);
 
+typedef int (*dht_refresh_layout_unlock)(call_frame_t *frame, xlator_t *this,
+                                         int op_ret, int invoke_cbk);
+
+typedef int (*dht_refresh_layout_done_handle)(call_frame_t *frame);
 
 struct dht_layout {
-        int                spread_cnt;  /* layout spread count per directory,
-                                           is controlled by 'setxattr()' with
-                                           special key */
-        int                cnt;
-        int                preset;
-        int                gen;
-        int                type;
-        int                ref; /* use with dht_conf_t->layout_lock */
-        int                search_unhashed;
-        struct {
-                int        err;   /* 0 = normal
-                                     -1 = dir exists and no xattr
-                                     >0 = dir lookup failed with errno
-                                  */
-                uint32_t   start;
-                uint32_t   stop;
-                xlator_t  *xlator;
-        } list[];
+    int spread_cnt; /* layout spread count per directory,
+                       is controlled by 'setxattr()' with
+                       special key */
+    int cnt;
+    int preset;
+    /*
+     * The last *configuration* state for which this directory was known
+     * to be in balance.  The corresponding vol_commit_hash changes
+     * whenever bricks are added or removed.  This value changes when a
+     * (full) rebalance is complete.  If they match, it's safe to assume
+     * that every file is where it should be and there's no need to do
+     * lookups for files elsewhere.  If they don't, then we have to do a
+     * global lookup to be sure.
+     */
+    uint32_t commit_hash;
+    /*
+     * The *runtime* state of the volume, changes when connections to
+     * bricks are made or lost.
+     */
+    int gen;
+    int type;
+    gf_atomic_t ref; /* use with dht_conf_t->layout_lock */
+    uint32_t search_unhashed;
+    struct {
+        int err; /* 0 = normal
+                    -1 = dir exists and no xattr
+                    >0 = dir lookup failed with errno
+                 */
+        uint32_t start;
+        uint32_t stop;
+        uint32_t commit_hash;
+        xlator_t *xlator;
+    } list[];
 };
-typedef struct dht_layout  dht_layout_t;
+typedef struct dht_layout dht_layout_t;
 
 struct dht_stat_time {
-        uint32_t        atime;
-        uint32_t        atime_nsec;
-        uint32_t        ctime;
-        uint32_t        ctime_nsec;
-        uint32_t        mtime;
-        uint32_t        mtime_nsec;
+    uint32_t atime;
+    uint32_t atime_nsec;
+    uint32_t ctime;
+    uint32_t ctime_nsec;
+    uint32_t mtime;
+    uint32_t mtime_nsec;
 };
 
 typedef struct dht_stat_time dht_stat_time_t;
 
 struct dht_inode_ctx {
-        dht_layout_t    *layout;
-        dht_stat_time_t  time;
+    dht_layout_t *layout;
+    dht_stat_time_t time;
+    xlator_t *lock_subvol;
+    xlator_t *mds_subvol; /* This is only used for directories */
 };
 
 typedef struct dht_inode_ctx dht_inode_ctx_t;
 
-
 typedef enum {
-        DHT_HASH_TYPE_DM,
-        DHT_HASH_TYPE_DM_USER,
+    DHT_HASH_TYPE_DM,
+    DHT_HASH_TYPE_DM_USER,
 } dht_hashfn_type_t;
 
+typedef enum {
+    DHT_INODELK,
+    DHT_ENTRYLK,
+} dht_lock_type_t;
+
 /* rebalance related */
 struct dht_rebalance_ {
-        xlator_t            *from_subvol;
-        xlator_t            *target_node;
-        off_t                offset;
-        size_t               size;
-        int32_t              flags;
-        int                  count;
-        struct iobref       *iobref;
-        struct iovec        *vector;
-        struct iatt          stbuf;
-        dht_defrag_cbk_fn_t  target_op_fn;
-        dict_t              *xdata;
+    xlator_t *from_subvol;
+    xlator_t *target_node;
+    off_t offset;
+    size_t size;
+    int32_t flags;
+    int count;
+    struct iobref *iobref;
+    struct iovec *vector;
+    struct iatt stbuf;
+    struct iatt prebuf;
+    struct iatt postbuf;
+    dht_defrag_cbk_fn_t target_op_fn;
+    dict_t *xdata;
+    dict_t *xattr;
+    dict_t *dict;
+    struct gf_flock flock;
+    int32_t set;
+    int lock_cmd;
 };
 
-struct dht_local {
-        int                      call_cnt;
-        loc_t                    loc;
-        loc_t                    loc2;
-        int                      op_ret;
-        int                      op_errno;
-        int                      layout_mismatch;
-        /* Use stbuf as the postbuf, when we require both
-         * pre and post attrs */
-        struct iatt              stbuf;
-        struct iatt              prebuf;
-        struct iatt              preoldparent;
-        struct iatt              postoldparent;
-        struct iatt              preparent;
-        struct iatt              postparent;
-        struct statvfs           statvfs;
-        fd_t                    *fd;
-        inode_t                 *inode;
-        dict_t                  *params;
-        dict_t                  *xattr;
-        dict_t                  *xattr_req;
-        dht_layout_t            *layout;
-        size_t                   size;
-        ino_t                    ia_ino;
-        xlator_t                *src_hashed, *src_cached;
-        xlator_t                *dst_hashed, *dst_cached;
-        xlator_t                *cached_subvol;
-        xlator_t                *hashed_subvol;
-        char                     need_selfheal;
-        int                      file_count;
-        int                      dir_count;
-        call_frame_t            *main_frame;
-        int                      fop_succeeded;
-        struct {
-                fop_mknod_cbk_t  linkfile_cbk;
-                struct iatt      stbuf;
-                loc_t            loc;
-                inode_t         *inode;
-                dict_t          *xattr;
-                xlator_t        *srcvol;
-        } linkfile;
-        struct {
-                uint32_t         hole_cnt;
-                uint32_t         overlaps_cnt;
-                uint32_t         down;
-                uint32_t         misc;
-                dht_selfheal_dir_cbk_t   dir_cbk;
-                dht_layout_t    *layout;
-        } selfheal;
-        uint32_t                 uid;
-        uint32_t                 gid;
-
-        /* needed by nufa */
-        int32_t flags;
-        mode_t  mode;
-        dev_t   rdev;
-        mode_t  umask;
-
-        /* need for file-info */
-        char   *xattr_val;
-        char   *key;
-
-        /* which xattr request? */
-        char xsel[256];
-        int32_t alloc_len;
-
-        char   *newpath;
-
-        /* gfid related */
-        uuid_t  gfid;
-
-        /*Marker Related*/
-        struct marker_str    marker;
-
-        /* flag used to make sure we need to return estale in
-           {lookup,revalidate}_cbk */
-        char return_estale;
-        char need_lookup_everywhere;
-
-        glusterfs_fop_t      fop;
+/**
+ * Enum to store decided action based on the qdstatfs (quota-deem-statfs)
+ * events
+ **/
+typedef enum {
+    qdstatfs_action_OFF = 0,
+    qdstatfs_action_REPLACE,
+    qdstatfs_action_NEGLECT,
+    qdstatfs_action_COMPARE,
+} qdstatfs_action_t;
 
-        gf_boolean_t     linked;
-        xlator_t        *link_subvol;
+typedef enum {
+    REACTION_INVALID,
+    FAIL_ON_ANY_ERROR,
+    IGNORE_ENOENT_ESTALE,
+    IGNORE_ENOENT_ESTALE_EIO,
+} dht_reaction_type_t;
+
+struct dht_skip_linkto_unlink {
+    xlator_t *hash_links_to;
+    uuid_t cached_gfid;
+    uuid_t hashed_gfid;
+    int opend_fd_count;
+    gf_boolean_t handle_valid_link;
+};
 
-        struct dht_rebalance_ rebalance;
-        xlator_t        *first_up_subvol;
+typedef struct {
+    xlator_t *xl;
+    loc_t loc;      /* contains/points to inode to lock on. */
+    char *domain;   /* Only locks within a single domain
+                     * contend with each other
+                     */
+    char *basename; /* Required for entrylk */
+    gf_boolean_t locked;
+    dht_reaction_type_t do_on_failure;
+    short type; /* read/write lock.                     */
+    gf_lkowner_t lk_owner;
+} dht_lock_t;
+
+/* The lock structure represents inodelk. */
+typedef struct {
+    fop_inodelk_cbk_t inodelk_cbk;
+    dht_lock_t **locks;
+    int lk_count;
+    dht_reaction_type_t reaction;
+
+    /* whether locking failed on _any_ of the "locks" above */
+    int op_ret;
+    int op_errno;
+} dht_ilock_wrap_t;
+
+/* The lock structure represents entrylk. */
+typedef struct {
+    fop_entrylk_cbk_t entrylk_cbk;
+    dht_lock_t **locks;
+    int lk_count;
+    dht_reaction_type_t reaction;
+
+    /* whether locking failed on _any_ of the "locks" above */
+    int op_ret;
+    int op_errno;
+} dht_elock_wrap_t;
+
+/* The first member of dht_dir_transaction_t should be of type dht_ilock_wrap_t.
+ * Otherwise it can result in subtle memory corruption issues as in most of the
+ * places we use lock[0].layout.my_layout or lock[0].layout.parent_layout and
+ * lock[0].ns.parent_layout (like in dht_local_wipe).
+ */
+typedef union {
+    union {
+        dht_ilock_wrap_t my_layout;
+        dht_ilock_wrap_t parent_layout;
+    } layout;
+    struct dht_namespace {
+        dht_ilock_wrap_t parent_layout;
+        dht_elock_wrap_t directory_ns;
+        fop_entrylk_cbk_t ns_cbk;
+    } ns;
+} dht_dir_transaction_t;
+
+typedef int (*dht_selfheal_layout_t)(call_frame_t *frame, loc_t *loc,
+                                     dht_layout_t *layout);
+
+typedef gf_boolean_t (*dht_need_heal_t)(call_frame_t *frame,
+                                        dht_layout_t **inmem,
+                                        dht_layout_t **ondisk);
 
+struct dht_local {
+    loc_t loc;
+    loc_t loc2;
+    int call_cnt;
+    int op_ret;
+    int op_errno;
+    int layout_mismatch;
+    /* Use stbuf as the postbuf, when we require both
+     * pre and post attrs */
+    struct iatt stbuf;
+    struct iatt mds_stbuf;
+    struct iatt prebuf;
+    struct iatt preoldparent;
+    struct iatt postoldparent;
+    struct iatt preparent;
+    struct iatt postparent;
+    struct statvfs statvfs;
+    fd_t *fd;
+    inode_t *inode;
+    dict_t *params;
+    dict_t *xattr;
+    dict_t *mds_xattr;
+    dict_t *xdata; /* dict used to save xdata response by xattr fop */
+    dict_t *xattr_req;
+    dht_layout_t *layout;
+    size_t size;
+    ino_t ia_ino;
+    xlator_t *src_hashed, *src_cached;
+    xlator_t *dst_hashed, *dst_cached;
+    xlator_t *cached_subvol;
+    xlator_t *hashed_subvol;
+    xlator_t *mds_subvol; /* This is use for dir only */
+    int file_count;
+    int dir_count;
+    call_frame_t *main_frame;
+    int fop_succeeded;
+    struct {
+        fop_mknod_cbk_t linkfile_cbk;
+        struct iatt stbuf;
+        loc_t loc;
+        inode_t *inode;
+        dict_t *xattr;
+        xlator_t *srcvol;
+    } linkfile;
+    struct {
+        uint32_t hole_cnt;
+        uint32_t overlaps_cnt;
+        uint32_t down;
+        uint32_t misc;
+        dht_selfheal_dir_cbk_t dir_cbk;
+        dht_selfheal_layout_t healer;
+        dht_need_heal_t should_heal;
+        dht_layout_t *layout, *refreshed_layout;
+        uint32_t missing_cnt;
+        gf_boolean_t force_mkdir;
+    } selfheal;
+
+    dht_refresh_layout_unlock refresh_layout_unlock;
+    dht_refresh_layout_done_handle refresh_layout_done;
+
+    uint32_t uid;
+    uint32_t gid;
+    pid_t pid;
+
+    glusterfs_fop_t fop;
+
+    /* need for file-info */
+    char *xattr_val;
+    char *key;
+
+    /* needed by nufa */
+    int32_t flags;
+    mode_t mode;
+    dev_t rdev;
+    mode_t umask;
+
+    /* which xattr request? */
+    char xsel[256];
+    int32_t alloc_len;
+
+    /* gfid related */
+    uuid_t gfid;
+    uuid_t gfid_req;
+
+    xlator_t *link_subvol;
+
+    struct dht_rebalance_ rebalance;
+    xlator_t *first_up_subvol;
+
+    struct dht_skip_linkto_unlink skip_unlink;
+
+    dht_dir_transaction_t lock[2], *current;
+
+    /* inodelks during filerename for backward compatibility */
+    dht_lock_t **rename_inodelk_backward_compatible;
+
+    call_stub_t *stub;
+    int32_t parent_disk_layout[4];
+
+    /* rename rollback */
+    int *ret_cache;
+
+    loc_t loc2_copy;
+
+    int rename_inodelk_bc_count;
+    /* This is use only for directory operation */
+    int32_t valid;
+    int32_t mds_heal_fresh_lookup;
+    short lock_type;
+    char need_selfheal;
+    char need_xattr_heal;
+    char need_attrheal;
+    /* flag used to make sure we need to return estale in
+       {lookup,revalidate}_cbk */
+    char return_estale;
+    char need_lookup_everywhere;
+    /* fd open check */
+    gf_boolean_t fd_checked;
+    gf_boolean_t linked;
+    gf_boolean_t added_link;
+    gf_boolean_t is_linkfile;
+    gf_boolean_t quota_deem_statfs;
+    gf_boolean_t heal_layout;
+    gf_boolean_t locked;
+    gf_boolean_t dont_create_linkto;
+    gf_boolean_t gfid_missing;
 };
 typedef struct dht_local dht_local_t;
 
 /* du - disk-usage */
 struct dht_du {
-        double   avail_percent;
-	double   avail_inodes;
-        uint64_t avail_space;
-        uint32_t log;
+    double avail_percent;
+    double avail_inodes;
+    uint64_t avail_space;
+    uint32_t log;
+    uint32_t chunks;
+    uint32_t total_blocks;
+    uint32_t avail_blocks;
+    uint32_t frsize; /*fragment size*/
 };
 typedef struct dht_du dht_du_t;
 
 enum gf_defrag_type {
-        GF_DEFRAG_CMD_START = 1,
-        GF_DEFRAG_CMD_STOP = 1 + 1,
-        GF_DEFRAG_CMD_STATUS = 1 + 2,
-        GF_DEFRAG_CMD_START_LAYOUT_FIX = 1 + 3,
-        GF_DEFRAG_CMD_START_FORCE = 1 + 4,
+    GF_DEFRAG_CMD_NONE = 0,
+    GF_DEFRAG_CMD_START = 1,
+    GF_DEFRAG_CMD_STOP = 1 + 1,
+    GF_DEFRAG_CMD_STATUS = 1 + 2,
+    GF_DEFRAG_CMD_START_LAYOUT_FIX = 1 + 3,
+    GF_DEFRAG_CMD_START_FORCE = 1 + 4,
+    GF_DEFRAG_CMD_DETACH_STATUS = 1 + 11,
+    GF_DEFRAG_CMD_DETACH_START = 1 + 13,
+    GF_DEFRAG_CMD_DETACH_COMMIT = 1 + 14,
+    GF_DEFRAG_CMD_DETACH_COMMIT_FORCE = 1 + 15,
+    GF_DEFRAG_CMD_DETACH_STOP = 1 + 16,
+    /* new labels are used so it will help
+     * while removing old labels by easily differentiating.
+     * A few labels are added so that the count remains same
+     * between this enum and the ones on the xdr file.
+     * different values for the same enum cause errors and
+     * confusion.
+     */
 };
 typedef enum gf_defrag_type gf_defrag_type;
 
 enum gf_defrag_status_t {
-        GF_DEFRAG_STATUS_NOT_STARTED,
-        GF_DEFRAG_STATUS_STARTED,
-        GF_DEFRAG_STATUS_STOPPED,
-        GF_DEFRAG_STATUS_COMPLETE,
-        GF_DEFRAG_STATUS_FAILED,
-        GF_DEFRAG_STATUS_LAYOUT_FIX_STARTED,
-        GF_DEFRAG_STATUS_LAYOUT_FIX_STOPPED,
-        GF_DEFRAG_STATUS_LAYOUT_FIX_COMPLETE,
-        GF_DEFRAG_STATUS_LAYOUT_FIX_FAILED,
+    GF_DEFRAG_STATUS_NOT_STARTED,
+    GF_DEFRAG_STATUS_STARTED,
+    GF_DEFRAG_STATUS_STOPPED,
+    GF_DEFRAG_STATUS_COMPLETE,
+    GF_DEFRAG_STATUS_FAILED,
+    GF_DEFRAG_STATUS_LAYOUT_FIX_STARTED,
+    GF_DEFRAG_STATUS_LAYOUT_FIX_STOPPED,
+    GF_DEFRAG_STATUS_LAYOUT_FIX_COMPLETE,
+    GF_DEFRAG_STATUS_LAYOUT_FIX_FAILED,
 };
 typedef enum gf_defrag_status_t gf_defrag_status_t;
 
 typedef struct gf_defrag_pattern_list gf_defrag_pattern_list_t;
 
 struct gf_defrag_pattern_list {
-        char                       path_pattern[256];
-        uint64_t                   size;
-        gf_defrag_pattern_list_t  *next;
+    char path_pattern[256];
+    uint64_t size;
+    gf_defrag_pattern_list_t *next;
 };
 
+struct dht_container {
+    union {
+        struct list_head list;
+        struct {
+            struct _gf_dirent_t *next;
+            struct _gf_dirent_t *prev;
+        };
+    };
+    gf_dirent_t *df_entry;
+    xlator_t *this;
+    loc_t *parent_loc;
+    dict_t *migrate_data;
+    int local_subvol_index;
+};
+
+typedef struct nodeuuid_info {
+    char info;   /* Set to 1 is this is my node's uuid*/
+    uuid_t uuid; /* Store the nodeuuid as well for debugging*/
+} nodeuuid_info_t;
+
+typedef struct subvol_nodeuuids_info {
+    nodeuuid_info_t *elements;
+    int count;
+} subvol_nodeuuids_info_t;
+
 struct gf_defrag_info_ {
-        uint64_t                     total_files;
-        uint64_t                     total_data;
-        uint64_t                     num_files_lookedup;
-        uint64_t                     total_failures;
-        uint64_t                     skipped;
-        gf_lock_t                    lock;
-        int                          cmd;
-        pthread_t                    th;
-        gf_defrag_status_t           defrag_status;
-        struct rpc_clnt             *rpc;
-        uint32_t                     connected;
-        uint32_t                     is_exiting;
-        pid_t                        pid;
-        inode_t                     *root_inode;
-        uuid_t                       node_uuid;
-        struct timeval               start_time;
-        gf_boolean_t                 stats;
-        gf_defrag_pattern_list_t    *defrag_pattern;
+    uint64_t total_files;
+    uint64_t total_data;
+    uint64_t num_files_lookedup;
+    uint64_t total_failures;
+    uint64_t skipped;
+    uint64_t num_dirs_processed;
+    uint64_t size_processed;
+    gf_lock_t lock;
+    pthread_t th;
+    struct rpc_clnt *rpc;
+    uint32_t connected;
+    uint32_t is_exiting;
+    pid_t pid;
+    int cmd;
+    inode_t *root_inode;
+    uuid_t node_uuid;
+    time_t start_time;
+    uint32_t new_commit_hash;
+    gf_defrag_status_t defrag_status;
+    gf_defrag_pattern_list_t *defrag_pattern;
+
+    pthread_cond_t parallel_migration_cond;
+    pthread_mutex_t dfq_mutex;
+    pthread_cond_t rebalance_crawler_alarm;
+    int32_t q_entry_count;
+    int32_t global_error;
+    struct dht_container *queue;
+    int32_t crawl_done;
+    int32_t abort;
+    int32_t wakeup_crawler;
+
+    /*Throttle params*/
+    /*stands for reconfigured thread count*/
+    int32_t recon_thread_count;
+    pthread_cond_t df_wakeup_thread;
+
+    /* backpointer to make it easier to write functions for rebalance */
+    xlator_t *this;
+
+    pthread_cond_t fc_wakeup_cond;
+    pthread_mutex_t fc_mutex;
+
+    /*stands for current running thread count*/
+    int32_t current_thread_count;
+
+    gf_boolean_t stats;
+    /* lock migration flag */
+    gf_boolean_t lock_migration_enabled;
 };
 
 typedef struct gf_defrag_info_ gf_defrag_info_t;
 
+struct dht_methods_s {
+    int32_t (*migration_get_dst_subvol)(xlator_t *this, dht_local_t *local);
+    int32_t (*migration_other)(xlator_t *this, gf_defrag_info_t *defrag);
+    xlator_t *(*layout_search)(xlator_t *this, dht_layout_t *layout,
+                               const char *name);
+};
+
+typedef struct dht_methods_s dht_methods_t;
+
 struct dht_conf {
-        gf_lock_t      subvolume_lock;
-        int            subvolume_cnt;
-        xlator_t     **subvolumes;
-        char          *subvolume_status;
-        int           *last_event;
-        dht_layout_t **file_layouts;
-        dht_layout_t **dir_layouts;
-        gf_boolean_t   search_unhashed;
-        int            gen;
-        dht_du_t      *du_stats;
-        double         min_free_disk;
-	double         min_free_inodes;
-        char           disk_unit;
-        int32_t        refresh_interval;
-        gf_boolean_t   unhashed_sticky_bit;
-        struct timeval last_stat_fetch;
-        gf_lock_t      layout_lock;
-        void          *private;     /* Can be used by wrapper xlators over
-                                       dht */
-        gf_boolean_t   use_readdirp;
-        char           vol_uuid[UUID_SIZE + 1];
-        gf_boolean_t   assert_no_child_down;
-        time_t        *subvol_up_time;
-
-        /* This is the count used as the distribute layout for a directory */
-        /* Will be a global flag to control the layout spread count */
-        uint32_t       dir_spread_cnt;
-
-        /* to keep track of nodes which are decomissioned */
-        xlator_t     **decommissioned_bricks;
-        int            decommission_in_progress;
-        int            decommission_subvols_cnt;
-
-        /* defrag related */
-        gf_defrag_info_t *defrag;
-
-        /* Request to filter directory entries in readdir request */
-
-        gf_boolean_t    readdir_optimize;
-
-        /* Support regex-based name reinterpretation. */
-        regex_t         rsync_regex;
-        gf_boolean_t    rsync_regex_valid;
-        regex_t         extra_regex;
-        gf_boolean_t    extra_regex_valid;
-
-        /* Support variable xattr names. */
-        char            *xattr_name;
-        char            *link_xattr_name;
-        char            *wild_xattr_name;
+    xlator_t **subvolumes;
+    char *subvolume_status;
+    int *last_event;
+    dht_layout_t **file_layouts;
+    dht_layout_t **dir_layouts;
+    unsigned int search_unhashed;
+    int gen;
+    dht_du_t *du_stats;
+    double min_free_disk;
+    double min_free_inodes;
+    int subvolume_cnt;
+    int32_t refresh_interval;
+    gf_lock_t subvolume_lock;
+    time_t last_stat_fetch;
+    gf_lock_t layout_lock;
+    dict_t *leaf_to_subvol;
+    void *private; /* Can be used by wrapper xlators over
+                      dht */
+    time_t *subvol_up_time;
+
+    /* to keep track of nodes which are decommissioned */
+    xlator_t **decommissioned_bricks;
+    int decommission_in_progress;
+    int decommission_subvols_cnt;
+
+    /* defrag related */
+    gf_defrag_info_t *defrag;
+
+    /* Support regex-based name reinterpretation. */
+    regex_t rsync_regex;
+    regex_t extra_regex;
+
+    /* Support variable xattr names. */
+    char *xattr_name;
+    char *mds_xattr_key;
+    char *link_xattr_name;
+    char *commithash_xattr_name;
+    char *wild_xattr_name;
+
+    dht_methods_t methods;
+
+    struct mem_pool *lock_pool;
+
+    /*local subvol storage for rebalance*/
+    xlator_t **local_subvols;
+    subvol_nodeuuids_info_t *local_nodeuuids;
+    int32_t local_subvols_cnt;
+
+    int dthrottle;
+
+    /* Hard link handle requirement for migration triggered from client*/
+    synclock_t link_lock;
+
+    /* lock migration */
+    gf_lock_t lock;
+
+    /* This is the count used as the distribute layout for a directory */
+    /* Will be a global flag to control the layout spread count */
+    uint32_t dir_spread_cnt;
+
+    /*
+     * "Commit hash" for this volume topology.  Changed whenever bricks
+     * are added or removed.
+     */
+    uint32_t vol_commit_hash;
+
+    char vol_uuid[UUID_SIZE + 1];
+
+    char disk_unit;
+
+    gf_boolean_t lock_migration_enabled;
+
+    gf_boolean_t vch_forced;
+
+    gf_boolean_t use_fallocate;
+
+    gf_boolean_t force_migration;
+
+    gf_boolean_t lookup_optimize;
+
+    gf_boolean_t unhashed_sticky_bit;
+
+    gf_boolean_t assert_no_child_down;
+
+    gf_boolean_t use_readdirp;
+
+    /* Request to filter directory entries in readdir request */
+    gf_boolean_t readdir_optimize;
+
+    gf_boolean_t rsync_regex_valid;
+
+    gf_boolean_t extra_regex_valid;
+
+    /* Support size-weighted rebalancing (heterogeneous bricks). */
+    gf_boolean_t do_weighting;
+
+    gf_boolean_t randomize_by_gfid;
 };
 typedef struct dht_conf dht_conf_t;
 
+struct dht_dfoffset_ctx {
+    xlator_t *this;
+    off_t offset;
+    int32_t readdir_done;
+};
+typedef struct dht_dfoffset_ctx dht_dfoffset_ctx_t;
 
 struct dht_disk_layout {
-        uint32_t           cnt;
-        uint32_t           type;
-        struct {
-                uint32_t   start;
-                uint32_t   stop;
-        } list[1];
+    uint32_t cnt;
+    uint32_t type;
+    struct {
+        uint32_t start;
+        uint32_t stop;
+    } list[1];
 };
 typedef struct dht_disk_layout dht_disk_layout_t;
 
 typedef enum {
-        GF_DHT_MIGRATE_DATA,
-        GF_DHT_MIGRATE_DATA_EVEN_IF_LINK_EXISTS,
-        GF_DHT_MIGRATE_HARDLINK,
-        GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS
+    GF_DHT_MIGRATE_DATA,
+    GF_DHT_MIGRATE_DATA_EVEN_IF_LINK_EXISTS,
+    GF_DHT_MIGRATE_HARDLINK,
+    GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS
 } gf_dht_migrate_data_type_t;
 
+typedef enum {
+    GF_DHT_EQUAL_DISTRIBUTION,
+    GF_DHT_WEIGHTED_DISTRIBUTION
+} dht_distribution_type_t;
+
+struct dir_dfmeta {
+    gf_dirent_t *equeue;
+    dht_dfoffset_ctx_t *offset_var;
+    struct list_head **head;
+    struct list_head **iterator;
+    int *fetch_entries;
+    /* fds corresponding to local subvols only */
+    fd_t **lfd;
+};
+
+typedef struct dht_migrate_info {
+    xlator_t *src_subvol;
+    xlator_t *dst_subvol;
+    GF_REF_DECL;
+} dht_migrate_info_t;
+
+typedef struct dht_fd_ctx {
+    uint64_t opened_on_dst;
+    GF_REF_DECL;
+} dht_fd_ctx_t;
+
 #define ENTRY_MISSING(op_ret, op_errno) (op_ret == -1 && op_errno == ENOENT)
 
-#define is_revalidate(loc) (dht_inode_ctx_layout_get (loc->inode, this, NULL) == 0)
+#define is_revalidate(loc)                                                     \
+    (dht_inode_ctx_layout_get((loc)->inode, this, NULL) == 0)
 
 #define is_last_call(cnt) (cnt == 0)
 
 #define DHT_MIGRATION_IN_PROGRESS 1
-#define DHT_MIGRATION_COMPLETED   2
+#define DHT_MIGRATION_COMPLETED 2
 
-#define DHT_LINKFILE_MODE        (S_ISVTX)
+#define check_is_linkfile(i, s, x, n)                                          \
+    (IS_DHT_LINKFILE_MODE(s) && dict_get(x, n))
 
-#define check_is_linkfile(i,s,x,n) (                                      \
-                ((st_mode_from_ia ((s)->ia_prot, (s)->ia_type) & ~S_IFMT) \
-                 == DHT_LINKFILE_MODE) &&                                 \
-                dict_get (x, n))
+#define IS_DHT_MIGRATION_PHASE2(buf)                                           \
+    (IA_ISREG((buf)->ia_type) &&                                               \
+     ((st_mode_from_ia((buf)->ia_prot, (buf)->ia_type) & ~S_IFMT) ==           \
+      DHT_LINKFILE_MODE))
 
-#define IS_DHT_MIGRATION_PHASE2(buf)  (                                 \
-                IA_ISREG ((buf)->ia_type) &&                            \
-                ((st_mode_from_ia ((buf)->ia_prot, (buf)->ia_type) &    \
-                  ~S_IFMT) == DHT_LINKFILE_MODE))
+#define IS_DHT_MIGRATION_PHASE1(buf)                                           \
+    (IA_ISREG((buf)->ia_type) && ((buf)->ia_prot.sticky == 1) &&               \
+     ((buf)->ia_prot.sgid == 1))
 
-#define IS_DHT_MIGRATION_PHASE1(buf)  (                                 \
-                IA_ISREG ((buf)->ia_type) &&                            \
-                ((buf)->ia_prot.sticky == 1) &&                         \
-                ((buf)->ia_prot.sgid == 1))
+#define DHT_STRIP_PHASE1_FLAGS(buf)                                            \
+    do {                                                                       \
+        if ((buf) && IS_DHT_MIGRATION_PHASE1(buf)) {                           \
+            (buf)->ia_prot.sticky = 0;                                         \
+            (buf)->ia_prot.sgid = 0;                                           \
+        }                                                                      \
+    } while (0)
 
-#define DHT_STRIP_PHASE1_FLAGS(buf)  do {                       \
-                if ((buf) && IS_DHT_MIGRATION_PHASE1(buf)) {    \
-                        (buf)->ia_prot.sticky = 0;              \
-                        (buf)->ia_prot.sgid = 0;                \
-                }                                               \
-        } while (0)
+#define dht_inode_missing(op_errno) (op_errno == ENOENT || op_errno == ESTALE)
 
-#define check_is_dir(i,s,x) (IA_ISDIR(s->ia_type))
+#define check_is_dir(i, s, x) (IA_ISDIR(s->ia_type))
 
 #define layout_is_sane(layout) ((layout) && (layout->cnt > 0))
 
-#define DHT_STACK_UNWIND(fop, frame, params ...) do {           \
-                dht_local_t *__local = NULL;                    \
-                xlator_t    *__xl    = NULL;                    \
-                if (frame) {                                    \
-                        __xl         = frame->this;             \
-                        __local      = frame->local;            \
-                        frame->local = NULL;                    \
-                }                                               \
-                STACK_UNWIND_STRICT (fop, frame, params);       \
-                dht_local_wipe (__xl, __local);                 \
-        } while (0)
-
-#define DHT_STACK_DESTROY(frame) do {                   \
-                dht_local_t *__local = NULL;            \
-                xlator_t    *__xl    = NULL;            \
-                __xl                 = frame->this;     \
-                __local              = frame->local;    \
-                frame->local         = NULL;            \
-                STACK_DESTROY (frame->root);            \
-                dht_local_wipe (__xl, __local);         \
-        } while (0)
-
-#define DHT_UPDATE_TIME(ctx_sec, ctx_nsec, new_sec, new_nsec, inode, post) do {\
-                int32_t sec = 0;                                        \
-                sec = new_sec;                                          \
-                LOCK (&inode->lock);                                    \
-                {                                                       \
-                        new_sec = max(new_sec, ctx_sec);                \
-                        if (sec < new_sec)                              \
-                                new_nsec = ctx_nsec;                    \
-                        if (sec == new_sec)                             \
-                                new_nsec = max (new_nsec, ctx_nsec);    \
-                        if (post) {                                     \
-                                ctx_sec = new_sec;                      \
-                                ctx_nsec = new_nsec;                    \
-                        }                                               \
-                }                                                       \
-                UNLOCK (&inode->lock);                                  \
-        } while (0)
-
-#define is_greater_time(a, an, b, bn) (((a) < (b)) || (((a) == (b)) && ((an) < (bn))))
-dht_layout_t                            *dht_layout_new (xlator_t *this, int cnt);
-dht_layout_t                            *dht_layout_get (xlator_t *this, inode_t *inode);
-dht_layout_t                            *dht_layout_for_subvol (xlator_t *this, xlator_t *subvol);
-xlator_t *dht_layout_search (xlator_t   *this, dht_layout_t *layout,
-                             const char *name);
-int                                      dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout);
-int dht_layout_anomalies (xlator_t      *this, loc_t *loc, dht_layout_t *layout,
-                          uint32_t      *holes_p, uint32_t *overlaps_p,
-                          uint32_t      *missing_p, uint32_t *down_p,
-                          uint32_t      *misc_p, uint32_t *no_space_p);
-int dht_layout_dir_mismatch (xlator_t   *this, dht_layout_t *layout,
-                             xlator_t   *subvol, loc_t *loc, dict_t *xattr);
-
-xlator_t *dht_linkfile_subvol (xlator_t *this, inode_t *inode,
-                               struct iatt *buf, dict_t *xattr);
-int dht_linkfile_unlink (call_frame_t      *frame, xlator_t *this,
-                         xlator_t          *subvol, loc_t *loc);
-
-int dht_layouts_init (xlator_t *this, dht_conf_t *conf);
-int dht_layout_merge (xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
-                      int       op_ret, int op_errno, dict_t *xattr);
-
-int dht_disk_layout_extract (xlator_t *this, dht_layout_t *layout,
-                             int       pos, int32_t **disk_layout_p);
-int dht_disk_layout_merge (xlator_t   *this, dht_layout_t *layout,
-                           int         pos, void *disk_layout_raw, int disk_layout_len);
-
-
-int dht_frame_return (call_frame_t *frame);
-
-int                             dht_itransform (xlator_t *this, xlator_t *subvol, uint64_t x, uint64_t *y);
-int dht_deitransform (xlator_t *this, uint64_t y, xlator_t **subvol,
-                      uint64_t *x);
-
-void dht_local_wipe (xlator_t *this, dht_local_t *local);
-dht_local_t *dht_local_init (call_frame_t    *frame, loc_t *loc, fd_t *fd,
-                             glusterfs_fop_t  fop);
-int dht_iatt_merge (xlator_t                 *this, struct iatt *to, struct iatt *from,
-                    xlator_t                 *subvol);
-
-xlator_t *dht_subvol_get_hashed (xlator_t *this, loc_t *loc);
-xlator_t *dht_subvol_get_cached (xlator_t *this, inode_t *inode);
-xlator_t *dht_subvol_next (xlator_t *this, xlator_t *prev);
-xlator_t *dht_subvol_next_available (xlator_t *this, xlator_t *prev);
-int       dht_subvol_cnt (xlator_t *this, xlator_t *subvol);
-
-int dht_hash_compute (xlator_t *this, int type, const char *name, uint32_t *hash_p);
-
-int dht_linkfile_create (call_frame_t    *frame, fop_mknod_cbk_t linkfile_cbk,
-                         xlator_t        *this, xlator_t *tovol,
-                         xlator_t        *fromvol, loc_t *loc);
-int                                       dht_lookup_directory (call_frame_t *frame, xlator_t *this, loc_t *loc);
-int                                       dht_lookup_everywhere (call_frame_t *frame, xlator_t *this, loc_t *loc);
-int
-dht_selfheal_directory (call_frame_t     *frame, dht_selfheal_dir_cbk_t cbk,
-                        loc_t            *loc, dht_layout_t *layout);
-int
-dht_selfheal_new_directory (call_frame_t *frame, dht_selfheal_dir_cbk_t cbk,
-                            dht_layout_t *layout);
-int
-dht_selfheal_restore (call_frame_t       *frame, dht_selfheal_dir_cbk_t cbk,
-                      loc_t              *loc, dht_layout_t *layout);
-int
-dht_layout_sort_volname (dht_layout_t *layout);
-
-int dht_get_du_info (call_frame_t *frame, xlator_t *this, loc_t *loc);
-
-gf_boolean_t dht_is_subvol_filled (xlator_t *this, xlator_t *subvol);
-xlator_t *dht_free_disk_available_subvol (xlator_t *this, xlator_t *subvol,
-                                          dht_local_t *layout);
-int       dht_get_du_info_for_subvol (xlator_t *this, int subvol_idx);
-
-int dht_layout_preset (xlator_t *this, xlator_t *subvol, inode_t *inode);
-int           dht_layout_set (xlator_t *this, inode_t *inode, dht_layout_t *layout);;
-void          dht_layout_unref (xlator_t *this, dht_layout_t *layout);
-dht_layout_t *dht_layout_ref (xlator_t *this, dht_layout_t *layout);
-xlator_t     *dht_first_up_subvol (xlator_t *this);
-xlator_t     *dht_last_up_subvol (xlator_t *this);
-
-int dht_build_child_loc (xlator_t *this, loc_t *child, loc_t *parent, char *name);
-
-int dht_filter_loc_subvol_key (xlator_t  *this, loc_t *loc, loc_t *new_loc,
-                               xlator_t **subvol);
-
-int                                     dht_rename_cleanup (call_frame_t *frame);
-int dht_rename_links_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                          int32_t           op_ret, int32_t op_errno,
-                          inode_t          *inode, struct iatt *stbuf,
-                          struct iatt      *preparent, struct iatt *postparent,
-                          dict_t *xdata);
-
-int dht_fix_directory_layout (call_frame_t *frame,
-                              dht_selfheal_dir_cbk_t  dir_cbk,
-                              dht_layout_t           *layout);
-
-int dht_init_subvolumes (xlator_t *this, dht_conf_t *conf);
+#define we_are_not_migrating(x) ((x) == 1)
+
+#define DHT_STACK_UNWIND(fop, frame, params...)                                \
+    do {                                                                       \
+        dht_local_t *__local = NULL;                                           \
+        xlator_t *__xl = NULL;                                                 \
+        if (frame) {                                                           \
+            __xl = frame->this;                                                \
+            __local = frame->local;                                            \
+            frame->local = NULL;                                               \
+        }                                                                      \
+        STACK_UNWIND_STRICT(fop, frame, params);                               \
+        dht_local_wipe(__xl, __local);                                         \
+    } while (0)
+
+#define DHT_STACK_DESTROY(frame)                                               \
+    do {                                                                       \
+        dht_local_t *__local = NULL;                                           \
+        xlator_t *__xl = NULL;                                                 \
+        __xl = frame->this;                                                    \
+        __local = frame->local;                                                \
+        frame->local = NULL;                                                   \
+        STACK_DESTROY(frame->root);                                            \
+        dht_local_wipe(__xl, __local);                                         \
+    } while (0)
+
+#define DHT_UPDATE_TIME(ctx_sec, ctx_nsec, new_sec, new_nsec, post)            \
+    do {                                                                       \
+        if (ctx_sec == new_sec)                                                \
+            new_nsec = max(new_nsec, ctx_nsec);                                \
+        else if (ctx_sec > new_sec) {                                          \
+            new_sec = ctx_sec;                                                 \
+            new_nsec = ctx_nsec;                                               \
+        }                                                                      \
+        if (post) {                                                            \
+            ctx_sec = new_sec;                                                 \
+            ctx_nsec = new_nsec;                                               \
+        }                                                                      \
+    } while (0)
+
+#define is_greater_time(a, an, b, bn)                                          \
+    (((a) < (b)) || (((a) == (b)) && ((an) < (bn))))
+
+#define DHT_MARK_FOP_INTERNAL(xattr)                                           \
+    do {                                                                       \
+        int tmp = -1;                                                          \
+        if (!xattr) {                                                          \
+            xattr = dict_new();                                                \
+            if (!xattr)                                                        \
+                break;                                                         \
+        }                                                                      \
+        tmp = dict_set_str(xattr, GLUSTERFS_INTERNAL_FOP_KEY, "yes");          \
+        if (tmp) {                                                             \
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,       \
+                   "Failed to set dictionary value: key = %s,"                 \
+                   " path = %s",                                               \
+                   GLUSTERFS_INTERNAL_FOP_KEY, local->loc.path);               \
+        }                                                                      \
+    } while (0)
+
+dht_layout_t *
+dht_layout_new(xlator_t *this, int cnt);
+dht_layout_t *
+dht_layout_get(xlator_t *this, inode_t *inode);
+dht_layout_t *
+dht_layout_for_subvol(xlator_t *this, xlator_t *subvol);
+xlator_t *
+dht_layout_search(xlator_t *this, dht_layout_t *layout, const char *name);
+int32_t
+dht_migration_get_dst_subvol(xlator_t *this, dht_local_t *local);
+int32_t
+dht_migration_needed(xlator_t *this);
+int
+dht_layout_normalize(xlator_t *this, loc_t *loc, dht_layout_t *layout);
+void
+dht_layout_anomalies(xlator_t *this, loc_t *loc, dht_layout_t *layout,
+                     uint32_t *holes_p, uint32_t *overlaps_p,
+                     uint32_t *missing_p, uint32_t *down_p, uint32_t *misc_p,
+                     uint32_t *no_space_p);
+int
+dht_layout_dir_mismatch(xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
+                        loc_t *loc, dict_t *xattr);
+xlator_t *
+dht_linkfile_subvol(xlator_t *this, inode_t *inode, struct iatt *buf,
+                    dict_t *xattr);
+int
+dht_linkfile_unlink(call_frame_t *frame, xlator_t *this, xlator_t *subvol,
+                    loc_t *loc);
+
+int
+dht_layouts_init(xlator_t *this, dht_conf_t *conf);
+int
+dht_layout_merge(xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
+                 int op_ret, int op_errno, dict_t *xattr);
+
+int
+dht_disk_layout_extract(xlator_t *this, dht_layout_t *layout, int pos,
+                        int32_t **disk_layout_p);
+int
+dht_disk_layout_extract_for_subvol(xlator_t *this, dht_layout_t *layout,
+                                   xlator_t *subvol, int32_t **disk_layout_p);
+
+int
+dht_frame_return(call_frame_t *frame);
+
+int
+dht_deitransform(xlator_t *this, uint64_t y, xlator_t **subvol);
+
+void
+dht_local_wipe(xlator_t *this, dht_local_t *local);
+dht_local_t *
+dht_local_init(call_frame_t *frame, loc_t *loc, fd_t *fd, glusterfs_fop_t fop);
+int
+dht_iatt_merge(xlator_t *this, struct iatt *to, struct iatt *from);
+
+xlator_t *
+dht_subvol_get_hashed(xlator_t *this, loc_t *loc);
+xlator_t *
+dht_subvol_get_cached(xlator_t *this, inode_t *inode);
+xlator_t *
+dht_subvol_next(xlator_t *this, xlator_t *prev);
+xlator_t *
+dht_subvol_next_available(xlator_t *this, xlator_t *prev);
+int
+dht_subvol_cnt(xlator_t *this, xlator_t *subvol);
+
+int
+dht_hash_compute(xlator_t *this, int type, const char *name, uint32_t *hash_p);
+
+int
+dht_linkfile_create(call_frame_t *frame, fop_mknod_cbk_t linkfile_cbk,
+                    xlator_t *this, xlator_t *tovol, xlator_t *fromvol,
+                    loc_t *loc);
+int
+dht_lookup_everywhere(call_frame_t *frame, xlator_t *this, loc_t *loc);
+int
+dht_selfheal_directory(call_frame_t *frame, dht_selfheal_dir_cbk_t cbk,
+                       loc_t *loc, dht_layout_t *layout);
+int
+dht_selfheal_new_directory(call_frame_t *frame, dht_selfheal_dir_cbk_t cbk,
+                           dht_layout_t *layout);
+int
+dht_selfheal_restore(call_frame_t *frame, dht_selfheal_dir_cbk_t cbk,
+                     loc_t *loc, dht_layout_t *layout);
+void
+dht_layout_sort_volname(dht_layout_t *layout);
+
+int
+dht_get_du_info(call_frame_t *frame, xlator_t *this, loc_t *loc);
+
+gf_boolean_t
+dht_is_subvol_filled(xlator_t *this, xlator_t *subvol);
+xlator_t *
+dht_free_disk_available_subvol(xlator_t *this, xlator_t *subvol,
+                               dht_local_t *layout);
+int
+dht_get_du_info_for_subvol(xlator_t *this, int subvol_idx);
+
+int
+dht_layout_preset(xlator_t *this, xlator_t *subvol, inode_t *inode);
+int
+dht_layout_set(xlator_t *this, inode_t *inode, dht_layout_t *layout);
+;
+void
+dht_layout_unref(xlator_t *this, dht_layout_t *layout);
+dht_layout_t *
+dht_layout_ref(xlator_t *this, dht_layout_t *layout);
+int
+dht_layout_index_for_subvol(dht_layout_t *layout, xlator_t *subvol);
+xlator_t *
+dht_first_up_subvol(xlator_t *this);
+xlator_t *
+dht_last_up_subvol(xlator_t *this);
+
+int
+dht_build_child_loc(xlator_t *this, loc_t *child, loc_t *parent, char *name);
+
+int
+dht_filter_loc_subvol_key(xlator_t *this, loc_t *loc, loc_t *new_loc,
+                          xlator_t **subvol);
+
+int
+dht_rename_cleanup(call_frame_t *frame);
+int
+dht_rename_link_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, inode_t *inode,
+                    struct iatt *stbuf, struct iatt *preparent,
+                    struct iatt *postparent, dict_t *xdata);
+
+int
+dht_update_commit_hash_for_layout(call_frame_t *frame);
+int
+dht_fix_directory_layout(call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk,
+                         dht_layout_t *layout);
+
+int
+dht_init_subvolumes(xlator_t *this, dht_conf_t *conf);
 
 /* migration/rebalance */
-int dht_start_rebalance_task (xlator_t *this, call_frame_t *frame);
+int
+dht_start_rebalance_task(xlator_t *this, call_frame_t *frame);
 
-int dht_rebalance_in_progress_check (xlator_t *this, call_frame_t *frame);
-int dht_rebalance_complete_check (xlator_t *this, call_frame_t *frame);
+int
+dht_rebalance_in_progress_check(xlator_t *this, call_frame_t *frame);
+int
+dht_rebalance_complete_check(xlator_t *this, call_frame_t *frame);
 
+int
+dht_init_local_subvolumes(xlator_t *this, dht_conf_t *conf);
 
 /* FOPS */
-int32_t dht_lookup (call_frame_t *frame,
-                    xlator_t *this,
-                    loc_t    *loc,
-                    dict_t   *xattr_req);
-
-int32_t dht_stat (call_frame_t *frame,
-                  xlator_t *this,
-                  loc_t    *loc, dict_t *xdata);
-
-int32_t dht_fstat (call_frame_t *frame,
-                   xlator_t *this,
-                   fd_t     *fd, dict_t *xdata);
-
-int32_t dht_truncate (call_frame_t *frame,
-                      xlator_t *this,
-                      loc_t    *loc,
-                      off_t     offset, dict_t *xdata);
-
-int32_t dht_ftruncate (call_frame_t *frame,
-                       xlator_t *this,
-                       fd_t     *fd,
-                       off_t     offset, dict_t *xdata);
-
-int32_t dht_access (call_frame_t *frame,
-                    xlator_t *this,
-                    loc_t    *loc,
-                    int32_t   mask, dict_t *xdata);
-
-int32_t dht_readlink (call_frame_t *frame,
-                      xlator_t *this,
-                      loc_t    *loc,
-                      size_t    size, dict_t *xdata);
-
-int32_t dht_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                   mode_t mode, dev_t rdev, mode_t umask, dict_t *xdata);
-
-int32_t dht_mkdir (call_frame_t *frame, xlator_t *this,
-                   loc_t *loc, mode_t mode, mode_t umask, dict_t *xdata);
-
-int32_t dht_unlink (call_frame_t *frame,
-                    xlator_t *this,
-                    loc_t    *loc, int xflag, dict_t *xdata);
-
-int32_t dht_rmdir (call_frame_t *frame, xlator_t *this,
-                   loc_t    *loc, int flags, dict_t *xdata);
-
-int32_t dht_symlink (call_frame_t   *frame, xlator_t *this,
-                     const char *linkpath, loc_t *loc, mode_t umask,
-                     dict_t *xdata);
-
-int32_t dht_rename (call_frame_t *frame,
-                    xlator_t *this,
-                    loc_t    *oldloc,
-                    loc_t    *newloc, dict_t *xdata);
-
-int32_t dht_link (call_frame_t *frame,
-                  xlator_t *this,
-                  loc_t    *oldloc,
-                  loc_t    *newloc, dict_t *xdata);
-
-int32_t dht_create (call_frame_t *frame, xlator_t *this,
-                    loc_t    *loc, int32_t flags, mode_t mode,
-                    mode_t umask, fd_t     *fd, dict_t *params);
-
-int32_t dht_open (call_frame_t *frame,
-                  xlator_t *this,
-                  loc_t    *loc,
-                  int32_t   flags, fd_t *fd, dict_t *xdata);
-
-int32_t dht_readv (call_frame_t *frame,
-                   xlator_t *this,
-                   fd_t     *fd,
-                   size_t    size,
-                   off_t     offset, uint32_t flags, dict_t *xdata);
-
-int32_t dht_writev (call_frame_t      *frame,
-                    xlator_t      *this,
-                    fd_t          *fd,
-                    struct iovec  *vector,
-                    int32_t        count,
-                    off_t          offset,
-                    uint32_t       flags,
-                    struct iobref *iobref, dict_t *xdata);
-
-int32_t dht_flush (call_frame_t *frame,
-                   xlator_t *this,
-                   fd_t     *fd, dict_t *xdata);
-
-int32_t dht_fsync (call_frame_t *frame,
-                   xlator_t *this,
-                   fd_t     *fd,
-                   int32_t   datasync, dict_t *xdata);
-
-int32_t dht_opendir (call_frame_t *frame,
-                     xlator_t *this,
-                     loc_t    *loc, fd_t *fd, dict_t *xdata);
-
-int32_t dht_fsyncdir (call_frame_t *frame,
-                      xlator_t *this,
-                      fd_t     *fd,
-                      int32_t   datasync, dict_t *xdata);
-
-int32_t dht_statfs (call_frame_t *frame,
-                    xlator_t *this,
-                    loc_t    *loc, dict_t *xdata);
-
-int32_t dht_setxattr (call_frame_t *frame,
-                      xlator_t *this,
-                      loc_t    *loc,
-                      dict_t   *dict,
-                      int32_t   flags, dict_t *xdata);
-
-int32_t dht_getxattr (call_frame_t   *frame,
-                      xlator_t   *this,
-                      loc_t      *loc,
-                      const char *name, dict_t *xdata);
-
-int32_t dht_fsetxattr (call_frame_t *frame,
-                       xlator_t *this,
-                       fd_t     *fd,
-                       dict_t   *dict,
-                       int32_t   flags, dict_t *xdata);
-
-int32_t dht_fgetxattr (call_frame_t   *frame,
-                       xlator_t   *this,
-                       fd_t       *fd,
-                       const char *name, dict_t *xdata);
-
-int32_t dht_removexattr (call_frame_t   *frame,
-                         xlator_t   *this,
-                         loc_t      *loc,
-                         const char *name, dict_t *xdata);
-int32_t dht_fremovexattr (call_frame_t   *frame,
-                          xlator_t   *this,
-                          fd_t      *fd,
-                          const char *name, dict_t *xdata);
-
-int32_t dht_lk (call_frame_t        *frame,
-                xlator_t        *this,
-                fd_t            *fd,
-                int32_t          cmd,
-                struct gf_flock *flock, dict_t *xdata);
-
-int32_t dht_inodelk (call_frame_t *frame, xlator_t *this,
-                     const char      *volume, loc_t *loc, int32_t cmd,
-                     struct gf_flock *flock, dict_t *xdata);
-
-int32_t dht_finodelk (call_frame_t        *frame, xlator_t *this,
-                      const char      *volume, fd_t *fd, int32_t cmd,
-                      struct gf_flock *flock, dict_t *xdata);
-
-int32_t dht_entrylk (call_frame_t    *frame, xlator_t *this,
-                     const char  *volume, loc_t *loc, const char *basename,
-                     entrylk_cmd  cmd, entrylk_type type, dict_t *xdata);
-
-int32_t dht_fentrylk (call_frame_t    *frame, xlator_t *this,
-                      const char  *volume, fd_t *fd, const char *basename,
-                      entrylk_cmd  cmd, entrylk_type type, dict_t *xdata);
-
-int32_t dht_readdir (call_frame_t  *frame,
-                     xlator_t *this,
-                     fd_t     *fd,
-                     size_t    size, off_t off, dict_t *xdata);
-
-int32_t dht_readdirp (call_frame_t *frame,
-                      xlator_t *this,
-                      fd_t     *fd,
-                      size_t    size, off_t off, dict_t *dict);
-
-int32_t dht_xattrop (call_frame_t           *frame,
-                     xlator_t           *this,
-                     loc_t              *loc,
-                     gf_xattrop_flags_t  flags,
-                     dict_t             *dict, dict_t *xdata);
-
-int32_t dht_fxattrop (call_frame_t *frame,
-                      xlator_t           *this,
-                      fd_t               *fd,
-                      gf_xattrop_flags_t  flags,
-                      dict_t             *dict, dict_t *xdata);
-
-int32_t dht_forget (xlator_t *this, inode_t *inode);
-int32_t dht_setattr (call_frame_t  *frame, xlator_t *this, loc_t *loc,
-                     struct iatt   *stbuf, int32_t valid, dict_t *xdata);
-int32_t dht_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                      struct iatt  *stbuf, int32_t valid, dict_t *xdata);
-int32_t dht_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd,
-		      int32_t mode, off_t offset, size_t len, dict_t *xdata);
-int32_t dht_discard(call_frame_t *frame, xlator_t *this, fd_t *fd,
-		    off_t offset, size_t len, dict_t *xdata);
-
-int32_t dht_init (xlator_t *this);
-void    dht_fini (xlator_t *this);
-int     dht_reconfigure (xlator_t *this, dict_t *options);
-int32_t dht_notify (xlator_t *this, int32_t event, void *data, ...);
+int32_t
+dht_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req);
+
+int32_t
+dht_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata);
+
+int32_t
+dht_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata);
+
+int32_t
+dht_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+             dict_t *xdata);
+
+int32_t
+dht_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+              dict_t *xdata);
+
+int32_t
+dht_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+           dict_t *xdata);
+
+int32_t
+dht_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+             dict_t *xdata);
+
+int32_t
+dht_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+          dev_t rdev, mode_t umask, dict_t *xdata);
+
+int32_t
+dht_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+          mode_t umask, dict_t *xdata);
+
+int32_t
+dht_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+           dict_t *xdata);
+
+int32_t
+dht_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+          dict_t *xdata);
+
+int32_t
+dht_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath,
+            loc_t *loc, mode_t umask, dict_t *xdata);
+
+int32_t
+dht_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+           dict_t *xdata);
+
+int32_t
+dht_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+         dict_t *xdata);
+
+int32_t
+dht_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+           mode_t mode, mode_t umask, fd_t *fd, dict_t *params);
+
+int32_t
+dht_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+         fd_t *fd, dict_t *xdata);
+
+int32_t
+dht_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+          off_t offset, uint32_t flags, dict_t *xdata);
+
+int32_t
+dht_writev(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,
+           int32_t count, off_t offset, uint32_t flags, struct iobref *iobref,
+           dict_t *xdata);
+
+int32_t
+dht_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata);
+
+int32_t
+dht_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+          dict_t *xdata);
+
+int32_t
+dht_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+            dict_t *xdata);
+
+int32_t
+dht_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+             dict_t *xdata);
+
+int32_t
+dht_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata);
+
+int32_t
+dht_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+             int32_t flags, dict_t *xdata);
+
+int32_t
+dht_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, const char *name,
+             dict_t *xdata);
+
+int32_t
+dht_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+              int32_t flags, dict_t *xdata);
+
+int32_t
+dht_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+              dict_t *xdata);
+
+int32_t
+dht_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                const char *name, dict_t *xdata);
+int32_t
+dht_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                 const char *name, dict_t *xdata);
+
+int32_t
+dht_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
+       struct gf_flock *flock, dict_t *xdata);
+
+int32_t
+dht_lease(call_frame_t *frame, xlator_t *this, loc_t *loc,
+          struct gf_lease *lease, dict_t *xdata);
+
+int32_t
+dht_inodelk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+            int32_t cmd, struct gf_flock *flock, dict_t *xdata);
+
+int32_t
+dht_finodelk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+             int32_t cmd, struct gf_flock *flock, dict_t *xdata);
+
+int32_t
+dht_entrylk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+            const char *basename, entrylk_cmd cmd, entrylk_type type,
+            dict_t *xdata);
+
+int32_t
+dht_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+             const char *basename, entrylk_cmd cmd, entrylk_type type,
+             dict_t *xdata);
+
+int32_t
+dht_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t off, dict_t *xdata);
+
+int32_t
+dht_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+             off_t off, dict_t *dict);
+
+int32_t
+dht_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+            gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata);
+
+int32_t
+dht_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+             gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata);
+
+int32_t
+dht_forget(xlator_t *this, inode_t *inode);
+int32_t
+dht_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *stbuf,
+            int32_t valid, dict_t *xdata);
+int32_t
+dht_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf,
+             int32_t valid, dict_t *xdata);
+int32_t
+dht_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+              off_t offset, size_t len, dict_t *xdata);
+int32_t
+dht_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+            size_t len, dict_t *xdata);
+int32_t
+dht_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             off_t len, dict_t *xdata);
+int32_t
+dht_ipc(call_frame_t *frame, xlator_t *this, int32_t op, dict_t *xdata);
+
+int
+dht_set_subvol_range(xlator_t *this);
+int32_t
+dht_init(xlator_t *this);
+void
+dht_fini(xlator_t *this);
+int
+dht_reconfigure(xlator_t *this, dict_t *options);
+int32_t
+dht_notify(xlator_t *this, int32_t event, void *data, ...);
 
 /* definitions for nufa/switch */
-int dht_revalidate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int op_ret, int op_errno, inode_t *inode,
-                        struct iatt *stbuf, dict_t *xattr,
-                        struct iatt *postparent);
-int dht_lookup_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+int
+dht_revalidate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int op_ret, int op_errno, inode_t *inode, struct iatt *stbuf,
+                   dict_t *xattr, struct iatt *postparent);
+int
+dht_lookup_dir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int op_ret, int op_errno, inode_t *inode, struct iatt *stbuf,
+                   dict_t *xattr, struct iatt *postparent);
+int
+dht_lookup_linkfile_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
                         int op_ret, int op_errno, inode_t *inode,
                         struct iatt *stbuf, dict_t *xattr,
                         struct iatt *postparent);
-int dht_lookup_linkfile_cbk (call_frame_t *frame, void *cookie,
-                             xlator_t *this, int op_ret, int op_errno,
-                             inode_t *inode, struct iatt *stbuf, dict_t *xattr,
-                             struct iatt *postparent);
-int dht_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int op_ret, int op_errno,
-                    inode_t *inode, struct iatt *stbuf, dict_t *xattr,
-                    struct iatt *postparent);
-int dht_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int op_ret, int op_errno,
-                    fd_t *fd, inode_t *inode, struct iatt *stbuf,
-                    struct iatt *preparent, struct iatt *postparent,
-                    dict_t *xdata);
-int dht_newfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int op_ret, int op_errno,
-                     inode_t *inode, struct iatt *stbuf, struct iatt *preparent,
-                     struct iatt *postparent, dict_t *xdata);
+int
+dht_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, inode_t *inode, struct iatt *stbuf, dict_t *xattr,
+               struct iatt *postparent);
+int
+dht_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, fd_t *fd, inode_t *inode, struct iatt *stbuf,
+               struct iatt *preparent, struct iatt *postparent, dict_t *xdata);
+int
+dht_newfile_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                int op_errno, inode_t *inode, struct iatt *stbuf,
+                struct iatt *preparent, struct iatt *postparent, dict_t *xdata);
 
 int
-gf_defrag_status_get (gf_defrag_info_t *defrag, dict_t *dict);
+dht_finodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata);
 
 int
-gf_defrag_stop (gf_defrag_info_t *defrag, dict_t *output);
+dht_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, dict_t *xattr, dict_t *xdata);
 
-void*
-gf_defrag_start (void *this);
+int
+dht_common_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *dict,
+                       dict_t *xdata);
+int
+gf_defrag_status_get(dht_conf_t *conf, dict_t *dict);
+
+int
+gf_defrag_stop(dht_conf_t *conf, gf_defrag_status_t status, dict_t *output);
+
+void *
+gf_defrag_start(void *this);
 
 int32_t
-gf_defrag_handle_hardlink (xlator_t *this, loc_t *loc, dict_t  *xattrs,
-                           struct iatt *stbuf);
+gf_defrag_handle_hardlink(xlator_t *this, loc_t *loc, int *fop_errno);
 int
-dht_migrate_file (xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to,
-                 int flag);
+dht_migrate_file(xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to,
+                 int flag, int *fop_errno);
 int
-dht_inode_ctx_layout_get (inode_t *inode, xlator_t *this,
-                          dht_layout_t **layout_int);
+dht_inode_ctx_layout_get(inode_t *inode, xlator_t *this,
+                         dht_layout_t **layout_int);
 int
-dht_inode_ctx_layout_set (inode_t *inode, xlator_t *this,
-                          dht_layout_t* layout_int);
+dht_inode_ctx_layout_set(inode_t *inode, xlator_t *this,
+                         dht_layout_t *layout_int);
 int
-dht_inode_ctx_time_update (inode_t *inode, xlator_t *this, struct iatt *stat,
-                           int32_t update_ctx);
+dht_inode_ctx_time_update(inode_t *inode, xlator_t *this, struct iatt *stat,
+                          int32_t update_ctx);
+void
+dht_inode_ctx_time_set(inode_t *inode, xlator_t *this, struct iatt *stat);
 
-int dht_inode_ctx_get (inode_t *inode, xlator_t *this, dht_inode_ctx_t **ctx);
-int dht_inode_ctx_set (inode_t *inode, xlator_t *this, dht_inode_ctx_t *ctx);
 int
-dht_dir_attr_heal (void *data);
+dht_inode_ctx_get(inode_t *inode, xlator_t *this, dht_inode_ctx_t **ctx);
 int
-dht_dir_attr_heal_done (int ret, call_frame_t *sync_frame, void *data);
+dht_inode_ctx_set(inode_t *inode, xlator_t *this, dht_inode_ctx_t *ctx);
 int
-dht_dir_has_layout (dict_t *xattr, char *name);
-gf_boolean_t
-dht_is_subvol_in_layout (dht_layout_t *layout, xlator_t *xlator);
+dht_dir_attr_heal(void *data);
+int
+dht_dir_attr_heal_done(int ret, call_frame_t *sync_frame, void *data);
+xlator_t *
+dht_subvol_with_free_space_inodes(xlator_t *this, xlator_t *subvol,
+                                  xlator_t *ignore, dht_layout_t *layout,
+                                  uint64_t filesize);
 xlator_t *
-dht_subvol_with_free_space_inodes (xlator_t *this, xlator_t *subvol,
-                                   dht_layout_t *layout);
+dht_subvol_maxspace_nonzeroinode(xlator_t *this, xlator_t *subvol,
+                                 dht_layout_t *layout);
+int
+dht_dir_has_layout(dict_t *xattr, char *name);
+int
+dht_linkfile_attr_heal(call_frame_t *frame, xlator_t *this);
+
+int32_t
+dht_priv_dump(xlator_t *this);
+int32_t
+dht_inodectx_dump(xlator_t *this, inode_t *inode);
+
+gf_boolean_t
+dht_is_subvol_in_layout(dht_layout_t *layout, xlator_t *xlator);
+
+int
+dht_inode_ctx_get_mig_info(xlator_t *this, inode_t *inode,
+                           xlator_t **src_subvol, xlator_t **dst_subvol);
+gf_boolean_t
+dht_mig_info_is_invalid(xlator_t *current, xlator_t *src_subvol,
+                        xlator_t *dst_subvol);
+
+int
+dht_subvol_status(dht_conf_t *conf, xlator_t *subvol);
+
+void
+dht_log_new_layout_for_dir_selfheal(xlator_t *this, loc_t *loc,
+                                    dht_layout_t *layout);
+
+int
+dht_layout_sort(dht_layout_t *layout);
+
+int
+dht_heal_full_path(void *data);
+
+int
+dht_heal_full_path_done(int op_ret, call_frame_t *frame, void *data);
+
+int
+dht_layout_missing_dirs(dht_layout_t *layout);
+
+int
+dht_refresh_layout(call_frame_t *frame);
+
+int
+dht_build_parent_loc(xlator_t *this, loc_t *parent, loc_t *child,
+                     int32_t *op_errno);
+
+int32_t
+dht_set_local_rebalance(xlator_t *this, dht_local_t *local, struct iatt *stbuf,
+                        struct iatt *prebuf, struct iatt *postbuf,
+                        dict_t *xdata);
+void
+dht_build_root_loc(inode_t *inode, loc_t *loc);
+
+gf_boolean_t
+dht_fd_open_on_dst(xlator_t *this, fd_t *fd, xlator_t *dst);
+
+int32_t
+dht_fd_ctx_destroy(xlator_t *this, fd_t *fd);
+
+int32_t
+dht_release(xlator_t *this, fd_t *fd);
+
+int32_t
+dht_set_fixed_dir_stat(struct iatt *stat);
+
 xlator_t *
-dht_subvol_maxspace_nonzeroinode (xlator_t *this, xlator_t *subvol,
-                                  dht_layout_t *layout);
+dht_get_lock_subvolume(xlator_t *this, struct gf_flock *lock,
+                       dht_local_t *local);
+
+int
+dht_lk_inode_unref(call_frame_t *frame, int32_t op_ret);
+
+int
+dht_fd_ctx_set(xlator_t *this, fd_t *fd, xlator_t *subvol);
+
+int
+dht_check_and_open_fd_on_subvol(xlator_t *this, call_frame_t *frame);
+
+/* FD fop callbacks */
+
+int
+dht_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+               dict_t *xdata);
+
+int
+dht_flush_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, dict_t *xdata);
+
+int
+dht_file_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata);
+
+int
+dht_zerofill_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+                 dict_t *xdata);
+
+int
+dht_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+                dict_t *xdata);
+
+int
+dht_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                  int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+                  dict_t *xdata);
+
+int
+dht_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+                 dict_t *xdata);
+
+int
+dht_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+              dict_t *xdata);
+
+int
+dht_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, struct iovec *vector, int count, struct iatt *stbuf,
+              struct iobref *iobref, dict_t *xdata);
+
+int
+dht_file_attr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                  int op_errno, struct iatt *stbuf, dict_t *xdata);
+
+int
+dht_file_removexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int op_ret, int op_errno, dict_t *xdata);
+
+int
+dht_file_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, dict_t *xdata);
+
+/* All custom xattr heal functions */
+int
+dht_dir_heal_xattrs(void *data);
+
 int
-dht_linkfile_attr_heal (call_frame_t *frame, xlator_t *this);
+dht_dir_heal_xattrs_done(int ret, call_frame_t *sync_frame, void *data);
+
+int32_t
+dht_dict_set_array(dict_t *dict, char *key, int32_t value[], int32_t size);
+
+int
+dht_set_user_xattr(dict_t *dict, char *k, data_t *v, void *data);
 
 void
-dht_layout_dump (dht_layout_t  *layout, const char *prefix);
+dht_dir_set_heal_xattr(xlator_t *this, dht_local_t *local, dict_t *dst,
+                       dict_t *src, int *uret, int *uflag);
+
+int
+dht_dir_xattr_heal(xlator_t *this, dht_local_t *local, int *op_errno);
+
+int
+dht_common_mark_mdsxattr(call_frame_t *frame, int *errst, int flag);
+
+int
+dht_inode_ctx_mdsvol_get(inode_t *inode, xlator_t *this, xlator_t **mdsvol);
+
+int
+dht_selfheal_dir_setattr(call_frame_t *frame, loc_t *loc, struct iatt *stbuf,
+                         int32_t valid, dht_layout_t *layout);
+
+/* Abstract out the DHT-IATT-IN-DICT */
+
+void
+dht_selfheal_layout_new_directory(call_frame_t *frame, loc_t *loc,
+                                  dht_layout_t *new_layout);
+
+int
+dht_pt_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *key,
+                 dict_t *xdata);
+
+int
+dht_pt_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                const char *key, dict_t *xdata);
+
 int32_t
-dht_priv_dump (xlator_t *this);
+dht_pt_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+             mode_t umask, dict_t *xdata);
+
+int
+dht_pt_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+              dict_t *xdata);
+
+int32_t
+dht_check_remote_fd_failed_error(dht_local_t *local, int op_ret, int op_errno);
+
+int
+dht_common_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *dict,
+                       dict_t *xdata);
+
 int32_t
-dht_inodectx_dump (xlator_t *this, inode_t *inode);
+dht_create_lock(call_frame_t *frame, xlator_t *subvol);
 
-#endif/* _DHT_H */
+int
+dht_set_parent_layout_in_dict(loc_t *loc, xlator_t *this, dht_local_t *local);
+
+int
+dht_dir_layout_error_check(xlator_t *this, inode_t *inode);
+
+int
+dht_inode_ctx_mdsvol_set(inode_t *inode, xlator_t *this, xlator_t *mds_subvol);
+#endif /* _DHT_H */
diff --git a/xlators/cluster/dht/src/dht-diskusage.c b/xlators/cluster/dht/src/dht-diskusage.c
index fe3955ecbb7..c0588828fdb 100644
--- a/xlators/cluster/dht/src/dht-diskusage.c
+++ b/xlators/cluster/dht/src/dht-diskusage.c
@@ -8,407 +8,480 @@
   cases as published by the Free Software Foundation.
 */
 
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
 /* TODO: add NS locking */
 
-#include "glusterfs.h"
-#include "xlator.h"
 #include "dht-common.h"
-#include "defaults.h"
 
 #include <sys/time.h>
-
+#include <glusterfs/events.h>
 
 int
-dht_du_info_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		 int op_ret, int op_errno, struct statvfs *statvfs,
-                 dict_t *xdata)
+dht_du_info_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                int op_errno, struct statvfs *statvfs, dict_t *xdata)
 {
-	dht_conf_t    *conf         = NULL;
-	call_frame_t  *prev          = NULL;
-	int            this_call_cnt = 0;
-	int            i = 0;
-	double         percent = 0;
-	double         percent_inodes = 0;
-	uint64_t       bytes = 0;
-
-	conf = this->private;
-	prev = cookie;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"failed to get disk info from %s", prev->this->name);
-		goto out;
-	}
-
-	if (statvfs && statvfs->f_blocks) {
-		percent = (statvfs->f_bavail * 100) / statvfs->f_blocks;
-		bytes = (statvfs->f_bavail * statvfs->f_frsize);
-	}
-
-	if (statvfs && statvfs->f_files) {
-		percent_inodes = (statvfs->f_ffree * 100) / statvfs->f_files;
-	} else {
-		/* set percent inodes to 100 for dynamically allocated inode filesystems
-		   this logic holds good so that, distribute has nothing to worry about
-		   total inodes rather let the 'create()' to be scheduled on the hashed
-		   subvol regardless of the total inodes. since we have no awareness on
-		   loosing inodes this logic fits well
-		*/
-		percent_inodes = 100;
-	}
-
-	LOCK (&conf->subvolume_lock);
-	{
-		for (i = 0; i < conf->subvolume_cnt; i++)
-			if (prev->this == conf->subvolumes[i]) {
-				conf->du_stats[i].avail_percent = percent;
-				conf->du_stats[i].avail_space   = bytes;
-				conf->du_stats[i].avail_inodes  = percent_inodes;
-				gf_log (this->name, GF_LOG_DEBUG,
-					"on subvolume '%s': avail_percent is: "
-					"%.2f and avail_space is: %"PRIu64" "
-					"and avail_inodes is: %.2f",
-					prev->this->name,
-					conf->du_stats[i].avail_percent,
-					conf->du_stats[i].avail_space,
-					conf->du_stats[i].avail_inodes);
-			}
-	}
-	UNLOCK (&conf->subvolume_lock);
+    dht_conf_t *conf = NULL;
+    xlator_t *prev = NULL;
+    int this_call_cnt = 0;
+    int i = 0;
+    double percent = 0;
+    double percent_inodes = 0;
+    uint64_t bytes = 0;
+    uint32_t bpc; /* blocks per chunk */
+    uint32_t chunks = 0;
+
+    conf = this->private;
+    prev = cookie;
+
+    if (op_ret == -1 || !statvfs) {
+        gf_msg(this->name, GF_LOG_WARNING, op_errno,
+               DHT_MSG_GET_DISK_INFO_ERROR, "failed to get disk info from %s",
+               prev->name);
+        goto out;
+    }
+
+    if (statvfs->f_blocks) {
+        percent = (statvfs->f_bavail * 100) / statvfs->f_blocks;
+        bytes = (statvfs->f_bavail * statvfs->f_frsize);
+        /*
+         * A 32-bit count of 1MB chunks allows a maximum brick size of
+         * ~4PB.  It's possible that we could see a single local FS
+         * bigger than that some day, but this code is likely to be
+         * irrelevant by then.  Meanwhile, it's more important to keep
+         * the chunk size small so the layout-calculation code that
+         * uses this value can be tested on normal machines.
+         */
+        bpc = (1 << 20) / statvfs->f_bsize;
+        chunks = (statvfs->f_blocks + bpc - 1) / bpc;
+    }
+
+    if (statvfs->f_files) {
+        percent_inodes = (statvfs->f_ffree * 100) / statvfs->f_files;
+    } else {
+        /*
+         * Set percent inodes to 100 for dynamically allocated inode
+         * filesystems. The rationale is that distribute need not
+         * worry about total inodes; rather, let the 'create()' be
+         * scheduled on the hashed subvol regardless of the total
+         * inodes.
+         */
+        percent_inodes = 100;
+    }
+
+    LOCK(&conf->subvolume_lock);
+    {
+        for (i = 0; i < conf->subvolume_cnt; i++)
+            if (prev == conf->subvolumes[i]) {
+                conf->du_stats[i].avail_percent = percent;
+                conf->du_stats[i].avail_space = bytes;
+                conf->du_stats[i].avail_inodes = percent_inodes;
+                conf->du_stats[i].chunks = chunks;
+                conf->du_stats[i].total_blocks = statvfs->f_blocks;
+                conf->du_stats[i].avail_blocks = statvfs->f_bavail;
+                conf->du_stats[i].frsize = statvfs->f_frsize;
+
+                gf_msg_debug(this->name, 0,
+                             "subvolume '%s': avail_percent "
+                             "is: %.2f and avail_space "
+                             "is: %" PRIu64
+                             " and avail_inodes"
+                             " is: %.2f",
+                             prev->name, conf->du_stats[i].avail_percent,
+                             conf->du_stats[i].avail_space,
+                             conf->du_stats[i].avail_inodes);
+                break; /* no point in looping further */
+            }
+    }
+    UNLOCK(&conf->subvolume_lock);
 
 out:
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt))
-		DHT_STACK_DESTROY (frame);
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt))
+        DHT_STACK_DESTROY(frame);
 
-	return 0;
+    return 0;
 }
 
 int
-dht_get_du_info_for_subvol (xlator_t *this, int subvol_idx)
+dht_get_du_info_for_subvol(xlator_t *this, int subvol_idx)
 {
-	dht_conf_t    *conf         = NULL;
-	call_frame_t  *statfs_frame = NULL;
-	dht_local_t   *statfs_local = NULL;
-	call_pool_t   *pool         = NULL;
-	loc_t          tmp_loc      = {0,};
-
-	conf = this->private;
-	pool = this->ctx->pool;
-
-	statfs_frame = create_frame (this, pool);
-	if (!statfs_frame) {
-		goto err;
-	}
-
-	/* local->fop value is not used in this case */
-	statfs_local = dht_local_init (statfs_frame, NULL, NULL,
-				       GF_FOP_MAXVALUE);
-	if (!statfs_local) {
-		goto err;
-	}
-
-        /* make it root gfid, should be enough to get the proper info back */
-        tmp_loc.gfid[15] = 1;
-
-	statfs_local->call_cnt = 1;
-	STACK_WIND (statfs_frame, dht_du_info_cbk,
-		    conf->subvolumes[subvol_idx],
-		    conf->subvolumes[subvol_idx]->fops->statfs,
-		    &tmp_loc, NULL);
-
-	return 0;
+    dht_conf_t *conf = NULL;
+    call_frame_t *statfs_frame = NULL;
+    dht_local_t *statfs_local = NULL;
+    call_pool_t *pool = NULL;
+    loc_t tmp_loc = {
+        0,
+    };
+
+    conf = this->private;
+    pool = this->ctx->pool;
+
+    statfs_frame = create_frame(this, pool);
+    if (!statfs_frame) {
+        goto err;
+    }
+
+    /* local->fop value is not used in this case */
+    statfs_local = dht_local_init(statfs_frame, NULL, NULL, GF_FOP_MAXVALUE);
+    if (!statfs_local) {
+        goto err;
+    }
+
+    /* make it root gfid, should be enough to get the proper info back */
+    tmp_loc.gfid[15] = 1;
+
+    statfs_local->call_cnt = 1;
+    STACK_WIND_COOKIE(
+        statfs_frame, dht_du_info_cbk, conf->subvolumes[subvol_idx],
+        conf->subvolumes[subvol_idx],
+        conf->subvolumes[subvol_idx]->fops->statfs, &tmp_loc, NULL);
+
+    return 0;
 err:
-	if (statfs_frame)
-		DHT_STACK_DESTROY (statfs_frame);
+    if (statfs_frame)
+        DHT_STACK_DESTROY(statfs_frame);
 
-	return -1;
+    return -1;
 }
 
 int
-dht_get_du_info (call_frame_t *frame, xlator_t *this, loc_t *loc)
+dht_get_du_info(call_frame_t *frame, xlator_t *this, loc_t *loc)
 {
-	int            i            = 0;
-	dht_conf_t    *conf         = NULL;
-	call_frame_t  *statfs_frame = NULL;
-	dht_local_t   *statfs_local = NULL;
-	struct timeval tv           = {0,};
-        loc_t          tmp_loc      = {0,};
-
-	conf  = this->private;
-
-	gettimeofday (&tv, NULL);
-
-        /* make it root gfid, should be enough to get the proper
-           info back */
-        tmp_loc.gfid[15] = 1;
-
-	if (tv.tv_sec > (conf->refresh_interval
-			 + conf->last_stat_fetch.tv_sec)) {
-
-		statfs_frame = copy_frame (frame);
-		if (!statfs_frame) {
-			goto err;
-		}
-
-		/* In this case, 'local->fop' is not used */
-		statfs_local = dht_local_init (statfs_frame, loc, NULL,
-					       GF_FOP_MAXVALUE);
-		if (!statfs_local) {
-			goto err;
-		}
-
-		statfs_local->call_cnt = conf->subvolume_cnt;
-		for (i = 0; i < conf->subvolume_cnt; i++) {
-			STACK_WIND (statfs_frame, dht_du_info_cbk,
-				    conf->subvolumes[i],
-				    conf->subvolumes[i]->fops->statfs,
-				    &tmp_loc, NULL);
-		}
-
-		conf->last_stat_fetch.tv_sec = tv.tv_sec;
-	}
-	return 0;
+    int i = 0;
+    int ret = -1;
+    dht_conf_t *conf = NULL;
+    call_frame_t *statfs_frame = NULL;
+    dht_local_t *statfs_local = NULL;
+    loc_t tmp_loc = {
+        0,
+    };
+    time_t now;
+
+    conf = this->private;
+    now = gf_time();
+    /* make it root gfid, should be enough to get the proper
+       info back */
+    tmp_loc.gfid[15] = 1;
+
+    if (now > (conf->refresh_interval + conf->last_stat_fetch)) {
+        statfs_frame = copy_frame(frame);
+        if (!statfs_frame) {
+            goto err;
+        }
+
+        /* In this case, 'local->fop' is not used */
+        statfs_local = dht_local_init(statfs_frame, loc, NULL, GF_FOP_MAXVALUE);
+        if (!statfs_local) {
+            goto err;
+        }
+
+        statfs_local->params = dict_new();
+        if (!statfs_local->params)
+            goto err;
+
+        ret = dict_set_int8(statfs_local->params,
+                            GF_INTERNAL_IGNORE_DEEM_STATFS, 1);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                   "Failed to set " GF_INTERNAL_IGNORE_DEEM_STATFS " in dict");
+            goto err;
+        }
+
+        statfs_local->call_cnt = conf->subvolume_cnt;
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            STACK_WIND_COOKIE(statfs_frame, dht_du_info_cbk,
+                              conf->subvolumes[i], conf->subvolumes[i],
+                              conf->subvolumes[i]->fops->statfs, &tmp_loc,
+                              statfs_local->params);
+        }
+
+        conf->last_stat_fetch = now;
+    }
+    return 0;
 err:
-	if (statfs_frame)
-		DHT_STACK_DESTROY (statfs_frame);
+    if (statfs_frame)
+        DHT_STACK_DESTROY(statfs_frame);
 
-	return -1;
+    return -1;
 }
 
-
 gf_boolean_t
-dht_is_subvol_filled (xlator_t *this, xlator_t *subvol)
+dht_is_subvol_filled(xlator_t *this, xlator_t *subvol)
 {
-	int         i = 0;
-	dht_conf_t *conf = NULL;
-	gf_boolean_t subvol_filled_inodes = _gf_false;
-	gf_boolean_t subvol_filled_space = _gf_false;
-	gf_boolean_t is_subvol_filled = _gf_false;
-
-	conf = this->private;
-
-	/* Check for values above specified percent or free disk */
-	LOCK (&conf->subvolume_lock);
-	{
-		for (i = 0; i < conf->subvolume_cnt; i++) {
-			if (subvol == conf->subvolumes[i]) {
-				if (conf->disk_unit == 'p') {
-					if (conf->du_stats[i].avail_percent <
-					    conf->min_free_disk) {
-						subvol_filled_space = _gf_true;
-						break;
-					}
-
-				} else {
-					if (conf->du_stats[i].avail_space <
-					    conf->min_free_disk) {
-						subvol_filled_space = _gf_true;
-						break;
-					}
-				}
-				if (conf->du_stats[i].avail_inodes <
-				    conf->min_free_inodes) {
-					subvol_filled_inodes = _gf_true;
-					break;
-				}
-			}
-		}
-	}
-	UNLOCK (&conf->subvolume_lock);
-
-	if (subvol_filled_space && conf->subvolume_status[i]) {
-		if (!(conf->du_stats[i].log++ % (GF_UNIVERSAL_ANSWER * 10))) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"disk space on subvolume '%s' is getting "
-				"full (%.2f %%), consider adding more nodes",
-				subvol->name,
-				(100 - conf->du_stats[i].avail_percent));
-		}
-	}
-
-	if (subvol_filled_inodes && conf->subvolume_status[i]) {
-		if (!(conf->du_stats[i].log++ % (GF_UNIVERSAL_ANSWER * 10))) {
-			gf_log (this->name, GF_LOG_CRITICAL,
-				"inodes on subvolume '%s' are at "
-				"(%.2f %%), consider adding more nodes",
-				subvol->name,
-				(100 - conf->du_stats[i].avail_inodes));
-		}
-	}
-
-	is_subvol_filled = (subvol_filled_space || subvol_filled_inodes);
-
-	return is_subvol_filled;
-}
+    int i = 0;
+    char vol_name[256];
+    dht_conf_t *conf = NULL;
+    gf_boolean_t subvol_filled_inodes = _gf_false;
+    gf_boolean_t subvol_filled_space = _gf_false;
+    gf_boolean_t is_subvol_filled = _gf_false;
+    double usage = 0;
+
+    conf = this->private;
+
+    /* Check for values above specified percent or free disk */
+    LOCK(&conf->subvolume_lock);
+    {
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            if (subvol == conf->subvolumes[i]) {
+                if (conf->disk_unit == 'p') {
+                    if (conf->du_stats[i].avail_percent < conf->min_free_disk) {
+                        subvol_filled_space = _gf_true;
+                        break;
+                    }
+
+                } else {
+                    if (conf->du_stats[i].avail_space < conf->min_free_disk) {
+                        subvol_filled_space = _gf_true;
+                        break;
+                    }
+                }
+                if (conf->du_stats[i].avail_inodes < conf->min_free_inodes) {
+                    subvol_filled_inodes = _gf_true;
+                    break;
+                }
+            }
+        }
+    }
+    UNLOCK(&conf->subvolume_lock);
 
+    if (subvol_filled_space && conf->subvolume_status[i]) {
+        if (!(conf->du_stats[i].log++ % (GF_UNIVERSAL_ANSWER * 10))) {
+            usage = 100 - conf->du_stats[i].avail_percent;
 
-/*Get the best subvolume to create the file in*/
-xlator_t *
-dht_free_disk_available_subvol (xlator_t *this, xlator_t *subvol,
-                                dht_local_t *local)
-{
-	xlator_t   *avail_subvol = NULL;
-	dht_conf_t *conf = NULL;
-        dht_layout_t *layout = NULL;
-        loc_t      *loc = NULL;
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_SUBVOL_INSUFF_SPACE,
+                   "disk space on subvolume '%s' is getting "
+                   "full (%.2f %%), consider adding more bricks",
+                   subvol->name, usage);
 
-	conf = this->private;
-        if (!local)
-                goto out;
-        loc = &local->loc;
-        if (!local->layout) {
-                layout = dht_layout_get (this, loc->parent);
-
-                if (!layout) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "layout missing path=%s parent=%s",
-                                loc->path, uuid_utoa (loc->parent->gfid));
-                        goto out;
-                }
-        } else {
-                layout = dht_layout_ref (this, local->layout);
+            (void)snprintf(vol_name, sizeof(vol_name), "%s", this->name);
+            vol_name[(strlen(this->name) - 4)] = '\0';
+
+            gf_event(EVENT_DHT_DISK_USAGE, "volume=%s;subvol=%s;usage=%.2f %%",
+                     vol_name, subvol->name, usage);
         }
+    }
+
+    if (subvol_filled_inodes && conf->subvolume_status[i]) {
+        if (!(conf->du_stats[i].log++ % (GF_UNIVERSAL_ANSWER * 10))) {
+            usage = 100 - conf->du_stats[i].avail_inodes;
+            gf_msg(this->name, GF_LOG_CRITICAL, 0, DHT_MSG_SUBVOL_INSUFF_INODES,
+                   "inodes on subvolume '%s' are at "
+                   "(%.2f %%), consider adding more bricks",
+                   subvol->name, usage);
+
+            (void)snprintf(vol_name, sizeof(vol_name), "%s", this->name);
+            vol_name[(strlen(this->name) - 4)] = '\0';
+
+            gf_event(EVENT_DHT_INODES_USAGE,
+                     "volume=%s;subvol=%s;usage=%.2f %%", vol_name,
+                     subvol->name, usage);
+        }
+    }
 
-        LOCK (&conf->subvolume_lock);
-	{
-                avail_subvol = dht_subvol_with_free_space_inodes(this, subvol,
-                                                                 layout);
-                if(!avail_subvol)
-                {
-                        avail_subvol = dht_subvol_maxspace_nonzeroinode(this,
-                                                                        subvol,
-                                                                        layout);
-                }
+    is_subvol_filled = (subvol_filled_space || subvol_filled_inodes);
+
+    return is_subvol_filled;
+}
 
-	}
-	UNLOCK (&conf->subvolume_lock);
+/*Get the best subvolume to create the file in*/
+xlator_t *
+dht_free_disk_available_subvol(xlator_t *this, xlator_t *subvol,
+                               dht_local_t *local)
+{
+    xlator_t *avail_subvol = NULL;
+    dht_conf_t *conf = NULL;
+    dht_layout_t *layout = NULL;
+    loc_t *loc = NULL;
+
+    conf = this->private;
+    if (!local)
+        goto out;
+    loc = &local->loc;
+    if (!local->layout) {
+        layout = dht_layout_get(this, loc->parent);
+
+        if (!layout) {
+            gf_msg_debug(this->name, 0,
+                         "Missing layout. path=%s,"
+                         " parent gfid = %s",
+                         loc->path, uuid_utoa(loc->parent->gfid));
+            goto out;
+        }
+    } else {
+        layout = dht_layout_ref(this, local->layout);
+    }
+
+    LOCK(&conf->subvolume_lock);
+    {
+        avail_subvol = dht_subvol_with_free_space_inodes(this, subvol, NULL,
+                                                         layout, 0);
+        if (!avail_subvol) {
+            avail_subvol = dht_subvol_maxspace_nonzeroinode(this, subvol,
+                                                            layout);
+        }
+    }
+    UNLOCK(&conf->subvolume_lock);
 out:
-	if (!avail_subvol) {
-		gf_log (this->name,
-                        GF_LOG_DEBUG,
-			"no subvolume has enough free space and/or inodes\
-                         to create");
-                avail_subvol = subvol;
-	}
-
-        if (layout)
-                dht_layout_unref (this, layout);
-	return avail_subvol;
+    if (!avail_subvol) {
+        gf_msg_debug(this->name, 0,
+                     "No subvolume has enough free space \
+                              and/or inodes to create");
+        avail_subvol = subvol;
+    }
+
+    if (layout)
+        dht_layout_unref(this, layout);
+    return avail_subvol;
 }
 
-static inline
-int32_t dht_subvol_has_err (xlator_t *this, dht_layout_t *layout)
+static inline int32_t
+dht_subvol_has_err(dht_conf_t *conf, xlator_t *this, xlator_t *ignore,
+                   dht_layout_t *layout)
 {
-        int ret = -1;
-        int i   = 0;
+    int ret = -1;
+    int i = 0;
+
+    if (!this || !layout)
+        goto out;
+
+    /* this check is meant for rebalance process. The source of the file
+     * should be ignored for space check */
+    if (this == ignore) {
+        goto out;
+    }
+
+    /* check if subvol has layout errors, before selecting it */
+    for (i = 0; i < layout->cnt; i++) {
+        if (!strcmp(layout->list[i].xlator->name, this->name) &&
+            (layout->list[i].err != 0)) {
+            ret = -1;
+            goto out;
+        }
+    }
 
-        if (!this || !layout)
+    /* discard decommissioned subvol */
+    if (conf->decommission_subvols_cnt) {
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            if (conf->decommissioned_bricks[i] &&
+                conf->decommissioned_bricks[i] == this) {
+                ret = -1;
                 goto out;
-
-        /* check if subvol has layout errors, before selecting it */
-        for (i = 0; i < layout->cnt; i++) {
-                if (!strcmp (layout->list[i].xlator->name, this->name) &&
-                     (layout->list[i].err != 0)) {
-                        ret = -1;
-                        goto out;
-                }
+            }
         }
-        ret = 0;
+    }
+
+    ret = 0;
 out:
-        return ret;
+    return ret;
 }
 
 /*Get subvolume which has both space and inodes more than the min criteria*/
 xlator_t *
 dht_subvol_with_free_space_inodes(xlator_t *this, xlator_t *subvol,
-                                  dht_layout_t *layout)
+                                  xlator_t *ignore, dht_layout_t *layout,
+                                  uint64_t filesize)
 {
-        int i = 0;
-        double max = 0;
-        double max_inodes = 0;
-        int    ignore_subvol = 0;
-
-        xlator_t *avail_subvol = NULL;
-        dht_conf_t *conf = NULL;
-
-        conf = this->private;
-
-        for(i=0; i < conf->subvolume_cnt; i++) {
-                /* check if subvol has layout errors, before selecting it */
-                ignore_subvol = dht_subvol_has_err (conf->subvolumes[i],
-                                                    layout);
-                if (ignore_subvol)
-                        continue;
-
-                if ((conf->disk_unit == 'p') &&
-                    (conf->du_stats[i].avail_percent > conf->min_free_disk) &&
-                    (conf->du_stats[i].avail_inodes  > conf->min_free_inodes)) {
-                        if ((conf->du_stats[i].avail_inodes > max_inodes) ||
-                            (conf->du_stats[i].avail_percent > max)) {
-                                max = conf->du_stats[i].avail_percent;
-                                max_inodes = conf->du_stats[i].avail_inodes;
-                                avail_subvol = conf->subvolumes[i];
-                        }
-                }
+    int i = 0;
+    double max = 0;
+    double max_inodes = 0;
+    int ignore_subvol = 0;
+    uint64_t total_blocks = 0;
+    uint64_t avail_blocks = 0;
+    uint64_t frsize = 0;
+    double post_availspace = 0;
+    double post_percent = 0;
+
+    xlator_t *avail_subvol = NULL;
+    dht_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        /* check if subvol has layout errors and also it is not a
+         * decommissioned brick, before selecting it */
+        ignore_subvol = dht_subvol_has_err(conf, conf->subvolumes[i], ignore,
+                                           layout);
+        if (ignore_subvol)
+            continue;
+
+        if ((conf->disk_unit == 'p') &&
+            (conf->du_stats[i].avail_percent > conf->min_free_disk) &&
+            (conf->du_stats[i].avail_inodes > conf->min_free_inodes)) {
+            if ((conf->du_stats[i].avail_inodes > max_inodes) ||
+                (conf->du_stats[i].avail_percent > max)) {
+                max = conf->du_stats[i].avail_percent;
+                max_inodes = conf->du_stats[i].avail_inodes;
+                avail_subvol = conf->subvolumes[i];
+                total_blocks = conf->du_stats[i].total_blocks;
+                avail_blocks = conf->du_stats[i].avail_blocks;
+                frsize = conf->du_stats[i].frsize;
+            }
+        }
 
-                if ((conf->disk_unit != 'p') &&
-                    (conf->du_stats[i].avail_space > conf->min_free_disk) &&
-                    (conf->du_stats[i].avail_inodes  > conf->min_free_inodes)) {
-                        if ((conf->du_stats[i].avail_inodes > max_inodes) ||
-                            (conf->du_stats[i].avail_space > max)) {
-                                max = conf->du_stats[i].avail_space;
-                                max_inodes = conf->du_stats[i].avail_inodes;
-                                avail_subvol = conf->subvolumes[i];
-                        }
-                }
+        if ((conf->disk_unit != 'p') &&
+            (conf->du_stats[i].avail_space > conf->min_free_disk) &&
+            (conf->du_stats[i].avail_inodes > conf->min_free_inodes)) {
+            if ((conf->du_stats[i].avail_inodes > max_inodes) ||
+                (conf->du_stats[i].avail_space > max)) {
+                max = conf->du_stats[i].avail_space;
+                max_inodes = conf->du_stats[i].avail_inodes;
+                avail_subvol = conf->subvolumes[i];
+            }
+        }
+    }
+
+    if (avail_subvol) {
+        if (conf->disk_unit == 'p') {
+            post_availspace = (avail_blocks * frsize) - filesize;
+            post_percent = (post_availspace * 100) / (total_blocks * frsize);
+            if (post_percent < conf->min_free_disk)
+                avail_subvol = NULL;
         }
+        if (conf->disk_unit != 'p') {
+            if ((max - filesize) < conf->min_free_disk)
+                avail_subvol = NULL;
+        }
+    }
 
-        return avail_subvol;
+    return avail_subvol;
 }
 
-
-/* Get subvol which has atleast one inode and maximum space */
+/* Get subvol which has at least one inode and maximum space */
 xlator_t *
-dht_subvol_maxspace_nonzeroinode (xlator_t *this, xlator_t *subvol,
-                                  dht_layout_t *layout)
+dht_subvol_maxspace_nonzeroinode(xlator_t *this, xlator_t *subvol,
+                                 dht_layout_t *layout)
 {
-        int         i = 0;
-        double      max = 0;
-        int         ignore_subvol = 0;
-
-        xlator_t   *avail_subvol = NULL;
-        dht_conf_t *conf = NULL;
-
-        conf = this->private;
-
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                /* check if subvol has layout errors, before selecting it */
-                ignore_subvol = dht_subvol_has_err (conf->subvolumes[i],
-                                                    layout);
-                if (ignore_subvol)
-                        continue;
-
-                if (conf->disk_unit == 'p') {
-                        if ((conf->du_stats[i].avail_percent > max)
-                            && (conf->du_stats[i].avail_inodes > 0 )) {
-                                max = conf->du_stats[i].avail_percent;
-                                avail_subvol = conf->subvolumes[i];
-                        }
-               } else {
-                         if ((conf->du_stats[i].avail_space > max)
-                            && (conf->du_stats[i].avail_inodes > 0)) {
-                                 max = conf->du_stats[i].avail_space;
-                                 avail_subvol = conf->subvolumes[i];
-                         }
-               }
+    int i = 0;
+    double max = 0;
+    int ignore_subvol = 0;
+
+    xlator_t *avail_subvol = NULL;
+    dht_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        /* check if subvol has layout errors and also it is not a
+         * decommissioned brick, before selecting it*/
+
+        ignore_subvol = dht_subvol_has_err(conf, conf->subvolumes[i], NULL,
+                                           layout);
+        if (ignore_subvol)
+            continue;
+
+        if (conf->disk_unit == 'p') {
+            if ((conf->du_stats[i].avail_percent > max) &&
+                (conf->du_stats[i].avail_inodes > 0)) {
+                max = conf->du_stats[i].avail_percent;
+                avail_subvol = conf->subvolumes[i];
+            }
+        } else {
+            if ((conf->du_stats[i].avail_space > max) &&
+                (conf->du_stats[i].avail_inodes > 0)) {
+                max = conf->du_stats[i].avail_space;
+                avail_subvol = conf->subvolumes[i];
+            }
         }
+    }
 
-        return avail_subvol;
+    return avail_subvol;
 }
diff --git a/xlators/cluster/dht/src/dht-hashfn.c b/xlators/cluster/dht/src/dht-hashfn.c
index 656cf23a058..acda67c312a 100644
--- a/xlators/cluster/dht/src/dht-hashfn.c
+++ b/xlators/cluster/dht/src/dht-hashfn.c
@@ -8,104 +8,103 @@
   cases as published by the Free Software Foundation.
 */
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-
-#include "glusterfs.h"
-#include "xlator.h"
 #include "dht-common.h"
-#include "hashfn.h"
+#include <glusterfs/hashfn.h>
 
-
-int
-dht_hash_compute_internal (int type, const char *name, uint32_t *hash_p)
+static int
+dht_hash_compute_internal(int type, const char *name, const int len,
+                          uint32_t *hash_p)
 {
-        int      ret = 0;
-        uint32_t hash = 0;
+    int ret = 0;
+    uint32_t hash = 0;
 
-        switch (type) {
+    switch (type) {
         case DHT_HASH_TYPE_DM:
         case DHT_HASH_TYPE_DM_USER:
-                hash = gf_dm_hashfn (name, strlen (name));
-                break;
+            hash = gf_dm_hashfn(name, len);
+            break;
         default:
-                ret = -1;
-                break;
-        }
+            ret = -1;
+            break;
+    }
 
-        if (ret == 0) {
-                *hash_p = hash;
-        }
+    if (ret == 0) {
+        *hash_p = hash;
+    }
 
-        return ret;
+    return ret;
 }
 
-
-static inline
-gf_boolean_t
-dht_munge_name (const char *original, char *modified, size_t len, regex_t *re)
+/* The function returns:
+ * 0  : in case no munge took place
+ * >0 : the length (inc. terminating NULL!) of the newly modified string,
+ *      if it was munged.
+ */
+static int
+dht_munge_name(const char *original, char *modified, size_t len, regex_t *re)
 {
-        regmatch_t      matches[2];
-        size_t          new_len;
-
-        if (regexec(re,original,2,matches,0) != REG_NOMATCH) {
-                if (matches[1].rm_so != -1) {
-                        new_len = matches[1].rm_eo - matches[1].rm_so;
-                        /* Equal would fail due to the NUL at the end. */
-                        if (new_len < len) {
-                                memcpy (modified,original+matches[1].rm_so,
-                                        new_len);
-                                modified[new_len] = '\0';
-                                return _gf_true;
-                        }
-                }
+    regmatch_t matches[2] = {
+        {0},
+    };
+    size_t new_len = 0;
+    int ret = 0;
+
+    ret = regexec(re, original, 2, matches, 0);
+
+    if (ret != REG_NOMATCH) {
+        if (matches[1].rm_so != -1) {
+            new_len = matches[1].rm_eo - matches[1].rm_so;
+            /* Equal would fail due to the NUL at the end. */
+            if (new_len < len) {
+                memcpy(modified, original + matches[1].rm_so, new_len);
+                modified[new_len] = '\0';
+                return new_len + 1; /* +1 for the terminating NULL */
+            }
         }
+    }
 
-        /* This is guaranteed safe because of how the dest was allocated. */
-        strcpy(modified,original);
-        return _gf_false;
+    /* This is guaranteed safe because of how the dest was allocated. */
+    strcpy(modified, original);
+    return 0;
 }
 
 int
-dht_hash_compute (xlator_t *this, int type, const char *name, uint32_t *hash_p)
+dht_hash_compute(xlator_t *this, int type, const char *name, uint32_t *hash_p)
 {
-        char            *rsync_friendly_name    = NULL;
-        dht_conf_t      *priv                   = this->private;
-        size_t           len                    = 0;
-        gf_boolean_t     munged                 = _gf_false;
-
-        /*
-         * It wouldn't be safe to use alloca in an inline function that doesn't
-         * actually get inlined, and it wouldn't be efficient to do a real
-         * allocation, so we use alloca here (if needed) and pass that to the
-         * inline.
-         */
+    char *rsync_friendly_name = NULL;
+    dht_conf_t *priv = NULL;
+    size_t len = 0;
+    int munged = 0;
 
+    priv = this->private;
+
+    if (name == NULL)
+        return -1;
+
+    len = strlen(name) + 1;
+    rsync_friendly_name = alloca(len);
+
+    LOCK(&priv->lock);
+    {
         if (priv->extra_regex_valid) {
-                len = strlen(name) + 1;
-                rsync_friendly_name = alloca(len);
-                munged = dht_munge_name (name, rsync_friendly_name, len,
-                                         &priv->extra_regex);
+            munged = dht_munge_name(name, rsync_friendly_name, len,
+                                    &priv->extra_regex);
         }
 
         if (!munged && priv->rsync_regex_valid) {
-                len = strlen(name) + 1;
-                rsync_friendly_name = alloca(len);
-                gf_log (this->name, GF_LOG_TRACE, "trying regex for %s", name);
-                munged = dht_munge_name (name, rsync_friendly_name, len,
-                                         &priv->rsync_regex);
-                if (munged) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "munged down to %s", rsync_friendly_name);
-                }
+            gf_msg_trace(this->name, 0, "trying regex for %s", name);
+            munged = dht_munge_name(name, rsync_friendly_name, len,
+                                    &priv->rsync_regex);
         }
-
-        if (!munged) {
-                rsync_friendly_name = (char *)name;
-        }
-
-        return dht_hash_compute_internal (type, rsync_friendly_name, hash_p);
+    }
+    UNLOCK(&priv->lock);
+    if (munged) {
+        gf_msg_debug(this->name, 0, "munged down to %s", rsync_friendly_name);
+        len = munged;
+    } else {
+        rsync_friendly_name = (char *)name;
+    }
+
+    return dht_hash_compute_internal(type, rsync_friendly_name, len - 1,
+                                     hash_p);
 }
diff --git a/xlators/cluster/dht/src/dht-helper.c b/xlators/cluster/dht/src/dht-helper.c
index 1cf343d24ac..3f2fe43d5f3 100644
--- a/xlators/cluster/dht/src/dht-helper.c
+++ b/xlators/cluster/dht/src/dht-helper.c
@@ -8,1110 +8,2297 @@
   cases as published by the Free Software Foundation.
 */
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+#include "dht-common.h"
+#include "dht-lock.h"
+#include "glusterfs/compat-errno.h"  // for ENODATA on BSD
 
+static void
+dht_free_fd_ctx(dht_fd_ctx_t *fd_ctx)
+{
+    GF_FREE(fd_ctx);
+}
 
-#include "glusterfs.h"
-#include "xlator.h"
-#include "dht-common.h"
+int32_t
+dht_fd_ctx_destroy(xlator_t *this, fd_t *fd)
+{
+    dht_fd_ctx_t *fd_ctx = NULL;
+    uint64_t value = 0;
+    int32_t ret = -1;
+
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    ret = fd_ctx_del(fd, this, &value);
+    if (ret) {
+        goto out;
+    }
+
+    fd_ctx = (dht_fd_ctx_t *)(uintptr_t)value;
+    if (fd_ctx) {
+        GF_REF_PUT(fd_ctx);
+    }
+out:
+    return ret;
+}
+
+static int
+__dht_fd_ctx_set(xlator_t *this, fd_t *fd, xlator_t *dst)
+{
+    dht_fd_ctx_t *fd_ctx = NULL;
+    uint64_t value = 0;
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    fd_ctx = GF_CALLOC(1, sizeof(*fd_ctx), gf_dht_mt_fd_ctx_t);
 
+    if (!fd_ctx) {
+        goto out;
+    }
+
+    fd_ctx->opened_on_dst = (uint64_t)(uintptr_t)dst;
+    GF_REF_INIT(fd_ctx, dht_free_fd_ctx);
+
+    value = (uint64_t)(uintptr_t)fd_ctx;
+
+    ret = __fd_ctx_set(fd, this, value);
+    if (ret < 0) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_FD_CTX_SET_FAILED,
+                "fd=0x%p", fd, NULL);
+        GF_REF_PUT(fd_ctx);
+    }
+out:
+    return ret;
+}
 
 int
-dht_frame_return (call_frame_t *frame)
+dht_fd_ctx_set(xlator_t *this, fd_t *fd, xlator_t *dst)
 {
-        dht_local_t *local = NULL;
-        int          this_call_cnt = -1;
+    dht_fd_ctx_t *fd_ctx = NULL;
+    uint64_t value = 0;
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    LOCK(&fd->lock);
+    {
+        ret = __fd_ctx_get(fd, this, &value);
+        if (ret && value) {
+            fd_ctx = (dht_fd_ctx_t *)(uintptr_t)value;
+            if (fd_ctx->opened_on_dst == (uint64_t)(uintptr_t)dst) {
+                /* This could happen due to racing
+                 * check_progress tasks*/
+                goto unlock;
+            } else {
+                /* This would be a big problem*/
+                /* Overwrite and hope for the best*/
+                fd_ctx->opened_on_dst = (uint64_t)(uintptr_t)dst;
+                UNLOCK(&fd->lock);
+                gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_INVALID_VALUE,
+                        NULL);
 
-        if (!frame)
-                return -1;
+                goto out;
+            }
+        }
+        ret = __dht_fd_ctx_set(this, fd, dst);
+    }
+unlock:
+    UNLOCK(&fd->lock);
+out:
+    return ret;
+}
 
-        local = frame->local;
+static dht_fd_ctx_t *
+dht_fd_ctx_get(xlator_t *this, fd_t *fd)
+{
+    dht_fd_ctx_t *fd_ctx = NULL;
+    int ret = -1;
+    uint64_t tmp_val = 0;
+
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    LOCK(&fd->lock);
+    {
+        ret = __fd_ctx_get(fd, this, &tmp_val);
+        if ((ret < 0) || (tmp_val == 0)) {
+            goto unlock;
+        }
 
-        LOCK (&frame->lock);
-        {
-                this_call_cnt = --local->call_cnt;
+        fd_ctx = (dht_fd_ctx_t *)(uintptr_t)tmp_val;
+        GF_REF_GET(fd_ctx);
+    }
+unlock:
+    UNLOCK(&fd->lock);
+
+out:
+    return fd_ctx;
+}
+
+gf_boolean_t
+dht_fd_open_on_dst(xlator_t *this, fd_t *fd, xlator_t *dst)
+{
+    dht_fd_ctx_t *fd_ctx = NULL;
+    gf_boolean_t opened = _gf_false;
+
+    fd_ctx = dht_fd_ctx_get(this, fd);
+
+    if (fd_ctx) {
+        if (fd_ctx->opened_on_dst == (uint64_t)(uintptr_t)dst) {
+            opened = _gf_true;
         }
-        UNLOCK (&frame->lock);
+        GF_REF_PUT(fd_ctx);
+    }
 
-        return this_call_cnt;
+    return opened;
 }
 
+void
+dht_free_mig_info(void *data)
+{
+    dht_migrate_info_t *miginfo = NULL;
+
+    miginfo = data;
+    GF_FREE(miginfo);
+
+    return;
+}
 
-static uint64_t
-dht_bits_for (uint64_t num)
+static int
+dht_inode_ctx_set_mig_info(xlator_t *this, inode_t *inode, xlator_t *src_subvol,
+                           xlator_t *dst_subvol)
 {
-	uint64_t bits = 0, ctrl = 1;
+    dht_migrate_info_t *miginfo = NULL;
+    uint64_t value = 0;
+    int ret = -1;
+
+    miginfo = GF_CALLOC(1, sizeof(*miginfo), gf_dht_mt_miginfo_t);
+    if (miginfo == NULL)
+        goto out;
 
-	while (ctrl < num) {
-		ctrl *= 2;
-		bits ++;
-	}
+    miginfo->src_subvol = src_subvol;
+    miginfo->dst_subvol = dst_subvol;
+    GF_REF_INIT(miginfo, dht_free_mig_info);
 
-	return bits;
+    value = (uint64_t)(uintptr_t)miginfo;
+
+    ret = inode_ctx_set1(inode, this, &value);
+    if (ret < 0) {
+        GF_REF_PUT(miginfo);
+    }
+
+out:
+    return ret;
 }
 
-/*
- * A slightly "updated" version of the algorithm described in the commit log
- * is used here.
- *
- * The only enhancement is that:
+int
+dht_inode_ctx_get_mig_info(xlator_t *this, inode_t *inode,
+                           xlator_t **src_subvol, xlator_t **dst_subvol)
+{
+    int ret = -1;
+    uint64_t tmp_miginfo = 0;
+    dht_migrate_info_t *miginfo = NULL;
+
+    LOCK(&inode->lock);
+    {
+        ret = __inode_ctx_get1(inode, this, &tmp_miginfo);
+        if ((ret < 0) || (tmp_miginfo == 0)) {
+            UNLOCK(&inode->lock);
+            goto out;
+        }
+
+        miginfo = (dht_migrate_info_t *)(uintptr_t)tmp_miginfo;
+        GF_REF_GET(miginfo);
+    }
+    UNLOCK(&inode->lock);
+
+    if (src_subvol)
+        *src_subvol = miginfo->src_subvol;
+
+    if (dst_subvol)
+        *dst_subvol = miginfo->dst_subvol;
+
+    GF_REF_PUT(miginfo);
+
+out:
+    return ret;
+}
+
+gf_boolean_t
+dht_mig_info_is_invalid(xlator_t *current, xlator_t *src_subvol,
+                        xlator_t *dst_subvol)
+{
+    /* Not set
+     */
+    if (!src_subvol || !dst_subvol)
+        return _gf_true;
+
+    /* Invalid scenarios:
+     * The src_subvol does not match the subvol on which the current op was sent
+     * so the cached subvol has changed between the last mig_info_set and now.
+     * src_subvol == dst_subvol. The file was migrated without any FOP detecting
+     * a P2 so the old dst is now the current subvol.
+     *
+     * There is still one scenario where the info could be outdated - if
+     * file has undergone multiple migrations and ends up on the same src_subvol
+     * on which the mig_info was first set.
+     */
+    if ((current == dst_subvol) || (current != src_subvol))
+        return _gf_true;
+
+    return _gf_false;
+}
+
+/* Used to check if fd fops have the fd opened on the cached subvol
+ * This is required when:
+ * 1. an fd is opened on FILE1 on subvol1
+ * 2. the file is migrated to subvol2
+ * 3. a lookup updates the cached subvol in the inode_ctx to subvol2
+ * 4. a write comes on the fd
+ * The write is sent to subvol2 on an fd which has been opened only on fd1
+ * Since the migration phase checks don't kick in, the fop fails with EBADF
  *
- * - The number of bits used by the backend filesystem for HUGE d_off which
- *   is described as 63, and
- * - The number of bits used by the d_off presented by the transformation
- *   upwards which is described as 64, are both made "configurable."
  */
 
+int
+dht_check_and_open_fd_on_subvol_complete(int ret, call_frame_t *frame,
+                                         void *data)
+{
+    glusterfs_fop_t fop = 0;
+    dht_local_t *local = NULL;
+    xlator_t *subvol = NULL;
+    xlator_t *this = NULL;
+    fd_t *fd = NULL;
+    int op_errno = -1;
+
+    local = frame->local;
+    this = frame->this;
+    fop = local->fop;
+    subvol = local->cached_subvol;
+    fd = local->fd;
+
+    if (ret) {
+        op_errno = local->op_errno;
+        goto handle_err;
+    }
+
+    switch (fop) {
+        case GF_FOP_WRITE:
+            STACK_WIND_COOKIE(frame, dht_writev_cbk, subvol, subvol,
+                              subvol->fops->writev, fd, local->rebalance.vector,
+                              local->rebalance.count, local->rebalance.offset,
+                              local->rebalance.flags, local->rebalance.iobref,
+                              local->xattr_req);
+            break;
+
+        case GF_FOP_FLUSH:
+            STACK_WIND(frame, dht_flush_cbk, subvol, subvol->fops->flush, fd,
+                       local->xattr_req);
+            break;
+
+        case GF_FOP_FSETATTR:
+            STACK_WIND_COOKIE(frame, dht_file_setattr_cbk, subvol, subvol,
+                              subvol->fops->fsetattr, fd,
+                              &local->rebalance.stbuf, local->rebalance.flags,
+                              local->xattr_req);
+            break;
+
+        case GF_FOP_ZEROFILL:
+            STACK_WIND_COOKIE(frame, dht_zerofill_cbk, subvol, subvol,
+                              subvol->fops->zerofill, fd,
+                              local->rebalance.offset, local->rebalance.size,
+                              local->xattr_req);
+
+            break;
+
+        case GF_FOP_DISCARD:
+            STACK_WIND_COOKIE(frame, dht_discard_cbk, subvol, subvol,
+                              subvol->fops->discard, local->fd,
+                              local->rebalance.offset, local->rebalance.size,
+                              local->xattr_req);
+            break;
+
+        case GF_FOP_FALLOCATE:
+            STACK_WIND_COOKIE(frame, dht_fallocate_cbk, subvol, subvol,
+                              subvol->fops->fallocate, fd,
+                              local->rebalance.flags, local->rebalance.offset,
+                              local->rebalance.size, local->xattr_req);
+            break;
+
+        case GF_FOP_FTRUNCATE:
+            STACK_WIND_COOKIE(frame, dht_truncate_cbk, subvol, subvol,
+                              subvol->fops->ftruncate, fd,
+                              local->rebalance.offset, local->xattr_req);
+            break;
+
+        case GF_FOP_FSYNC:
+            STACK_WIND_COOKIE(frame, dht_fsync_cbk, subvol, subvol,
+                              subvol->fops->fsync, local->fd,
+                              local->rebalance.flags, local->xattr_req);
+            break;
+
+        case GF_FOP_READ:
+            STACK_WIND(frame, dht_readv_cbk, subvol, subvol->fops->readv,
+                       local->fd, local->rebalance.size,
+                       local->rebalance.offset, local->rebalance.flags,
+                       local->xattr_req);
+            break;
+
+        case GF_FOP_FSTAT:
+            STACK_WIND_COOKIE(frame, dht_file_attr_cbk, subvol, subvol,
+                              subvol->fops->fstat, fd, local->xattr_req);
+            break;
+
+        case GF_FOP_FSETXATTR:
+            STACK_WIND_COOKIE(frame, dht_file_setxattr_cbk, subvol, subvol,
+                              subvol->fops->fsetxattr, local->fd,
+                              local->rebalance.xattr, local->rebalance.flags,
+                              local->xattr_req);
+            break;
+
+        case GF_FOP_FREMOVEXATTR:
+            STACK_WIND_COOKIE(frame, dht_file_removexattr_cbk, subvol, subvol,
+                              subvol->fops->fremovexattr, local->fd, local->key,
+                              local->xattr_req);
+
+            break;
+
+        case GF_FOP_FXATTROP:
+            STACK_WIND(frame, dht_common_xattrop_cbk, subvol,
+                       subvol->fops->fxattrop, local->fd,
+                       local->rebalance.flags, local->rebalance.xattr,
+                       local->xattr_req);
+            break;
+
+        case GF_FOP_FGETXATTR:
+            STACK_WIND(frame, dht_getxattr_cbk, subvol, subvol->fops->fgetxattr,
+                       local->fd, local->key, NULL);
+            break;
+
+        case GF_FOP_FINODELK:
+            STACK_WIND(frame, dht_finodelk_cbk, subvol, subvol->fops->finodelk,
+                       local->key, local->fd, local->rebalance.lock_cmd,
+                       &local->rebalance.flock, local->xattr_req);
+            break;
+        default:
+            gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_UNKNOWN_FOP, "fd=%p",
+                    fd, "gfid=%s", uuid_utoa(fd->inode->gfid), "name=%s",
+                    subvol->name, NULL);
+            break;
+    }
+
+    goto out;
+
+    /* Could not open the fd on the dst. Unwind */
+
+handle_err:
+
+    switch (fop) {
+        case GF_FOP_WRITE:
+            DHT_STACK_UNWIND(writev, frame, -1, op_errno, NULL, NULL, NULL);
+            break;
+
+        case GF_FOP_FLUSH:
+            DHT_STACK_UNWIND(flush, frame, -1, op_errno, NULL);
+            break;
+
+        case GF_FOP_FSETATTR:
+            DHT_STACK_UNWIND(fsetattr, frame, -1, op_errno, NULL, NULL, NULL);
+            break;
+
+        case GF_FOP_ZEROFILL:
+            DHT_STACK_UNWIND(zerofill, frame, -1, op_errno, NULL, NULL, NULL);
+            break;
+
+        case GF_FOP_DISCARD:
+            DHT_STACK_UNWIND(discard, frame, -1, op_errno, NULL, NULL, NULL);
+            break;
+
+        case GF_FOP_FALLOCATE:
+            DHT_STACK_UNWIND(fallocate, frame, -1, op_errno, NULL, NULL, NULL);
+            break;
+
+        case GF_FOP_FTRUNCATE:
+            DHT_STACK_UNWIND(ftruncate, frame, -1, op_errno, NULL, NULL, NULL);
+            break;
+
+        case GF_FOP_FSYNC:
+            DHT_STACK_UNWIND(fsync, frame, -1, op_errno, NULL, NULL, NULL);
+            break;
+
+        case GF_FOP_READ:
+            DHT_STACK_UNWIND(readv, frame, -1, op_errno, NULL, 0, NULL, NULL,
+                             NULL);
+            break;
+
+        case GF_FOP_FSTAT:
+            DHT_STACK_UNWIND(fstat, frame, -1, op_errno, NULL, NULL);
+            break;
+
+        case GF_FOP_FSETXATTR:
+            DHT_STACK_UNWIND(fsetxattr, frame, -1, op_errno, NULL);
+            break;
+
+        case GF_FOP_FREMOVEXATTR:
+            DHT_STACK_UNWIND(fremovexattr, frame, -1, op_errno, NULL);
+            break;
+
+        case GF_FOP_FXATTROP:
+            DHT_STACK_UNWIND(fxattrop, frame, -1, op_errno, NULL, NULL);
+            break;
+
+        case GF_FOP_FGETXATTR:
+            DHT_STACK_UNWIND(fgetxattr, frame, -1, op_errno, NULL, NULL);
+            break;
+
+        case GF_FOP_FINODELK:
+            DHT_STACK_UNWIND(finodelk, frame, -1, op_errno, NULL);
+            break;
+
+        default:
+            gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_UNKNOWN_FOP, "fd=%p",
+                    fd, "gfid=%s", uuid_utoa(fd->inode->gfid), "name=%s",
+                    subvol->name, NULL);
+            break;
+    }
 
-#define BACKEND_D_OFF_BITS 63
-#define PRESENT_D_OFF_BITS 63
+out:
 
-#define ONE 1ULL
-#define MASK (~0ULL)
-#define PRESENT_MASK (MASK >> (64 - PRESENT_D_OFF_BITS))
-#define BACKEND_MASK (MASK >> (64 - BACKEND_D_OFF_BITS))
+    return 0;
+}
 
-#define TOP_BIT (ONE << (PRESENT_D_OFF_BITS - 1))
-#define SHIFT_BITS (max (0, (BACKEND_D_OFF_BITS - PRESENT_D_OFF_BITS + 1)))
+/* Check once again if the fd has been opened on the cached subvol.
+ * If not, open and update the fd_ctx.
+ */
 
 int
-dht_itransform (xlator_t *this, xlator_t *subvol, uint64_t x, uint64_t *y_p)
+dht_check_and_open_fd_on_subvol_task(void *data)
 {
-        dht_conf_t *conf = NULL;
-        int         cnt = 0;
-        int         max = 0;
-        uint64_t    y = 0;
-        uint64_t    hi_mask = 0;
-        uint64_t    off_mask = 0;
-        int         max_bits = 0;
-
-        if (x == ((uint64_t) -1)) {
-                y = (uint64_t) -1;
-                goto out;
-        }
+    loc_t loc = {
+        0,
+    };
+    int ret = -1;
+    call_frame_t *frame = NULL;
+    dht_local_t *local = NULL;
+    fd_t *fd = NULL;
+    xlator_t *this = NULL;
+    xlator_t *subvol = NULL;
+
+    frame = data;
+    local = frame->local;
+    this = THIS;
+    fd = local->fd;
+    subvol = local->cached_subvol;
+
+    local->fd_checked = _gf_true;
+
+    if (fd_is_anonymous(fd) || dht_fd_open_on_dst(this, fd, subvol)) {
+        ret = 0;
+        goto out;
+    }
 
-        conf = this->private;
-        if (!conf)
-                goto out;
+    gf_msg_debug(this->name, 0, "Opening fd (%p, flags=0%o) on file %s @ %s",
+                 fd, fd->flags, uuid_utoa(fd->inode->gfid), subvol->name);
 
-        max = conf->subvolume_cnt;
-        cnt = dht_subvol_cnt (this, subvol);
+    loc.inode = inode_ref(fd->inode);
+    gf_uuid_copy(loc.gfid, fd->inode->gfid);
 
-	if (max == 1) {
-		y = x;
-		goto out;
-	}
+    /* Open this on the dst subvol */
 
-        max_bits = dht_bits_for (max);
+    SYNCTASK_SETID(0, 0);
 
-        hi_mask = ~(PRESENT_MASK >> (max_bits + 1));
+    ret = syncop_open(subvol, &loc, (fd->flags & ~(O_CREAT | O_EXCL | O_TRUNC)),
+                      fd, NULL, NULL);
 
-        if (x & hi_mask) {
-                /* HUGE d_off */
-                off_mask = MASK << max_bits;
-                y = TOP_BIT | ((x >> SHIFT_BITS) & off_mask) | cnt;
+    if (ret < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_OPEN_FD_ON_DST_FAILED,
+                "fd=%p", fd, "flags=0%o", fd->flags, "gfid=%s",
+                uuid_utoa(fd->inode->gfid), "name=%s", subvol->name, NULL);
+        /* This can happen if the cached subvol was updated in the
+         * inode_ctx and the fd was opened on the new cached suvol
+         * after this fop was wound on the old cached subvol.
+         * As we do not close the fd on the old subvol (a leak)
+         * don't treat ENOENT as an error and allow the phase1/phase2
+         * checks to handle it.
+         */
+
+        if ((-ret != ENOENT) && (-ret != ESTALE)) {
+            local->op_errno = -ret;
+            ret = -1;
         } else {
-                /* small d_off */
-                y = ((x * max) + cnt);
+            ret = 0;
         }
 
+        local->op_errno = -ret;
+        ret = -1;
+
+    } else {
+        dht_fd_ctx_set(this, fd, subvol);
+    }
+
+    SYNCTASK_SETID(frame->root->uid, frame->root->gid);
 out:
-        if (y_p)
-                *y_p = y;
+    loc_wipe(&loc);
 
-        return 0;
+    return ret;
 }
 
 int
-dht_filter_loc_subvol_key (xlator_t *this, loc_t *loc, loc_t *new_loc,
-                           xlator_t **subvol)
+dht_check_and_open_fd_on_subvol(xlator_t *this, call_frame_t *frame)
 {
-        char          *new_name  = NULL;
-        char          *new_path  = NULL;
-        xlator_list_t *trav      = NULL;
-        char           key[1024] = {0,};
-        int            ret       = 0; /* not found */
-
-        /* Why do other tasks if first required 'char' itself is not there */
-        if (!new_loc || !loc || !loc->name || !strchr (loc->name, '@'))
-                goto out;
+    int ret = -1;
+    dht_local_t *local = NULL;
 
-        trav = this->children;
-        while (trav) {
-                snprintf (key, 1024, "*@%s:%s", this->name, trav->xlator->name);
-                if (fnmatch (key, loc->name, FNM_NOESCAPE) == 0) {
-                        new_name = GF_CALLOC(strlen (loc->name),
-                                             sizeof (char),
-                                             gf_common_mt_char);
-                        if (!new_name)
-                                goto out;
-                        if (fnmatch (key, loc->path, FNM_NOESCAPE) == 0) {
-                                new_path = GF_CALLOC(strlen (loc->path),
-                                                     sizeof (char),
-                                                     gf_common_mt_char);
-                                if (!new_path)
-                                        goto out;
-                                strncpy (new_path, loc->path, (strlen (loc->path) -
-                                                               strlen (key) + 1));
-                        }
-                        strncpy (new_name, loc->name, (strlen (loc->name) -
-                                                       strlen (key) + 1));
-
-                        if (new_loc) {
-                                new_loc->path   = ((new_path) ? new_path:
-                                                   gf_strdup (loc->path));
-                                new_loc->name   = new_name;
-                                new_loc->inode  = inode_ref (loc->inode);
-                                new_loc->parent = inode_ref (loc->parent);
-                        }
-                        *subvol         = trav->xlator;
-                        ret = 1;  /* success */
-                        goto out;
-                }
-                trav = trav->next;
-        }
-out:
-        if (!ret) {
-                /* !success */
-                GF_FREE (new_path);
-                GF_FREE (new_name);
-        }
-        return ret;
+    /*
+            if (dht_fd_open_on_dst (this, fd, subvol))
+                    goto out;
+    */
+    local = frame->local;
+
+    ret = synctask_new(this->ctx->env, dht_check_and_open_fd_on_subvol_task,
+                       dht_check_and_open_fd_on_subvol_complete, frame, frame);
+
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SYNCTASK_CREATE_FAILED,
+                "to-check-and-open fd=%p", local->fd, NULL);
+    }
+
+    return ret;
 }
 
 int
-dht_deitransform (xlator_t *this, uint64_t y, xlator_t **subvol_p,
-                  uint64_t *x_p)
+dht_frame_return(call_frame_t *frame)
 {
-        dht_conf_t *conf = NULL;
-        int         cnt = 0;
-        int         max = 0;
-        uint64_t    x = 0;
-        xlator_t   *subvol = 0;
-        int         max_bits = 0;
-        uint64_t    off_mask = 0;
-        uint64_t    host_mask = 0;
-
-        if (!this->private)
-                return -1;
-
-        conf = this->private;
-        max = conf->subvolume_cnt;
-
-	if (max == 1) {
-		x = y;
-		cnt = 0;
-		goto out;
-	}
-
-        if (y & TOP_BIT) {
-                /* HUGE d_off */
-                max_bits = dht_bits_for (max);
-                off_mask = (MASK << max_bits);
-                host_mask = ~(off_mask);
-
-                x = ((y & ~TOP_BIT) & off_mask) << SHIFT_BITS;
-
-                cnt = y & host_mask;
-	} else {
-                /* small d_off */
-                cnt = y % max;
-                x = y / max;
-        }
+    dht_local_t *local = NULL;
+    int this_call_cnt = -1;
+
+    if (!frame)
+        return -1;
 
+    local = frame->local;
+
+    LOCK(&frame->lock);
+    {
+        this_call_cnt = --local->call_cnt;
+    }
+    UNLOCK(&frame->lock);
+
+    return this_call_cnt;
+}
+
+/*
+ * Use this function to specify which subvol you want the file created
+ * on - this need not be the hashed subvol.
+ * Format: <filename>@<this->name>:<subvol-name>
+ * Eg: file-1@vol1-dht:vol1-client-0
+ *     where vol1 is a pure distribute volume
+ *     will create file-1 on vol1-client-0
+ */
+
+int
+dht_filter_loc_subvol_key(xlator_t *this, loc_t *loc, loc_t *new_loc,
+                          xlator_t **subvol)
+{
+    char *new_name = NULL;
+    char *new_path = NULL;
+    xlator_list_t *trav = NULL;
+    char key[1024] = {
+        0,
+    };
+    int ret = 0; /* not found */
+    int keylen = 0;
+    int name_len = 0;
+    int path_len = 0;
+
+    /* Why do other tasks if first required 'char' itself is not there */
+    if (!new_loc || !loc || !loc->name || !strchr(loc->name, '@')) {
+        /* Skip the GF_FREE checks here */
+        return ret;
+    }
+
+    trav = this->children;
+    while (trav) {
+        keylen = snprintf(key, sizeof(key), "*@%s:%s", this->name,
+                          trav->xlator->name);
+        /* Ignore '*' */
+        keylen = keylen - 1;
+        if (fnmatch(key, loc->name, FNM_NOESCAPE) == 0) {
+            name_len = strlen(loc->name) - keylen;
+            new_name = GF_MALLOC(name_len + 1, gf_common_mt_char);
+            if (!new_name)
+                goto out;
+            if (fnmatch(key, loc->path, FNM_NOESCAPE) == 0) {
+                path_len = strlen(loc->path) - keylen;
+                new_path = GF_MALLOC(path_len + 1, gf_common_mt_char);
+                if (!new_path)
+                    goto out;
+                snprintf(new_path, path_len + 1, "%s", loc->path);
+            }
+            snprintf(new_name, name_len + 1, "%s", loc->name);
+
+            if (new_loc) {
+                new_loc->path = ((new_path) ? new_path : gf_strdup(loc->path));
+                new_loc->name = new_name;
+                new_loc->inode = inode_ref(loc->inode);
+                new_loc->parent = inode_ref(loc->parent);
+            }
+            *subvol = trav->xlator;
+            ret = 1; /* success */
+            goto out;
+        }
+        trav = trav->next;
+    }
 out:
-        subvol = conf->subvolumes[cnt];
+    if (!ret) {
+        /* !success */
+        GF_FREE(new_path);
+        GF_FREE(new_name);
+    }
+    return ret;
+}
+
+static xlator_t *
+dht_get_subvol_from_id(xlator_t *this, int client_id)
+{
+    xlator_t *xl = NULL;
+    dht_conf_t *conf = NULL;
+    char *sid = NULL;
+    int32_t ret = -1;
 
-        if (subvol_p)
-                *subvol_p = subvol;
+    conf = this->private;
 
-        if (x_p)
-                *x_p = x;
+    ret = gf_asprintf(&sid, "%d", client_id);
+    if (ret == -1) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_ASPRINTF_FAILED, NULL);
+        goto out;
+    }
 
-        return 0;
+    if (dict_get_ptr(conf->leaf_to_subvol, sid, (void **)&xl))
+        xl = NULL;
+
+    GF_FREE(sid);
+
+out:
+    return xl;
 }
 
+int
+dht_deitransform(xlator_t *this, uint64_t y, xlator_t **subvol_p)
+{
+    int client_id = 0;
+    xlator_t *subvol = 0;
+    dht_conf_t *conf = NULL;
+
+    if (!this->private)
+        return -1;
+
+    conf = this->private;
+
+    client_id = gf_deitransform(this, y);
+
+    subvol = dht_get_subvol_from_id(this, client_id);
+
+    if (!subvol)
+        subvol = conf->subvolumes[0];
+
+    if (subvol_p)
+        *subvol_p = subvol;
+
+    return 0;
+}
 
 void
-dht_local_wipe (xlator_t *this, dht_local_t *local)
+dht_local_wipe(xlator_t *this, dht_local_t *local)
 {
-        if (!local)
-                return;
+    int i = 0;
 
-        loc_wipe (&local->loc);
-        loc_wipe (&local->loc2);
+    if (!local)
+        return;
 
-        if (local->xattr)
-                dict_unref (local->xattr);
+    loc_wipe(&local->loc);
+    loc_wipe(&local->loc2);
+    loc_wipe(&local->loc2_copy);
 
-        if (local->inode)
-                inode_unref (local->inode);
+    if (local->xattr)
+        dict_unref(local->xattr);
 
-        if (local->layout) {
-                dht_layout_unref (this, local->layout);
-                local->layout = NULL;
-        }
+    if (local->inode)
+        inode_unref(local->inode);
 
-        loc_wipe (&local->linkfile.loc);
+    if (local->layout) {
+        dht_layout_unref(this, local->layout);
+        local->layout = NULL;
+    }
 
-        if (local->linkfile.xattr)
-                dict_unref (local->linkfile.xattr);
+    loc_wipe(&local->linkfile.loc);
 
-        if (local->linkfile.inode)
-                inode_unref (local->linkfile.inode);
+    if (local->linkfile.xattr)
+        dict_unref(local->linkfile.xattr);
 
-        if (local->fd) {
-                fd_unref (local->fd);
-                local->fd = NULL;
-        }
+    if (local->linkfile.inode)
+        inode_unref(local->linkfile.inode);
 
-        if (local->params) {
-                dict_unref (local->params);
-                local->params = NULL;
-        }
+    if (local->fd) {
+        fd_unref(local->fd);
+        local->fd = NULL;
+    }
 
-        if (local->xattr_req)
-                dict_unref (local->xattr_req);
+    if (local->params) {
+        dict_unref(local->params);
+        local->params = NULL;
+    }
 
-        if (local->selfheal.layout) {
-                dht_layout_unref (this, local->selfheal.layout);
-                local->selfheal.layout = NULL;
-        }
+    if (local->xattr_req)
+        dict_unref(local->xattr_req);
+    if (local->mds_xattr)
+        dict_unref(local->mds_xattr);
+    if (local->xdata)
+        dict_unref(local->xdata);
 
-        GF_FREE (local->newpath);
+    if (local->selfheal.layout) {
+        dht_layout_unref(this, local->selfheal.layout);
+        local->selfheal.layout = NULL;
+    }
 
-        GF_FREE (local->key);
+    if (local->selfheal.refreshed_layout) {
+        dht_layout_unref(this, local->selfheal.refreshed_layout);
+        local->selfheal.refreshed_layout = NULL;
+    }
 
-        GF_FREE (local->rebalance.vector);
+    for (i = 0; i < 2; i++) {
+        dht_lock_array_free(local->lock[i].ns.parent_layout.locks,
+                            local->lock[i].ns.parent_layout.lk_count);
 
-        if (local->rebalance.iobref)
-                iobref_unref (local->rebalance.iobref);
+        GF_FREE(local->lock[i].ns.parent_layout.locks);
 
-        mem_put (local);
-}
+        dht_lock_array_free(local->lock[i].ns.directory_ns.locks,
+                            local->lock[i].ns.directory_ns.lk_count);
+        GF_FREE(local->lock[i].ns.directory_ns.locks);
+    }
 
+    GF_FREE(local->key);
 
-dht_local_t *
-dht_local_init (call_frame_t *frame, loc_t *loc, fd_t *fd, glusterfs_fop_t fop)
-{
-        dht_local_t *local = NULL;
-        inode_t     *inode = NULL;
-        int          ret   = 0;
+    if (local->rebalance.xdata)
+        dict_unref(local->rebalance.xdata);
 
-        local = mem_get0 (THIS->local_pool);
-        if (!local)
-                goto out;
+    if (local->rebalance.xattr)
+        dict_unref(local->rebalance.xattr);
 
-        if (loc) {
-                ret = loc_copy (&local->loc, loc);
-                if (ret)
-                        goto out;
+    if (local->rebalance.dict)
+        dict_unref(local->rebalance.dict);
 
-                inode = loc->inode;
-        }
+    GF_FREE(local->rebalance.vector);
 
-        if (fd) {
-                local->fd = fd_ref (fd);
-                if (!inode)
-                        inode = fd->inode;
-        }
+    if (local->rebalance.iobref)
+        iobref_unref(local->rebalance.iobref);
 
-        local->op_ret   = -1;
-        local->op_errno = EUCLEAN;
-        local->fop      = fop;
-
-        if (inode) {
-                local->layout   = dht_layout_get (frame->this, inode);
-                local->cached_subvol = dht_subvol_get_cached (frame->this,
-                                                              inode);
-        }
+    if (local->stub) {
+        call_stub_destroy(local->stub);
+        local->stub = NULL;
+    }
 
-        frame->local = local;
+    if (local->ret_cache)
+        GF_FREE(local->ret_cache);
 
-out:
-        if (ret) {
-                if (local)
-                        mem_put (local);
-                local = NULL;
-        }
-        return local;
+    mem_put(local);
 }
 
-
-char *
-basestr (const char *str)
+dht_local_t *
+dht_local_init(call_frame_t *frame, loc_t *loc, fd_t *fd, glusterfs_fop_t fop)
 {
-        char *basestr = NULL;
+    dht_local_t *local = NULL;
+    inode_t *inode = NULL;
+    int ret = 0;
 
-        basestr = strrchr (str, '/');
-        if (basestr)
-                basestr ++;
+    local = mem_get0(THIS->local_pool);
+    if (!local)
+        goto out;
 
-        return basestr;
-}
+    if (loc) {
+        ret = loc_copy(&local->loc, loc);
+        if (ret)
+            goto out;
+
+        inode = loc->inode;
+    }
+
+    if (fd) {
+        local->fd = fd_ref(fd);
+        if (!inode)
+            inode = fd->inode;
+    }
+
+    local->op_ret = -1;
+    local->op_errno = EUCLEAN;
+    local->fop = fop;
+
+    if (inode) {
+        local->layout = dht_layout_get(frame->this, inode);
+        local->cached_subvol = dht_subvol_get_cached(frame->this, inode);
+    }
 
+    frame->local = local;
+
+out:
+    if (ret) {
+        if (local)
+            mem_put(local);
+        local = NULL;
+    }
+    return local;
+}
 
 xlator_t *
-dht_first_up_subvol (xlator_t *this)
+dht_first_up_subvol(xlator_t *this)
 {
-        dht_conf_t *conf = NULL;
-        xlator_t   *child = NULL;
-        int         i = 0;
-        time_t      time = 0;
+    dht_conf_t *conf = NULL;
+    xlator_t *child = NULL;
+    int i = 0;
+    time_t time = 0;
 
-        conf = this->private;
-        if (!conf)
-                goto out;
+    conf = this->private;
+    if (!conf)
+        goto out;
 
-        LOCK (&conf->subvolume_lock);
-        {
-                for (i = 0; i < conf->subvolume_cnt; i++) {
-                        if (conf->subvol_up_time[i]) {
-                                if (!time) {
-                                        time = conf->subvol_up_time[i];
-                                        child = conf->subvolumes[i];
-                                } else if (time > conf->subvol_up_time[i]) {
-                                        time  = conf->subvol_up_time[i];
-                                        child = conf->subvolumes[i];
-                                }
-                        }
+    LOCK(&conf->subvolume_lock);
+    {
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            if (conf->subvol_up_time[i]) {
+                if (!time) {
+                    time = conf->subvol_up_time[i];
+                    child = conf->subvolumes[i];
+                } else if (time > conf->subvol_up_time[i]) {
+                    time = conf->subvol_up_time[i];
+                    child = conf->subvolumes[i];
                 }
+            }
         }
-        UNLOCK (&conf->subvolume_lock);
+    }
+    UNLOCK(&conf->subvolume_lock);
 
 out:
-        return child;
+    return child;
 }
 
 xlator_t *
-dht_last_up_subvol (xlator_t *this)
+dht_last_up_subvol(xlator_t *this)
 {
-        dht_conf_t *conf = NULL;
-        xlator_t   *child = NULL;
-        int         i = 0;
-
-        conf = this->private;
-        if (!conf)
-                goto out;
-
-        LOCK (&conf->subvolume_lock);
-        {
-                for (i = conf->subvolume_cnt-1; i >= 0; i--) {
-                        if (conf->subvolume_status[i]) {
-                                child = conf->subvolumes[i];
-                                break;
-                        }
-                }
+    dht_conf_t *conf = NULL;
+    xlator_t *child = NULL;
+    int i = 0;
+
+    conf = this->private;
+    if (!conf)
+        goto out;
+
+    LOCK(&conf->subvolume_lock);
+    {
+        for (i = conf->subvolume_cnt - 1; i >= 0; i--) {
+            if (conf->subvolume_status[i]) {
+                child = conf->subvolumes[i];
+                break;
+            }
         }
-        UNLOCK (&conf->subvolume_lock);
+    }
+    UNLOCK(&conf->subvolume_lock);
 
 out:
-        return child;
+    return child;
 }
 
 xlator_t *
-dht_subvol_get_hashed (xlator_t *this, loc_t *loc)
+dht_subvol_get_hashed(xlator_t *this, loc_t *loc)
 {
-        dht_layout_t *layout = NULL;
-        xlator_t     *subvol = NULL;
+    dht_layout_t *layout = NULL;
+    xlator_t *subvol = NULL;
+    dht_conf_t *conf = NULL;
+    dht_methods_t *methods = NULL;
 
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-        GF_VALIDATE_OR_GOTO (this->name, loc, out);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
 
-        if (__is_root_gfid (loc->gfid)) {
-                subvol = dht_first_up_subvol (this);
-                goto out;
-        }
+    conf = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, conf, out);
 
-        GF_VALIDATE_OR_GOTO (this->name, loc->parent, out);
-        GF_VALIDATE_OR_GOTO (this->name, loc->name, out);
+    methods = &(conf->methods);
 
-        layout = dht_layout_get (this, loc->parent);
+    if (__is_root_gfid(loc->gfid)) {
+        subvol = dht_first_up_subvol(this);
+        goto out;
+    }
 
-        if (!layout) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "layout missing path=%s parent=%s",
-                        loc->path, uuid_utoa (loc->parent->gfid));
-                goto out;
-        }
+    GF_VALIDATE_OR_GOTO(this->name, loc->parent, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->name, out);
 
-        subvol = dht_layout_search (this, layout, loc->name);
+    layout = dht_layout_get(this, loc->parent);
 
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "could not find subvolume for path=%s",
-                        loc->path);
-                goto out;
-        }
+    if (!layout) {
+        gf_msg_debug(this->name, 0, "Missing layout. path=%s, parent gfid =%s",
+                     loc->path, uuid_utoa(loc->parent->gfid));
+        goto out;
+    }
+
+    subvol = methods->layout_search(this, layout, loc->name);
+
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "No hashed subvolume for path=%s",
+                     loc->path);
+        goto out;
+    }
 
 out:
-        if (layout) {
-                dht_layout_unref (this, layout);
-        }
+    if (layout) {
+        dht_layout_unref(this, layout);
+    }
 
-        return subvol;
+    return subvol;
 }
 
-
 xlator_t *
-dht_subvol_get_cached (xlator_t *this, inode_t *inode)
+dht_subvol_get_cached(xlator_t *this, inode_t *inode)
 {
-        dht_layout_t *layout = NULL;
-        xlator_t     *subvol = NULL;
+    dht_layout_t *layout = NULL;
+    xlator_t *subvol = NULL;
 
-        GF_VALIDATE_OR_GOTO (this->name, this, out);
-        GF_VALIDATE_OR_GOTO (this->name, inode, out);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
 
-        layout = dht_layout_get (this, inode);
+    layout = dht_layout_get(this, inode);
 
-        if (!layout) {
-                goto out;
-        }
+    if (!layout) {
+        goto out;
+    }
 
-        subvol = layout->list[0].xlator;
+    subvol = layout->list[0].xlator;
 
 out:
-        if (layout) {
-                dht_layout_unref (this, layout);
-        }
+    if (layout) {
+        dht_layout_unref(this, layout);
+    }
 
-        return subvol;
+    return subvol;
 }
 
-
 xlator_t *
-dht_subvol_next (xlator_t *this, xlator_t *prev)
+dht_subvol_next(xlator_t *this, xlator_t *prev)
 {
-        dht_conf_t *conf = NULL;
-        int         i = 0;
-        xlator_t   *next = NULL;
-
-        conf = this->private;
-        if (!conf)
-                goto out;
-
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                if (conf->subvolumes[i] == prev) {
-                        if ((i + 1) < conf->subvolume_cnt)
-                                next = conf->subvolumes[i + 1];
-                        break;
-                }
+    dht_conf_t *conf = NULL;
+    int i = 0;
+    xlator_t *next = NULL;
+
+    conf = this->private;
+    if (!conf)
+        goto out;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (conf->subvolumes[i] == prev) {
+            if ((i + 1) < conf->subvolume_cnt)
+                next = conf->subvolumes[i + 1];
+            break;
         }
+    }
 
 out:
-        return next;
+    return next;
 }
 
 /* This func wraps around, if prev is actually the last subvol.
  */
 xlator_t *
-dht_subvol_next_available (xlator_t *this, xlator_t *prev)
+dht_subvol_next_available(xlator_t *this, xlator_t *prev)
 {
-        dht_conf_t *conf = NULL;
-        int         i = 0;
-        xlator_t   *next = NULL;
-
-        conf = this->private;
-        if (!conf)
-                goto out;
-
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                if (conf->subvolumes[i] == prev) {
-                        /* if prev is last in conf->subvolumes, then wrap
-                         * around.
-                         */
-                        if ((i + 1) < conf->subvolume_cnt) {
-                                next = conf->subvolumes[i + 1];
-                        } else {
-                                next = conf->subvolumes[0];
-                        }
-                        break;
-                }
+    dht_conf_t *conf = NULL;
+    int i = 0;
+    xlator_t *next = NULL;
+
+    conf = this->private;
+    if (!conf)
+        goto out;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (conf->subvolumes[i] == prev) {
+            /* if prev is last in conf->subvolumes, then wrap
+             * around.
+             */
+            if ((i + 1) < conf->subvolume_cnt) {
+                next = conf->subvolumes[i + 1];
+            } else {
+                next = conf->subvolumes[0];
+            }
+            break;
         }
+    }
 
 out:
-        return next;
+    return next;
 }
 int
-dht_subvol_cnt (xlator_t *this, xlator_t *subvol)
+dht_subvol_cnt(xlator_t *this, xlator_t *subvol)
 {
-        int i = 0;
-        int ret = -1;
-        dht_conf_t *conf = NULL;
-
-        conf = this->private;
-        if (!conf)
-                goto out;
-
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                if (subvol == conf->subvolumes[i]) {
-                        ret = i;
-                        break;
-                }
+    int i = 0;
+    int ret = -1;
+    dht_conf_t *conf = NULL;
+
+    conf = this->private;
+    if (!conf)
+        goto out;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (subvol == conf->subvolumes[i]) {
+            ret = i;
+            break;
         }
+    }
 
 out:
-        return ret;
+    return ret;
 }
 
+#define set_if_greater(a, b)                                                   \
+    do {                                                                       \
+        if ((a) < (b))                                                         \
+            (a) = (b);                                                         \
+    } while (0)
 
-#define set_if_greater(a, b) do {               \
-                if ((a) < (b))                  \
-                        (a) = (b);              \
-        } while (0)
-
+#define set_if_greater_time(a, an, b, bn)                                      \
+    do {                                                                       \
+        if (((a) < (b)) || (((a) == (b)) && ((an) < (bn)))) {                  \
+            (a) = (b);                                                         \
+            (an) = (bn);                                                       \
+        }                                                                      \
+    } while (0)
 
-#define set_if_greater_time(a, an, b, bn) do {                          \
-                if (((a) < (b)) || (((a) == (b)) && ((an) < (bn)))){    \
-                        (a) = (b);                                      \
-                        (an) = (bn);                                    \
-                }                                                       \
-        } while (0)                                                     \
+int
+dht_iatt_merge(xlator_t *this, struct iatt *to, struct iatt *from)
+{
+    if (!from || !to)
+        return 0;
 
+    to->ia_dev = from->ia_dev;
+
+    gf_uuid_copy(to->ia_gfid, from->ia_gfid);
+
+    to->ia_ino = from->ia_ino;
+    to->ia_prot = from->ia_prot;
+    to->ia_type = from->ia_type;
+    to->ia_nlink = from->ia_nlink;
+    to->ia_rdev = from->ia_rdev;
+    to->ia_size += from->ia_size;
+    to->ia_blksize = from->ia_blksize;
+    to->ia_blocks += from->ia_blocks;
+
+    if (IA_ISDIR(from->ia_type)) {
+        to->ia_blocks = DHT_DIR_STAT_BLOCKS;
+        to->ia_size = DHT_DIR_STAT_SIZE;
+    }
+    set_if_greater(to->ia_uid, from->ia_uid);
+    set_if_greater(to->ia_gid, from->ia_gid);
+
+    set_if_greater_time(to->ia_atime, to->ia_atime_nsec, from->ia_atime,
+                        from->ia_atime_nsec);
+    set_if_greater_time(to->ia_mtime, to->ia_mtime_nsec, from->ia_mtime,
+                        from->ia_mtime_nsec);
+    set_if_greater_time(to->ia_ctime, to->ia_ctime_nsec, from->ia_ctime,
+                        from->ia_ctime_nsec);
+
+    return 0;
+}
 
 int
-dht_iatt_merge (xlator_t *this, struct iatt *to,
-                struct iatt *from, xlator_t *subvol)
+dht_build_child_loc(xlator_t *this, loc_t *child, loc_t *parent, char *name)
 {
-        if (!from || !to)
-                return 0;
+    if (!child) {
+        goto err;
+    }
 
-        to->ia_dev      = from->ia_dev;
+    if (strcmp(parent->path, "/") == 0)
+        gf_asprintf((char **)&child->path, "/%s", name);
+    else
+        gf_asprintf((char **)&child->path, "%s/%s", parent->path, name);
 
-        uuid_copy (to->ia_gfid, from->ia_gfid);
+    if (!child->path) {
+        goto err;
+    }
 
-        to->ia_ino      = from->ia_ino;
-        to->ia_prot     = from->ia_prot;
-        to->ia_type     = from->ia_type;
-        to->ia_nlink    = from->ia_nlink;
-        to->ia_rdev     = from->ia_rdev;
-        to->ia_size    += from->ia_size;
-        to->ia_blksize  = from->ia_blksize;
-        to->ia_blocks  += from->ia_blocks;
+    child->name = strrchr(child->path, '/');
+    if (child->name)
+        child->name++;
 
-        set_if_greater (to->ia_uid, from->ia_uid);
-        set_if_greater (to->ia_gid, from->ia_gid);
+    child->parent = inode_ref(parent->inode);
+    child->inode = inode_new(parent->inode->table);
 
-        set_if_greater_time(to->ia_atime, to->ia_atime_nsec,
-                            from->ia_atime, from->ia_atime_nsec);
-        set_if_greater_time (to->ia_mtime, to->ia_mtime_nsec,
-                             from->ia_mtime, from->ia_mtime_nsec);
-        set_if_greater_time (to->ia_ctime, to->ia_ctime_nsec,
-                             from->ia_ctime, from->ia_ctime_nsec);
+    if (!child->inode) {
+        goto err;
+    }
 
-        return 0;
+    return 0;
+err:
+    if (child) {
+        loc_wipe(child);
+    }
+    return -1;
 }
 
 int
-dht_build_child_loc (xlator_t *this, loc_t *child, loc_t *parent, char *name)
+dht_init_local_subvolumes(xlator_t *this, dht_conf_t *conf)
 {
-        if (!child) {
-                goto err;
-        }
-
-        if (strcmp (parent->path, "/") == 0)
-                gf_asprintf ((char **)&child->path, "/%s", name);
-        else
-                gf_asprintf ((char **)&child->path, "%s/%s", parent->path, name);
+    xlator_list_t *subvols = NULL;
+    int cnt = 0;
 
-        if (!child->path) {
-                goto err;
-        }
+    if (!conf)
+        return -1;
 
-        child->name = strrchr (child->path, '/');
-        if (child->name)
-                child->name++;
+    for (subvols = this->children; subvols; subvols = subvols->next)
+        cnt++;
 
-        child->parent = inode_ref (parent->inode);
-        child->inode = inode_new (parent->inode->table);
+    conf->local_subvols = GF_CALLOC(cnt, sizeof(xlator_t *),
+                                    gf_dht_mt_xlator_t);
 
-        if (!child->inode) {
-                goto err;
-        }
+    /* FIX FIX : do this dynamically*/
+    conf->local_nodeuuids = GF_CALLOC(cnt, sizeof(subvol_nodeuuids_info_t),
+                                      gf_dht_nodeuuids_t);
 
-        return 0;
-err:
-        loc_wipe (child);
+    if (!conf->local_subvols || !conf->local_nodeuuids) {
         return -1;
-}
+    }
 
+    conf->local_subvols_cnt = 0;
 
+    return 0;
+}
 
 int
-dht_init_subvolumes (xlator_t *this, dht_conf_t *conf)
+dht_init_subvolumes(xlator_t *this, dht_conf_t *conf)
 {
-        xlator_list_t *subvols = NULL;
-        int            cnt = 0;
+    xlator_list_t *subvols = NULL;
+    int cnt = 0;
 
-        if (!conf)
-                return -1;
+    if (!conf)
+        return -1;
 
-        for (subvols = this->children; subvols; subvols = subvols->next)
-                cnt++;
+    for (subvols = this->children; subvols; subvols = subvols->next)
+        cnt++;
 
-        conf->subvolumes = GF_CALLOC (cnt, sizeof (xlator_t *),
-                                      gf_dht_mt_xlator_t);
-        if (!conf->subvolumes) {
-                return -1;
-        }
-        conf->subvolume_cnt = cnt;
+    conf->subvolumes = GF_CALLOC(cnt, sizeof(xlator_t *), gf_dht_mt_xlator_t);
+    if (!conf->subvolumes) {
+        return -1;
+    }
+    conf->subvolume_cnt = cnt;
+    /* Doesn't make sense to do any dht layer tasks
+       if the subvol count is 1. Set it as pass_through */
+    if (cnt == 1)
+        this->pass_through = _gf_true;
 
-        cnt = 0;
-        for (subvols = this->children; subvols; subvols = subvols->next)
-                conf->subvolumes[cnt++] = subvols->xlator;
+    conf->local_subvols_cnt = 0;
 
-        conf->subvolume_status = GF_CALLOC (cnt, sizeof (char),
-                                            gf_dht_mt_char);
-        if (!conf->subvolume_status) {
-                return -1;
-        }
+    dht_set_subvol_range(this);
 
-        conf->last_event = GF_CALLOC (cnt, sizeof (int),
-                                      gf_dht_mt_char);
-        if (!conf->last_event) {
-                return -1;
-        }
+    cnt = 0;
+    for (subvols = this->children; subvols; subvols = subvols->next)
+        conf->subvolumes[cnt++] = subvols->xlator;
 
-        conf->subvol_up_time = GF_CALLOC (cnt, sizeof (time_t),
-                                          gf_dht_mt_subvol_time);
-        if (!conf->subvol_up_time) {
-                return -1;
-        }
+    conf->subvolume_status = GF_CALLOC(cnt, sizeof(char), gf_dht_mt_char);
+    if (!conf->subvolume_status) {
+        return -1;
+    }
 
-        conf->du_stats = GF_CALLOC (conf->subvolume_cnt, sizeof (dht_du_t),
-                                    gf_dht_mt_dht_du_t);
-        if (!conf->du_stats) {
-                return -1;
-        }
+    conf->last_event = GF_CALLOC(cnt, sizeof(int), gf_dht_mt_char);
+    if (!conf->last_event) {
+        return -1;
+    }
 
-        conf->decommissioned_bricks = GF_CALLOC (cnt, sizeof (xlator_t *),
-                                                 gf_dht_mt_xlator_t);
-        if (!conf->decommissioned_bricks) {
-                return -1;
-        }
+    conf->subvol_up_time = GF_CALLOC(cnt, sizeof(time_t),
+                                     gf_dht_mt_subvol_time);
+    if (!conf->subvol_up_time) {
+        return -1;
+    }
 
-        return 0;
-}
+    conf->du_stats = GF_CALLOC(conf->subvolume_cnt, sizeof(dht_du_t),
+                               gf_dht_mt_dht_du_t);
+    if (!conf->du_stats) {
+        return -1;
+    }
 
+    conf->decommissioned_bricks = GF_CALLOC(cnt, sizeof(xlator_t *),
+                                            gf_dht_mt_xlator_t);
+    if (!conf->decommissioned_bricks) {
+        return -1;
+    }
 
+    return 0;
+}
 
+/*
+ op_ret values :
+  0 : Success.
+ -1 : Failure.
+  1 : File is being migrated but not by this DHT layer.
+*/
 
 static int
-dht_migration_complete_check_done (int op_ret, call_frame_t *frame, void *data)
+dht_migration_complete_check_done(int op_ret, call_frame_t *frame, void *data)
 {
-        dht_local_t *local = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *subvol = NULL;
 
-        local = frame->local;
+    local = frame->local;
 
-        local->rebalance.target_op_fn (THIS, frame, op_ret);
+    if (op_ret != 0)
+        goto out;
 
-        return 0;
-}
+    if (local->cached_subvol == NULL) {
+        local->op_errno = EINVAL;
+        goto out;
+    }
+
+    subvol = local->cached_subvol;
 
+out:
+    local->rebalance.target_op_fn(THIS, subvol, frame, op_ret);
+
+    return 0;
+}
 
 int
-dht_migration_complete_check_task (void *data)
+dht_migration_complete_check_task(void *data)
 {
-        int           ret      = -1;
-        xlator_t     *src_node = NULL;
-        xlator_t     *dst_node = NULL;
-        dht_local_t  *local    = NULL;
-        dict_t       *dict     = NULL;
-        dht_layout_t *layout   = NULL;
-        struct iatt   stbuf    = {0,};
-        xlator_t     *this     = NULL;
-        call_frame_t *frame    = NULL;
-        loc_t         tmp_loc  = {0,};
-        char         *path     = NULL;
-        dht_conf_t   *conf     = NULL;
-
-        this  = THIS;
-        frame = data;
-        local = frame->local;
-        conf = this->private;
-
-        src_node = local->cached_subvol;
-
-        if (!local->loc.inode && !local->fd)
-                goto out;
+    int ret = -1;
+    xlator_t *src_node = NULL;
+    xlator_t *dst_node = NULL, *linkto_target = NULL;
+    dht_local_t *local = NULL;
+    dict_t *dict = NULL;
+    struct iatt stbuf = {
+        0,
+    };
+    xlator_t *this = NULL;
+    call_frame_t *frame = NULL;
+    loc_t tmp_loc = {
+        0,
+    };
+    char *path = NULL;
+    dht_conf_t *conf = NULL;
+    inode_t *inode = NULL;
+    fd_t *iter_fd = NULL;
+    fd_t *tmp = NULL;
+    uint64_t tmp_miginfo = 0;
+    dht_migrate_info_t *miginfo = NULL;
+    gf_boolean_t skip_open = _gf_false;
+    int open_failed = 0;
+
+    this = THIS;
+    frame = data;
+    local = frame->local;
+    conf = this->private;
+
+    src_node = local->cached_subvol;
+
+    if (!local->loc.inode && !local->fd) {
+        local->op_errno = EINVAL;
+        goto out;
+    }
+
+    inode = (!local->fd) ? local->loc.inode : local->fd->inode;
+
+    /* getxattr on cached_subvol for 'linkto' value. Do path based getxattr
+     * as root:root. If a fd is already open, access check won't be done*/
+
+    if (!local->loc.inode) {
+        ret = syncop_fgetxattr(src_node, local->fd, &dict,
+                               conf->link_xattr_name, NULL, NULL);
+    } else {
+        SYNCTASK_SETID(0, 0);
+        ret = syncop_getxattr(src_node, &local->loc, &dict,
+                              conf->link_xattr_name, NULL, NULL);
+        SYNCTASK_SETID(frame->root->uid, frame->root->gid);
+    }
+
+    /*
+     * Each DHT xlator layer has its own name for the linkto xattr.
+     * If the file mode bits indicate the the file is being migrated but
+     * this layer's linkto xattr is not set, it means that another
+     * DHT layer is migrating the file. In this case, return 1 so
+     * the mode bits can be passed on to the higher layer for appropriate
+     * action.
+     */
+    if (-ret == ENODATA) {
+        /* This DHT translator is not migrating this file */
+
+        ret = inode_ctx_reset1(inode, this, &tmp_miginfo);
+        if (tmp_miginfo) {
+            /* This can be a problem if the file was
+             * migrated by two different layers. Raise
+             * a warning here.
+             */
+            gf_smsg(
+                this->name, GF_LOG_WARNING, 0, DHT_MSG_HAS_MIGINFO, "tmp=%s",
+                tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid), NULL);
+
+            miginfo = (void *)(uintptr_t)tmp_miginfo;
+            GF_REF_PUT(miginfo);
+        }
+        ret = 1;
+        goto out;
+    }
+
+    if (!ret)
+        linkto_target = dht_linkfile_subvol(this, NULL, NULL, dict);
+
+    if (local->loc.inode) {
+        loc_copy(&tmp_loc, &local->loc);
+    } else {
+        tmp_loc.inode = inode_ref(inode);
+        gf_uuid_copy(tmp_loc.gfid, inode->gfid);
+    }
+
+    ret = syncop_lookup(this, &tmp_loc, &stbuf, 0, 0, 0);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_FILE_LOOKUP_FAILED,
+                "tmp=%s", tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid),
+                "name=%s", this->name, NULL);
+        local->op_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    dst_node = dht_subvol_get_cached(this, tmp_loc.inode);
+    if (linkto_target && dst_node != linkto_target) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_INVALID_LINKFILE,
+                "linkto_target_name=%s", linkto_target->name, "dst_name=%s",
+                dst_node->name, NULL);
+    }
+
+    if (gf_uuid_compare(stbuf.ia_gfid, tmp_loc.inode->gfid)) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_GFID_MISMATCH, "tmp=%s",
+                tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid),
+                "dst_name=%s", dst_node->name, NULL);
+        ret = -1;
+        local->op_errno = EIO;
+        goto out;
+    }
+
+    /* update local. A layout is set in inode-ctx in lookup already */
+
+    dht_layout_unref(this, local->layout);
+
+    local->layout = dht_layout_get(frame->this, inode);
+    local->cached_subvol = dst_node;
+
+    ret = 0;
+
+    /* once we detect the migration complete, the inode-ctx2 is no more
+       required.. delete the ctx and also, it means, open() already
+       done on all the fd of inode */
+    ret = inode_ctx_reset1(inode, this, &tmp_miginfo);
+    if (tmp_miginfo) {
+        miginfo = (void *)(uintptr_t)tmp_miginfo;
+        GF_REF_PUT(miginfo);
+        goto out;
+    }
+
+    /* perform 'open()' on all the fd's present on the inode */
+    if (tmp_loc.path == NULL) {
+        inode_path(inode, NULL, &path);
+        if (path)
+            tmp_loc.path = path;
+    }
+
+    LOCK(&inode->lock);
+
+    if (list_empty(&inode->fd_list))
+        goto unlock;
+
+    /* perform open as root:root. There is window between linkfile
+     * creation(root:root) and setattr with the correct uid/gid
+     */
+    SYNCTASK_SETID(0, 0);
+
+    /* It's possible that we are the last user of iter_fd after each
+     * iteration. In this case the fd_unref() of iter_fd at the end of
+     * the loop will cause the destruction of the fd. So we need to
+     * iterate the list safely because iter_fd cannot be trusted.
+     */
+    iter_fd = list_entry((&inode->fd_list)->next, typeof(*iter_fd), inode_list);
+    while (&iter_fd->inode_list != (&inode->fd_list)) {
+        if (fd_is_anonymous(iter_fd) ||
+            (dht_fd_open_on_dst(this, iter_fd, dst_node))) {
+            if (!tmp) {
+                iter_fd = list_entry(iter_fd->inode_list.next, typeof(*iter_fd),
+                                     inode_list);
+                continue;
+            }
+            skip_open = _gf_true;
+        }
+        /* We need to release the inode->lock before calling
+         * syncop_open() to avoid possible deadlocks. However this
+         * can cause the iter_fd to be released by other threads.
+         * To avoid this, we take a reference before releasing the
+         * lock.
+         */
+        fd_ref(iter_fd);
 
-        /* getxattr on cached_subvol for 'linkto' value. Do path based getxattr
-         * as root:root. If a fd is already open, access check wont be done*/
+        UNLOCK(&inode->lock);
 
-        if (!local->loc.inode) {
-                ret = syncop_fgetxattr (src_node, local->fd, &dict,
-                                        conf->link_xattr_name);
+        if (tmp) {
+            fd_unref(tmp);
+            tmp = NULL;
+        }
+        if (skip_open)
+            goto next;
+
+        /* flags for open are stripped down to allow following the
+         * new location of the file, otherwise we can get EEXIST or
+         * truncate the file again as rebalance is moving the data */
+        ret = syncop_open(dst_node, &tmp_loc,
+                          (iter_fd->flags & ~(O_CREAT | O_EXCL | O_TRUNC)),
+                          iter_fd, NULL, NULL);
+        if (ret < 0) {
+            gf_smsg(this->name, GF_LOG_ERROR, -ret,
+                    DHT_MSG_OPEN_FD_ON_DST_FAILED, "id=%p", iter_fd,
+                    "flags=0%o", iter_fd->flags, "path=%s", path, "name=%s",
+                    dst_node->name, NULL);
+
+            open_failed = 1;
+            local->op_errno = -ret;
+            ret = -1;
         } else {
-                SYNCTASK_SETID (0, 0);
-                ret = syncop_getxattr (src_node, &local->loc, &dict,
-                                       conf->link_xattr_name);
-                SYNCTASK_SETID (frame->root->uid, frame->root->gid);
+            dht_fd_ctx_set(this, iter_fd, dst_node);
         }
 
-        if (!ret)
-                dst_node = dht_linkfile_subvol (this, NULL, NULL, dict);
+    next:
+        LOCK(&inode->lock);
+        skip_open = _gf_false;
+        tmp = iter_fd;
+        iter_fd = list_entry(tmp->inode_list.next, typeof(*tmp), inode_list);
+    }
 
-        if (ret) {
-                if ((errno != ENOENT) || (!local->loc.inode)) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "%s: failed to get the 'linkto' xattr %s",
-                                local->loc.path, strerror (errno));
-                        goto out;
-                }
-                /* Need to do lookup on hashed subvol, then get the file */
-                ret = syncop_lookup (this, &local->loc, NULL, &stbuf, NULL,
-                                     NULL);
-                if (ret)
-                        goto out;
-                dst_node = dht_subvol_get_cached (this, local->loc.inode);
-        }
+    SYNCTASK_SETID(frame->root->uid, frame->root->gid);
 
-        if (!dst_node) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s: failed to get the destination node",
-                        local->loc.path);
-                ret = -1;
-                goto out;
-        }
+    if (open_failed) {
+        ret = -1;
+        goto unlock;
+    }
+    ret = 0;
 
-        /* lookup on dst */
-        if (local->loc.inode) {
-                ret = syncop_lookup (dst_node, &local->loc, NULL, &stbuf, NULL, NULL);
+unlock:
+    UNLOCK(&inode->lock);
+    if (tmp) {
+        fd_unref(tmp);
+        tmp = NULL;
+    }
 
-                if (ret) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "%s: failed to lookup the file on %s",
-                                local->loc.path, dst_node->name);
-                        goto out;
-                }
+out:
+    if (dict) {
+        dict_unref(dict);
+    }
 
-                if (uuid_compare (stbuf.ia_gfid, local->loc.inode->gfid)) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "%s: gfid different on the target file on %s",
-                                local->loc.path, dst_node->name);
-                        ret = -1;
-                        goto out;
-                }
-        }
+    loc_wipe(&tmp_loc);
 
-        /* update inode ctx (the layout) */
-        dht_layout_unref (this, local->layout);
-
-        if (!local->loc.inode)
-                ret = dht_layout_preset (this, dst_node, local->fd->inode);
-        else
-                ret = dht_layout_preset (this, dst_node, local->loc.inode);
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "%s: could not set preset layout for subvol %s",
-                        local->loc.path, dst_node->name);
-                ret   = -1;
-                goto out;
-        }
+    return ret;
+}
 
-        layout = dht_layout_for_subvol (this, dst_node);
-        if (!layout) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "%s: no pre-set layout for subvolume %s",
-                        local->loc.path, dst_node ? dst_node->name : "<nil>");
-                ret = -1;
-                goto out;
-        }
+int
+dht_rebalance_complete_check(xlator_t *this, call_frame_t *frame)
+{
+    int ret = -1;
 
-        if (!local->loc.inode)
-                ret = dht_layout_set (this, local->fd->inode, layout);
-        else
-                ret = dht_layout_set (this, local->loc.inode, layout);
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s: failed to set the new layout",
-                        local->loc.path);
-                goto out;
+    ret = synctask_new(this->ctx->env, dht_migration_complete_check_task,
+                       dht_migration_complete_check_done, frame, frame);
+    return ret;
+}
+
+/* During 'in-progress' state, both nodes should have the file */
+/*
+ op_ret values :
+  0 : Success
+ -1 : Failure.
+  1 : File is being migrated but not by this DHT layer.
+*/
+static int
+dht_inprogress_check_done(int op_ret, call_frame_t *frame, void *data)
+{
+    dht_local_t *local = NULL;
+    xlator_t *dst_subvol = NULL, *src_subvol = NULL;
+    inode_t *inode = NULL;
+
+    local = frame->local;
+
+    if (op_ret != 0)
+        goto out;
+
+    inode = local->loc.inode ? local->loc.inode : local->fd->inode;
+
+    dht_inode_ctx_get_mig_info(THIS, inode, &src_subvol, &dst_subvol);
+    if (dht_mig_info_is_invalid(local->cached_subvol, src_subvol, dst_subvol)) {
+        dst_subvol = dht_subvol_get_cached(THIS, inode);
+        if (!dst_subvol) {
+            local->op_errno = EINVAL;
+            goto out;
         }
+    }
 
-        local->cached_subvol = dst_node;
-        ret = 0;
+out:
+    local->rebalance.target_op_fn(THIS, dst_subvol, frame, op_ret);
 
-        if (!local->fd)
-                goto out;
-        /* once we detect the migration complete, the fd-ctx is no more
-           required.. delete the ctx */
-        ret = fd_ctx_del (local->fd, this, NULL);
-        if (!ret)
-                goto out;
+    return 0;
+}
 
-        /* perform open as root:root. There is window between linkfile
-         * creation(root:root) and setattr with the correct uid/gid
-         */
+static int
+dht_rebalance_inprogress_task(void *data)
+{
+    int ret = -1;
+    xlator_t *src_node = NULL;
+    xlator_t *dst_node = NULL;
+    dht_local_t *local = NULL;
+    dict_t *dict = NULL;
+    call_frame_t *frame = NULL;
+    xlator_t *this = NULL;
+    char *path = NULL;
+    struct iatt stbuf = {
+        0,
+    };
+    loc_t tmp_loc = {
+        0,
+    };
+    dht_conf_t *conf = NULL;
+    inode_t *inode = NULL;
+    fd_t *iter_fd = NULL;
+    fd_t *tmp = NULL;
+    int open_failed = 0;
+    uint64_t tmp_miginfo = 0;
+    dht_migrate_info_t *miginfo = NULL;
+    gf_boolean_t skip_open = _gf_false;
+
+    this = THIS;
+    frame = data;
+    local = frame->local;
+    conf = this->private;
+
+    src_node = local->cached_subvol;
+
+    if (!local->loc.inode && !local->fd)
+        goto out;
+
+    inode = (!local->fd) ? local->loc.inode : local->fd->inode;
+
+    /* getxattr on cached_subvol for 'linkto' value. Do path based getxattr
+     * as root:root. If a fd is already open, access check won't be done*/
+    if (local->loc.inode) {
         SYNCTASK_SETID(0, 0);
-        /* if 'local->fd' (ie, fd based operation), send a 'open()' on
-           destination if not already done */
-        if (local->loc.inode) {
-                ret = syncop_open (dst_node, &local->loc,
-                                   local->fd->flags, local->fd);
-        } else {
-                tmp_loc.inode = local->fd->inode;
-                inode_path (local->fd->inode, NULL, &path);
-                if (path)
-                        tmp_loc.path = path;
-                ret = syncop_open (dst_node, &tmp_loc,
-                                   local->fd->flags, local->fd);
-                GF_FREE (path);
+        ret = syncop_getxattr(src_node, &local->loc, &dict,
+                              conf->link_xattr_name, NULL, NULL);
+        SYNCTASK_SETID(frame->root->uid, frame->root->gid);
+    } else {
+        ret = syncop_fgetxattr(src_node, local->fd, &dict,
+                               conf->link_xattr_name, NULL, NULL);
+    }
+
+    /*
+     * Each DHT xlator layer has its own name for the linkto xattr.
+     * If the file mode bits indicate the the file is being migrated but
+     * this layer's linkto xattr is not present, it means that another
+     * DHT layer is migrating the file. In this case, return 1 so
+     * the mode bits can be passed on to the higher layer for appropriate
+     * action.
+     */
+
+    if (-ret == ENODATA) {
+        /* This DHT layer is not migrating this file */
+        ret = inode_ctx_reset1(inode, this, &tmp_miginfo);
+        if (tmp_miginfo) {
+            /* This can be a problem if the file was
+             * migrated by two different layers. Raise
+             * a warning here.
+             */
+            gf_smsg(
+                this->name, GF_LOG_WARNING, 0, DHT_MSG_HAS_MIGINFO, "tmp=%s",
+                tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid), NULL);
+            miginfo = (void *)(uintptr_t)tmp_miginfo;
+            GF_REF_PUT(miginfo);
+        }
+        ret = 1;
+        goto out;
+    }
+
+    if (ret < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_GET_XATTR_FAILED,
+                "path=%s", local->loc.path, NULL);
+        ret = -1;
+        goto out;
+    }
+
+    dst_node = dht_linkfile_subvol(this, NULL, NULL, dict);
+    if (!dst_node) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_GET_XATTR_FAILED,
+                "path=%s", local->loc.path, NULL);
+        ret = -1;
+        goto out;
+    }
+
+    local->rebalance.target_node = dst_node;
+
+    if (local->loc.inode) {
+        loc_copy(&tmp_loc, &local->loc);
+    } else {
+        tmp_loc.inode = inode_ref(inode);
+        gf_uuid_copy(tmp_loc.gfid, inode->gfid);
+    }
+
+    /* lookup on dst */
+    ret = syncop_lookup(dst_node, &tmp_loc, &stbuf, NULL, NULL, NULL);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_FILE_LOOKUP_FAILED,
+                "tmp=%s", tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid),
+                "name=%s", dst_node->name, NULL);
+        ret = -1;
+        goto out;
+    }
+
+    if (gf_uuid_compare(stbuf.ia_gfid, tmp_loc.inode->gfid)) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_GFID_MISMATCH, "tmp=%s",
+                tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid),
+                "name=%s", dst_node->name, NULL);
+        ret = -1;
+        goto out;
+    }
+    ret = 0;
+
+    if (tmp_loc.path == NULL) {
+        inode_path(inode, NULL, &path);
+        if (path)
+            tmp_loc.path = path;
+    }
+
+    LOCK(&inode->lock);
+
+    if (list_empty(&inode->fd_list))
+        goto unlock;
+
+    /* perform open as root:root. There is window between linkfile
+     * creation(root:root) and setattr with the correct uid/gid
+     */
+    SYNCTASK_SETID(0, 0);
+
+    /* It's possible that we are the last user of iter_fd after each
+     * iteration. In this case the fd_unref() of iter_fd at the end of
+     * the loop will cause the destruction of the fd. So we need to
+     * iterate the list safely because iter_fd cannot be trusted.
+     */
+    iter_fd = list_entry((&inode->fd_list)->next, typeof(*iter_fd), inode_list);
+    while (&iter_fd->inode_list != (&inode->fd_list)) {
+        /* We need to release the inode->lock before calling
+         * syncop_open() to avoid possible deadlocks. However this
+         * can cause the iter_fd to be released by other threads.
+         * To avoid this, we take a reference before releasing the
+         * lock.
+         */
 
+        if (fd_is_anonymous(iter_fd) ||
+            (dht_fd_open_on_dst(this, iter_fd, dst_node))) {
+            if (!tmp) {
+                iter_fd = list_entry(iter_fd->inode_list.next, typeof(*iter_fd),
+                                     inode_list);
+                continue;
+            }
+            skip_open = _gf_true;
         }
-        SYNCTASK_SETID (frame->root->uid, frame->root->gid);
-        if (ret == -1) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s: failed to send open() on target file at %s",
-                        local->loc.path, dst_node->name);
-                goto out;
+
+        /* Yes, this is ugly but there isn't a cleaner way to do this
+         * the fd_ref is an atomic increment so not too bad. We want to
+         * reduce the number of inode locks and unlocks.
+         */
+
+        fd_ref(iter_fd);
+        UNLOCK(&inode->lock);
+
+        if (tmp) {
+            fd_unref(tmp);
+            tmp = NULL;
+        }
+        if (skip_open)
+            goto next;
+
+        /* flags for open are stripped down to allow following the
+         * new location of the file, otherwise we can get EEXIST or
+         * truncate the file again as rebalance is moving the data */
+        ret = syncop_open(dst_node, &tmp_loc,
+                          (iter_fd->flags & ~(O_CREAT | O_EXCL | O_TRUNC)),
+                          iter_fd, NULL, NULL);
+        if (ret < 0) {
+            gf_smsg(this->name, GF_LOG_ERROR, -ret,
+                    DHT_MSG_OPEN_FD_ON_DST_FAILED, "fd=%p", iter_fd,
+                    "flags=0%o", iter_fd->flags, "path=%s", path, "name=%s",
+                    dst_node->name, NULL);
+            ret = -1;
+            open_failed = 1;
+        } else {
+            /* Potential fd leak if this fails here as it will be
+               reopened at the next Phase1/2 check */
+            dht_fd_ctx_set(this, iter_fd, dst_node);
         }
 
-        ret = 0;
+    next:
+        LOCK(&inode->lock);
+        skip_open = _gf_false;
+        tmp = iter_fd;
+        iter_fd = list_entry(tmp->inode_list.next, typeof(*tmp), inode_list);
+    }
+
+    SYNCTASK_SETID(frame->root->uid, frame->root->gid);
+
+unlock:
+    UNLOCK(&inode->lock);
+
+    if (tmp) {
+        fd_unref(tmp);
+        tmp = NULL;
+    }
+    if (open_failed) {
+        ret = -1;
+        goto out;
+    }
+
+    ret = dht_inode_ctx_set_mig_info(this, inode, src_node, dst_node);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SET_INODE_CTX_FAILED,
+                "path=%s", local->loc.path, "name=%s", dst_node->name, NULL);
+        goto out;
+    }
+
+    ret = 0;
 out:
+    if (dict) {
+        dict_unref(dict);
+    }
 
-        return ret;
+    loc_wipe(&tmp_loc);
+    return ret;
 }
 
 int
-dht_rebalance_complete_check (xlator_t *this, call_frame_t *frame)
+dht_rebalance_in_progress_check(xlator_t *this, call_frame_t *frame)
 {
-        int         ret     = -1;
+    int ret = -1;
 
-        ret = synctask_new (this->ctx->env, dht_migration_complete_check_task,
-                            dht_migration_complete_check_done,
-                            frame, frame);
-        return ret;
+    ret = synctask_new(this->ctx->env, dht_rebalance_inprogress_task,
+                       dht_inprogress_check_done, frame, frame);
+    return ret;
 }
 
-/* During 'in-progress' state, both nodes should have the file */
-static int
-dht_inprogress_check_done (int op_ret, call_frame_t *sync_frame, void *data)
+int
+dht_inode_ctx_layout_set(inode_t *inode, xlator_t *this,
+                         dht_layout_t *layout_int)
 {
-        dht_local_t *local = NULL;
-
-        local = sync_frame->local;
-
-        local->rebalance.target_op_fn (THIS, sync_frame, op_ret);
-
-        return 0;
+    dht_inode_ctx_t *ctx = NULL;
+    int ret = -1;
+
+    ret = dht_inode_ctx_get(inode, this, &ctx);
+    if (!ret && ctx) {
+        ctx->layout = layout_int;
+    } else {
+        ctx = GF_CALLOC(1, sizeof(*ctx), gf_dht_mt_inode_ctx_t);
+        if (!ctx)
+            return ret;
+        ctx->layout = layout_int;
+    }
+
+    ret = dht_inode_ctx_set(inode, this, ctx);
+
+    return ret;
 }
 
-static int
-dht_rebalance_inprogress_task (void *data)
+void
+dht_inode_ctx_time_set(inode_t *inode, xlator_t *this, struct iatt *stat)
 {
-        int           ret      = -1;
-        xlator_t     *src_node = NULL;
-        xlator_t     *dst_node = NULL;
-        dht_local_t  *local    = NULL;
-        dict_t       *dict     = NULL;
-        call_frame_t *frame    = NULL;
-        xlator_t     *this     = NULL;
-        char         *path     = NULL;
-        struct iatt   stbuf    = {0,};
-        loc_t         tmp_loc  = {0,};
-        dht_conf_t   *conf     = NULL;
-
-        this  = THIS;
-        frame = data;
-        local = frame->local;
-        conf = this->private;
-
-        src_node = local->cached_subvol;
-
-        if (!local->loc.inode && !local->fd)
-                goto out;
+    dht_inode_ctx_t *ctx = NULL;
+    dht_stat_time_t *time = 0;
+    int ret = -1;
 
-        /* getxattr on cached_subvol for 'linkto' value. Do path based getxattr
-         * as root:root. If a fd is already open, access check wont be done*/
-        if (local->loc.inode) {
-                SYNCTASK_SETID (0, 0);
-                ret = syncop_getxattr (src_node, &local->loc, &dict,
-                                       conf->link_xattr_name);
-                SYNCTASK_SETID (frame->root->uid, frame->root->gid);
-        } else {
-                ret = syncop_fgetxattr (src_node, local->fd, &dict,
-                                        conf->link_xattr_name);
-        }
+    ret = dht_inode_ctx_get(inode, this, &ctx);
 
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s: failed to get the 'linkto' xattr %s",
-                        local->loc.path, strerror (errno));
-                        goto out;
-        }
+    if (ret)
+        return;
 
-        dst_node = dht_linkfile_subvol (this, NULL, NULL, dict);
-        if (!dst_node) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s: failed to get the 'linkto' xattr from dict",
-                        local->loc.path);
-                ret = -1;
-                goto out;
-        }
+    time = &ctx->time;
 
-        local->rebalance.target_node = dst_node;
-
-        if (local->loc.inode) {
-                /* lookup on dst */
-                ret = syncop_lookup (dst_node, &local->loc, NULL,
-                                     &stbuf, NULL, NULL);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "%s: failed to lookup the file on %s",
-                                local->loc.path, dst_node->name);
-                        goto out;
-                }
+    time->mtime = stat->ia_mtime;
+    time->mtime_nsec = stat->ia_mtime_nsec;
 
-                if (uuid_compare (stbuf.ia_gfid, local->loc.inode->gfid)) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "%s: gfid different on the target file on %s",
-                                local->loc.path, dst_node->name);
-                        ret = -1;
-                        goto out;
-                }
-        }
+    time->ctime = stat->ia_ctime;
+    time->ctime_nsec = stat->ia_ctime_nsec;
 
-        ret = 0;
-        /* perform open as root:root. There is window between linkfile
-         * creation(root:root) and setattr with the correct uid/gid
-         */
-        SYNCTASK_SETID (0, 0);
-        if (local->loc.inode) {
-                ret = syncop_open (dst_node, &local->loc,
-                                   local->fd->flags, local->fd);
-        } else {
-                tmp_loc.inode = local->fd->inode;
-                inode_path (local->fd->inode, NULL, &path);
-                if (path)
-                        tmp_loc.path = path;
-                ret = syncop_open (dst_node, &tmp_loc,
-                                   local->fd->flags, local->fd);
-                GF_FREE (path);
-        }
+    time->atime = stat->ia_atime;
+    time->atime_nsec = stat->ia_atime_nsec;
 
-        if (ret == -1) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s: failed to send open() on target file at %s",
-                        local->loc.path, dst_node->name);
-                goto out;
-        }
-
-        SYNCTASK_SETID (frame->root->uid, frame->root->gid);
-        ret = fd_ctx_set (local->fd, this, (uint64_t)(long)dst_node);
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s: failed to set fd-ctx target file at %s",
-                        local->loc.path, dst_node->name);
-                goto out;
-        }
+    return;
+}
 
-        ret = 0;
+int
+dht_inode_ctx_time_update(inode_t *inode, xlator_t *this, struct iatt *stat,
+                          int32_t post)
+{
+    dht_inode_ctx_t *ctx = NULL;
+    dht_stat_time_t *time = 0;
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO(this->name, stat, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+
+    ret = dht_inode_ctx_get(inode, this, &ctx);
+
+    if (ret) {
+        ctx = GF_CALLOC(1, sizeof(*ctx), gf_dht_mt_inode_ctx_t);
+        if (!ctx)
+            return -1;
+    }
+
+    time = &ctx->time;
+
+    LOCK(&inode->lock);
+    {
+        DHT_UPDATE_TIME(time->mtime, time->mtime_nsec, stat->ia_mtime,
+                        stat->ia_mtime_nsec, post);
+        DHT_UPDATE_TIME(time->ctime, time->ctime_nsec, stat->ia_ctime,
+                        stat->ia_ctime_nsec, post);
+        DHT_UPDATE_TIME(time->atime, time->atime_nsec, stat->ia_atime,
+                        stat->ia_atime_nsec, post);
+    }
+    UNLOCK(&inode->lock);
+
+    ret = dht_inode_ctx_set(inode, this, ctx);
 out:
-        return ret;
+    return 0;
 }
 
 int
-dht_rebalance_in_progress_check (xlator_t *this, call_frame_t *frame)
+dht_inode_ctx_get(inode_t *inode, xlator_t *this, dht_inode_ctx_t **ctx)
 {
+    int ret = -1;
+    uint64_t ctx_int = 0;
 
-        int         ret     = -1;
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
 
-        ret = synctask_new (this->ctx->env, dht_rebalance_inprogress_task,
-                            dht_inprogress_check_done,
-                            frame, frame);
+    ret = inode_ctx_get(inode, this, &ctx_int);
+
+    if (ret)
         return ret;
+
+    if (ctx)
+        *ctx = (dht_inode_ctx_t *)(uintptr_t)ctx_int;
+out:
+    return ret;
 }
 
 int
-dht_inode_ctx_layout_set (inode_t *inode, xlator_t *this,
-                          dht_layout_t *layout_int)
+dht_inode_ctx_set(inode_t *inode, xlator_t *this, dht_inode_ctx_t *ctx)
 {
-        dht_inode_ctx_t         *ctx            = NULL;
-        int                      ret            = -1;
-
-        ret = dht_inode_ctx_get (inode, this, &ctx);
-        if (!ret && ctx) {
-                ctx->layout = layout_int;
-        } else {
-                ctx = GF_CALLOC (1, sizeof (*ctx), gf_dht_mt_inode_ctx_t);
-                if (!ctx)
-                        return ret;
-                ctx->layout = layout_int;
-        }
+    int ret = -1;
+    uint64_t ctx_int = 0;
 
-        ret = dht_inode_ctx_set (inode, this, ctx);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+    GF_VALIDATE_OR_GOTO(this->name, ctx, out);
 
-        return ret;
+    ctx_int = (long)ctx;
+    ret = inode_ctx_set(inode, this, &ctx_int);
+out:
+    return ret;
 }
 
 int
-dht_inode_ctx_time_update (inode_t *inode, xlator_t *this, struct iatt *stat,
-                           int32_t post)
+dht_subvol_status(dht_conf_t *conf, xlator_t *subvol)
 {
-        dht_inode_ctx_t         *ctx            = NULL;
-        dht_stat_time_t         *time           = 0;
-        int                      ret            = -1;
+    int i;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (conf->subvolumes[i] == subvol) {
+            return conf->subvolume_status[i];
+        }
+    }
+    return 0;
+}
 
-        GF_VALIDATE_OR_GOTO (this->name, stat, out);
-        GF_VALIDATE_OR_GOTO (this->name, inode, out);
+inode_t *
+dht_heal_path(xlator_t *this, char *path, inode_table_t *itable)
+{
+    int ret = -1;
+    struct iatt iatt = {
+        0,
+    };
+    inode_t *linked_inode = NULL;
+    loc_t loc = {
+        0,
+    };
+    char *bname = NULL;
+    char *save_ptr = NULL;
+    static uuid_t gfid = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1};
+    char *tmp_path = NULL;
+
+    tmp_path = gf_strdup(path);
+    if (!tmp_path) {
+        goto out;
+    }
+
+    gf_uuid_copy(loc.pargfid, gfid);
+    loc.parent = inode_ref(itable->root);
+
+    bname = strtok_r(tmp_path, "/", &save_ptr);
+
+    /* sending a lookup on parent directory,
+     * Eg:  if  path is like /a/b/c/d/e/f/g/
+     * then we will send a lookup on a first and then b,c,d,etc
+     */
+
+    while (bname) {
+        linked_inode = NULL;
+        loc.inode = inode_grep(itable, loc.parent, bname);
+        if (loc.inode == NULL) {
+            loc.inode = inode_new(itable);
+            if (loc.inode == NULL) {
+                ret = -ENOMEM;
+                goto out;
+            }
+        } else {
+            /*
+             * Inode is already populated in the inode table.
+             * Which means we already looked up the inode and
+             * linked with a dentry. So that we will skip
+             * lookup on this entry, and proceed to next.
+             */
+            linked_inode = loc.inode;
+            bname = strtok_r(NULL, "/", &save_ptr);
+            if (!bname) {
+                goto out;
+            }
+            inode_unref(loc.parent);
+            loc.parent = loc.inode;
+            gf_uuid_copy(loc.pargfid, loc.inode->gfid);
+            loc.inode = NULL;
+            continue;
+        }
 
-        ret = dht_inode_ctx_get (inode, this, &ctx);
+        loc.name = bname;
+        ret = loc_path(&loc, bname);
 
+        ret = syncop_lookup(this, &loc, &iatt, NULL, NULL, NULL);
         if (ret) {
-                ctx = GF_CALLOC (1, sizeof (*ctx), gf_dht_mt_inode_ctx_t);
-                if (!ctx)
-                        return -1;
+            gf_smsg(this->name, GF_LOG_INFO, -ret, DHT_MSG_DIR_SELFHEAL_FAILED,
+                    "path=%s", path, "subvolume=%s", this->name, "bname=%s",
+                    bname, NULL);
+            goto out;
         }
 
-        time = &ctx->time;
+        linked_inode = inode_link(loc.inode, loc.parent, bname, &iatt);
+        if (!linked_inode)
+            goto out;
 
-        DHT_UPDATE_TIME(time->mtime, time->mtime_nsec,
-                        stat->ia_mtime, stat->ia_mtime_nsec, inode, post);
-        DHT_UPDATE_TIME(time->ctime, time->ctime_nsec,
-                        stat->ia_ctime, stat->ia_ctime_nsec, inode, post);
-        DHT_UPDATE_TIME(time->atime, time->atime_nsec,
-                        stat->ia_atime, stat->ia_atime_nsec, inode, post);
+        loc_wipe(&loc);
+        gf_uuid_copy(loc.pargfid, linked_inode->gfid);
+        loc.inode = NULL;
 
-        ret = dht_inode_ctx_set (inode, this, ctx);
+        bname = strtok_r(NULL, "/", &save_ptr);
+        if (bname)
+            loc.parent = linked_inode;
+    }
 out:
-        return 0;
+    inode_ref(linked_inode);
+    loc_wipe(&loc);
+    GF_FREE(tmp_path);
+
+    return linked_inode;
 }
 
 int
-dht_inode_ctx_get (inode_t *inode, xlator_t *this, dht_inode_ctx_t **ctx)
+dht_heal_full_path(void *data)
 {
-        int             ret     = -1;
-        uint64_t        ctx_int = 0;
+    call_frame_t *heal_frame = data;
+    dht_local_t *local = NULL;
+    loc_t loc = {
+        0,
+    };
+    dict_t *dict = NULL;
+    char *path = NULL;
+    int ret = -1;
+    xlator_t *source = NULL;
+    xlator_t *this = NULL;
+    inode_table_t *itable = NULL;
+    inode_t *inode = NULL;
+    inode_t *tmp_inode = NULL;
+
+    GF_VALIDATE_OR_GOTO("DHT", heal_frame, out);
+
+    local = heal_frame->local;
+    this = heal_frame->this;
+    source = heal_frame->cookie;
+    heal_frame->cookie = NULL;
+    gf_uuid_copy(loc.gfid, local->gfid);
+
+    if (local->loc.inode)
+        loc.inode = inode_ref(local->loc.inode);
+    else
+        goto out;
+
+    itable = loc.inode->table;
+    ret = syncop_getxattr(source, &loc, &dict, GET_ANCESTRY_PATH_KEY, NULL,
+                          NULL);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_INFO, -ret, DHT_MSG_DIR_HEAL_ABORT,
+                "subvol=%s", source->name, NULL);
+        goto out;
+    }
+
+    ret = dict_get_str(dict, GET_ANCESTRY_PATH_KEY, &path);
+    if (path) {
+        inode = dht_heal_path(this, path, itable);
+        if (inode && inode != local->inode) {
+            /*
+             * if inode returned by heal function is different
+             * from what we passed, which means a racing thread
+             * already linked a different inode for dentry.
+             * So we will update our local->inode, so that we can
+             * retrurn proper inode.
+             */
+            tmp_inode = local->inode;
+            local->inode = inode;
+            inode_unref(tmp_inode);
+            tmp_inode = NULL;
+        } else {
+            inode_unref(inode);
+        }
+    }
 
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-        GF_VALIDATE_OR_GOTO (this->name, inode, out);
+out:
+    loc_wipe(&loc);
+    if (dict)
+        dict_unref(dict);
+    return 0;
+}
 
-        ret = inode_ctx_get (inode, this, &ctx_int);
+int
+dht_heal_full_path_done(int op_ret, call_frame_t *heal_frame, void *data)
+{
+    call_frame_t *main_frame = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *this = NULL;
+    int ret = -1;
+    int op_errno = 0;
+
+    local = heal_frame->local;
+    main_frame = local->main_frame;
+    local->main_frame = NULL;
+    this = heal_frame->this;
+
+    dht_set_fixed_dir_stat(&local->postparent);
+    if (local->need_xattr_heal) {
+        local->need_xattr_heal = 0;
+        ret = dht_dir_xattr_heal(this, local, &op_errno);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                    DHT_MSG_DIR_XATTR_HEAL_FAILED, "path=%s", local->loc.path,
+                    NULL);
+        }
+    }
 
-        if (ret)
-                return ret;
+    DHT_STACK_UNWIND(lookup, main_frame, 0, 0, local->inode, &local->stbuf,
+                     local->xattr, &local->postparent);
 
-        if (ctx)
-                *ctx = (dht_inode_ctx_t *) ctx_int;
-out:
-        return ret;
+    DHT_STACK_DESTROY(heal_frame);
+    return 0;
 }
 
-int dht_inode_ctx_set (inode_t *inode, xlator_t *this, dht_inode_ctx_t *ctx)
+/* This function must be called inside an inode lock */
+int
+__dht_lock_subvol_set(inode_t *inode, xlator_t *this, xlator_t *lock_subvol)
 {
-        int             ret = -1;
-        uint64_t        ctx_int = 0;
+    dht_inode_ctx_t *ctx = NULL;
+    int ret = -1;
+    uint64_t value = 0;
+
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
 
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-        GF_VALIDATE_OR_GOTO (this->name, inode, out);
-        GF_VALIDATE_OR_GOTO (this->name, ctx, out);
+    ret = __inode_ctx_get0(inode, this, &value);
+    if (ret || !value) {
+        return -1;
+    }
 
-        ctx_int = (long)ctx;
-        ret = inode_ctx_set (inode, this, &ctx_int);
+    ctx = (dht_inode_ctx_t *)(uintptr_t)value;
+    ctx->lock_subvol = lock_subvol;
 out:
-        return ret;
+    return ret;
+}
+
+xlator_t *
+dht_get_lock_subvolume(xlator_t *this, struct gf_flock *lock,
+                       dht_local_t *local)
+{
+    xlator_t *subvol = NULL;
+    inode_t *inode = NULL;
+    int32_t ret = -1;
+    uint64_t value = 0;
+    xlator_t *cached_subvol = NULL;
+    dht_inode_ctx_t *ctx = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    GF_VALIDATE_OR_GOTO(this->name, lock, out);
+    GF_VALIDATE_OR_GOTO(this->name, local, out);
+
+    cached_subvol = local->cached_subvol;
+
+    if (local->loc.inode || local->fd) {
+        inode = local->loc.inode ? local->loc.inode : local->fd->inode;
+    }
+
+    if (!inode)
+        goto out;
+
+    if (!(IA_ISDIR(inode->ia_type) || IA_ISINVAL(inode->ia_type))) {
+        /*
+         * We may get non-linked inode for directories as part
+         * of the selfheal code path. So checking  for IA_INVAL
+         * type also. This will only happen for directory.
+         */
+        subvol = local->cached_subvol;
+        goto out;
+    }
+
+    if (lock->l_type != F_UNLCK) {
+        /*
+         * inode purging might happen on NFS between a lk
+         * and unlk. Due to this lk and unlk might be sent
+         * to different subvols.
+         * So during a lock request, taking a ref on inode
+         * to prevent inode purging. inode unref will happen
+         * in unlock cbk code path.
+         */
+        inode_ref(inode);
+    }
+
+    LOCK(&inode->lock);
+    ret = __inode_ctx_get0(inode, this, &value);
+    if (!ret && value) {
+        ctx = (dht_inode_ctx_t *)(uintptr_t)value;
+        subvol = ctx->lock_subvol;
+    }
+    if (!subvol && lock->l_type != F_UNLCK && cached_subvol) {
+        ret = __dht_lock_subvol_set(inode, this, cached_subvol);
+        if (ret) {
+            gf_uuid_unparse(inode->gfid, gfid);
+            UNLOCK(&inode->lock);
+            gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_SET_INODE_CTX_FAILED,
+                    "lock_subvol gfid=%s", gfid, NULL);
+            goto post_unlock;
+        }
+        subvol = cached_subvol;
+    }
+    UNLOCK(&inode->lock);
+post_unlock:
+    if (!subvol && inode && lock->l_type != F_UNLCK) {
+        inode_unref(inode);
+    }
+out:
+    return subvol;
+}
+
+int
+dht_lk_inode_unref(call_frame_t *frame, int32_t op_ret)
+{
+    int ret = -1;
+    dht_local_t *local = NULL;
+    inode_t *inode = NULL;
+    xlator_t *this = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    this = frame->this;
+
+    if (local->loc.inode || local->fd) {
+        inode = local->loc.inode ? local->loc.inode : local->fd->inode;
+    }
+    if (!inode) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LOCK_INODE_UNREF_FAILED,
+                NULL);
+        goto out;
+    }
+
+    if (!(IA_ISDIR(inode->ia_type) || IA_ISINVAL(inode->ia_type))) {
+        ret = 0;
+        goto out;
+    }
+
+    switch (local->lock_type) {
+        case F_RDLCK:
+        case F_WRLCK:
+            if (op_ret) {
+                gf_uuid_unparse(inode->gfid, gfid);
+                gf_msg_debug(this->name, 0, "lock request failed for gfid %s",
+                             gfid);
+                inode_unref(inode);
+                goto out;
+            }
+            break;
+
+        case F_UNLCK:
+            if (!op_ret) {
+                inode_unref(inode);
+            } else {
+                gf_uuid_unparse(inode->gfid, gfid);
+                gf_smsg(this->name, GF_LOG_WARNING, 0,
+                        DHT_MSG_LOCK_INODE_UNREF_FAILED, "gfid=%s", gfid, NULL);
+                goto out;
+            }
+        default:
+            break;
+    }
+    ret = 0;
+out:
+    return ret;
+}
+
+/* Code to update custom extended attributes from src dict to dst dict
+ */
+void
+dht_dir_set_heal_xattr(xlator_t *this, dht_local_t *local, dict_t *dst,
+                       dict_t *src, int *uret, int *uflag)
+{
+    int ret = -1;
+    data_t *keyval = NULL;
+    int luret = -1;
+    int luflag = -1;
+    int i = 0;
+    char **xattrs_to_heal;
+
+    if (!src || !dst) {
+        gf_smsg(this->name, GF_LOG_WARNING, EINVAL, DHT_MSG_DST_NULL_SET_FAILED,
+                "path=%s", local->loc.path, NULL);
+        return;
+    }
+    /* Check if any user xattr present in src dict and set
+       it to dst dict
+    */
+    luret = dict_foreach_fnmatch(src, "user.*", dht_set_user_xattr, dst);
+    /* Check if any other custom xattr present in src dict
+       and set it to dst dict, here index start from 1 because
+       user xattr already checked in previous statement
+    */
+
+    xattrs_to_heal = get_xattrs_to_heal();
+
+    for (i = 1; xattrs_to_heal[i]; i++) {
+        keyval = dict_get(src, xattrs_to_heal[i]);
+        if (keyval) {
+            luflag = 1;
+            ret = dict_set(dst, xattrs_to_heal[i], keyval);
+            if (ret)
+                gf_smsg(this->name, GF_LOG_WARNING, ENOMEM,
+                        DHT_MSG_DICT_SET_FAILED, "key=%s", xattrs_to_heal[i],
+                        "path=%s", local->loc.path, NULL);
+            keyval = NULL;
+        }
+    }
+    if (uret)
+        (*uret) = luret;
+    if (uflag)
+        (*uflag) = luflag;
 }
diff --git a/xlators/cluster/dht/src/dht-inode-read.c b/xlators/cluster/dht/src/dht-inode-read.c
index 10f59ab26a1..dbb8070b0da 100644
--- a/xlators/cluster/dht/src/dht-inode-read.c
+++ b/xlators/cluster/dht/src/dht-inode-read.c
@@ -8,1132 +8,1651 @@
   cases as published by the Free Software Foundation.
 */
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
 #include "dht-common.h"
 
-int dht_access2 (xlator_t *this, call_frame_t *frame, int ret);
-int dht_readv2 (xlator_t *this, call_frame_t *frame, int ret);
-int dht_attr2 (xlator_t *this, call_frame_t *frame, int ret);
-int dht_open2 (xlator_t *this, call_frame_t *frame, int ret);
-int dht_flush2 (xlator_t *this, call_frame_t *frame, int ret);
-int dht_lk2 (xlator_t *this, call_frame_t *frame, int ret);
-int dht_fsync2 (xlator_t *this, call_frame_t *frame, int ret);
-
-int
-dht_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-              int op_ret, int op_errno, fd_t *fd, dict_t *xdata)
+static int
+dht_access2(xlator_t *this, xlator_t *dst_node, call_frame_t *frame, int ret);
+static int
+dht_readv2(xlator_t *this, xlator_t *dst_node, call_frame_t *frame, int ret);
+static int
+dht_attr2(xlator_t *this, xlator_t *dst_node, call_frame_t *frame, int ret);
+static int
+dht_open2(xlator_t *this, xlator_t *dst_node, call_frame_t *frame, int ret);
+static int
+dht_flush2(xlator_t *this, xlator_t *dst_node, call_frame_t *frame, int ret);
+static int
+dht_lk2(xlator_t *this, xlator_t *dst_node, call_frame_t *frame, int ret);
+static int
+dht_fsync2(xlator_t *this, xlator_t *dst_node, call_frame_t *frame, int ret);
+static int
+dht_common_xattrop2(xlator_t *this, xlator_t *subvol, call_frame_t *frame,
+                    int ret);
+
+static int
+dht_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+             int op_errno, fd_t *fd, dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
-        call_frame_t *prev = NULL;
-        int           ret = 0;
-
-        local = frame->local;
-        prev = cookie;
-
-        local->op_errno = op_errno;
-        if ((op_ret == -1) && (op_errno != ENOENT)) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "subvolume %s returned -1 (%s)",
-                        prev->this->name, strerror (op_errno));
-                goto out;
-        }
-
-        if (!op_ret || (local->call_cnt != 1))
-                goto out;
-
-        /* rebalance would have happened */
-        local->rebalance.target_op_fn = dht_open2;
-        ret = dht_rebalance_complete_check (this, frame);
-        if (!ret)
-                return 0;
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    int ret = 0;
+
+    local = frame->local;
+    prev = cookie;
+
+    local->op_errno = op_errno;
+    if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->name);
+        goto out;
+    }
+
+    /* Update ctx if the fd has been opened on the target*/
+    if (!op_ret && (local->call_cnt == 1)) {
+        dht_fd_ctx_set(this, fd, prev);
+        goto out;
+    }
+
+    if (!op_ret || (local->call_cnt != 1))
+        goto out;
+
+    /* rebalance would have happened */
+    local->rebalance.target_op_fn = dht_open2;
+    ret = dht_rebalance_complete_check(this, frame);
+    if (!ret)
+        return 0;
 
 out:
-        DHT_STACK_UNWIND (open, frame, op_ret, op_errno, local->fd, xdata);
+    DHT_STACK_UNWIND(open, frame, op_ret, op_errno, local->fd, xdata);
 
-        return 0;
+    return 0;
 }
 
-int
-dht_open2 (xlator_t *this, call_frame_t *frame, int op_ret)
+static int
+dht_open2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
 {
-        dht_local_t *local  = NULL;
-        xlator_t    *subvol = NULL;
-        int          op_errno = EINVAL;
+    dht_local_t *local = NULL;
+    int op_errno = EINVAL;
 
-        local = frame->local;
-        if (!local)
-                goto out;
+    if (!frame || !frame->local)
+        goto out;
 
-        op_errno = ENOENT;
-        if (op_ret)
-                goto out;
-
-        local->call_cnt = 2;
-        subvol = local->cached_subvol;
+    local = frame->local;
+    op_errno = local->op_errno;
 
-        STACK_WIND (frame, dht_open_cbk, subvol, subvol->fops->open,
-                    &local->loc, local->rebalance.flags, local->fd,
-                    NULL);
+    if (we_are_not_migrating(ret)) {
+        /* This DHT layer is not migrating the file */
+        DHT_STACK_UNWIND(open, frame, -1, local->op_errno, NULL,
+                         local->rebalance.xdata);
         return 0;
+    }
+
+    if (subvol == NULL)
+        goto out;
+
+    local->call_cnt = 2;
+
+    STACK_WIND_COOKIE(frame, dht_open_cbk, subvol, subvol, subvol->fops->open,
+                      &local->loc, local->rebalance.flags, local->fd,
+                      local->xattr_req);
+    return 0;
 
 out:
-        DHT_STACK_UNWIND (stat, frame, -1, op_errno, NULL, NULL);
-        return 0;
+    DHT_STACK_UNWIND(open, frame, -1, op_errno, NULL, NULL);
+    return 0;
 }
 
 int
-dht_open (call_frame_t *frame, xlator_t *this,
-          loc_t *loc, int flags, fd_t *fd, dict_t *xdata)
+dht_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags, fd_t *fd,
+         dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        local = dht_local_init (frame, loc, fd, GF_FOP_OPEN);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
 
-        subvol = local->cached_subvol;
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
-        }
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
 
-        local->rebalance.flags = flags;
-        local->call_cnt = 1;
+    local = dht_local_init(frame, loc, fd, GF_FOP_OPEN);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
 
-        STACK_WIND (frame, dht_open_cbk, subvol, subvol->fops->open,
-                    loc, flags, fd, xdata);
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
 
-        return 0;
+    local->rebalance.flags = flags;
+    local->call_cnt = 1;
+
+    STACK_WIND_COOKIE(frame, dht_open_cbk, subvol, subvol, subvol->fops->open,
+                      loc, flags, fd, xdata);
+
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (open, frame, -1, op_errno, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(open, frame, -1, op_errno, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
 int
-dht_file_attr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int op_ret, int op_errno, struct iatt *stbuf, dict_t *xdata)
+dht_file_attr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                  int op_errno, struct iatt *stbuf, dict_t *xdata)
 {
-        uint64_t      tmp_subvol = 0;
-        dht_local_t  *local = NULL;
-        call_frame_t *prev = NULL;
-        int           ret = -1;
-
-        GF_VALIDATE_OR_GOTO ("dht", frame, err);
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
-        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
-
-        local = frame->local;
-        prev = cookie;
-
-        if ((op_ret == -1) && (op_errno != ENOENT)) {
-                local->op_errno = op_errno;
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "subvolume %s returned -1 (%s)",
-                        prev->this->name, strerror (op_errno));
-                goto out;
-        }
-
-        if (local->call_cnt != 1)
-                goto out;
+    xlator_t *subvol1 = 0;
+    xlator_t *subvol2 = 0;
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    int ret = -1;
+    inode_t *inode = NULL;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", cookie, out);
+
+    local = frame->local;
+    prev = cookie;
+
+    if ((local->fop == GF_FOP_FSTAT) &&
+        dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
 
+    if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
         local->op_errno = op_errno;
-        /* Check if the rebalance phase2 is true */
-        if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (stbuf)) {
-                if (local->fd)
-                        ret = fd_ctx_get (local->fd, this, &tmp_subvol);
-                if (ret) {
-                        /* Phase 2 of migration */
-                        local->rebalance.target_op_fn = dht_attr2;
-                        ret = dht_rebalance_complete_check (this, frame);
-                } else {
-                        /* value is already set in fd_ctx, that means no need
-                           to check for whether its complete or not. */
-                        dht_attr2 (this, frame, 0);
-                }
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->name);
+        goto out;
+    }
+
+    if (local->call_cnt != 1)
+        goto out;
+
+    local->op_errno = op_errno;
+    local->op_ret = op_ret;
+
+    /* Check if the rebalance phase2 is true */
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(stbuf)) {
+        local->rebalance.target_op_fn = dht_attr2;
+        dht_set_local_rebalance(this, local, NULL, NULL, stbuf, xdata);
+        inode = (local->fd) ? local->fd->inode : local->loc.inode;
+
+        dht_inode_ctx_get_mig_info(this, inode, &subvol1, &subvol2);
+        if (dht_mig_info_is_invalid(local->cached_subvol, subvol1, subvol2)) {
+            /* Phase 2 of migration */
+            ret = dht_rebalance_complete_check(this, frame);
+            if (!ret)
+                return 0;
+        } else {
+            /* it is a non-fd op or it is an fd based Fop and
+               opened on the dst.*/
+            if (local->fd && !dht_fd_open_on_dst(this, local->fd, subvol2)) {
+                ret = dht_rebalance_complete_check(this, frame);
                 if (!ret)
-                        return 0;
+                    return 0;
+            } else {
+                dht_attr2(this, subvol2, frame, 0);
+                return 0;
+            }
         }
+    }
 
 out:
-        DHT_STRIP_PHASE1_FLAGS (stbuf);
-        DHT_STACK_UNWIND (stat, frame, op_ret, op_errno, stbuf, xdata);
+    DHT_STRIP_PHASE1_FLAGS(stbuf);
+    DHT_STACK_UNWIND(stat, frame, op_ret, op_errno, stbuf, xdata);
 err:
-        return 0;
+    return 0;
 }
 
-int
-dht_attr2 (xlator_t *this, call_frame_t *frame, int op_ret)
+static int
+dht_attr2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
 {
-        dht_local_t *local  = NULL;
-        xlator_t    *subvol = NULL;
-        int          op_errno = EINVAL;
+    dht_local_t *local = NULL;
+    int op_errno = EINVAL;
+
+    local = frame->local;
+    if (!local)
+        goto out;
+
+    op_errno = local->op_errno;
+
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        DHT_STACK_UNWIND(stat, frame, local->op_ret, op_errno,
+                         &local->rebalance.postbuf, local->rebalance.xdata);
+        return 0;
+    }
 
-        local = frame->local;
-        if (!local)
-                goto out;
+    if (subvol == NULL)
+        goto out;
 
-        op_errno = local->op_errno;
-        if (op_ret == -1)
-                goto out;
+    local->call_cnt = 2;
 
-        subvol = local->cached_subvol;
-        local->call_cnt = 2;
+    if (local->fop == GF_FOP_FSTAT) {
+        STACK_WIND_COOKIE(frame, dht_file_attr_cbk, subvol, subvol,
+                          subvol->fops->fstat, local->fd, local->xattr_req);
+    } else {
+        STACK_WIND_COOKIE(frame, dht_file_attr_cbk, subvol, subvol,
+                          subvol->fops->stat, &local->loc, local->xattr_req);
+    }
 
-        if (local->fop == GF_FOP_FSTAT) {
-                STACK_WIND (frame, dht_file_attr_cbk, subvol,
-                            subvol->fops->fstat, local->fd, NULL);
-        } else {
-                STACK_WIND (frame, dht_file_attr_cbk, subvol,
-                            subvol->fops->stat, &local->loc, NULL);
-        }
-        return 0;
+    return 0;
 
 out:
-        DHT_STACK_UNWIND (stat, frame, -1, op_errno, NULL, NULL);
-        return 0;
+    DHT_STACK_UNWIND(stat, frame, -1, op_errno, NULL, NULL);
+    return 0;
+}
+
+static int
+dht_attr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+             int op_errno, struct iatt *stbuf, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+    local = frame->local;
+    prev = cookie;
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            local->op_errno = op_errno;
+            UNLOCK(&frame->lock);
+            gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                         prev->name);
+
+            goto post_unlock;
+        }
+
+        dht_iatt_merge(this, &local->stbuf, stbuf);
+
+        local->op_ret = 0;
+    }
+    UNLOCK(&frame->lock);
+post_unlock:
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        DHT_STACK_UNWIND(stat, frame, local->op_ret, local->op_errno,
+                         &local->stbuf, xdata);
+    }
+
+    return 0;
 }
 
 int
-dht_attr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-              int op_ret, int op_errno, struct iatt *stbuf, dict_t *xdata)
+dht_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
-        int           this_call_cnt = 0;
-        call_frame_t *prev = NULL;
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
+    int i = 0;
+    int call_cnt = 0;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+    VALIDATE_OR_GOTO(loc->path, err);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_STAT);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    layout = local->layout;
+    if (!layout) {
+        gf_msg_debug(this->name, 0, "no layout for path=%s", loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    if (IA_ISREG(loc->inode->ia_type)) {
+        local->call_cnt = 1;
 
-        GF_VALIDATE_OR_GOTO ("dht", frame, err);
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
-        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
+        subvol = local->cached_subvol;
 
-        local = frame->local;
-        prev = cookie;
+        STACK_WIND_COOKIE(frame, dht_file_attr_cbk, subvol, subvol,
+                          subvol->fops->stat, loc, xdata);
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        local->op_errno = op_errno;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "subvolume %s returned -1 (%s)",
-                                prev->this->name, strerror (op_errno));
+        return 0;
+    }
 
-                        goto unlock;
-                }
+    local->call_cnt = call_cnt = layout->cnt;
 
-                dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
+    for (i = 0; i < call_cnt; i++) {
+        subvol = layout->list[i].xlator;
+
+        STACK_WIND_COOKIE(frame, dht_attr_cbk, subvol, subvol,
+                          subvol->fops->stat, loc, xdata);
+    }
+
+    return 0;
 
-                local->op_ret = 0;
-        }
-unlock:
-        UNLOCK (&frame->lock);
-out:
-        this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt)) {
-                DHT_STACK_UNWIND (stat, frame, local->op_ret, local->op_errno,
-                                  &local->stbuf, xdata);
-        }
 err:
-        return 0;
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(stat, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
 }
 
 int
-dht_stat (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+dht_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-        dht_layout_t *layout = NULL;
-        int           i = 0;
-        int           call_cnt = 0;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-
-        local = dht_local_init (frame, loc, NULL, GF_FOP_STAT);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
+    int i = 0;
+    int call_cnt = 0;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_FSTAT);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    layout = local->layout;
+    if (!layout) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "no layout for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    if (IA_ISREG(fd->inode->ia_type)) {
+        local->call_cnt = 1;
 
-        layout = local->layout;
-        if (!layout) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no layout for path=%s", loc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
+        subvol = local->cached_subvol;
+
+        STACK_WIND_COOKIE(frame, dht_file_attr_cbk, subvol, subvol,
+                          subvol->fops->fstat, fd, xdata);
+        return 0;
+    }
+
+    local->call_cnt = call_cnt = layout->cnt;
 
-        if (IA_ISREG (loc->inode->ia_type)) {
-                local->call_cnt = 1;
+    for (i = 0; i < call_cnt; i++) {
+        subvol = layout->list[i].xlator;
+        STACK_WIND_COOKIE(frame, dht_attr_cbk, subvol, subvol,
+                          subvol->fops->fstat, fd, xdata);
+    }
 
-                subvol = local->cached_subvol;
+    return 0;
 
-                STACK_WIND (frame, dht_file_attr_cbk, subvol,
-                            subvol->fops->stat, loc, xdata);
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(fstat, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
+}
 
+int
+dht_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, struct iovec *vector, int count, struct iatt *stbuf,
+              struct iobref *iobref, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int ret = 0;
+    xlator_t *src_subvol = 0;
+    xlator_t *dst_subvol = 0;
+
+    local = frame->local;
+    if (!local) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    /* This is already second try, no need for re-check */
+    if (local->call_cnt != 1)
+        goto out;
+
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
+
+    if ((op_ret == -1) && !dht_inode_missing(op_errno))
+        goto out;
+
+    local->op_errno = op_errno;
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(stbuf)) {
+        local->op_ret = op_ret;
+        local->rebalance.target_op_fn = dht_readv2;
+        dht_set_local_rebalance(this, local, NULL, NULL, stbuf, xdata);
+        /* File would be migrated to other node */
+        ret = dht_inode_ctx_get_mig_info(this, local->fd->inode, &src_subvol,
+                                         &dst_subvol);
+
+        if (dht_mig_info_is_invalid(local->cached_subvol, src_subvol,
+                                    dst_subvol) ||
+            !dht_fd_open_on_dst(this, local->fd, dst_subvol)) {
+            ret = dht_rebalance_complete_check(this, frame);
+            if (!ret)
                 return 0;
+        } else {
+            /* value is already set in fd_ctx, that means no need
+               to check for whether its complete or not. */
+            dht_readv2(this, dst_subvol, frame, 0);
+            return 0;
         }
+    }
 
-        local->call_cnt = call_cnt = layout->cnt;
+out:
+    DHT_STRIP_PHASE1_FLAGS(stbuf);
 
-        for (i = 0; i < call_cnt; i++) {
-                subvol = layout->list[i].xlator;
+    DHT_STACK_UNWIND(readv, frame, op_ret, op_errno, vector, count, stbuf,
+                     iobref, xdata);
 
-                STACK_WIND (frame, dht_attr_cbk,
-                            subvol, subvol->fops->stat,
-                            loc, xdata);
-        }
+    return 0;
+}
 
+static int
+dht_readv2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
+{
+    dht_local_t *local = NULL;
+    int op_errno = EINVAL;
+
+    local = frame->local;
+    if (!local)
+        goto out;
+
+    op_errno = local->op_errno;
+
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        DHT_STACK_UNWIND(readv, frame, local->op_ret, op_errno, NULL, 0,
+                         &local->rebalance.postbuf, NULL,
+                         local->rebalance.xdata);
         return 0;
+    }
 
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (stat, frame, -1, op_errno, NULL, NULL);
+    if (subvol == NULL)
+        goto out;
 
-        return 0;
-}
+    local->call_cnt = 2;
+
+    STACK_WIND(frame, dht_readv_cbk, subvol, subvol->fops->readv, local->fd,
+               local->rebalance.size, local->rebalance.offset,
+               local->rebalance.flags, local->xattr_req);
 
+    return 0;
+
+out:
+    DHT_STACK_UNWIND(readv, frame, -1, op_errno, NULL, 0, NULL, NULL, NULL);
+    return 0;
+}
 
 int
-dht_fstat (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+dht_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size, off_t off,
+          uint32_t flags, dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-        dht_layout_t *layout = NULL;
-        int           i = 0;
-        int           call_cnt = 0;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        local = dht_local_init (frame, NULL, fd, GF_FOP_FSTAT);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
 
-        layout = local->layout;
-        if (!layout) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no layout for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
-        }
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
 
-        if (IA_ISREG (fd->inode->ia_type)) {
-                local->call_cnt = 1;
+    local = dht_local_init(frame, NULL, fd, GF_FOP_READ);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
 
-                subvol = local->cached_subvol;
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
 
-                STACK_WIND (frame, dht_file_attr_cbk, subvol,
-                            subvol->fops->fstat, fd, xdata);
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
 
-                return 0;
-        }
+    local->rebalance.offset = off;
+    local->rebalance.size = size;
+    local->rebalance.flags = flags;
+    local->call_cnt = 1;
 
-        local->call_cnt = call_cnt = layout->cnt;
+    STACK_WIND(frame, dht_readv_cbk, subvol, subvol->fops->readv, local->fd,
+               local->rebalance.size, local->rebalance.offset,
+               local->rebalance.flags, local->xattr_req);
 
-        for (i = 0; i < call_cnt; i++) {
-                subvol = layout->list[i].xlator;
-                STACK_WIND (frame, dht_attr_cbk,
-                            subvol, subvol->fops->fstat,
-                            fd, xdata);
-        }
-
-        return 0;
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (fstat, frame, -1, op_errno, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(readv, frame, -1, op_errno, NULL, 0, NULL, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
-int
-dht_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-               int op_ret, int op_errno,
-               struct iovec *vector, int count, struct iatt *stbuf,
-               struct iobref *iobref, dict_t *xdata)
+static int
+dht_access_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, dict_t *xdata)
 {
-        dht_local_t *local      = NULL;
-        int          ret        = 0;
-
-        local = frame->local;
-        if (!local) {
-                op_ret = -1;
-                op_errno = EINVAL;
-                goto out;
-        }
-
-        /* This is already second try, no need for re-check */
-        if (local->call_cnt != 1)
-                goto out;
+    int ret = -1;
+    dht_local_t *local = NULL;
+    xlator_t *subvol = NULL;
+    xlator_t *prev = NULL;
+
+    local = frame->local;
+    prev = cookie;
+
+    if (!prev)
+        goto out;
+    if (local->call_cnt != 1)
+        goto out;
+    if ((op_ret == -1) &&
+        ((op_errno == ENOTCONN) || dht_inode_missing(op_errno)) &&
+        IA_ISDIR(local->loc.inode->ia_type)) {
+        subvol = dht_subvol_next_available(this, prev);
+        if (!subvol)
+            goto out;
 
-        if ((op_ret == -1) && (op_errno != ENOENT))
-                goto out;
+        /* check if we are done with visiting every node */
+        if (subvol == local->cached_subvol) {
+            goto out;
+        }
 
+        STACK_WIND_COOKIE(frame, dht_access_cbk, subvol, subvol,
+                          subvol->fops->access, &local->loc,
+                          local->rebalance.flags, NULL);
+        return 0;
+    }
+    if ((op_ret == -1) && dht_inode_missing(op_errno) &&
+        !(IA_ISDIR(local->loc.inode->ia_type))) {
+        /* File would be migrated to other node */
         local->op_errno = op_errno;
-        if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (stbuf)) {
-                /* File would be migrated to other node */
-                ret = fd_ctx_get (local->fd, this, NULL);
-                if (ret) {
-                        local->rebalance.target_op_fn = dht_readv2;
-                        ret = dht_rebalance_complete_check (this, frame);
-                } else {
-                        /* value is already set in fd_ctx, that means no need
-                           to check for whether its complete or not. */
-                        dht_readv2 (this, frame, 0);
-                }
-                if (!ret)
-                        return 0;
-        }
+        local->rebalance.target_op_fn = dht_access2;
+        ret = dht_rebalance_complete_check(frame->this, frame);
+        if (!ret)
+            return 0;
+    }
 
 out:
-        DHT_STRIP_PHASE1_FLAGS (stbuf);
-        DHT_STACK_UNWIND (readv, frame, op_ret, op_errno, vector, count, stbuf,
-                          iobref, xdata);
-
-        return 0;
+    DHT_STACK_UNWIND(access, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
-int
-dht_readv2 (xlator_t *this, call_frame_t *frame, int op_ret)
+static int
+dht_access2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
 {
-        dht_local_t *local  = NULL;
-        xlator_t    *subvol = NULL;
-        int          op_errno = EINVAL;
+    dht_local_t *local = NULL;
+    int op_errno = EINVAL;
 
-        local = frame->local;
-        if (!local)
-                goto out;
+    local = frame->local;
+    if (!local)
+        goto out;
 
-        op_errno = local->op_errno;
-        if (op_ret == -1)
-                goto out;
+    op_errno = local->op_errno;
 
-        local->call_cnt = 2;
-        subvol = local->cached_subvol;
-
-        STACK_WIND (frame, dht_readv_cbk, subvol, subvol->fops->readv,
-                    local->fd, local->rebalance.size, local->rebalance.offset,
-                    local->rebalance.flags, NULL);
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
 
+        DHT_STACK_UNWIND(access, frame, -1, op_errno, NULL);
         return 0;
+    }
+
+    if (subvol == NULL)
+        goto out;
+
+    local->call_cnt = 2;
+
+    STACK_WIND_COOKIE(frame, dht_access_cbk, subvol, subvol,
+                      subvol->fops->access, &local->loc, local->rebalance.flags,
+                      local->xattr_req);
+
+    return 0;
 
 out:
-        DHT_STACK_UNWIND (readv, frame, -1, op_errno, NULL, 0, NULL, NULL, NULL);
-        return 0;
+    DHT_STACK_UNWIND(access, frame, -1, op_errno, NULL);
+    return 0;
 }
 
 int
-dht_readv (call_frame_t *frame, xlator_t *this,
-           fd_t *fd, size_t size, off_t off, uint32_t flags, dict_t *xdata)
+dht_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+           dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        local = dht_local_init (frame, NULL, fd, GF_FOP_READ);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+    VALIDATE_OR_GOTO(loc->path, err);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_ACCESS);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->rebalance.flags = mask;
+    local->call_cnt = 1;
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for path=%s",
+                     loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    STACK_WIND_COOKIE(frame, dht_access_cbk, subvol, subvol,
+                      subvol->fops->access, loc, mask, xdata);
+
+    return 0;
 
-        subvol = local->cached_subvol;
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
-        }
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(access, frame, -1, op_errno, NULL);
 
-        local->rebalance.offset = off;
-        local->rebalance.size   = size;
-        local->rebalance.flags  = flags;
-        local->call_cnt = 1;
+    return 0;
+}
+
+int
+dht_flush_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *subvol = 0;
+    int ret = 0;
+
+    local = frame->local;
 
-        STACK_WIND (frame, dht_readv_cbk,
-                    subvol, subvol->fops->readv,
-                    fd, size, off, flags, xdata);
+    local->op_errno = op_errno;
 
+    if (local->call_cnt != 1)
+        goto out;
+
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
         return 0;
+    }
 
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (readv, frame, -1, op_errno, NULL, 0, NULL, NULL, NULL);
+    local->rebalance.target_op_fn = dht_flush2;
+
+    local->op_ret = op_ret;
+    local->op_errno = op_errno;
 
+    /* If context is set, then send flush() it to the destination */
+    dht_inode_ctx_get_mig_info(this, local->fd->inode, NULL, &subvol);
+    if (subvol && dht_fd_open_on_dst(this, local->fd, subvol)) {
+        dht_flush2(this, subvol, frame, 0);
         return 0;
-}
+    }
 
-int
-dht_access_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                int op_ret, int op_errno, dict_t *xdata)
-{
-        int          ret = -1;
-        dht_local_t *local = NULL;
-        xlator_t    *subvol = NULL;
-        call_frame_t *prev = NULL;
-
-        local = frame->local;
-        prev = cookie;
-
-        if (!prev || !prev->this)
-                goto out;
-        if (local->call_cnt != 1)
-                goto out;
-        if ((op_ret == -1) && (op_errno == ENOTCONN) &&
-            IA_ISDIR(local->loc.inode->ia_type)) {
-
-                subvol = dht_subvol_next_available (this, prev->this);
-                if (!subvol)
-                        goto out;
-
-                /* check if we are done with visiting every node */
-                if (subvol == local->cached_subvol) {
-                        goto out;
-                }
-
-                STACK_WIND (frame, dht_access_cbk, subvol, subvol->fops->access,
-                            &local->loc, local->rebalance.flags, NULL);
-                return 0;
-        }
-        if ((op_ret == -1) && (op_errno == ENOENT)) {
-                /* File would be migrated to other node */
-                local->op_errno = op_errno;
-                local->rebalance.target_op_fn = dht_access2;
-                ret = dht_rebalance_complete_check (frame->this, frame);
-                if (!ret)
-                        return 0;
+    if (op_errno == EREMOTE) {
+        ret = dht_rebalance_complete_check(this, frame);
+        if (!ret) {
+            return 0;
         }
+    }
 
 out:
-        DHT_STACK_UNWIND (access, frame, op_ret, op_errno, xdata);
-        return 0;
+    DHT_STACK_UNWIND(flush, frame, op_ret, op_errno, xdata);
+
+    return 0;
 }
 
-int
-dht_access2 (xlator_t *this, call_frame_t *frame, int op_ret)
+static int
+dht_flush2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
 {
-        dht_local_t *local  = NULL;
-        xlator_t    *subvol = NULL;
-        int          op_errno = EINVAL;
+    dht_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
 
-        local = frame->local;
-        if (!local)
-                goto out;
+    if ((frame == NULL) || (frame->local == NULL))
+        goto out;
 
-        op_errno = local->op_errno;
-        if (op_ret == -1)
-                goto out;
+    local = frame->local;
 
-        local->call_cnt = 2;
-        subvol = local->cached_subvol;
+    op_errno = local->op_errno;
 
-        STACK_WIND (frame, dht_access_cbk, subvol, subvol->fops->access,
-                    &local->loc, local->rebalance.flags, NULL);
+    if (subvol == NULL)
+        goto out;
 
-        return 0;
+    local->call_cnt = 2; /* This is the second attempt */
+
+    STACK_WIND(frame, dht_flush_cbk, subvol, subvol->fops->flush, local->fd,
+               local->xattr_req);
+
+    return 0;
 
 out:
-        DHT_STACK_UNWIND (access, frame, -1, op_errno, NULL);
-        return 0;
+    DHT_STACK_UNWIND(flush, frame, -1, op_errno, NULL);
+    return 0;
 }
 
-
 int
-dht_access (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
-            dict_t *xdata)
+dht_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-        local = dht_local_init (frame, loc, NULL, GF_FOP_ACCESS);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
 
-        local->rebalance.flags = mask;
-        local->call_cnt = 1;
-        subvol = local->cached_subvol;
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for path=%s", loc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
 
-        STACK_WIND (frame, dht_access_cbk, subvol, subvol->fops->access,
-                    loc, mask, xdata);
+    local = dht_local_init(frame, NULL, fd, GF_FOP_FLUSH);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
 
-        return 0;
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    local->call_cnt = 1;
+
+    STACK_WIND(frame, dht_flush_cbk, subvol, subvol->fops->flush, fd,
+               local->xattr_req);
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (access, frame, -1, op_errno, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(flush, frame, -1, op_errno, NULL);
 
-        return 0;
+    return 0;
 }
 
-
 int
-dht_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-               int op_ret, int op_errno, dict_t *xdata)
+dht_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+              dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
-        int           ret = -1;
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    int ret = -1;
+    inode_t *inode = NULL;
+    xlator_t *src_subvol = 0;
+    xlator_t *dst_subvol = 0;
+
+    local = frame->local;
+    prev = cookie;
+
+    local->op_errno = op_errno;
+
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
+
+    if (op_ret == -1 && !dht_inode_missing(op_errno)) {
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->name);
+        goto out;
+    }
+
+    if (local->call_cnt != 1) {
+        if (local->stbuf.ia_blocks) {
+            dht_iatt_merge(this, postbuf, &local->stbuf);
+            dht_iatt_merge(this, prebuf, &local->prebuf);
+        }
+        goto out;
+    }
 
-        local = frame->local;
+    local->op_ret = op_ret;
+    inode = local->fd->inode;
 
-        local->op_errno = op_errno;
+    local->rebalance.target_op_fn = dht_fsync2;
+    dht_set_local_rebalance(this, local, NULL, prebuf, postbuf, xdata);
 
-        if (local->call_cnt != 1)
-                goto out;
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(postbuf)) {
+        ret = dht_rebalance_complete_check(this, frame);
+        if (!ret)
+            return 0;
+    }
 
-        /* If context is set, then send flush() it to the destination */
-        ret = fd_ctx_get (local->fd, this, NULL);
-        if (!ret) {
-                dht_flush2 (this, frame, 0);
+    /* Check if the rebalance phase1 is true */
+    if (IS_DHT_MIGRATION_PHASE1(postbuf)) {
+        dht_iatt_merge(this, &local->stbuf, postbuf);
+        dht_iatt_merge(this, &local->prebuf, prebuf);
+
+        dht_inode_ctx_get_mig_info(this, inode, &src_subvol, &dst_subvol);
+
+        if (dht_mig_info_is_invalid(local->cached_subvol, src_subvol,
+                                    dst_subvol) ||
+            !dht_fd_open_on_dst(this, local->fd, dst_subvol)) {
+            ret = dht_rebalance_in_progress_check(this, frame);
+            if (!ret)
                 return 0;
+        } else {
+            dht_fsync2(this, dst_subvol, frame, 0);
+            return 0;
         }
+    }
 
 out:
-        DHT_STACK_UNWIND (flush, frame, op_ret, op_errno, xdata);
+    DHT_STRIP_PHASE1_FLAGS(postbuf);
+    DHT_STRIP_PHASE1_FLAGS(prebuf);
 
-        return 0;
+    DHT_STACK_UNWIND(fsync, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+
+    return 0;
 }
 
-int
-dht_flush2 (xlator_t *this, call_frame_t *frame, int op_ret)
+static int
+dht_fsync2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
 {
-        dht_local_t  *local  = NULL;
-        xlator_t     *subvol = NULL;
-        uint64_t      tmp_subvol = 0;
-        int           ret = -1;
-
-        local = frame->local;
+    dht_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
+
+    if ((frame == NULL) || (frame->local == NULL))
+        goto out;
+
+    local = frame->local;
+    op_errno = local->op_errno;
+
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        DHT_STACK_UNWIND(fsync, frame, local->op_ret, op_errno,
+                         &local->rebalance.prebuf, &local->rebalance.postbuf,
+                         local->rebalance.xdata);
+        return 0;
+    }
 
-        ret = fd_ctx_get (local->fd, this, &tmp_subvol);
-        if (!ret)
-                subvol = (xlator_t *)(long)tmp_subvol;
+    if (subvol == NULL)
+        goto out;
 
-        if (!subvol)
-                subvol = local->cached_subvol;
+    local->call_cnt = 2; /* This is the second attempt */
 
-        local->call_cnt = 2; /* This is the second attempt */
+    STACK_WIND_COOKIE(frame, dht_fsync_cbk, subvol, subvol, subvol->fops->fsync,
+                      local->fd, local->rebalance.flags, local->xattr_req);
 
-        STACK_WIND (frame, dht_flush_cbk,
-                    subvol, subvol->fops->flush, local->fd, NULL);
+    return 0;
 
-        return 0;
+out:
+    DHT_STACK_UNWIND(fsync, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
-
 int
-dht_flush (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+dht_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int datasync,
+          dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        local = dht_local_init (frame, NULL, fd, GF_FOP_FLUSH);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
 
-        subvol = local->cached_subvol;
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
-        }
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
 
-        local->call_cnt = 1;
+    local = dht_local_init(frame, NULL, fd, GF_FOP_FSYNC);
+    if (!local) {
+        op_errno = ENOMEM;
 
-        STACK_WIND (frame, dht_flush_cbk,
-                    subvol, subvol->fops->flush, fd, xdata);
+        goto err;
+    }
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
 
-        return 0;
+    local->call_cnt = 1;
+    local->rebalance.flags = datasync;
+
+    subvol = local->cached_subvol;
+
+    STACK_WIND_COOKIE(frame, dht_fsync_cbk, subvol, subvol, subvol->fops->fsync,
+                      local->fd, local->rebalance.flags, local->xattr_req);
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (flush, frame, -1, op_errno, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(fsync, frame, -1, op_errno, NULL, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
-
-int
-dht_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
-               int op_errno, struct iatt *prebuf, struct iatt *postbuf,
-               dict_t *xdata)
+/* TODO: for 'lk()' call, we need some other special error, may be ESTALE to
+   indicate that lock migration happened on the fd, so we can consider it as
+   phase 2 of migration */
+static int
+dht_lk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+           int op_errno, struct gf_flock *flock, dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
-        call_frame_t *prev = NULL;
-        int           ret = -1;
+    dht_local_t *local = NULL;
+    int ret = -1;
+    xlator_t *subvol = NULL;
 
-        local = frame->local;
-        prev = cookie;
+    local = frame->local;
 
-        local->op_errno = op_errno;
-        if (op_ret == -1 && (op_errno != ENOENT)) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "subvolume %s returned -1 (%s)",
-                        prev->this->name, strerror (op_errno));
-                goto out;
-        }
+    if (!local) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        goto out;
+    }
 
-        if (local->call_cnt != 1) {
-                if (local->stbuf.ia_blocks) {
-                        dht_iatt_merge (this, postbuf, &local->stbuf, NULL);
-                        dht_iatt_merge (this, prebuf, &local->prebuf, NULL);
-                }
-                goto out;
-        }
+    if (local->call_cnt != 1)
+        goto out;
 
-        local->op_errno = op_errno;
-        ret = fd_ctx_get (local->fd, this, NULL);
-        if (ret) {
-                local->rebalance.target_op_fn = dht_fsync2;
+    local->rebalance.target_op_fn = dht_lk2;
 
-                /* Check if the rebalance phase1 is true */
-                if (IS_DHT_MIGRATION_PHASE1 (postbuf)) {
-                        dht_iatt_merge (this, &local->stbuf, postbuf, NULL);
-                        dht_iatt_merge (this, &local->prebuf, prebuf, NULL);
+    local->op_ret = op_ret;
+    local->op_errno = op_errno;
 
-                        ret = dht_rebalance_in_progress_check (this, frame);
-                }
+    if (xdata)
+        local->rebalance.xdata = dict_ref(xdata);
 
-                /* Check if the rebalance phase2 is true */
-                if (IS_DHT_MIGRATION_PHASE2 (postbuf)) {
-                        ret = dht_rebalance_complete_check (this, frame);
-                }
+    if (op_errno == EREMOTE) {
+        dht_inode_ctx_get_mig_info(this, local->fd->inode, NULL, &subvol);
+        if (subvol && dht_fd_open_on_dst(this, local->fd, subvol)) {
+            dht_lk2(this, subvol, frame, 0);
+            return 0;
         } else {
-                dht_fsync2 (this, frame, 0);
-        }
-        if (!ret)
+            ret = dht_rebalance_complete_check(this, frame);
+            if (!ret) {
                 return 0;
+            }
+        }
+    }
 
 out:
-        DHT_STRIP_PHASE1_FLAGS (postbuf);
-        DHT_STRIP_PHASE1_FLAGS (prebuf);
-        DHT_STACK_UNWIND (fsync, frame, op_ret, op_errno,
-                          prebuf, postbuf, xdata);
+    dht_lk_inode_unref(frame, op_ret);
+    DHT_STACK_UNWIND(lk, frame, op_ret, op_errno, flock, xdata);
 
-        return 0;
+    return 0;
 }
 
-int
-dht_fsync2 (xlator_t *this, call_frame_t *frame, int op_ret)
+static int
+dht_lk2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
 {
-        dht_local_t  *local  = NULL;
-        xlator_t     *subvol = NULL;
-        uint64_t      tmp_subvol = 0;
-        int           ret = -1;
+    dht_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
 
-        local = frame->local;
+    if ((frame == NULL) || (frame->local == NULL))
+        goto out;
 
-        ret = fd_ctx_get (local->fd, this, &tmp_subvol);
-        if (!ret)
-                subvol = (xlator_t *)(long)tmp_subvol;
+    local = frame->local;
 
-        if (!subvol)
-                subvol = local->cached_subvol;
+    op_errno = local->op_errno;
 
-        local->call_cnt = 2; /* This is the second attempt */
+    if (subvol == NULL)
+        goto out;
 
-        STACK_WIND (frame, dht_fsync_cbk, subvol, subvol->fops->fsync,
-                    local->fd, local->rebalance.flags, NULL);
+    local->call_cnt = 2; /* This is the second attempt */
 
-        return 0;
+    STACK_WIND(frame, dht_lk_cbk, subvol, subvol->fops->lk, local->fd,
+               local->rebalance.lock_cmd, &local->rebalance.flock,
+               local->xattr_req);
+
+    return 0;
+
+out:
+    DHT_STACK_UNWIND(lk, frame, -1, op_errno, NULL, NULL);
+    return 0;
 }
 
 int
-dht_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int datasync,
-           dict_t *xdata)
+dht_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int cmd,
+       struct gf_flock *flock, dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
+    xlator_t *lock_subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_LK);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->lock_type = flock->l_type;
+    lock_subvol = dht_get_lock_subvolume(this, flock, local);
+    if (!lock_subvol) {
+        gf_msg_debug(this->name, 0, "no lock subvolume for path=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    /*
+            local->cached_subvol = lock_subvol;
+            ret = dht_check_and_open_fd_on_subvol (this, frame);
+            if (ret)
+                    goto err;
+    */
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    local->rebalance.flock = *flock;
+    local->rebalance.lock_cmd = cmd;
+
+    local->call_cnt = 1;
+
+    STACK_WIND(frame, dht_lk_cbk, lock_subvol, lock_subvol->fops->lk, fd, cmd,
+               flock, xdata);
+
+    return 0;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(lk, frame, -1, op_errno, NULL, NULL);
 
-        local = dht_local_init (frame, NULL, fd, GF_FOP_FSYNC);
-        if (!local) {
-                op_errno = ENOMEM;
+    return 0;
+}
 
-                goto err;
-        }
+static int
+dht_lease_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, struct gf_lease *lease, dict_t *xdata)
+{
+    DHT_STACK_UNWIND(lease, frame, op_ret, op_errno, lease, xdata);
 
-        local->call_cnt = 1;
-        local->rebalance.flags = datasync;
+    return 0;
+}
 
-        subvol = local->cached_subvol;
+int
+dht_lease(call_frame_t *frame, xlator_t *this, loc_t *loc,
+          struct gf_lease *lease, dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
 
-        STACK_WIND (frame, dht_fsync_cbk, subvol, subvol->fops->fsync,
-                    fd, datasync, xdata);
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
 
-        return 0;
+    subvol = dht_subvol_get_cached(this, loc->inode);
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for path=%s",
+                     loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    /* TODO: for rebalance, we need to preserve the fop arguments */
+    STACK_WIND(frame, dht_lease_cbk, subvol, subvol->fops->lease, loc, lease,
+               xdata);
+
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (fsync, frame, -1, op_errno, NULL, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(lease, frame, -1, op_errno, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
-
-/* TODO: for 'lk()' call, we need some other special error, may be ESTALE to
-   indicate that lock migration happened on the fd, so we can consider it as
-   phase 2 of migration */
-int
-dht_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-            int op_ret, int op_errno, struct gf_flock *flock, dict_t *xdata)
+/* Symlinks are currently not migrated, so no need for any check here */
+static int
+dht_readlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, const char *path, struct iatt *stbuf,
+                 dict_t *xdata)
 {
-        DHT_STACK_UNWIND (lk, frame, op_ret, op_errno, flock, xdata);
+    dht_local_t *local = NULL;
 
-        return 0;
-}
+    local = frame->local;
+    if (op_ret == -1)
+        goto err;
+
+    if (!local) {
+        op_ret = -1;
+        op_errno = EINVAL;
+    }
+
+err:
+    DHT_STRIP_PHASE1_FLAGS(stbuf);
+    DHT_STACK_UNWIND(readlink, frame, op_ret, op_errno, path, stbuf, xdata);
 
+    return 0;
+}
 
 int
-dht_lk (call_frame_t *frame, xlator_t *this,
-        fd_t *fd, int cmd, struct gf_flock *flock, dict_t *xdata)
+dht_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+             dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+    VALIDATE_OR_GOTO(loc->path, err);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_READLINK);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for path=%s",
+                     loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    STACK_WIND(frame, dht_readlink_cbk, subvol, subvol->fops->readlink, loc,
+               size, xdata);
+
+    return 0;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(readlink, frame, -1, op_errno, NULL, NULL, NULL);
 
-        subvol = dht_subvol_get_cached (this, fd->inode);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
-        }
+    return 0;
+}
 
-        /* TODO: for rebalance, we need to preserve the fop arguments */
-        STACK_WIND (frame, dht_lk_cbk, subvol, subvol->fops->lk, fd,
-                    cmd, flock, xdata);
+/* Get both DHT_IATT_IN_XDATA_KEY and DHT_MODE_IN_XDATA_KEY
+ * Use DHT_MODE_IN_XDATA_KEY if available, else fall back to
+ * DHT_IATT_IN_XDATA_KEY
+ * This will return a dummy iatt with only the mode and type set
+ */
+static int
+dht_read_iatt_from_xdata(dict_t *xdata, struct iatt *stbuf)
+{
+    int ret = -1;
+    int32_t mode = 0;
 
-        return 0;
+    ret = dict_get_int32(xdata, DHT_MODE_IN_XDATA_KEY, &mode);
 
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (lk, frame, -1, op_errno, NULL, NULL);
+    if (ret) {
+        ret = dict_get_bin(xdata, DHT_IATT_IN_XDATA_KEY, (void **)&stbuf);
+    } else {
+        stbuf->ia_prot = ia_prot_from_st_mode(mode);
+        stbuf->ia_type = ia_type_from_st_mode(mode);
+    }
 
-        return 0;
+    return ret;
 }
 
-/* Symlinks are currently not migrated, so no need for any check here */
 int
-dht_readlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int op_ret, int op_errno, const char *path,
-                  struct iatt *stbuf, dict_t *xdata)
+dht_common_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *dict,
+                       dict_t *xdata)
 {
-        dht_local_t *local = NULL;
+    dht_local_t *local = NULL;
+    call_frame_t *call_frame = NULL;
+    xlator_t *prev = NULL;
+    xlator_t *src_subvol = NULL;
+    xlator_t *dst_subvol = NULL;
+    struct iatt stbuf = {
+        0,
+    };
+    int ret = -1;
+    inode_t *inode = NULL;
+
+    local = frame->local;
+    call_frame = cookie;
+    prev = call_frame->this;
+
+    local->op_errno = op_errno;
+
+    if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1.",
+                     prev->name);
+        goto out;
+    }
+
+    if (local->call_cnt != 1)
+        goto out;
+
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
+
+    ret = dht_read_iatt_from_xdata(xdata, &stbuf);
+
+    if ((!op_ret) && (ret)) {
+        /* This is a potential problem and can cause corruption
+         * with sharding.
+         * Oh well. We tried.
+         */
+        goto out;
+    }
 
-        local = frame->local;
-        if (op_ret == -1)
-                goto err;
+    local->op_ret = op_ret;
+    local->rebalance.target_op_fn = dht_common_xattrop2;
+    if (xdata)
+        local->rebalance.xdata = dict_ref(xdata);
 
-        if (!local) {
-                op_ret = -1;
-                op_errno = EINVAL;
+    if (dict)
+        local->rebalance.dict = dict_ref(dict);
+
+    /* Phase 2 of migration */
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(&stbuf)) {
+        ret = dht_rebalance_complete_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+    /* Check if the rebalance phase1 is true */
+    if (IS_DHT_MIGRATION_PHASE1(&stbuf)) {
+        inode = local->loc.inode ? local->loc.inode : local->fd->inode;
+        dht_inode_ctx_get_mig_info(this, inode, &src_subvol, &dst_subvol);
+
+        if (dht_mig_info_is_invalid(local->cached_subvol, src_subvol,
+                                    dst_subvol) ||
+            !dht_fd_open_on_dst(this, local->fd, dst_subvol)) {
+            ret = dht_rebalance_in_progress_check(this, frame);
+            if (!ret)
+                return 0;
+        } else {
+            dht_common_xattrop2(this, dst_subvol, frame, 0);
+            return 0;
         }
+    }
 
-err:
-        DHT_STRIP_PHASE1_FLAGS (stbuf);
-        DHT_STACK_UNWIND (readlink, frame, op_ret, op_errno, path, stbuf, xdata);
+out:
+    if (local->fop == GF_FOP_XATTROP) {
+        DHT_STACK_UNWIND(xattrop, frame, op_ret, op_errno, dict, xdata);
+    } else {
+        DHT_STACK_UNWIND(fxattrop, frame, op_ret, op_errno, dict, xdata);
+    }
 
-        return 0;
+    return 0;
 }
 
-
-int
-dht_readlink (call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
-              dict_t *xdata)
+static int
+dht_common_xattrop2(xlator_t *this, xlator_t *subvol, call_frame_t *frame,
+                    int ret)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-        local = dht_local_init (frame, loc, NULL, GF_FOP_READLINK);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
+    dht_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
+
+    if ((frame == NULL) || (frame->local == NULL))
+        goto out;
+
+    local = frame->local;
+    op_errno = local->op_errno;
+
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        if (local->fop == GF_FOP_XATTROP) {
+            DHT_STACK_UNWIND(xattrop, frame, local->op_ret, op_errno,
+                             local->rebalance.dict, local->rebalance.xdata);
+        } else {
+            DHT_STACK_UNWIND(fxattrop, frame, local->op_ret, op_errno,
+                             local->rebalance.dict, local->rebalance.xdata);
         }
 
-        subvol = local->cached_subvol;
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for path=%s", loc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
+        return 0;
+    }
 
-        STACK_WIND (frame, dht_readlink_cbk,
-                    subvol, subvol->fops->readlink,
-                    loc, size, xdata);
+    if (subvol == NULL)
+        goto out;
 
-        return 0;
+    local->call_cnt = 2; /* This is the second attempt */
 
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (readlink, frame, -1, op_errno, NULL, NULL, NULL);
+    if (local->fop == GF_FOP_XATTROP) {
+        STACK_WIND(frame, dht_common_xattrop_cbk, subvol, subvol->fops->xattrop,
+                   &local->loc, local->rebalance.flags, local->rebalance.xattr,
+                   local->xattr_req);
+    } else {
+        STACK_WIND(frame, dht_common_xattrop_cbk, subvol,
+                   subvol->fops->fxattrop, local->fd, local->rebalance.flags,
+                   local->rebalance.xattr, local->xattr_req);
+    }
 
-        return 0;
-}
+    return 0;
 
-/* Currently no translators on top of 'distribute' will be using
- * below fops, hence not implementing 'migration' related checks
- */
+out:
 
-int
-dht_xattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+    /* If local is unavailable we could be unwinding the wrong
+     * function here */
+
+    if (local && (local->fop == GF_FOP_XATTROP)) {
+        DHT_STACK_UNWIND(xattrop, frame, -1, op_errno, NULL, NULL);
+    } else {
+        DHT_STACK_UNWIND(fxattrop, frame, -1, op_errno, NULL, NULL);
+    }
+    return 0;
+}
+
+static int
+dht_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
 {
-        DHT_STACK_UNWIND (xattrop, frame, op_ret, op_errno, dict, xdata);
-        return 0;
+    DHT_STACK_UNWIND(xattrop, frame, op_ret, op_errno, dict, xdata);
+    return 0;
 }
 
+/* Set both DHT_IATT_IN_XDATA_KEY and DHT_MODE_IN_XDATA_KEY
+ * Use DHT_MODE_IN_XDATA_KEY if available. Else fall back to
+ * DHT_IATT_IN_XDATA_KEY
+ */
+static int
+dht_request_iatt_in_xdata(dict_t *xattr_req)
+{
+    int ret = -1;
+
+    ret = dict_set_int8(xattr_req, DHT_MODE_IN_XDATA_KEY, 1);
+    ret = dict_set_int8(xattr_req, DHT_IATT_IN_XDATA_KEY, 1);
+
+    /* At least one call succeeded */
+    return ret;
+}
 
 int
-dht_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc,
-             gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+dht_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+            gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-        local = dht_local_init (frame, loc, NULL, GF_FOP_XATTROP);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+    int ret = -1;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_XATTROP);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for gfid=%s",
+                     uuid_utoa(loc->inode->gfid));
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    /* Todo : Handle dirs as well. At the moment the only xlator above dht
+     * that uses xattrop is sharding and that is only for files */
+
+    if (IA_ISDIR(loc->inode->ia_type)) {
+        STACK_WIND(frame, dht_xattrop_cbk, subvol, subvol->fops->xattrop, loc,
+                   flags, dict, xdata);
+
+    } else {
+        local->xattr_req = xdata ? dict_ref(xdata) : dict_new();
+        local->call_cnt = 1;
 
-        subvol = local->cached_subvol;
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for path=%s", loc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
+        local->rebalance.xattr = dict_ref(dict);
+        local->rebalance.flags = flags;
 
-        local->call_cnt = 1;
+        ret = dht_request_iatt_in_xdata(local->xattr_req);
+
+        if (ret) {
+            gf_msg_debug(this->name, 0,
+                         "Failed to set dictionary key %s file=%s",
+                         DHT_IATT_IN_XDATA_KEY, loc->path);
+        }
 
-        STACK_WIND (frame,
-                    dht_xattrop_cbk,
-                    subvol, subvol->fops->xattrop,
-                    loc, flags, dict, xdata);
+        STACK_WIND(frame, dht_common_xattrop_cbk, subvol, subvol->fops->xattrop,
+                   loc, local->rebalance.flags, local->rebalance.xattr,
+                   local->xattr_req);
+    }
 
-        return 0;
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (xattrop, frame, -1, op_errno, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(xattrop, frame, -1, op_errno, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
-
-int
-dht_fxattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+static int
+dht_fxattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
 {
-        DHT_STACK_UNWIND (fxattrop, frame, op_ret, op_errno, dict, xdata);
-        return 0;
+    DHT_STACK_UNWIND(fxattrop, frame, op_ret, op_errno, dict, xdata);
+    return 0;
 }
 
-
 int
-dht_fxattrop (call_frame_t *frame, xlator_t *this,
-              fd_t *fd, gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+dht_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+             gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        subvol = dht_subvol_get_cached (this, fd->inode);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+    int ret = -1;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    subvol = dht_subvol_get_cached(this, fd->inode);
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_FXATTROP);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    /* Todo : Handle dirs as well. At the moment the only xlator above dht
+     * that uses xattrop is sharding and that is only for files */
+
+    if (IA_ISDIR(fd->inode->ia_type)) {
+        STACK_WIND(frame, dht_fxattrop_cbk, subvol, subvol->fops->fxattrop, fd,
+                   flags, dict, xdata);
+
+    } else {
+        local->xattr_req = xdata ? dict_ref(xdata) : dict_new();
+        local->call_cnt = 1;
+
+        local->rebalance.xattr = dict_ref(dict);
+        local->rebalance.flags = flags;
+
+        ret = dht_request_iatt_in_xdata(local->xattr_req);
+
+        if (ret) {
+            gf_msg_debug(this->name, 0, "Failed to set dictionary key %s fd=%p",
+                         DHT_IATT_IN_XDATA_KEY, fd);
         }
 
-        STACK_WIND (frame,
-                    dht_fxattrop_cbk,
-                    subvol, subvol->fops->fxattrop,
-                    fd, flags, dict, xdata);
+        STACK_WIND(frame, dht_common_xattrop_cbk, subvol,
+                   subvol->fops->fxattrop, fd, local->rebalance.flags,
+                   local->rebalance.xattr, local->xattr_req);
+    }
 
-        return 0;
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (fxattrop, frame, -1, op_errno, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(fxattrop, frame, -1, op_errno, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
+/* Currently no translators on top of 'distribute' will be using
+ * below fops, hence not implementing 'migration' related checks
+ */
 
-int
-dht_inodelk_cbk (call_frame_t *frame, void *cookie,
-                 xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata)
+static int
+dht_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *xdata)
 
 {
-        DHT_STACK_UNWIND (inodelk, frame, op_ret, op_errno, xdata);
-        return 0;
+    dht_lk_inode_unref(frame, op_ret);
+    DHT_STACK_UNWIND(inodelk, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
-
 int32_t
-dht_inodelk (call_frame_t *frame, xlator_t *this, const char *volume,
-             loc_t *loc, int32_t cmd, struct gf_flock *lock, dict_t *xdata)
+dht_inodelk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+            int32_t cmd, struct gf_flock *lock, dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
+    xlator_t *lock_subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
 
-        local = dht_local_init (frame, loc, NULL, GF_FOP_INODELK);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+    local = dht_local_init(frame, loc, NULL, GF_FOP_INODELK);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
 
-        subvol = local->cached_subvol;
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for path=%s", loc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
+    local->lock_type = lock->l_type;
+    lock_subvol = dht_get_lock_subvolume(this, lock, local);
+    if (!lock_subvol) {
+        gf_msg_debug(this->name, 0, "no lock subvolume for path=%s", loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
 
-        local->call_cnt = 1;
+    local->call_cnt = 1;
 
-        STACK_WIND (frame,
-                    dht_inodelk_cbk,
-                    subvol, subvol->fops->inodelk,
-                    volume, loc, cmd, lock, xdata);
+    STACK_WIND(frame, dht_inodelk_cbk, lock_subvol, lock_subvol->fops->inodelk,
+               volume, loc, cmd, lock, xdata);
 
-        return 0;
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (inodelk, frame, -1, op_errno, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(inodelk, frame, -1, op_errno, NULL);
 
-        return 0;
+    return 0;
 }
 
-
 int
-dht_finodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno, dict_t *xdata)
+dht_finodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
 
 {
-        DHT_STACK_UNWIND (finodelk, frame, op_ret, op_errno, xdata);
-        return 0;
-}
+    dht_local_t *local = NULL;
+    int ret = 0;
 
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
 
-int
-dht_finodelk (call_frame_t *frame, xlator_t *this, const char *volume,
-              fd_t *fd, int32_t cmd, struct gf_flock *lock, dict_t *xdata)
-{
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        subvol = dht_subvol_get_cached (this, fd->inode);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
-        }
+    local = frame->local;
 
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
 
-        STACK_WIND (frame, dht_finodelk_cbk, subvol, subvol->fops->finodelk,
-                    volume, fd, cmd, lock, xdata);
+out:
+    dht_lk_inode_unref(frame, op_ret);
+    DHT_STACK_UNWIND(finodelk, frame, op_ret, op_errno, xdata);
 
-        return 0;
+    return 0;
+}
+
+int
+dht_finodelk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+             int32_t cmd, struct gf_flock *lock, dict_t *xdata)
+{
+    xlator_t *lock_subvol = NULL;
+    dht_local_t *local = NULL;
+    int op_errno = -1;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_INODELK);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->call_cnt = 1;
+    local->lock_type = lock->l_type;
+
+    lock_subvol = dht_get_lock_subvolume(this, lock, local);
+    if (!lock_subvol) {
+        gf_msg_debug(this->name, 0, "no lock subvolume for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    /*
+            local->cached_subvol = lock_subvol;
+            ret = dht_check_and_open_fd_on_subvol (this, frame);
+            if (ret)
+                    goto err;
+    */
+    local->rebalance.flock = *lock;
+    local->rebalance.lock_cmd = cmd;
+    local->key = gf_strdup(volume);
+
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    STACK_WIND(frame, dht_finodelk_cbk, lock_subvol,
+               lock_subvol->fops->finodelk, volume, fd, cmd, lock, xdata);
+
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (finodelk, frame, -1, op_errno, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(finodelk, frame, -1, op_errno, NULL);
 
-        return 0;
+    return 0;
 }
diff --git a/xlators/cluster/dht/src/dht-inode-write.c b/xlators/cluster/dht/src/dht-inode-write.c
index 9bcd84ae15a..2f23ce90fbd 100644
--- a/xlators/cluster/dht/src/dht-inode-write.c
+++ b/xlators/cluster/dht/src/dht-inode-write.c
@@ -8,883 +8,1397 @@
   cases as published by the Free Software Foundation.
 */
 
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
 #include "dht-common.h"
 
-int dht_writev2 (xlator_t *this, call_frame_t *frame, int ret);
-int dht_truncate2 (xlator_t *this, call_frame_t *frame, int ret);
-int dht_setattr2 (xlator_t *this, call_frame_t *frame, int ret);
-int dht_fallocate2(xlator_t *this, call_frame_t *frame, int op_ret);
-int dht_discard2(xlator_t *this, call_frame_t *frame, int op_ret);
+static int
+dht_writev2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret);
+static int
+dht_truncate2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret);
+static int
+dht_setattr2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret);
+static int
+dht_fallocate2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret);
+static int
+dht_discard2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret);
+static int
+dht_zerofill2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret);
 
 int
-dht_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                int op_ret, int op_errno, struct iatt *prebuf,
-                struct iatt *postbuf, dict_t *xdata)
+dht_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+               dict_t *xdata)
 {
-        dht_local_t *local = NULL;
-        int          ret   = -1;
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    int ret = -1;
+    xlator_t *subvol1 = NULL;
+    xlator_t *subvol2 = NULL;
+
+    local = frame->local;
+    prev = cookie;
+
+    if (!local) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    /* writev fails with EBADF if dht has not yet opened the fd
+     * on the cached subvol. This could happen if the file was migrated
+     * and a lookup updated the cached subvol in the inode ctx.
+     * We only check once as this could be a valid bad fd error.
+     */
+
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
 
-        if (op_ret == -1 && (op_errno != ENOENT)) {
-                goto out;
+    if (op_ret == -1 && !dht_inode_missing(op_errno)) {
+        local->op_errno = op_errno;
+        local->op_ret = -1;
+        gf_msg_debug(this->name, 0, "subvolume %s returned -1 (%s)", prev->name,
+                     strerror(op_errno));
+        goto out;
+    }
+
+    if (local->call_cnt != 1) {
+        /* preserve the modes of source */
+        if (local->stbuf.ia_blocks) {
+            dht_iatt_merge(this, postbuf, &local->stbuf);
+            dht_iatt_merge(this, prebuf, &local->prebuf);
         }
+        goto out;
+    }
 
-        local = frame->local;
-        if (!local) {
-                op_ret = -1;
-                op_errno = EINVAL;
-                goto out;
-        }
+    local->rebalance.target_op_fn = dht_writev2;
 
-        if (local->call_cnt != 1) {
-                /* preserve the modes of source */
-                if (local->stbuf.ia_blocks) {
-                        dht_iatt_merge (this, postbuf, &local->stbuf, NULL);
-                        dht_iatt_merge (this, prebuf, &local->prebuf, NULL);
-                }
+    local->op_ret = op_ret;
+    local->op_errno = op_errno;
+
+    /* We might need to pass the stbuf information to the higher DHT
+     * layer for appropriate handling.
+     */
+
+    dht_set_local_rebalance(this, local, NULL, prebuf, postbuf, xdata);
+
+    /* Phase 2 of migration */
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(postbuf)) {
+        ret = dht_rebalance_complete_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+    /* Check if the rebalance phase1 is true */
+    if (IS_DHT_MIGRATION_PHASE1(postbuf)) {
+        if (!local->xattr_req) {
+            local->xattr_req = dict_new();
+            if (!local->xattr_req) {
+                gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_NO_MEMORY, ENOMEM,
+                       "insufficient memory");
+                local->op_errno = ENOMEM;
+                local->op_ret = -1;
                 goto out;
+            }
         }
 
-        local->rebalance.target_op_fn = dht_writev2;
-
-        local->op_errno = op_errno;
-        /* Phase 2 of migration */
-        if (IS_DHT_MIGRATION_PHASE2 (postbuf)) {
-                ret = dht_rebalance_complete_check (this, frame);
-                if (!ret)
-                        return 0;
+        ret = dict_set_uint32(local->xattr_req, GF_PROTECT_FROM_EXTERNAL_WRITES,
+                              1);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_DICT_SET_FAILED, 0,
+                   "Failed to set key %s in dictionary",
+                   GF_PROTECT_FROM_EXTERNAL_WRITES);
+            local->op_errno = ENOMEM;
+            local->op_ret = -1;
+            goto out;
         }
 
-        /* Check if the rebalance phase1 is true */
-        if (IS_DHT_MIGRATION_PHASE1 (postbuf)) {
-                dht_iatt_merge (this, &local->stbuf, postbuf, NULL);
-                dht_iatt_merge (this, &local->prebuf, prebuf, NULL);
+        dht_iatt_merge(this, &local->stbuf, postbuf);
+        dht_iatt_merge(this, &local->prebuf, prebuf);
 
-                ret = fd_ctx_get (local->fd, this, NULL);
-                if (!ret) {
-                        dht_writev2 (this, frame, 0);
-                        return 0;
-                }
-                ret = dht_rebalance_in_progress_check (this, frame);
-                if (!ret)
-                        return 0;
+        ret = dht_inode_ctx_get_mig_info(this, local->fd->inode, &subvol1,
+                                         &subvol2);
+        if (!dht_mig_info_is_invalid(local->cached_subvol, subvol1, subvol2)) {
+            if (dht_fd_open_on_dst(this, local->fd, subvol2)) {
+                dht_writev2(this, subvol2, frame, 0);
+                return 0;
+            }
         }
+        ret = dht_rebalance_in_progress_check(this, frame);
+        if (!ret)
+            return 0;
+    }
 
 out:
-        DHT_STRIP_PHASE1_FLAGS (postbuf);
-        DHT_STRIP_PHASE1_FLAGS (prebuf);
+    DHT_STRIP_PHASE1_FLAGS(postbuf);
+    DHT_STRIP_PHASE1_FLAGS(prebuf);
 
-        DHT_STACK_UNWIND (writev, frame, op_ret, op_errno, prebuf, postbuf,
-                          xdata);
+    DHT_STACK_UNWIND(writev, frame, op_ret, op_errno, prebuf, postbuf, xdata);
 
-        return 0;
+    return 0;
 }
 
-int
-dht_writev2 (xlator_t *this, call_frame_t *frame, int op_ret)
+static int
+dht_writev2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
 {
-        dht_local_t  *local  = NULL;
-        xlator_t     *subvol = NULL;
-        uint64_t      tmp_subvol = 0;
-        int           ret = -1;
+    dht_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
+
+    if ((frame == NULL) || (frame->local == NULL))
+        goto out;
+
+    local = frame->local;
+    op_errno = local->op_errno;
+
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        DHT_STACK_UNWIND(writev, frame, local->op_ret, local->op_errno,
+                         &local->rebalance.prebuf, &local->rebalance.postbuf,
+                         local->rebalance.xdata);
+        return 0;
+    }
 
-        local = frame->local;
+    if (subvol == NULL)
+        goto out;
 
-        ret = fd_ctx_get (local->fd, this, &tmp_subvol);
-        if (!ret)
-                subvol = (xlator_t *)(long)tmp_subvol;
+    local->call_cnt = 2; /* This is the second attempt */
 
-        if (!subvol)
-                subvol = local->cached_subvol;
+    STACK_WIND_COOKIE(frame, dht_writev_cbk, subvol, subvol,
+                      subvol->fops->writev, local->fd, local->rebalance.vector,
+                      local->rebalance.count, local->rebalance.offset,
+                      local->rebalance.flags, local->rebalance.iobref,
+                      local->xattr_req);
 
-        local->call_cnt = 2; /* This is the second attempt */
+    return 0;
 
-        STACK_WIND (frame, dht_writev_cbk,
-                    subvol, subvol->fops->writev,
-                    local->fd, local->rebalance.vector, local->rebalance.count,
-                    local->rebalance.offset, local->rebalance.flags,
-                    local->rebalance.iobref, NULL);
+out:
+    DHT_STACK_UNWIND(writev, frame, -1, op_errno, NULL, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
 int
-dht_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
-            struct iovec *vector, int count, off_t off, uint32_t flags,
-            struct iobref *iobref, dict_t *xdata)
+dht_writev(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,
+           int count, off_t off, uint32_t flags, struct iobref *iobref,
+           dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        local = dht_local_init (frame, NULL, fd, GF_FOP_WRITE);
-        if (!local) {
-
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-        subvol = local->cached_subvol;
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
-        }
-
-
-        local->rebalance.vector = iov_dup (vector, count);
-        local->rebalance.offset = off;
-        local->rebalance.count = count;
-        local->rebalance.flags = flags;
-        local->rebalance.iobref = iobref_ref (iobref);
-        local->call_cnt = 1;
-
-        STACK_WIND (frame, dht_writev_cbk,
-                    subvol, subvol->fops->writev,
-                    fd, vector, count, off, flags, iobref, xdata);
-
-        return 0;
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_WRITE);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    local->rebalance.vector = iov_dup(vector, count);
+    local->rebalance.offset = off;
+    local->rebalance.count = count;
+    local->rebalance.flags = flags;
+    local->rebalance.iobref = iobref_ref(iobref);
+    local->call_cnt = 1;
+
+    STACK_WIND_COOKIE(frame, dht_writev_cbk, subvol, subvol,
+                      subvol->fops->writev, fd, local->rebalance.vector,
+                      local->rebalance.count, local->rebalance.offset,
+                      local->rebalance.flags, local->rebalance.iobref,
+                      local->xattr_req);
+
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (writev, frame, -1, op_errno, NULL, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(writev, frame, -1, op_errno, NULL, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
-
-
 int
-dht_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int op_ret, int op_errno, struct iatt *prebuf,
-                  struct iatt *postbuf, dict_t *xdata)
+dht_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+                 dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
-        call_frame_t *prev = NULL;
-        int           ret = -1;
-
-        GF_VALIDATE_OR_GOTO ("dht", frame, err);
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
-        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    int ret = -1;
+    xlator_t *src_subvol = NULL;
+    xlator_t *dst_subvol = NULL;
+    inode_t *inode = NULL;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", cookie, out);
+
+    local = frame->local;
+    prev = cookie;
+
+    /* Needs to be checked only for ftruncate.
+     * ftruncate fails with EBADF/EINVAL if dht has not yet opened the fd
+     * on the cached subvol. This could happen if the file was migrated
+     * and a lookup updated the cached subvol in the inode ctx.
+     * We only check once as this could actually be a valid error.
+     */
+
+    if ((local->fop == GF_FOP_FTRUNCATE) &&
+        dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
 
-        local = frame->local;
-        prev = cookie;
+    if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+        local->op_errno = op_errno;
+        local->op_ret = -1;
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->name);
 
-        if ((op_ret == -1) && (op_errno != ENOENT)) {
-                local->op_errno = op_errno;
-                local->op_ret = -1;
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "subvolume %s returned -1 (%s)",
-                        prev->this->name, strerror (op_errno));
+        goto out;
+    }
 
-                goto out;
+    if (local->call_cnt != 1) {
+        if (local->stbuf.ia_blocks) {
+            dht_iatt_merge(this, postbuf, &local->stbuf);
+            dht_iatt_merge(this, prebuf, &local->prebuf);
         }
+        goto out;
+    }
 
-        if (local->call_cnt != 1) {
-                if (local->stbuf.ia_blocks) {
-                        dht_iatt_merge (this, postbuf, &local->stbuf, NULL);
-                        dht_iatt_merge (this, prebuf, &local->prebuf, NULL);
-                }
-                goto out;
-        }
+    local->rebalance.target_op_fn = dht_truncate2;
 
-        local->rebalance.target_op_fn = dht_truncate2;
+    local->op_ret = op_ret;
+    local->op_errno = op_errno;
 
-        local->op_errno = op_errno;
-        /* Phase 2 of migration */
-        if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (postbuf)) {
-                ret = dht_rebalance_complete_check (this, frame);
-                if (!ret)
-                        return 0;
-        }
+    /* We might need to pass the stbuf information to the higher DHT
+     * layer for appropriate handling.
+     */
 
-        /* Check if the rebalance phase1 is true */
-        if (IS_DHT_MIGRATION_PHASE1 (postbuf)) {
-                dht_iatt_merge (this, &local->stbuf, postbuf, NULL);
-                dht_iatt_merge (this, &local->prebuf, prebuf, NULL);
-                ret = fd_ctx_get (local->fd, this, NULL);
-                if (!ret || !local->fd) {
-                        dht_truncate2 (this, frame, 0);
-                        return 0;
-                }
-                ret = dht_rebalance_in_progress_check (this, frame);
-                if (!ret)
-                        return 0;
+    dht_set_local_rebalance(this, local, NULL, prebuf, postbuf, xdata);
+
+    /* Phase 2 of migration */
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(postbuf)) {
+        ret = dht_rebalance_complete_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+    /* Check if the rebalance phase1 is true */
+    if (IS_DHT_MIGRATION_PHASE1(postbuf)) {
+        dht_iatt_merge(this, &local->stbuf, postbuf);
+        dht_iatt_merge(this, &local->prebuf, prebuf);
+
+        inode = (local->fd) ? local->fd->inode : local->loc.inode;
+
+        dht_inode_ctx_get_mig_info(this, inode, &src_subvol, &dst_subvol);
+        if (!dht_mig_info_is_invalid(local->cached_subvol, src_subvol,
+                                     dst_subvol)) {
+            if ((!local->fd) ||
+                ((local->fd) &&
+                 dht_fd_open_on_dst(this, local->fd, dst_subvol))) {
+                dht_truncate2(this, dst_subvol, frame, 0);
+                return 0;
+            }
         }
+        ret = dht_rebalance_in_progress_check(this, frame);
+        if (!ret)
+            return 0;
+    }
 
 out:
-        DHT_STRIP_PHASE1_FLAGS (postbuf);
-        DHT_STRIP_PHASE1_FLAGS (prebuf);
-        DHT_STACK_UNWIND (truncate, frame, op_ret, op_errno,
-                          prebuf, postbuf, xdata);
+    DHT_STRIP_PHASE1_FLAGS(postbuf);
+    DHT_STRIP_PHASE1_FLAGS(prebuf);
+
+    DHT_STACK_UNWIND(truncate, frame, op_ret, op_errno, prebuf, postbuf, xdata);
 err:
-        return 0;
+    return 0;
 }
 
-
-int
-dht_truncate2 (xlator_t *this, call_frame_t *frame, int op_ret)
+static int
+dht_truncate2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
 {
-        dht_local_t  *local  = NULL;
-        xlator_t     *subvol = NULL;
-        uint64_t      tmp_subvol = 0;
-        int           ret = -1;
+    dht_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
 
-        local = frame->local;
+    if (!frame || !frame->local)
+        goto out;
 
-        if (local->fd)
-                ret = fd_ctx_get (local->fd, this, &tmp_subvol);
-        if (!ret)
-                subvol = (xlator_t *)(long)tmp_subvol;
+    local = frame->local;
+    op_errno = local->op_errno;
+
+    /* This dht xlator is not migrating the file  */
+    if (we_are_not_migrating(ret)) {
+        DHT_STACK_UNWIND(truncate, frame, local->op_ret, local->op_errno,
+                         &local->rebalance.prebuf, &local->rebalance.postbuf,
+                         local->rebalance.xdata);
+        return 0;
+    }
 
-        if (!subvol)
-                subvol = local->cached_subvol;
+    if (subvol == NULL)
+        goto out;
 
-        local->call_cnt = 2; /* This is the second attempt */
+    local->call_cnt = 2; /* This is the second attempt */
 
-        if (local->fop == GF_FOP_TRUNCATE) {
-                STACK_WIND (frame, dht_truncate_cbk, subvol,
-                            subvol->fops->truncate, &local->loc,
-                            local->rebalance.offset, NULL);
-        } else {
-                STACK_WIND (frame, dht_truncate_cbk, subvol,
-                            subvol->fops->ftruncate, local->fd,
-                            local->rebalance.offset, NULL);
-        }
+    if (local->fop == GF_FOP_TRUNCATE) {
+        STACK_WIND_COOKIE(frame, dht_truncate_cbk, subvol, subvol,
+                          subvol->fops->truncate, &local->loc,
+                          local->rebalance.offset, local->xattr_req);
+    } else {
+        STACK_WIND_COOKIE(frame, dht_truncate_cbk, subvol, subvol,
+                          subvol->fops->ftruncate, local->fd,
+                          local->rebalance.offset, local->xattr_req);
+    }
 
-        return 0;
+    return 0;
+
+out:
+    DHT_STACK_UNWIND(truncate, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
 int
-dht_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
-              dict_t *xdata)
+dht_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+             dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-        local = dht_local_init (frame, loc, NULL, GF_FOP_TRUNCATE);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_TRUNCATE);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->rebalance.offset = offset;
+    local->call_cnt = 1;
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for gfid=%s",
+                     uuid_utoa(loc->inode->gfid));
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    STACK_WIND_COOKIE(frame, dht_truncate_cbk, subvol, subvol,
+                      subvol->fops->truncate, loc, offset, xdata);
+
+    return 0;
 
-        local->rebalance.offset = offset;
-        local->call_cnt = 1;
-        subvol = local->cached_subvol;
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for path=%s", loc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(truncate, frame, -1, op_errno, NULL, NULL, NULL);
 
-        STACK_WIND (frame, dht_truncate_cbk,
-                    subvol, subvol->fops->truncate,
-                    loc, offset, xdata);
+    return 0;
+}
 
-        return 0;
+int
+dht_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+              dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_FTRUNCATE);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->rebalance.offset = offset;
+    local->call_cnt = 1;
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    STACK_WIND_COOKIE(frame, dht_truncate_cbk, subvol, subvol,
+                      subvol->fops->ftruncate, fd, local->rebalance.offset,
+                      local->xattr_req);
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (truncate, frame, -1, op_errno, NULL, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(ftruncate, frame, -1, op_errno, NULL, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
 int
-dht_ftruncate (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
-               dict_t *xdata)
+dht_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                  int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+                  dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        local = dht_local_init (frame, NULL, fd, GF_FOP_FTRUNCATE);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    int ret = -1;
+    xlator_t *src_subvol = NULL;
+    xlator_t *dst_subvol = NULL;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", cookie, out);
+
+    local = frame->local;
+    prev = cookie;
+
+    /* fallocate fails with EBADF if dht has not yet opened the fd
+     * on the cached subvol. This could happen if the file was migrated
+     * and a lookup updated the cached subvol in the inode ctx.
+     * We only check once as this could actually be a valid error.
+     */
+
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
 
-        local->rebalance.offset = offset;
-        local->call_cnt = 1;
-        subvol = local->cached_subvol;
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
+    if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+        local->op_errno = op_errno;
+        local->op_ret = -1;
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->name);
+
+        goto out;
+    }
+
+    if (local->call_cnt != 1) {
+        if (local->stbuf.ia_blocks) {
+            dht_iatt_merge(this, postbuf, &local->stbuf);
+            dht_iatt_merge(this, prebuf, &local->prebuf);
         }
+        goto out;
+    }
 
-        STACK_WIND (frame, dht_truncate_cbk,
-                    subvol, subvol->fops->ftruncate,
-                    fd, offset, xdata);
+    local->op_ret = op_ret;
+    local->op_errno = op_errno;
+    local->rebalance.target_op_fn = dht_fallocate2;
 
-        return 0;
+    dht_set_local_rebalance(this, local, NULL, prebuf, postbuf, xdata);
+
+    /* Phase 2 of migration */
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(postbuf)) {
+        ret = dht_rebalance_complete_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+    /* Check if the rebalance phase1 is true */
+    if (IS_DHT_MIGRATION_PHASE1(postbuf)) {
+        dht_iatt_merge(this, &local->stbuf, postbuf);
+        dht_iatt_merge(this, &local->prebuf, prebuf);
+
+        dht_inode_ctx_get_mig_info(this, local->fd->inode, &src_subvol,
+                                   &dst_subvol);
+        if (!dht_mig_info_is_invalid(local->cached_subvol, src_subvol,
+                                     dst_subvol)) {
+            if (dht_fd_open_on_dst(this, local->fd, dst_subvol)) {
+                dht_fallocate2(this, dst_subvol, frame, 0);
+                return 0;
+            }
+        }
+        ret = dht_rebalance_in_progress_check(this, frame);
+        if (!ret)
+            return 0;
+    }
 
+out:
+    DHT_STRIP_PHASE1_FLAGS(postbuf);
+    DHT_STRIP_PHASE1_FLAGS(prebuf);
+
+    DHT_STACK_UNWIND(fallocate, frame, op_ret, op_errno, prebuf, postbuf,
+                     xdata);
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (ftruncate, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
+}
 
+static int
+dht_fallocate2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
+{
+    dht_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
+
+    if (!frame || !frame->local)
+        goto out;
+
+    local = frame->local;
+    op_errno = local->op_errno;
+
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        DHT_STACK_UNWIND(fallocate, frame, local->op_ret, local->op_errno,
+                         &local->rebalance.prebuf, &local->rebalance.postbuf,
+                         local->rebalance.xdata);
         return 0;
-}
+    }
+
+    if (subvol == NULL)
+        goto out;
+
+    local->call_cnt = 2; /* This is the second attempt */
+
+    STACK_WIND_COOKIE(frame, dht_fallocate_cbk, subvol, subvol,
+                      subvol->fops->fallocate, local->fd,
+                      local->rebalance.flags, local->rebalance.offset,
+                      local->rebalance.size, local->xattr_req);
 
+    return 0;
+
+out:
+    DHT_STACK_UNWIND(fallocate, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
+}
 
 int
-dht_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
-                  int op_ret, int op_errno, struct iatt *prebuf,
-                  struct iatt *postbuf, dict_t *xdata)
+dht_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+              off_t offset, size_t len, dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
-        call_frame_t *prev = NULL;
-        int           ret = -1;
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_FALLOCATE);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->rebalance.flags = mode;
+    local->rebalance.offset = offset;
+    local->rebalance.size = len;
+
+    local->call_cnt = 1;
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    STACK_WIND_COOKIE(frame, dht_fallocate_cbk, subvol, subvol,
+                      subvol->fops->fallocate, fd, local->rebalance.flags,
+                      local->rebalance.offset, local->rebalance.size,
+                      local->xattr_req);
+
+    return 0;
 
-        GF_VALIDATE_OR_GOTO ("dht", frame, err);
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
-        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(fallocate, frame, -1, op_errno, NULL, NULL, NULL);
 
-        local = frame->local;
-        prev = cookie;
+    return 0;
+}
 
-        if ((op_ret == -1) && (op_errno != ENOENT)) {
-                local->op_errno = op_errno;
-                local->op_ret = -1;
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "subvolume %s returned -1 (%s)",
-                        prev->this->name, strerror (op_errno));
+int
+dht_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+                dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    int ret = -1;
+    xlator_t *src_subvol = NULL;
+    xlator_t *dst_subvol = NULL;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", cookie, out);
+
+    local = frame->local;
+    prev = cookie;
+
+    /* discard fails with EBADF if dht has not yet opened the fd
+     * on the cached subvol. This could happen if the file was migrated
+     * and a lookup updated the cached subvol in the inode ctx.
+     * We only check once as this could actually be a valid error.
+     */
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
 
-                goto out;
-        }
+    if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+        local->op_errno = op_errno;
+        local->op_ret = -1;
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->name);
 
-        if (local->call_cnt != 1) {
-                if (local->stbuf.ia_blocks) {
-                        dht_iatt_merge (this, postbuf, &local->stbuf, NULL);
-                        dht_iatt_merge (this, prebuf, &local->prebuf, NULL);
-                }
-                goto out;
-        }
-        local->rebalance.target_op_fn = dht_fallocate2;
+        goto out;
+    }
 
-        /* Phase 2 of migration */
-        if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (postbuf)) {
-                ret = dht_rebalance_complete_check (this, frame);
-                if (!ret)
-                        return 0;
+    if (local->call_cnt != 1) {
+        if (local->stbuf.ia_blocks) {
+            dht_iatt_merge(this, postbuf, &local->stbuf);
+            dht_iatt_merge(this, prebuf, &local->prebuf);
         }
+        goto out;
+    }
 
-        /* Check if the rebalance phase1 is true */
-        if (IS_DHT_MIGRATION_PHASE1 (postbuf)) {
-                dht_iatt_merge (this, &local->stbuf, postbuf, NULL);
-                dht_iatt_merge (this, &local->prebuf, prebuf, NULL);
-                ret = fd_ctx_get (local->fd, this, NULL);
-                if (!ret) {
-                        dht_fallocate2 (this, frame, 0);
-                        return 0;
-                }
-                ret = dht_rebalance_in_progress_check (this, frame);
-                if (!ret)
-                        return 0;
+    local->rebalance.target_op_fn = dht_discard2;
+    local->op_ret = op_ret;
+    local->op_errno = op_errno;
+
+    dht_set_local_rebalance(this, local, NULL, prebuf, postbuf, xdata);
+
+    /* Phase 2 of migration */
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(postbuf)) {
+        ret = dht_rebalance_complete_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+    /* Check if the rebalance phase1 is true */
+    if (IS_DHT_MIGRATION_PHASE1(postbuf)) {
+        dht_iatt_merge(this, &local->stbuf, postbuf);
+        dht_iatt_merge(this, &local->prebuf, prebuf);
+
+        dht_inode_ctx_get_mig_info(this, local->fd->inode, &src_subvol,
+                                   &dst_subvol);
+        if (!dht_mig_info_is_invalid(local->cached_subvol, src_subvol,
+                                     dst_subvol)) {
+            if (dht_fd_open_on_dst(this, local->fd, dst_subvol)) {
+                dht_discard2(this, dst_subvol, frame, 0);
+                return 0;
+            }
         }
+        ret = dht_rebalance_in_progress_check(this, frame);
+        if (!ret)
+            return 0;
+    }
 
 out:
-        DHT_STRIP_PHASE1_FLAGS (postbuf);
-        DHT_STRIP_PHASE1_FLAGS (prebuf);
-        DHT_STACK_UNWIND (fallocate, frame, op_ret, op_errno,
-                          prebuf, postbuf, xdata);
+    DHT_STRIP_PHASE1_FLAGS(postbuf);
+    DHT_STRIP_PHASE1_FLAGS(prebuf);
+
+    DHT_STACK_UNWIND(discard, frame, op_ret, op_errno, prebuf, postbuf, xdata);
 err:
-        return 0;
+    return 0;
 }
 
-int
-dht_fallocate2(xlator_t *this, call_frame_t *frame, int op_ret)
+static int
+dht_discard2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
 {
-        dht_local_t  *local  = NULL;
-        xlator_t     *subvol = NULL;
-        uint64_t      tmp_subvol = 0;
-        int           ret = -1;
-
-        local = frame->local;
+    dht_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
+
+    if (!frame || !frame->local)
+        goto out;
+
+    local = frame->local;
+    op_errno = local->op_errno;
+
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        DHT_STACK_UNWIND(discard, frame, local->op_ret, local->op_errno,
+                         &local->rebalance.prebuf, &local->rebalance.postbuf,
+                         local->rebalance.xdata);
+        return 0;
+    }
 
-        if (local->fd)
-                ret = fd_ctx_get (local->fd, this, &tmp_subvol);
-        if (!ret)
-                subvol = (xlator_t *)(long)tmp_subvol;
+    if (subvol == NULL)
+        goto out;
 
-        if (!subvol)
-                subvol = local->cached_subvol;
+    local->call_cnt = 2; /* This is the second attempt */
 
-        local->call_cnt = 2; /* This is the second attempt */
+    STACK_WIND_COOKIE(frame, dht_discard_cbk, subvol, subvol,
+                      subvol->fops->discard, local->fd, local->rebalance.offset,
+                      local->rebalance.size, local->xattr_req);
 
-	STACK_WIND(frame, dht_fallocate_cbk, subvol, subvol->fops->fallocate,
-		   local->fd, local->rebalance.flags, local->rebalance.offset,
-		   local->rebalance.size, NULL);
+    return 0;
 
-        return 0;
+out:
+    DHT_STACK_UNWIND(discard, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
 int
-dht_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
-	      off_t offset, size_t len, dict_t *xdata)
+dht_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+            size_t len, dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        local = dht_local_init (frame, NULL, fd, GF_FOP_FALLOCATE);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
 
-	local->rebalance.flags = mode;
-        local->rebalance.offset = offset;
-	local->rebalance.size = len;
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
 
-        local->call_cnt = 1;
-        subvol = local->cached_subvol;
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
-        }
+    local = dht_local_init(frame, NULL, fd, GF_FOP_DISCARD);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
 
-        STACK_WIND (frame, dht_fallocate_cbk,
-                    subvol, subvol->fops->fallocate,
-                    fd, mode, offset, len, xdata);
+    local->rebalance.offset = offset;
+    local->rebalance.size = len;
 
-        return 0;
+    local->call_cnt = 1;
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    STACK_WIND_COOKIE(frame, dht_discard_cbk, subvol, subvol,
+                      subvol->fops->discard, fd, local->rebalance.offset,
+                      local->rebalance.size, local->xattr_req);
+
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (fallocate, frame, -1, op_errno, NULL, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(discard, frame, -1, op_errno, NULL, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
-
 int
-dht_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
-                int op_ret, int op_errno, struct iatt *prebuf,
-                struct iatt *postbuf, dict_t *xdata)
+dht_zerofill_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+                 dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
-        call_frame_t *prev = NULL;
-        int           ret = -1;
-
-        GF_VALIDATE_OR_GOTO ("dht", frame, err);
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
-        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
-
-        local = frame->local;
-        prev = cookie;
-
-        if ((op_ret == -1) && (op_errno != ENOENT)) {
-                local->op_errno = op_errno;
-                local->op_ret = -1;
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "subvolume %s returned -1 (%s)",
-                        prev->this->name, strerror (op_errno));
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    int ret = -1;
+    xlator_t *subvol1 = NULL, *subvol2 = NULL;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", cookie, out);
+
+    local = frame->local;
+    prev = cookie;
+
+    /* zerofill fails with EBADF if dht has not yet opened the fd
+     * on the cached subvol. This could happen if the file was migrated
+     * and a lookup updated the cached subvol in the inode ctx.
+     * We only check once as this could actually be a valid error.
+     */
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
 
-                goto out;
+    if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+        local->op_errno = op_errno;
+        local->op_ret = -1;
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->name);
+        goto out;
+    }
+
+    if (local->call_cnt != 1) {
+        if (local->stbuf.ia_blocks) {
+            dht_iatt_merge(this, postbuf, &local->stbuf);
+            dht_iatt_merge(this, prebuf, &local->prebuf);
         }
+        goto out;
+    }
 
-        if (local->call_cnt != 1) {
-                if (local->stbuf.ia_blocks) {
-                        dht_iatt_merge (this, postbuf, &local->stbuf, NULL);
-                        dht_iatt_merge (this, prebuf, &local->prebuf, NULL);
-                }
-                goto out;
-        }
-        local->rebalance.target_op_fn = dht_discard2;
+    local->rebalance.target_op_fn = dht_zerofill2;
+    local->op_ret = op_ret;
+    local->op_errno = op_errno;
 
-        /* Phase 2 of migration */
-        if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (postbuf)) {
-                ret = dht_rebalance_complete_check (this, frame);
-                if (!ret)
-                        return 0;
-        }
+    dht_set_local_rebalance(this, local, NULL, prebuf, postbuf, xdata);
 
-        /* Check if the rebalance phase1 is true */
-        if (IS_DHT_MIGRATION_PHASE1 (postbuf)) {
-                dht_iatt_merge (this, &local->stbuf, postbuf, NULL);
-                dht_iatt_merge (this, &local->prebuf, prebuf, NULL);
-                ret = fd_ctx_get (local->fd, this, NULL);
-                if (!ret) {
-                        dht_discard2 (this, frame, 0);
-                        return 0;
-                }
-                ret = dht_rebalance_in_progress_check (this, frame);
-                if (!ret)
-                        return 0;
+    /* Phase 2 of migration */
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(postbuf)) {
+        ret = dht_rebalance_complete_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+    /* Check if the rebalance phase1 is true */
+    if (IS_DHT_MIGRATION_PHASE1(postbuf)) {
+        dht_iatt_merge(this, &local->stbuf, postbuf);
+        dht_iatt_merge(this, &local->prebuf, prebuf);
+
+        ret = dht_inode_ctx_get_mig_info(this, local->fd->inode, &subvol1,
+                                         &subvol2);
+        if (!dht_mig_info_is_invalid(local->cached_subvol, subvol1, subvol2)) {
+            if (dht_fd_open_on_dst(this, local->fd, subvol2)) {
+                dht_zerofill2(this, subvol2, frame, 0);
+                return 0;
+            }
         }
 
+        ret = dht_rebalance_in_progress_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
 out:
-        DHT_STRIP_PHASE1_FLAGS (postbuf);
-        DHT_STRIP_PHASE1_FLAGS (prebuf);
-        DHT_STACK_UNWIND (discard, frame, op_ret, op_errno,
-                          prebuf, postbuf, xdata);
+    DHT_STRIP_PHASE1_FLAGS(postbuf);
+    DHT_STRIP_PHASE1_FLAGS(prebuf);
+
+    DHT_STACK_UNWIND(zerofill, frame, op_ret, op_errno, prebuf, postbuf, xdata);
 err:
-        return 0;
+    return 0;
 }
 
-int
-dht_discard2(xlator_t *this, call_frame_t *frame, int op_ret)
+static int
+dht_zerofill2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
 {
-        dht_local_t  *local  = NULL;
-        xlator_t     *subvol = NULL;
-        uint64_t      tmp_subvol = 0;
-        int           ret = -1;
-
-        local = frame->local;
+    dht_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
 
-        if (local->fd)
-                ret = fd_ctx_get (local->fd, this, &tmp_subvol);
-        if (!ret)
-                subvol = (xlator_t *)(long)tmp_subvol;
+    if (!frame || !frame->local)
+        goto out;
 
-        if (!subvol)
-                subvol = local->cached_subvol;
+    local = frame->local;
 
-        local->call_cnt = 2; /* This is the second attempt */
+    op_errno = local->op_errno;
 
-	STACK_WIND(frame, dht_discard_cbk, subvol, subvol->fops->discard,
-		   local->fd, local->rebalance.offset, local->rebalance.size,
-		   NULL);
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        DHT_STACK_UNWIND(zerofill, frame, local->op_ret, local->op_errno,
+                         &local->rebalance.prebuf, &local->rebalance.postbuf,
+                         local->rebalance.xdata);
 
         return 0;
+    }
+
+    if (subvol == NULL)
+        goto out;
+
+    local->call_cnt = 2; /* This is the second attempt */
+
+    STACK_WIND_COOKIE(frame, dht_zerofill_cbk, subvol, subvol,
+                      subvol->fops->zerofill, local->fd,
+                      local->rebalance.offset, local->rebalance.size,
+                      local->xattr_req);
+
+    return 0;
+
+out:
+
+    DHT_STACK_UNWIND(zerofill, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
 int
-dht_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
-	    size_t len, dict_t *xdata)
+dht_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             off_t len, dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        local = dht_local_init (frame, NULL, fd, GF_FOP_DISCARD);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
 
-        local->rebalance.offset = offset;
-	local->rebalance.size = len;
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
 
-        local->call_cnt = 1;
-        subvol = local->cached_subvol;
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
-        }
+    local = dht_local_init(frame, NULL, fd, GF_FOP_ZEROFILL);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
 
-        STACK_WIND (frame, dht_discard_cbk, subvol, subvol->fops->discard,
-                    fd, offset, len, xdata);
+    local->rebalance.offset = offset;
+    local->rebalance.size = len;
 
-        return 0;
+    local->call_cnt = 1;
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    STACK_WIND_COOKIE(frame, dht_zerofill_cbk, subvol, subvol,
+                      subvol->fops->zerofill, fd, local->rebalance.offset,
+                      local->rebalance.size, local->xattr_req);
+
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (discard, frame, -1, op_errno, NULL, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(zerofill, frame, -1, op_errno, NULL, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
 /* handle cases of migration here for 'setattr()' calls */
 int
-dht_file_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int op_ret, int op_errno, struct iatt *prebuf,
-                      struct iatt *postbuf, dict_t *xdata)
+dht_file_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
-        call_frame_t *prev = NULL;
-        int           ret = -1;
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    int ret = -1;
 
-        local = frame->local;
-        prev = cookie;
+    local = frame->local;
+    prev = cookie;
 
-        local->op_errno = op_errno;
-        if ((op_ret == -1) && (op_errno != ENOENT)) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "subvolume %s returned -1 (%s)",
-                        prev->this->name, strerror (op_errno));
-                goto out;
-        }
+    local->op_errno = op_errno;
 
-        if (local->call_cnt != 1)
-                goto out;
+    if ((local->fop == GF_FOP_FSETATTR) &&
+        dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
 
-        local->rebalance.target_op_fn = dht_setattr2;
+    if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->name);
+        goto out;
+    }
 
-        /* Phase 2 of migration */
-        if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (postbuf)) {
-                ret = dht_rebalance_complete_check (this, frame);
-                if (!ret)
-                        return 0;
-        }
+    if (local->call_cnt != 1)
+        goto out;
+
+    local->op_ret = op_ret;
+    local->op_errno = op_errno;
+
+    local->rebalance.target_op_fn = dht_setattr2;
+
+    /* Phase 2 of migration */
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(postbuf)) {
+        dht_set_local_rebalance(this, local, NULL, prebuf, postbuf, xdata);
 
-        /* At the end of the migration process, whatever 'attr' we
-           have on source file will be migrated to destination file
-           in one shot, hence we don't need to check for in progress
-           state here (ie, PHASE1) */
+        ret = dht_rebalance_complete_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+    /* At the end of the migration process, whatever 'attr' we
+       have on source file will be migrated to destination file
+       in one shot, hence we don't need to check for in progress
+       state here (ie, PHASE1) */
 out:
-        DHT_STRIP_PHASE1_FLAGS (postbuf);
-        DHT_STRIP_PHASE1_FLAGS (prebuf);
-        DHT_STACK_UNWIND (setattr, frame, op_ret, op_errno,
-                          prebuf, postbuf, xdata);
+    DHT_STRIP_PHASE1_FLAGS(postbuf);
+    DHT_STRIP_PHASE1_FLAGS(prebuf);
 
-        return 0;
+    DHT_STACK_UNWIND(setattr, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+
+    return 0;
 }
 
-int
-dht_setattr2 (xlator_t *this, call_frame_t *frame, int op_ret)
+static int
+dht_setattr2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
 {
-        dht_local_t  *local  = NULL;
-        xlator_t     *subvol = NULL;
-        uint64_t      tmp_subvol = 0;
-        int           ret = -1;
+    dht_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
+
+    if (!frame || !frame->local)
+        goto out;
+
+    local = frame->local;
+    op_errno = local->op_errno;
+
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        DHT_STACK_UNWIND(setattr, frame, local->op_ret, local->op_errno,
+                         &local->rebalance.prebuf, &local->rebalance.postbuf,
+                         local->rebalance.xdata);
+        return 0;
+    }
 
-        local = frame->local;
+    if (subvol == NULL)
+        goto out;
 
-        if (local->fd)
-                ret = fd_ctx_get (local->fd, this, &tmp_subvol);
-        if (!ret)
-                subvol = (xlator_t *)(long)tmp_subvol;
-
-        if (!subvol)
-                subvol = local->cached_subvol;
-
-        local->call_cnt = 2; /* This is the second attempt */
-
-        if (local->fop == GF_FOP_SETATTR) {
-                STACK_WIND (frame, dht_file_setattr_cbk, subvol,
-                            subvol->fops->setattr, &local->loc,
-                            &local->rebalance.stbuf, local->rebalance.flags,
-                            NULL);
-        } else {
-                STACK_WIND (frame, dht_file_setattr_cbk, subvol,
-                            subvol->fops->fsetattr, local->fd,
-                            &local->rebalance.stbuf, local->rebalance.flags,
-                            NULL);
-        }
+    local->call_cnt = 2; /* This is the second attempt */
 
-        return 0;
-}
+    if (local->fop == GF_FOP_SETATTR) {
+        STACK_WIND_COOKIE(frame, dht_file_setattr_cbk, subvol, subvol,
+                          subvol->fops->setattr, &local->loc,
+                          &local->rebalance.stbuf, local->rebalance.flags,
+                          local->xattr_req);
+    } else {
+        STACK_WIND_COOKIE(frame, dht_file_setattr_cbk, subvol, subvol,
+                          subvol->fops->fsetattr, local->fd,
+                          &local->rebalance.stbuf, local->rebalance.flags,
+                          local->xattr_req);
+    }
 
+    return 0;
+
+out:
+    DHT_STACK_UNWIND(setattr, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
+}
 
 /* Keep the existing code same for all the cases other than regular file */
 int
-dht_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int op_ret, int op_errno, struct iatt *statpre,
-                 struct iatt *statpost, dict_t *xdata)
+dht_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                int op_errno, struct iatt *statpre, struct iatt *statpost,
+                dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
-        int           this_call_cnt = 0;
-        call_frame_t *prev = NULL;
-
-
-        local = frame->local;
-        prev = cookie;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        local->op_errno = op_errno;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "subvolume %s returned -1 (%s)",
-                                prev->this->name, strerror (op_errno));
-                        goto unlock;
-                }
-
-                dht_iatt_merge (this, &local->prebuf, statpre, prev->this);
-                dht_iatt_merge (this, &local->stbuf, statpost, prev->this);
-
-                local->op_ret = 0;
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+
+    local = frame->local;
+    prev = cookie;
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            local->op_errno = op_errno;
+            UNLOCK(&frame->lock);
+            gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                         prev->name);
+            goto post_unlock;
         }
-unlock:
-        UNLOCK (&frame->lock);
 
-        this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt))
-                DHT_STACK_UNWIND (setattr, frame, local->op_ret, local->op_errno,
-                                  &local->prebuf, &local->stbuf, xdata);
-
-        return 0;
+        dht_iatt_merge(this, &local->prebuf, statpre);
+        dht_iatt_merge(this, &local->stbuf, statpost);
+
+        local->op_ret = 0;
+        local->op_errno = 0;
+    }
+    UNLOCK(&frame->lock);
+post_unlock:
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        if (local->op_ret == 0)
+            dht_inode_ctx_time_set(local->loc.inode, this, &local->stbuf);
+        DHT_STACK_UNWIND(setattr, frame, local->op_ret, local->op_errno,
+                         &local->prebuf, &local->stbuf, xdata);
+    }
+
+    return 0;
 }
 
-
+/* Keep the existing code same for all the cases other than regular file */
 int
-dht_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-             struct iatt *stbuf, int32_t valid, dict_t *xdata)
+dht_non_mds_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, struct iatt *statpre,
+                        struct iatt *statpost, dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        dht_layout_t *layout = NULL;
-        dht_local_t  *local  = NULL;
-        int           op_errno = -1;
-        int           i = -1;
-        int           call_cnt = 0;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-        local = dht_local_init (frame, loc, NULL, GF_FOP_SETATTR);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+
+    local = frame->local;
+    prev = cookie;
+
+    if (op_ret == -1) {
+        gf_msg(this->name, op_errno, 0, 0, "subvolume %s returned -1",
+               prev->name);
+        goto post_unlock;
+    }
+
+    LOCK(&frame->lock);
+    {
+        dht_iatt_merge(this, &local->prebuf, statpre);
+        dht_iatt_merge(this, &local->stbuf, statpost);
+
+        local->op_ret = 0;
+        local->op_errno = 0;
+    }
+    UNLOCK(&frame->lock);
+post_unlock:
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        dht_inode_ctx_time_set(local->loc.inode, this, &local->stbuf);
+        DHT_STACK_UNWIND(setattr, frame, 0, 0, &local->prebuf, &local->stbuf,
+                         xdata);
+    }
+
+    return 0;
+}
 
-        layout = local->layout;
-        if (!layout) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no layout for path=%s", loc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
+int
+dht_mds_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int op_ret, int op_errno, struct iatt *statpre,
+                    struct iatt *statpost, dict_t *xdata)
 
-        if (!layout_is_sane (layout)) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "layout is not sane for path=%s", loc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
+{
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    xlator_t *prev = NULL;
+    xlator_t *mds_subvol = NULL;
+    struct iatt loc_stbuf = {
+        0,
+    };
+    int i = 0;
+
+    local = frame->local;
+    prev = cookie;
+    conf = this->private;
+    mds_subvol = local->mds_subvol;
+
+    if (op_ret == -1) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->name);
+        goto out;
+    }
+
+    local->op_ret = 0;
+    loc_stbuf = local->stbuf;
+    dht_iatt_merge(this, &local->prebuf, statpre);
+    dht_iatt_merge(this, &local->stbuf, statpost);
+
+    local->call_cnt = conf->subvolume_cnt - 1;
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (mds_subvol == conf->subvolumes[i])
+            continue;
+        STACK_WIND_COOKIE(frame, dht_non_mds_setattr_cbk, conf->subvolumes[i],
+                          conf->subvolumes[i],
+                          conf->subvolumes[i]->fops->setattr, &local->loc,
+                          &loc_stbuf, local->valid, local->xattr_req);
+    }
+
+    return 0;
+out:
+    DHT_STACK_UNWIND(setattr, frame, local->op_ret, local->op_errno,
+                     &local->prebuf, &local->stbuf, xdata);
+
+    return 0;
+}
 
-        if (IA_ISREG (loc->inode->ia_type)) {
-                /* in the regular file _cbk(), we need to check for
-                   migration possibilities */
-                local->rebalance.stbuf = *stbuf;
-                local->rebalance.flags = valid;
-                local->call_cnt = 1;
-                subvol = local->cached_subvol;
+int
+dht_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *stbuf,
+            int32_t valid, dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    xlator_t *mds_subvol = NULL;
+    dht_layout_t *layout = NULL;
+    dht_local_t *local = NULL;
+    int op_errno = -1;
+    int i = -1;
+    int ret = -1;
+    int call_cnt = 0;
+    dht_conf_t *conf = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+    VALIDATE_OR_GOTO(loc->path, err);
+
+    conf = this->private;
+    local = dht_local_init(frame, loc, NULL, GF_FOP_SETATTR);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    layout = local->layout;
+    if (!layout) {
+        gf_msg_debug(this->name, 0, "no layout for path=%s", loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (!layout_is_sane(layout)) {
+        gf_msg_debug(this->name, 0, "layout is not sane for path=%s",
+                     loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    if (IA_ISREG(loc->inode->ia_type)) {
+        /* in the regular file _cbk(), we need to check for
+           migration possibilities */
+        local->rebalance.stbuf = *stbuf;
+        local->rebalance.flags = valid;
+        local->call_cnt = 1;
+        subvol = local->cached_subvol;
 
-                STACK_WIND (frame, dht_file_setattr_cbk, subvol,
-                            subvol->fops->setattr,
-                            loc, stbuf, valid, xdata);
+        STACK_WIND_COOKIE(frame, dht_file_setattr_cbk, subvol, subvol,
+                          subvol->fops->setattr, loc, stbuf, valid, xdata);
 
-                return 0;
+        return 0;
+    }
+
+    local->call_cnt = call_cnt = layout->cnt;
+
+    if (IA_ISDIR(loc->inode->ia_type) && !__is_root_gfid(loc->inode->gfid) &&
+        call_cnt != 1) {
+        ret = dht_inode_ctx_mdsvol_get(loc->inode, this, &mds_subvol);
+        if (ret || !mds_subvol) {
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+                   "Failed to get mds subvol for path %s", local->loc.path);
+            op_errno = EINVAL;
+            goto err;
         }
 
-        local->call_cnt = call_cnt = layout->cnt;
+        local->mds_subvol = mds_subvol;
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            if (conf->subvolumes[i] == mds_subvol) {
+                if (!conf->subvolume_status[i]) {
+                    gf_msg(this->name, GF_LOG_WARNING, layout->list[i].err,
+                           DHT_MSG_HASHED_SUBVOL_DOWN,
+                           "MDS subvol is down for path "
+                           " %s Unable to set attr ",
+                           local->loc.path);
+                    op_errno = ENOTCONN;
+                    goto err;
+                }
+            }
+        }
+        local->valid = valid;
+        local->stbuf = *stbuf;
 
+        STACK_WIND_COOKIE(frame, dht_mds_setattr_cbk, local->mds_subvol,
+                          local->mds_subvol, local->mds_subvol->fops->setattr,
+                          loc, stbuf, valid, xdata);
+        return 0;
+    } else {
         for (i = 0; i < call_cnt; i++) {
-                STACK_WIND (frame, dht_setattr_cbk,
-                            layout->list[i].xlator,
-                            layout->list[i].xlator->fops->setattr,
-                            loc, stbuf, valid, xdata);
+            STACK_WIND_COOKIE(frame, dht_setattr_cbk, layout->list[i].xlator,
+                              layout->list[i].xlator,
+                              layout->list[i].xlator->fops->setattr, loc, stbuf,
+                              valid, xdata);
         }
+    }
 
-        return 0;
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (setattr, frame, -1, op_errno, NULL, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(setattr, frame, -1, op_errno, NULL, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
-
 int
-dht_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf,
-              int32_t valid, dict_t *xdata)
+dht_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf,
+             int32_t valid, dict_t *xdata)
 {
-        xlator_t     *subvol = NULL;
-        dht_layout_t *layout = NULL;
-        dht_local_t  *local  = NULL;
-        int           op_errno = -1;
-        int           i = -1;
-        int           call_cnt = 0;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        local = dht_local_init (frame, NULL, fd, GF_FOP_FSETATTR);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-        layout = local->layout;
-        if (!layout) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no layout for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
-        }
-
-        if (!layout_is_sane (layout)) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "layout is not sane for fd=%p", fd);
-                op_errno = EINVAL;
-                goto err;
-        }
-
-        if (IA_ISREG (fd->inode->ia_type)) {
-                /* in the regular file _cbk(), we need to check for
-                   migration possibilities */
-                local->rebalance.stbuf = *stbuf;
-                local->rebalance.flags = valid;
-                local->call_cnt = 1;
-                subvol = local->cached_subvol;
-
-                STACK_WIND (frame, dht_file_setattr_cbk, subvol,
-                            subvol->fops->fsetattr,
-                            fd, stbuf, valid, xdata);
+    xlator_t *subvol = NULL;
+    dht_layout_t *layout = NULL;
+    dht_local_t *local = NULL;
+    int op_errno = -1;
+    int i = -1;
+    int call_cnt = 0;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_FSETATTR);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    layout = local->layout;
+    if (!layout) {
+        gf_msg_debug(this->name, 0, "no layout for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (!layout_is_sane(layout)) {
+        gf_msg_debug(this->name, 0, "layout is not sane for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    if (IA_ISREG(fd->inode->ia_type)) {
+        /* in the regular file _cbk(), we need to check for
+           migration possibilities */
+        local->rebalance.stbuf = *stbuf;
+        local->rebalance.flags = valid;
+        local->call_cnt = 1;
+        subvol = local->cached_subvol;
 
-                return 0;
-        }
+        STACK_WIND_COOKIE(frame, dht_file_setattr_cbk, subvol, subvol,
+                          subvol->fops->fsetattr, fd, &local->rebalance.stbuf,
+                          local->rebalance.flags, local->xattr_req);
+        return 0;
+    }
 
-        local->call_cnt = call_cnt = layout->cnt;
+    local->call_cnt = call_cnt = layout->cnt;
 
-        for (i = 0; i < call_cnt; i++) {
-                STACK_WIND (frame, dht_setattr_cbk,
-                            layout->list[i].xlator,
-                            layout->list[i].xlator->fops->fsetattr,
-                            fd, stbuf, valid, xdata);
-        }
+    for (i = 0; i < call_cnt; i++) {
+        STACK_WIND_COOKIE(frame, dht_setattr_cbk, layout->list[i].xlator,
+                          layout->list[i].xlator,
+                          layout->list[i].xlator->fops->fsetattr, fd, stbuf,
+                          valid, xdata);
+    }
 
-        return 0;
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (fsetattr, frame, -1, op_errno, NULL, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(fsetattr, frame, -1, op_errno, NULL, NULL, NULL);
 
-        return 0;
+    return 0;
 }
diff --git a/xlators/cluster/dht/src/dht-layout.c b/xlators/cluster/dht/src/dht-layout.c
index 38e9970a7a8..fda904c92c9 100644
--- a/xlators/cluster/dht/src/dht-layout.c
+++ b/xlators/cluster/dht/src/dht-layout.c
@@ -8,785 +8,801 @@
   cases as published by the Free Software Foundation.
 */
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-
-#include "glusterfs.h"
-#include "xlator.h"
 #include "dht-common.h"
-#include "byte-order.h"
+#include <glusterfs/byte-order.h>
+#include "unittest/unittest.h"
 
-#define layout_base_size (sizeof (dht_layout_t))
+#define layout_base_size (sizeof(dht_layout_t))
 
-#define layout_entry_size (sizeof ((dht_layout_t *)NULL)->list[0])
+#define layout_entry_size (sizeof((dht_layout_t *)NULL)->list[0])
 
 #define layout_size(cnt) (layout_base_size + (cnt * layout_entry_size))
 
-
 dht_layout_t *
-dht_layout_new (xlator_t *this, int cnt)
+dht_layout_new(xlator_t *this, int cnt)
 {
-        dht_layout_t *layout = NULL;
-        dht_conf_t   *conf = NULL;
+    dht_layout_t *layout = NULL;
+    dht_conf_t *conf = NULL;
 
+    REQUIRE(NULL != this);
+    REQUIRE(cnt >= 0);
 
-        conf = this->private;
+    conf = this->private;
 
-        layout = GF_CALLOC (1, layout_size (cnt),
-                            gf_dht_mt_dht_layout_t);
-        if (!layout) {
-                goto out;
-        }
+    layout = GF_CALLOC(1, layout_size(cnt), gf_dht_mt_dht_layout_t);
+    if (!layout) {
+        goto out;
+    }
 
-        layout->type = DHT_HASH_TYPE_DM;
-        layout->cnt = cnt;
+    layout->type = DHT_HASH_TYPE_DM;
+    layout->cnt = cnt;
 
-        if (conf) {
-                layout->spread_cnt = conf->dir_spread_cnt;
-                layout->gen = conf->gen;
-        }
+    if (conf) {
+        layout->spread_cnt = conf->dir_spread_cnt;
+        layout->gen = conf->gen;
+    }
 
-        layout->ref = 1;
+    GF_ATOMIC_INIT(layout->ref, 1);
+
+    ENSURE(NULL != layout);
+    ENSURE(layout->type == DHT_HASH_TYPE_DM);
+    ENSURE(layout->cnt == cnt);
+    ENSURE(GF_ATOMIC_GET(layout->ref) == 1);
 out:
-        return layout;
+    return layout;
 }
 
-
 dht_layout_t *
-dht_layout_get (xlator_t *this, inode_t *inode)
+dht_layout_get(xlator_t *this, inode_t *inode)
 {
-        dht_conf_t   *conf = NULL;
-        dht_layout_t *layout = NULL;
-
-        conf = this->private;
-        if (!conf)
-                goto out;
-
-        LOCK (&conf->layout_lock);
-        {
-                dht_inode_ctx_layout_get (inode, this, &layout);
-                if (layout) {
-                        layout->ref++;
-                }
-        }
-        UNLOCK (&conf->layout_lock);
-
-out:
-        return layout;
+    dht_layout_t *layout = NULL;
+    int ret = 0;
+
+    ret = dht_inode_ctx_layout_get(inode, this, &layout);
+    if ((!ret) && layout) {
+        GF_ATOMIC_INC(layout->ref);
+    }
+    return layout;
 }
 
-
 int
-dht_layout_set (xlator_t *this, inode_t *inode, dht_layout_t *layout)
+dht_layout_set(xlator_t *this, inode_t *inode, dht_layout_t *layout)
 {
-        dht_conf_t   *conf = NULL;
-        int           oldret = -1;
-        int           ret = 0;
-        dht_layout_t *old_layout;
-
-        conf = this->private;
-        if (!conf)
-                goto out;
-
-        LOCK (&conf->layout_lock);
-        {
-                oldret = dht_inode_ctx_layout_get (inode, this, &old_layout);
-                layout->ref++;
-                dht_inode_ctx_layout_set (inode, this, layout);
-        }
-        UNLOCK (&conf->layout_lock);
-
-        if (!oldret) {
-                dht_layout_unref (this, old_layout);
-        }
+    dht_conf_t *conf = NULL;
+    int oldret = -1;
+    int ret = -1;
+    dht_layout_t *old_layout;
+
+    conf = this->private;
+    if (!conf || !layout)
+        goto out;
+
+    LOCK(&conf->layout_lock);
+    {
+        oldret = dht_inode_ctx_layout_get(inode, this, &old_layout);
+        if (layout)
+            GF_ATOMIC_INC(layout->ref);
+        ret = dht_inode_ctx_layout_set(inode, this, layout);
+    }
+    UNLOCK(&conf->layout_lock);
+
+    if (!oldret) {
+        dht_layout_unref(this, old_layout);
+    }
+    if (ret)
+        GF_ATOMIC_DEC(layout->ref);
 
 out:
-        return ret;
+    return ret;
 }
 
-
 void
-dht_layout_unref (xlator_t *this, dht_layout_t *layout)
+dht_layout_unref(xlator_t *this, dht_layout_t *layout)
 {
-        dht_conf_t  *conf = NULL;
-        int          ref = 0;
-
-        if (!layout || layout->preset || !this->private)
-                return;
+    int ref = 0;
 
-        conf = this->private;
+    if (!layout || layout->preset || !this->private)
+        return;
 
-        LOCK (&conf->layout_lock);
-        {
-                ref = --layout->ref;
-        }
-        UNLOCK (&conf->layout_lock);
+    ref = GF_ATOMIC_DEC(layout->ref);
 
-        if (!ref)
-                GF_FREE (layout);
+    if (!ref)
+        GF_FREE(layout);
 }
 
-
 dht_layout_t *
-dht_layout_ref (xlator_t *this, dht_layout_t *layout)
+dht_layout_ref(xlator_t *this, dht_layout_t *layout)
 {
-        dht_conf_t  *conf = NULL;
-
-        if (layout->preset || !this->private)
-                return layout;
+    if (layout->preset || !this->private)
+        return layout;
 
-        conf = this->private;
-        LOCK (&conf->layout_lock);
-        {
-                layout->ref++;
-        }
-        UNLOCK (&conf->layout_lock);
+    GF_ATOMIC_INC(layout->ref);
 
-        return layout;
+    return layout;
 }
 
-
 xlator_t *
-dht_layout_search (xlator_t *this, dht_layout_t *layout, const char *name)
+dht_layout_search(xlator_t *this, dht_layout_t *layout, const char *name)
 {
-        uint32_t   hash = 0;
-        xlator_t  *subvol = NULL;
-        int        i = 0;
-        int        ret = 0;
-
-
-        ret = dht_hash_compute (this, layout->type, name, &hash);
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "hash computation failed for type=%d name=%s",
-                        layout->type, name);
-                goto out;
-        }
-
-        for (i = 0; i < layout->cnt; i++) {
-                if (layout->list[i].start <= hash
-                    && layout->list[i].stop >= hash) {
-                        subvol = layout->list[i].xlator;
-                        break;
-                }
-        }
-
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "no subvolume for hash (value) = %u", hash);
-        }
+    uint32_t hash = 0;
+    xlator_t *subvol = NULL;
+    int i = 0;
+    int ret = 0;
+
+    ret = dht_hash_compute(this, layout->type, name, &hash);
+    if (ret != 0) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_COMPUTE_HASH_FAILED,
+                "type=%d", layout->type, "name=%s", name, NULL);
+        goto out;
+    }
+
+    for (i = 0; i < layout->cnt; i++) {
+        if (layout->list[i].start <= hash && layout->list[i].stop >= hash) {
+            subvol = layout->list[i].xlator;
+            break;
+        }
+    }
+
+    if (!subvol) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+                "hash-value=0x%x", hash, NULL);
+    }
 
 out:
-        return subvol;
+    return subvol;
 }
 
-
 dht_layout_t *
-dht_layout_for_subvol (xlator_t *this, xlator_t *subvol)
+dht_layout_for_subvol(xlator_t *this, xlator_t *subvol)
 {
-        dht_conf_t   *conf = NULL;
-        dht_layout_t *layout = NULL;
-        int           i = 0;
-
-        conf = this->private;
-        if (!conf)
-                goto out;
-
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                if (conf->subvolumes[i] == subvol) {
-                        layout = conf->file_layouts[i];
-                        break;
-                }
+    dht_conf_t *conf = NULL;
+    dht_layout_t *layout = NULL;
+    int i = 0;
+
+    conf = this->private;
+    if (!conf)
+        goto out;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (conf->subvolumes[i] == subvol) {
+            layout = conf->file_layouts[i];
+            break;
         }
+    }
 
 out:
-        return layout;
+    return layout;
 }
 
-
 int
-dht_layouts_init (xlator_t *this, dht_conf_t *conf)
+dht_layouts_init(xlator_t *this, dht_conf_t *conf)
 {
-        dht_layout_t *layout = NULL;
-        int           i = 0;
-        int           ret = -1;
-
-        if (!conf)
-                goto out;
-
-        conf->file_layouts = GF_CALLOC (conf->subvolume_cnt,
-                                        sizeof (dht_layout_t *),
-                                        gf_dht_mt_dht_layout_t);
-        if (!conf->file_layouts) {
-                goto out;
-        }
+    dht_layout_t *layout = NULL;
+    int i = 0;
+    int ret = -1;
 
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                layout = dht_layout_new (this, 1);
+    if (!conf)
+        goto out;
 
-                if (!layout) {
-                        goto out;
-                }
-
-                layout->preset = 1;
+    conf->file_layouts = GF_CALLOC(conf->subvolume_cnt, sizeof(dht_layout_t *),
+                                   gf_dht_mt_dht_layout_t);
+    if (!conf->file_layouts) {
+        goto out;
+    }
 
-                layout->list[0].xlator = conf->subvolumes[i];
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        layout = dht_layout_new(this, 1);
 
-                conf->file_layouts[i] = layout;
+        if (!layout) {
+            goto out;
         }
 
-        ret = 0;
+        layout->preset = 1;
+
+        layout->list[0].xlator = conf->subvolumes[i];
+
+        conf->file_layouts[i] = layout;
+    }
+
+    ret = 0;
 out:
-        return ret;
+    return ret;
 }
 
-
 int
-dht_disk_layout_extract (xlator_t *this, dht_layout_t *layout,
-                         int pos, int32_t **disk_layout_p)
+dht_disk_layout_extract(xlator_t *this, dht_layout_t *layout, int pos,
+                        int32_t **disk_layout_p)
 {
-        int      ret = -1;
-        int32_t *disk_layout = NULL;
+    int ret = -1;
+    int32_t *disk_layout = NULL;
 
-        disk_layout = GF_CALLOC (5, sizeof (int),
-                                 gf_dht_mt_int32_t);
-        if (!disk_layout) {
-                goto out;
-        }
+    disk_layout = GF_CALLOC(5, sizeof(int), gf_dht_mt_int32_t);
+    if (!disk_layout) {
+        goto out;
+    }
 
-        disk_layout[0] = hton32 (1);
-        disk_layout[1] = hton32 (layout->type);
-        disk_layout[2] = hton32 (layout->list[pos].start);
-        disk_layout[3] = hton32 (layout->list[pos].stop);
+    disk_layout[0] = hton32(layout->list[pos].commit_hash);
+    disk_layout[1] = hton32(layout->type);
+    disk_layout[2] = hton32(layout->list[pos].start);
+    disk_layout[3] = hton32(layout->list[pos].stop);
 
-        if (disk_layout_p)
-                *disk_layout_p = disk_layout;
-        else
-                GF_FREE (disk_layout);
+    if (disk_layout_p)
+        *disk_layout_p = disk_layout;
+    else
+        GF_FREE(disk_layout);
 
-        ret = 0;
+    ret = 0;
 
 out:
-        return ret;
+    return ret;
 }
 
-
 int
-dht_disk_layout_merge (xlator_t *this, dht_layout_t *layout,
-		       int pos, void *disk_layout_raw, int disk_layout_len)
+dht_disk_layout_extract_for_subvol(xlator_t *this, dht_layout_t *layout,
+                                   xlator_t *subvol, int32_t **disk_layout_p)
 {
-        int      cnt = 0;
-        int      type = 0;
-        int      start_off = 0;
-        int      stop_off = 0;
-        int      disk_layout[4];
-
-	if (!disk_layout_raw) {
-		gf_log (this->name, GF_LOG_CRITICAL,
-                        "error no layout on disk for merge");
-		return -1;
-	}
-
-	GF_ASSERT (disk_layout_len == sizeof (disk_layout));
-
-        memcpy (disk_layout, disk_layout_raw, disk_layout_len);
-
-        cnt  = ntoh32 (disk_layout[0]);
-        if (cnt != 1) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "disk layout has invalid count %d", cnt);
-                return -1;
-        }
+    int i = 0;
+
+    for (i = 0; i < layout->cnt; i++) {
+        if (layout->list[i].xlator == subvol)
+            break;
+    }
+
+    if (i == layout->cnt)
+        return -1;
+
+    return dht_disk_layout_extract(this, layout, i, disk_layout_p);
+}
 
-        type = ntoh32 (disk_layout[1]);
-	switch (type) {
+static int
+dht_disk_layout_merge(xlator_t *this, dht_layout_t *layout, int pos,
+                      void *disk_layout_raw, int disk_layout_len)
+{
+    int type = 0;
+    int start_off = 0;
+    int stop_off = 0;
+    int commit_hash = 0;
+    int disk_layout[4];
+
+    if (!disk_layout_raw) {
+        gf_smsg(this->name, GF_LOG_CRITICAL, 0, DHT_MSG_LAYOUT_MERGE_FAILED,
+                NULL);
+        return -1;
+    }
+
+    GF_ASSERT(disk_layout_len == sizeof(disk_layout));
+
+    memcpy(disk_layout, disk_layout_raw, disk_layout_len);
+
+    type = ntoh32(disk_layout[1]);
+    switch (type) {
         case DHT_HASH_TYPE_DM_USER:
-                gf_log (this->name, GF_LOG_DEBUG, "found user-set layout");
-                layout->type = type;
-                /* Fall through. */
-	case DHT_HASH_TYPE_DM:
-		break;
+            gf_msg_debug(this->name, 0, "found user-set layout");
+            layout->type = type;
+            /* Fall through. */
+        case DHT_HASH_TYPE_DM:
+            break;
         default:
-		gf_log (this->name, GF_LOG_CRITICAL,
-			"Catastrophic error layout with unknown type found %d",
-			disk_layout[1]);
-		return -1;
-	}
+            gf_smsg(this->name, GF_LOG_CRITICAL, 0, DHT_MSG_INVALID_DISK_LAYOUT,
+                    "layout=%d", disk_layout[1], NULL);
+            return -1;
+    }
 
-        start_off = ntoh32 (disk_layout[2]);
-        stop_off  = ntoh32 (disk_layout[3]);
+    commit_hash = ntoh32(disk_layout[0]);
+    start_off = ntoh32(disk_layout[2]);
+    stop_off = ntoh32(disk_layout[3]);
 
-        layout->list[pos].start = start_off;
-        layout->list[pos].stop  = stop_off;
+    layout->list[pos].commit_hash = commit_hash;
+    layout->list[pos].start = start_off;
+    layout->list[pos].stop = stop_off;
 
-        gf_log (this->name, GF_LOG_TRACE,
-                "merged to layout: %u - %u (type %d) from %s",
-                start_off, stop_off, type,
-                layout->list[pos].xlator->name);
+    gf_msg_trace(this->name, 0,
+                 "merged to layout: 0x%x - 0x%x (hash 0x%x, type %d) from %s",
+                 start_off, stop_off, commit_hash, type,
+                 layout->list[pos].xlator->name);
 
-        return 0;
+    return 0;
 }
 
-
 int
-dht_layout_merge (xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
-                  int op_ret, int op_errno, dict_t *xattr)
+dht_layout_merge(xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
+                 int op_ret, int op_errno, dict_t *xattr)
 {
-        int         i     = 0;
-        int         ret   = -1;
-        int         err   = -1;
-        void       *disk_layout_raw = NULL;
-        int         disk_layout_len = 0;
-        dht_conf_t *conf = this->private;
-
-        if (op_ret != 0) {
-                err = op_errno;
-        }
+    int i = 0;
+    int ret = -1;
+    int err = -1;
+    void *disk_layout_raw = NULL;
+    int disk_layout_len = 0;
+    dht_conf_t *conf = this->private;
 
-        for (i = 0; i < layout->cnt; i++) {
-                if (layout->list[i].xlator == NULL) {
-                        layout->list[i].err    = err;
-                        layout->list[i].xlator = subvol;
-                        break;
-                }
-        }
+    if (op_ret != 0) {
+        err = op_errno;
+    }
 
-        if (op_ret != 0) {
-                ret = 0;
-                goto out;
-        }
+    if (!layout)
+        goto out;
 
-        if (xattr) {
-                /* during lookup and not mkdir */
-                ret = dict_get_ptr_and_len (xattr, conf->xattr_name,
-					    &disk_layout_raw, &disk_layout_len);
+    for (i = 0; i < layout->cnt; i++) {
+        if (layout->list[i].xlator == NULL) {
+            layout->list[i].err = err;
+            layout->list[i].xlator = subvol;
+            break;
         }
+    }
 
-        if (ret != 0) {
-                layout->list[i].err = 0;
-                gf_log (this->name, GF_LOG_TRACE,
-                        "missing disk layout on %s. err = %d",
-                        subvol->name, err);
-                ret = 0;
-                goto out;
-        }
+    if (op_ret != 0) {
+        ret = 0;
+        goto out;
+    }
 
-        ret = dht_disk_layout_merge (this, layout, i, disk_layout_raw,
-				     disk_layout_len);
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "layout merge from subvolume %s failed",
-                        subvol->name);
-                goto out;
-        }
+    if (xattr) {
+        /* during lookup and not mkdir */
+        ret = dict_get_ptr_and_len(xattr, conf->xattr_name, &disk_layout_raw,
+                                   &disk_layout_len);
+    }
+
+    if (ret != 0) {
         layout->list[i].err = 0;
+        gf_msg_trace(this->name, 0, "Missing disk layout on %s. err = %d",
+                     subvol->name, err);
+        ret = 0;
+        goto out;
+    }
+
+    ret = dht_disk_layout_merge(this, layout, i, disk_layout_raw,
+                                disk_layout_len);
+    if (ret != 0) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LAYOUT_MERGE_FAILED,
+                "subvolume=%s", subvol->name, NULL);
+        goto out;
+    }
+
+    if (layout->commit_hash == 0) {
+        layout->commit_hash = layout->list[i].commit_hash;
+    } else if (layout->commit_hash != layout->list[i].commit_hash) {
+        layout->commit_hash = DHT_LAYOUT_HASH_INVALID;
+    }
+
+    layout->list[i].err = 0;
 
 out:
-        return ret;
+    return ret;
 }
 
-
 void
-dht_layout_entry_swap (dht_layout_t *layout, int i, int j)
+dht_layout_entry_swap(dht_layout_t *layout, int i, int j)
 {
-        uint32_t  start_swap = 0;
-        uint32_t  stop_swap = 0;
-        xlator_t *xlator_swap = 0;
-        int       err_swap = 0;
-
-        start_swap  = layout->list[i].start;
-        stop_swap   = layout->list[i].stop;
-        xlator_swap = layout->list[i].xlator;
-        err_swap    = layout->list[i].err;
-
-        layout->list[i].start  = layout->list[j].start;
-        layout->list[i].stop   = layout->list[j].stop;
-        layout->list[i].xlator = layout->list[j].xlator;
-        layout->list[i].err    = layout->list[j].err;
-
-        layout->list[j].start  = start_swap;
-        layout->list[j].stop   = stop_swap;
-        layout->list[j].xlator = xlator_swap;
-        layout->list[j].err    = err_swap;
+    uint32_t start_swap = 0;
+    uint32_t stop_swap = 0;
+    uint32_t commit_hash_swap = 0;
+    xlator_t *xlator_swap = 0;
+    int err_swap = 0;
+
+    start_swap = layout->list[i].start;
+    stop_swap = layout->list[i].stop;
+    xlator_swap = layout->list[i].xlator;
+    err_swap = layout->list[i].err;
+    commit_hash_swap = layout->list[i].commit_hash;
+
+    layout->list[i].start = layout->list[j].start;
+    layout->list[i].stop = layout->list[j].stop;
+    layout->list[i].xlator = layout->list[j].xlator;
+    layout->list[i].err = layout->list[j].err;
+    layout->list[i].commit_hash = layout->list[j].commit_hash;
+
+    layout->list[j].start = start_swap;
+    layout->list[j].stop = stop_swap;
+    layout->list[j].xlator = xlator_swap;
+    layout->list[j].err = err_swap;
+    layout->list[j].commit_hash = commit_hash_swap;
 }
 
 void
-dht_layout_range_swap (dht_layout_t *layout, int i, int j)
+dht_layout_range_swap(dht_layout_t *layout, int i, int j)
 {
-        uint32_t  start_swap = 0;
-        uint32_t  stop_swap = 0;
+    uint32_t start_swap = 0;
+    uint32_t stop_swap = 0;
 
-        start_swap  = layout->list[i].start;
-        stop_swap   = layout->list[i].stop;
+    start_swap = layout->list[i].start;
+    stop_swap = layout->list[i].stop;
 
-        layout->list[i].start  = layout->list[j].start;
-        layout->list[i].stop   = layout->list[j].stop;
+    layout->list[i].start = layout->list[j].start;
+    layout->list[i].stop = layout->list[j].stop;
 
-        layout->list[j].start  = start_swap;
-        layout->list[j].stop   = stop_swap;
+    layout->list[j].start = start_swap;
+    layout->list[j].stop = stop_swap;
 }
-
-int64_t
-dht_layout_entry_cmp_volname (dht_layout_t *layout, int i, int j)
+static int64_t
+dht_layout_entry_cmp_volname(dht_layout_t *layout, int i, int j)
 {
-        return (strcmp (layout->list[i].xlator->name,
-                        layout->list[j].xlator->name));
+    return (strcmp(layout->list[i].xlator->name, layout->list[j].xlator->name));
 }
 
-
 gf_boolean_t
-dht_is_subvol_in_layout (dht_layout_t *layout, xlator_t *xlator)
+dht_is_subvol_in_layout(dht_layout_t *layout, xlator_t *xlator)
 {
-        int i = 0;
-
-        for (i = 0; i < layout->cnt; i++) {
-                if (!strcmp (layout->list[i].xlator->name, xlator->name))
-                        return _gf_true;
-        }
-        return _gf_false;
+    int i = 0;
+
+    for (i = 0; i < layout->cnt; i++) {
+        /* Check if xlator is already part of layout, and layout is
+         * non-zero. */
+        if (!strcmp(layout->list[i].xlator->name, xlator->name)) {
+            if (layout->list[i].start != layout->list[i].stop)
+                return _gf_true;
+            break;
+        }
+    }
+    return _gf_false;
 }
 
-int64_t
-dht_layout_entry_cmp (dht_layout_t *layout, int i, int j)
+static int64_t
+dht_layout_entry_cmp(dht_layout_t *layout, int i, int j)
 {
-        int64_t diff = 0;
+    int64_t diff = 0;
 
-        /* swap zero'ed out layouts to front, if needed */
-        if (!layout->list[j].start && !layout->list[j].stop) {
-                diff = (int64_t) layout->list[i].stop
-                       - (int64_t) layout->list[j].stop;
-                       goto out;
-        }
-        if (layout->list[i].err || layout->list[j].err)
-                diff = layout->list[i].err - layout->list[j].err;
-        else
-                diff = (int64_t) layout->list[i].start
-                        - (int64_t) layout->list[j].start;
+    /* swap zero'ed out layouts to front, if needed */
+    if (!layout->list[j].start && !layout->list[j].stop) {
+        diff = (int64_t)layout->list[i].stop - (int64_t)layout->list[j].stop;
+        goto out;
+    }
+    diff = (int64_t)layout->list[i].start - (int64_t)layout->list[j].start;
 
 out:
-        return diff;
+    return diff;
 }
 
-
 int
-dht_layout_sort (dht_layout_t *layout)
+dht_layout_sort(dht_layout_t *layout)
 {
-        int       i = 0;
-        int       j = 0;
-        int64_t   ret = 0;
+    int i = 0;
+    int j = 0;
+    int64_t ret = 0;
 
-        /* TODO: O(n^2) -- bad bad */
+    /* TODO: O(n^2) -- bad bad */
 
-        for (i = 0; i < layout->cnt - 1; i++) {
-                for (j = i + 1; j < layout->cnt; j++) {
-                        ret = dht_layout_entry_cmp (layout, i, j);
-                        if (ret > 0)
-                                dht_layout_entry_swap (layout, i, j);
-                }
+    for (i = 0; i < layout->cnt - 1; i++) {
+        for (j = i + 1; j < layout->cnt; j++) {
+            ret = dht_layout_entry_cmp(layout, i, j);
+            if (ret > 0)
+                dht_layout_entry_swap(layout, i, j);
         }
+    }
 
-        return 0;
+    return 0;
 }
 
-int
-dht_layout_sort_volname (dht_layout_t *layout)
+void
+dht_layout_sort_volname(dht_layout_t *layout)
 {
-        int       i = 0;
-        int       j = 0;
-        int64_t   ret = 0;
+    int i = 0;
+    int j = 0;
+    int64_t ret = 0;
 
-        /* TODO: O(n^2) -- bad bad */
+    /* TODO: O(n^2) -- bad bad */
 
-        for (i = 0; i < layout->cnt - 1; i++) {
-                for (j = i + 1; j < layout->cnt; j++) {
-                        ret = dht_layout_entry_cmp_volname (layout, i, j);
-                        if (ret > 0)
-                                dht_layout_entry_swap (layout, i, j);
-                }
+    for (i = 0; i < layout->cnt - 1; i++) {
+        for (j = i + 1; j < layout->cnt; j++) {
+            ret = dht_layout_entry_cmp_volname(layout, i, j);
+            if (ret > 0)
+                dht_layout_entry_swap(layout, i, j);
         }
-
-        return 0;
+    }
 }
 
-
-int
-dht_layout_anomalies (xlator_t *this, loc_t *loc, dht_layout_t *layout,
-                      uint32_t *holes_p, uint32_t *overlaps_p,
-                      uint32_t *missing_p, uint32_t *down_p, uint32_t *misc_p,
-                      uint32_t *no_space_p)
+void
+dht_layout_anomalies(xlator_t *this, loc_t *loc, dht_layout_t *layout,
+                     uint32_t *holes_p, uint32_t *overlaps_p,
+                     uint32_t *missing_p, uint32_t *down_p, uint32_t *misc_p,
+                     uint32_t *no_space_p)
 {
-        uint32_t    overlaps = 0;
-        uint32_t    missing  = 0;
-        uint32_t    down     = 0;
-        uint32_t    misc     = 0;
-        uint32_t    hole_cnt = 0;
-        uint32_t    overlap_cnt = 0;
-        int         i = 0;
-        int         ret = 0;
-        uint32_t    prev_stop = 0;
-        uint32_t    last_stop = 0;
-        char        is_virgin = 1;
-        uint32_t    no_space  = 0;
-
-        /* TODO: explain what is happening */
-
-        last_stop = layout->list[0].start - 1;
-        prev_stop = last_stop;
-
-        for (i = 0; i < layout->cnt; i++) {
-                switch (layout->list[i].err) {
-                case -1:
-                case ENOENT:
-                        missing++;
-                        continue;
-                case ENOTCONN:
-                        down++;
-                        continue;
-                case ENOSPC:
-                        no_space++;
-                        continue;
-                case 0:
-                        /* if err == 0 and start == stop, then it is a non misc++;
-                         * participating subvolume(spread-cnt). Then, do not
-                         * check for anomalies. If start != stop, then treat it
-                         * as misc err */
-                        if (layout->list[i].start == layout->list[i].stop) {
-                                continue;
-                        }
-                        break;
-                default:
-                        misc++;
-                        continue;
-                 }
-
-                is_virgin = 0;
-
-                if ((prev_stop + 1) < layout->list[i].start) {
-                        hole_cnt++;
+    uint32_t overlaps = 0;
+    uint32_t missing = 0;
+    uint32_t down = 0;
+    uint32_t misc = 0;
+    uint32_t hole_cnt = 0;
+    uint32_t overlap_cnt = 0;
+    int i = 0;
+    uint32_t prev_stop = 0;
+    uint32_t last_stop = 0;
+    char is_virgin = 1;
+    uint32_t no_space = 0;
+
+    /* This function scans through the layout spread of a directory to
+       check if there are any anomalies. Prior to calling this function
+       the layout entries should be sorted in the ascending order.
+
+       If the layout entry has err != 0
+            then increment the corresponding anomaly.
+       else
+            if (start of the current layout entry > stop + 1 of previous
+               non erroneous layout entry)
+                    then it indicates a hole in the layout
+            if (start of the current layout entry < stop + 1 of previous
+                non erroneous layout entry)
+                     then it indicates an overlap in the layout
+    */
+    last_stop = layout->list[0].start - 1;
+    prev_stop = last_stop;
+
+    for (i = 0; i < layout->cnt; i++) {
+        switch (layout->list[i].err) {
+            case -1:
+            case ENOENT:
+            case ESTALE:
+                missing++;
+                continue;
+            case ENOTCONN:
+                down++;
+                continue;
+            case ENOSPC:
+                no_space++;
+                continue;
+            case 0:
+                /* if err == 0 and start == stop, then it is a non misc++;
+                 * participating subvolume(spread-cnt). Then, do not
+                 * check for anomalies. If start != stop, then treat it
+                 * as misc err */
+                if (layout->list[i].start == layout->list[i].stop) {
+                    continue;
                 }
+                break;
+            default:
+                misc++;
+                continue;
+        }
 
-                if ((prev_stop + 1) > layout->list[i].start) {
-                        overlap_cnt++;
-                        overlaps += ((prev_stop + 1) - layout->list[i].start);
-                }
-                prev_stop = layout->list[i].stop;
+        is_virgin = 0;
+
+        if ((prev_stop + 1) < layout->list[i].start) {
+            hole_cnt++;
         }
 
-        if ((last_stop - prev_stop) || is_virgin)
-                hole_cnt++;
+        if ((prev_stop + 1) > layout->list[i].start) {
+            overlap_cnt++;
+            overlaps += ((prev_stop + 1) - layout->list[i].start);
+        }
+        prev_stop = layout->list[i].stop;
+    }
 
-        if (holes_p)
-                *holes_p = hole_cnt;
+    if ((last_stop - prev_stop) || is_virgin)
+        hole_cnt++;
 
-        if (overlaps_p)
-                *overlaps_p = overlap_cnt;
+    if (holes_p)
+        *holes_p = hole_cnt;
 
-        if (missing_p)
-                *missing_p = missing;
+    if (overlaps_p)
+        *overlaps_p = overlap_cnt;
 
-        if (down_p)
-                *down_p = down;
+    if (missing_p)
+        *missing_p = missing;
 
-        if (misc_p)
-                *misc_p = misc;
+    if (down_p)
+        *down_p = down;
 
-        if (no_space_p)
-                *no_space_p = no_space;
+    if (misc_p)
+        *misc_p = misc;
 
-        return ret;
+    if (no_space_p)
+        *no_space_p = no_space;
 }
 
-
 int
-dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout)
+dht_layout_missing_dirs(dht_layout_t *layout)
 {
-        int          ret   = 0;
-        int          i = 0;
-        uint32_t     holes = 0;
-        uint32_t     overlaps = 0;
-        uint32_t     missing = 0;
-        uint32_t     down = 0;
-        uint32_t     misc = 0;
-
-        ret = dht_layout_sort (layout);
-        if (ret == -1) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "sort failed?! how the ....");
-                goto out;
-        }
+    int i = 0, missing = 0;
 
-        ret = dht_layout_anomalies (this, loc, layout,
-                                    &holes, &overlaps,
-                                    &missing, &down, &misc, NULL);
-        if (ret == -1) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "error while finding anomalies in %s -- not good news",
-                        loc->path);
-                goto out;
-        }
+    if (layout == NULL)
+        goto out;
 
-        if (holes || overlaps) {
-                if (missing == layout->cnt) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "directory %s looked up first time",
-                                loc->path);
-                } else {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "found anomalies in %s. holes=%d overlaps=%d",
-                                loc->path, holes, overlaps);
-                }
-                ret = -1;
+    for (i = 0; i < layout->cnt; i++) {
+        if ((layout->list[i].err == ENOENT) ||
+            ((layout->list[i].err == -1) && (layout->list[i].start == 0) &&
+             (layout->list[i].stop == 0))) {
+            missing++;
         }
+    }
 
-        for (i = 0; i < layout->cnt; i++) {
-                /* TODO During DHT selfheal rewrite (almost) find a better place
-                 * to detect this - probably in dht_layout_anomalies()
-                 */
-                if (layout->list[i].err > 0) {
-                        gf_log_callingfn (this->name, GF_LOG_DEBUG,
-                                          "path=%s err=%s on subvol=%s",
-                                          loc->path,
-                                          strerror (layout->list[i].err),
-                                          (layout->list[i].xlator ?
-                                           layout->list[i].xlator->name
-                                           : "<>"));
-                        if ((layout->list[i].err == ENOENT) && (ret >= 0)) {
-                                ret++;
-                        }
-                }
-        }
+out:
+    return missing;
+}
 
+int
+dht_layout_normalize(xlator_t *this, loc_t *loc, dht_layout_t *layout)
+{
+    int ret = 0;
+    uint32_t holes = 0;
+    uint32_t overlaps = 0;
+    uint32_t missing = 0;
+    uint32_t down = 0;
+    uint32_t misc = 0, missing_dirs = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    ret = dht_layout_sort(layout);
+    if (ret == -1) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LAYOUT_SORT_FAILED,
+                NULL);
+        goto out;
+    }
+
+    gf_uuid_unparse(loc->gfid, gfid);
+
+    dht_layout_anomalies(this, loc, layout, &holes, &overlaps, &missing, &down,
+                         &misc, NULL);
+
+    if (holes || overlaps) {
+        if (missing == layout->cnt) {
+            gf_msg_debug(this->name, 0,
+                         "Directory %s looked up first time"
+                         " gfid = %s",
+                         loc->path, gfid);
+        } else {
+            gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_ANOMALIES_INFO,
+                    "path=%s", loc->path, "gfid=%s", gfid, "holes=%d", holes,
+                    "overlaps=%d", overlaps, NULL);
+        }
+        ret = -1;
+    }
+
+    if (ret >= 0) {
+        missing_dirs = dht_layout_missing_dirs(layout);
+        /* TODO During DHT selfheal rewrite (almost) find a better place
+         * to detect this - probably in dht_layout_anomalies()
+         */
+        if (missing_dirs > 0)
+            ret += missing_dirs;
+    }
 
 out:
-        return ret;
+    return ret;
 }
 
 int
-dht_dir_has_layout (dict_t *xattr, char *name)
+dht_dir_has_layout(dict_t *xattr, char *name)
 {
+    void *disk_layout_raw = NULL;
 
-        void     *disk_layout_raw = NULL;
-
-        return dict_get_ptr (xattr, name, &disk_layout_raw);
+    return dict_get_ptr(xattr, name, &disk_layout_raw);
 }
 
 int
-dht_layout_dir_mismatch (xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
-                         loc_t *loc, dict_t *xattr)
+dht_layout_dir_mismatch(xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
+                        loc_t *loc, dict_t *xattr)
 {
-        int         idx = 0;
-        int         pos = -1;
-        int         ret = 0;
-        int         err = 0;
-        int         dict_ret = 0;
-        int32_t     disk_layout[4];
-        void       *disk_layout_raw = NULL;
-        int32_t     count = -1;
-        uint32_t    start_off = -1;
-        uint32_t    stop_off = -1;
-        dht_conf_t *conf = this->private;
-
-
-        for (idx = 0; idx < layout->cnt; idx++) {
-                if (layout->list[idx].xlator == subvol) {
-                        pos = idx;
-                        break;
-                }
-        }
-
-        if (pos == -1) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "%s - no layout info for subvolume %s",
-                        loc->path, subvol->name);
-                ret = 1;
-                goto out;
-        }
+    int idx = 0;
+    int pos = -1;
+    int ret = 0;
+    int err = 0;
+    int dict_ret = 0;
+    int32_t disk_layout[4];
+    void *disk_layout_raw = NULL;
+    uint32_t start_off = -1;
+    uint32_t stop_off = -1;
+    uint32_t commit_hash = -1;
+    dht_conf_t *conf = this->private;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    if (loc && loc->inode)
+        gf_uuid_unparse(loc->inode->gfid, gfid);
+
+    for (idx = 0; idx < layout->cnt; idx++) {
+        if (layout->list[idx].xlator == subvol) {
+            pos = idx;
+            break;
+        }
+    }
+
+    if (pos == -1) {
+        if (loc) {
+            gf_msg_debug(this->name, 0, "%s - no layout info for subvolume %s",
+                         loc ? loc->path : "path not found", subvol->name);
+        }
+        ret = 1;
+        goto out;
+    }
+
+    err = layout->list[pos].err;
+
+    if (!xattr) {
+        if (err == 0) {
+            if (loc) {
+                gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_XATTR_DICT_NULL,
+                        "path=%s", loc->path, NULL);
+            } else {
+                gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_XATTR_DICT_NULL,
+                        "path not found", NULL);
+            }
+            ret = -1;
+        }
+        goto out;
+    }
+
+    dict_ret = dict_get_ptr(xattr, conf->xattr_name, &disk_layout_raw);
+
+    if (dict_ret < 0) {
+        if (err == 0 && layout->list[pos].stop) {
+            if (loc) {
+                gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_DISK_LAYOUT_MISSING,
+                        "path=%s", loc->path, "gfid=%s", gfid, NULL);
+            } else {
+                gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_DISK_LAYOUT_MISSING,
+                        "path not found"
+                        "gfid=%s",
+                        gfid, NULL);
+            }
+            ret = -1;
+        }
+        goto out;
+    }
+
+    memcpy(disk_layout, disk_layout_raw, sizeof(disk_layout));
+
+    start_off = ntoh32(disk_layout[2]);
+    stop_off = ntoh32(disk_layout[3]);
+    commit_hash = ntoh32(disk_layout[0]);
+
+    if ((layout->list[pos].start != start_off) ||
+        (layout->list[pos].stop != stop_off) ||
+        (layout->list[pos].commit_hash != commit_hash)) {
+        gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_LAYOUT_INFO, "subvol=%s",
+                layout->list[pos].xlator->name, "inode-layout:start=0x%x",
+                layout->list[pos].start, "inode-layout:stop=0x%x",
+                layout->list[pos].stop, "layout-commit-hash=0x%x; ",
+                layout->list[pos].commit_hash, "disk-layout:start-off=0x%x",
+                start_off, "disk-layout:top-off=0x%x", stop_off,
+                "commit-hash=0x%x", commit_hash, NULL);
+        ret = 1;
+    } else {
+        ret = 0;
+    }
+out:
+    return ret;
+}
 
-        err = layout->list[pos].err;
+int
+dht_layout_preset(xlator_t *this, xlator_t *subvol, inode_t *inode)
+{
+    dht_layout_t *layout = NULL;
+    int ret = -1;
+    dht_conf_t *conf = NULL;
 
-        if (!xattr) {
-                if (err == 0) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "%s - xattr dictionary is NULL",
-                                loc->path);
-                        ret = -1;
-                }
-                goto out;
-        }
+    conf = this->private;
+    if (!conf)
+        goto out;
 
-        dict_ret = dict_get_ptr (xattr, conf->xattr_name,
-                                 &disk_layout_raw);
+    layout = dht_layout_for_subvol(this, subvol);
+    if (!layout) {
+        gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_NO_LAYOUT_INFO,
+                "subvolume=%s", subvol ? subvol->name : "<nil>", NULL);
+        ret = -1;
+        goto out;
+    }
 
-        if (dict_ret < 0) {
-                if (err == 0) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "%s - disk layout missing", loc->path);
-                        ret = -1;
-                }
-                goto out;
-        }
+    gf_msg_debug(this->name, 0, "file = %s, subvol = %s",
+                 uuid_utoa(inode->gfid), subvol ? subvol->name : "<nil>");
 
-        memcpy (disk_layout, disk_layout_raw, sizeof (disk_layout));
+    LOCK(&conf->layout_lock);
+    {
+        dht_inode_ctx_layout_set(inode, this, layout);
+    }
 
-        count  = ntoh32 (disk_layout[0]);
-        if (count != 1) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s - disk layout has invalid count %d",
-                        loc->path, count);
-                ret = -1;
-                goto out;
-        }
+    UNLOCK(&conf->layout_lock);
 
-        start_off = ntoh32 (disk_layout[2]);
-        stop_off  = ntoh32 (disk_layout[3]);
-
-        if ((layout->list[pos].start != start_off)
-            || (layout->list[pos].stop != stop_off)) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "subvol: %s; inode layout - %"PRIu32" - %"PRIu32"; "
-                        "disk layout - %"PRIu32" - %"PRIu32,
-                        layout->list[pos].xlator->name,
-                        layout->list[pos].start, layout->list[pos].stop,
-                        start_off, stop_off);
-                ret = 1;
-        } else {
-                ret = 0;
-        }
+    ret = 0;
 out:
-        return ret;
+    return ret;
 }
 
-
 int
-dht_layout_preset (xlator_t *this, xlator_t *subvol, inode_t *inode)
+dht_layout_index_for_subvol(dht_layout_t *layout, xlator_t *subvol)
 {
-        dht_layout_t *layout = NULL;
-        int           ret = -1;
-        dht_conf_t   *conf = NULL;
-
-        conf = this->private;
-        if (!conf)
-                goto out;
-
-        layout = dht_layout_for_subvol (this, subvol);
-        if (!layout) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "no pre-set layout for subvolume %s",
-                        subvol ? subvol->name : "<nil>");
-                ret = -1;
-                goto out;
-        }
+    int i = 0, ret = -1;
 
-        LOCK (&conf->layout_lock);
-        {
-                dht_inode_ctx_layout_set (inode, this, layout);
+    for (i = 0; i < layout->cnt; i++) {
+        if (layout->list[i].xlator == subvol) {
+            ret = i;
+            break;
         }
-        UNLOCK (&conf->layout_lock);
+    }
 
-        ret = 0;
-out:
-        return ret;
+    return ret;
 }
diff --git a/xlators/cluster/dht/src/dht-linkfile.c b/xlators/cluster/dht/src/dht-linkfile.c
index dbc9d0b3c7b..89ec6cca56e 100644
--- a/xlators/cluster/dht/src/dht-linkfile.c
+++ b/xlators/cluster/dht/src/dht-linkfile.c
@@ -8,321 +8,321 @@
   cases as published by the Free Software Foundation.
 */
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-
-#include "glusterfs.h"
-#include "xlator.h"
-#include "compat.h"
+#include <glusterfs/compat.h>
 #include "dht-common.h"
 
-int
-dht_linkfile_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int op_ret, int op_errno,
-                         inode_t *inode, struct iatt *stbuf, dict_t *xattr,
-                         struct iatt *postparent)
+static int
+dht_linkfile_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, inode_t *inode,
+                        struct iatt *stbuf, dict_t *xattr,
+                        struct iatt *postparent)
 {
-        char          is_linkfile   = 0;
-        dht_conf_t   *conf          = NULL;
-        dht_local_t  *local         = NULL;
-        call_frame_t *prev          = NULL;
-
-        local = frame->local;
-        prev = cookie;
-        conf = this->private;
-
-        if (op_ret)
-                goto out;
-
-        is_linkfile = check_is_linkfile (inode, stbuf, xattr,
-                                         conf->link_xattr_name);
-        if (!is_linkfile)
-                gf_log (this->name, GF_LOG_WARNING, "got non-linkfile %s:%s",
-                        prev->this->name, local->loc.path);
+    char is_linkfile = 0;
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    prev = cookie;
+    conf = this->private;
+
+    if (op_ret)
+        goto out;
+
+    gf_uuid_unparse(local->loc.gfid, gfid);
+
+    is_linkfile = check_is_linkfile(inode, stbuf, xattr, conf->link_xattr_name);
+    if (!is_linkfile)
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_NOT_LINK_FILE_ERROR,
+                "name=%s", prev->name, "path=%s", local->loc.path, "gfid=%s",
+                gfid, NULL);
 out:
-        local->linkfile.linkfile_cbk (frame, cookie, this, op_ret, op_errno,
-                                      inode, stbuf, postparent, postparent,
-                                      xattr);
-        return 0;
+    local->linkfile.linkfile_cbk(frame, cookie, this, op_ret, op_errno, inode,
+                                 stbuf, postparent, postparent, xattr);
+    return 0;
 }
 
-#define is_equal(a, b) (a == b)
-int
-dht_linkfile_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int op_ret, int op_errno, inode_t *inode,
-                         struct iatt *stbuf, struct iatt *preparent,
-                         struct iatt *postparent, dict_t *xdata)
+static int
+dht_linkfile_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, inode_t *inode,
+                        struct iatt *stbuf, struct iatt *preparent,
+                        struct iatt *postparent, dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
-        xlator_t     *subvol = NULL;
-        call_frame_t *prev = NULL;
-        dict_t       *xattrs = NULL;
-        dht_conf_t   *conf = NULL;
-        int           ret = -1;
-
-        local = frame->local;
-
-        if (!op_ret)
-                local->linked = _gf_true;
-
-        FRAME_SU_UNDO (frame, dht_local_t);
-
-        if (op_ret && (op_errno == EEXIST)) {
-                conf = this->private;
-                prev = cookie;
-                subvol = prev->this;
-                if (!subvol)
-                        goto out;
-                xattrs = dict_new ();
-                if (!xattrs)
-                        goto out;
-                ret = dict_set_uint32 (xattrs, conf->link_xattr_name, 256);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Failed to set linkto key");
-                        goto out;
-                }
-
-                STACK_WIND (frame, dht_linkfile_lookup_cbk, subvol,
-                            subvol->fops->lookup, &local->loc, xattrs);
-                if (xattrs)
-                        dict_unref (xattrs);
-                return 0;
+    dht_local_t *local = NULL;
+    xlator_t *subvol = NULL;
+    dict_t *xattrs = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
+
+    local = frame->local;
+
+    if (!op_ret)
+        local->linked = _gf_true;
+
+    FRAME_SU_UNDO(frame, dht_local_t);
+
+    if (op_ret && (op_errno == EEXIST)) {
+        conf = this->private;
+        subvol = cookie;
+        if (!subvol)
+            goto out;
+        xattrs = dict_new();
+        if (!xattrs)
+            goto out;
+        ret = dict_set_uint32(xattrs, conf->link_xattr_name, 256);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                    "mame=%s", conf->link_xattr_name, NULL);
+            goto out;
         }
-out:
-        local->linkfile.linkfile_cbk (frame, cookie, this, op_ret, op_errno,
-                                      inode, stbuf, preparent, postparent,
-                                      xdata);
+
+        STACK_WIND_COOKIE(frame, dht_linkfile_lookup_cbk, subvol, subvol,
+                          subvol->fops->lookup, &local->linkfile.loc, xattrs);
         if (xattrs)
-                dict_unref (xattrs);
+            dict_unref(xattrs);
         return 0;
+    }
+out:
+    local->linkfile.linkfile_cbk(frame, cookie, this, op_ret, op_errno, inode,
+                                 stbuf, preparent, postparent, xdata);
+    if (xattrs)
+        dict_unref(xattrs);
+    return 0;
 }
 
-
 int
-dht_linkfile_create (call_frame_t *frame, fop_mknod_cbk_t linkfile_cbk,
-                     xlator_t *this,
-                     xlator_t *tovol, xlator_t *fromvol, loc_t *loc)
+dht_linkfile_create(call_frame_t *frame, fop_mknod_cbk_t linkfile_cbk,
+                    xlator_t *this, xlator_t *tovol, xlator_t *fromvol,
+                    loc_t *loc)
 {
-        dht_local_t *local = NULL;
-        dict_t      *dict = NULL;
-        int          need_unref = 0;
-        int          ret = 0;
-        dht_conf_t  *conf = this->private;
-
-        local = frame->local;
-        local->linkfile.linkfile_cbk = linkfile_cbk;
-        local->linkfile.srcvol = tovol;
-
-        local->linked = _gf_false;
-
-        dict = local->params;
-        if (!dict) {
-                dict = dict_new ();
-                if (!dict)
-                        goto out;
-                need_unref = 1;
-        }
-
-        if (!uuid_is_null (local->gfid)) {
-                ret = dict_set_static_bin (dict, "gfid-req", local->gfid, 16);
-                if (ret)
-                        gf_log ("dht-linkfile", GF_LOG_INFO,
-                                "%s: gfid set failed", loc->path);
-        }
-
-        ret = dict_set_str (dict, GLUSTERFS_INTERNAL_FOP_KEY, "yes");
+    dht_local_t *local = NULL;
+    dict_t *dict = NULL;
+    int need_unref = 0;
+    int ret = 0;
+    dht_conf_t *conf = this->private;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    local->linkfile.linkfile_cbk = linkfile_cbk;
+    local->linkfile.srcvol = tovol;
+    loc_copy(&local->linkfile.loc, loc);
+
+    local->linked = _gf_false;
+
+    dict = local->params;
+    if (!dict) {
+        dict = dict_new();
+        if (!dict)
+            goto out;
+        need_unref = 1;
+    }
+
+    if (!gf_uuid_is_null(local->gfid)) {
+        gf_uuid_unparse(local->gfid, gfid);
+
+        ret = dict_set_gfuuid(dict, "gfid-req", local->gfid, true);
         if (ret)
-                gf_log ("dht-linkfile", GF_LOG_INFO,
-                        "%s: internal-fop set failed", loc->path);
-
-        ret = dict_set_str (dict, conf->link_xattr_name, tovol->name);
-
-        if (ret < 0) {
-                gf_log (frame->this->name, GF_LOG_INFO,
-                        "%s: failed to initialize linkfile data",
-                        loc->path);
-                goto out;
-        }
-
-        local->link_subvol = fromvol;
-        /* Always create as root:root. dht_linkfile_attr_heal fixes the
-         * ownsership */
-        FRAME_SU_DO (frame, dht_local_t);
-        STACK_WIND (frame, dht_linkfile_create_cbk,
-                    fromvol, fromvol->fops->mknod, loc,
-                    S_IFREG | DHT_LINKFILE_MODE, 0, 0, dict);
-
-        if (need_unref && dict)
-                dict_unref (dict);
-
-        return 0;
+            gf_smsg("dht-linkfile", GF_LOG_INFO, 0, DHT_MSG_DICT_SET_FAILED,
+                    "path=%s", loc->path, "gfid=%s", gfid, NULL);
+    } else {
+        gf_uuid_unparse(loc->gfid, gfid);
+    }
+
+    ret = dict_set_str(dict, GLUSTERFS_INTERNAL_FOP_KEY, "yes");
+    if (ret)
+        gf_smsg("dht-linkfile", GF_LOG_INFO, 0, DHT_MSG_DICT_SET_FAILED,
+                "path=%s", loc->path, "key=%s", GLUSTERFS_INTERNAL_FOP_KEY,
+                "gfid=%s", gfid, NULL);
+
+    ret = dict_set_str(dict, conf->link_xattr_name, tovol->name);
+
+    if (ret < 0) {
+        gf_smsg(frame->this->name, GF_LOG_INFO, 0, DHT_MSG_CREATE_LINK_FAILED,
+                "path=%s", loc->path, "gfid=%s", gfid, NULL);
+        goto out;
+    }
+
+    local->link_subvol = fromvol;
+    /* Always create as root:root. dht_linkfile_attr_heal fixes the
+     * ownsership */
+    FRAME_SU_DO(frame, dht_local_t);
+    STACK_WIND_COOKIE(frame, dht_linkfile_create_cbk, fromvol, fromvol,
+                      fromvol->fops->mknod, loc, S_IFREG | DHT_LINKFILE_MODE, 0,
+                      0, dict);
+
+    if (need_unref && dict)
+        dict_unref(dict);
+
+    return 0;
 out:
-        local->linkfile.linkfile_cbk (frame, NULL, frame->this, -1, ENOMEM,
-                                      loc->inode, NULL, NULL, NULL, NULL);
+    local->linkfile.linkfile_cbk(frame, frame->this, frame->this, -1, ENOMEM,
+                                 loc->inode, NULL, NULL, NULL, NULL);
 
-        if (need_unref && dict)
-                dict_unref (dict);
+    if (need_unref && dict)
+        dict_unref(dict);
 
-        return 0;
+    return 0;
 }
 
-
 int
-dht_linkfile_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno,
-                         struct iatt *preparent, struct iatt *postparent,
-                         dict_t *xdata)
+dht_linkfile_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno,
+                        struct iatt *preparent, struct iatt *postparent,
+                        dict_t *xdata)
 {
-        dht_local_t   *local = NULL;
-        call_frame_t  *prev = NULL;
-        xlator_t      *subvol = NULL;
-
-        local = frame->local;
-        prev = cookie;
-        subvol = prev->this;
-
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "unlinking linkfile %s on %s failed (%s)",
-                        local->loc.path, subvol->name, strerror (op_errno));
-        }
+    dht_local_t *local = NULL;
+    xlator_t *subvol = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
 
-        DHT_STACK_DESTROY (frame);
+    local = frame->local;
+    subvol = cookie;
 
-        return 0;
-}
+    if (op_ret == -1) {
+        gf_uuid_unparse(local->loc.gfid, gfid);
+        gf_smsg(this->name, GF_LOG_INFO, op_errno, DHT_MSG_UNLINK_FAILED,
+                "path=%s", local->loc.path, "gfid=%s", gfid, "subvolume=%s",
+                subvol->name, NULL);
+    }
 
+    DHT_STACK_DESTROY(frame);
+
+    return 0;
+}
 
 int
-dht_linkfile_unlink (call_frame_t *frame, xlator_t *this,
-                     xlator_t *subvol, loc_t *loc)
+dht_linkfile_unlink(call_frame_t *frame, xlator_t *this, xlator_t *subvol,
+                    loc_t *loc)
 {
-        call_frame_t *unlink_frame = NULL;
-        dht_local_t  *unlink_local = NULL;
+    call_frame_t *unlink_frame = NULL;
+    dht_local_t *unlink_local = NULL;
 
-        unlink_frame = copy_frame (frame);
-        if (!unlink_frame) {
-                goto err;
-        }
+    unlink_frame = copy_frame(frame);
+    if (!unlink_frame) {
+        goto err;
+    }
 
-        /* Using non-fop value here, as anyways, 'local->fop' is not used in
-           this particular case */
-        unlink_local = dht_local_init (unlink_frame, loc, NULL,
-                                       GF_FOP_MAXVALUE);
-        if (!unlink_local) {
-                goto err;
-        }
+    /* Using non-fop value here, as anyways, 'local->fop' is not used in
+       this particular case */
+    unlink_local = dht_local_init(unlink_frame, loc, NULL, GF_FOP_MAXVALUE);
+    if (!unlink_local) {
+        goto err;
+    }
 
-        STACK_WIND (unlink_frame, dht_linkfile_unlink_cbk,
-                    subvol, subvol->fops->unlink,
-                    &unlink_local->loc, 0, NULL);
+    STACK_WIND_COOKIE(unlink_frame, dht_linkfile_unlink_cbk, subvol, subvol,
+                      subvol->fops->unlink, &unlink_local->loc, 0, NULL);
 
-        return 0;
+    return 0;
 err:
-        if (unlink_frame)
-                DHT_STACK_DESTROY (unlink_frame);
+    if (unlink_frame)
+        DHT_STACK_DESTROY(unlink_frame);
 
-        return -1;
+    return -1;
 }
 
-
 xlator_t *
-dht_linkfile_subvol (xlator_t *this, inode_t *inode, struct iatt *stbuf,
-                     dict_t *xattr)
+dht_linkfile_subvol(xlator_t *this, inode_t *inode, struct iatt *stbuf,
+                    dict_t *xattr)
 {
-        dht_conf_t *conf = NULL;
-        xlator_t   *subvol = NULL;
-        void       *volname = NULL;
-        int         i = 0, ret = 0;
+    dht_conf_t *conf = NULL;
+    xlator_t *subvol = NULL;
+    void *volname = NULL;
+    int i = 0, ret = 0;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!xattr)
-                goto out;
+    if (!xattr)
+        goto out;
 
-        ret = dict_get_ptr (xattr, conf->link_xattr_name, &volname);
+    ret = dict_get_ptr(xattr, conf->link_xattr_name, &volname);
 
-        if ((-1 == ret) || !volname)
-                goto out;
+    if ((-1 == ret) || !volname)
+        goto out;
 
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                if (strcmp (conf->subvolumes[i]->name, (char *)volname) == 0) {
-                        subvol = conf->subvolumes[i];
-                        break;
-                }
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (strcmp(conf->subvolumes[i]->name, (char *)volname) == 0) {
+            subvol = conf->subvolumes[i];
+            break;
         }
+    }
 
 out:
-        return subvol;
+    return subvol;
 }
 
-int
-dht_linkfile_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                          int op_ret, int op_errno, struct iatt *statpre,
-                          struct iatt *statpost, dict_t *xdata)
+static int
+dht_linkfile_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int op_ret, int op_errno, struct iatt *statpre,
+                         struct iatt *statpost, dict_t *xdata)
 {
-        dht_local_t *local = NULL;
-        loc_t *loc = NULL;
+    dht_local_t *local = NULL;
+    loc_t *loc = NULL;
 
-        local = frame->local;
-        loc = &local->loc;
+    local = frame->local;
+    loc = &local->loc;
 
-        if (op_ret)
-                gf_log (this->name, GF_LOG_ERROR, "setattr of uid/gid on %s"
-                        " :<gfid:%s> failed (%s)",
-                        (loc->path? loc->path: "NULL"),
-                        uuid_utoa(local->gfid), strerror(op_errno));
+    if (op_ret)
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno, DHT_MSG_SETATTR_FAILED,
+                "path=%s", (loc->path ? loc->path : "NULL"), "gfid=%s",
+                uuid_utoa(local->gfid), NULL);
 
-        DHT_STACK_DESTROY (frame);
+    DHT_STACK_DESTROY(frame);
 
-       return 0;
+    return 0;
 }
 
 int
-dht_linkfile_attr_heal (call_frame_t *frame, xlator_t *this)
+dht_linkfile_attr_heal(call_frame_t *frame, xlator_t *this)
 {
-        int     ret = -1;
-        call_frame_t *copy = NULL;
-        dht_local_t  *local = NULL;
-        dht_local_t  *copy_local = NULL;
-        xlator_t     *subvol = NULL;
-        struct iatt   stbuf = {0,};
+    int ret = -1;
+    call_frame_t *copy = NULL;
+    dht_local_t *local = NULL;
+    dht_local_t *copy_local = NULL;
+    xlator_t *subvol = NULL;
+    struct iatt stbuf = {
+        0,
+    };
+    dict_t *xattr = NULL;
+
+    local = frame->local;
+
+    GF_VALIDATE_OR_GOTO("dht", local, out);
+    GF_VALIDATE_OR_GOTO("dht", local->link_subvol, out);
+
+    if (local->stbuf.ia_type == IA_INVAL)
+        return 0;
 
-        local = frame->local;
+    DHT_MARK_FOP_INTERNAL(xattr);
 
-        GF_VALIDATE_OR_GOTO ("dht", local, out);
-        GF_VALIDATE_OR_GOTO ("dht", local->link_subvol, out);
+    gf_uuid_copy(local->loc.gfid, local->stbuf.ia_gfid);
 
-        if (local->stbuf.ia_type == IA_INVAL)
-                return 0;
+    copy = copy_frame(frame);
 
-        uuid_copy (local->loc.gfid, local->stbuf.ia_gfid);
+    if (!copy)
+        goto out;
 
-        copy = copy_frame (frame);
+    copy_local = dht_local_init(copy, &local->loc, NULL, 0);
 
-        if (!copy)
-                goto out;
+    if (!copy_local)
+        goto out;
 
-       copy_local = dht_local_init (copy, &local->loc, NULL, 0);
+    stbuf = local->stbuf;
+    subvol = local->link_subvol;
 
-       if (!copy_local)
-               goto out;
+    copy->local = copy_local;
 
-        stbuf = local->stbuf;
-        subvol = local->link_subvol;
+    FRAME_SU_DO(copy, dht_local_t);
 
-        copy->local = copy_local;
+    STACK_WIND(copy, dht_linkfile_setattr_cbk, subvol, subvol->fops->setattr,
+               &copy_local->loc, &stbuf, (GF_SET_ATTR_UID | GF_SET_ATTR_GID),
+               xattr);
+    ret = 0;
+out:
+    if ((ret < 0) && (copy))
+        DHT_STACK_DESTROY(copy);
 
-        FRAME_SU_DO (copy, dht_local_t);
+    if (xattr)
+        dict_unref(xattr);
 
-        STACK_WIND (copy, dht_linkfile_setattr_cbk, subvol,
-                    subvol->fops->setattr, &copy_local->loc,
-                    &stbuf, (GF_SET_ATTR_UID | GF_SET_ATTR_GID), NULL);
-        ret = 0;
-out:
-        return ret;
+    return ret;
 }
diff --git a/xlators/cluster/dht/src/dht-lock.c b/xlators/cluster/dht/src/dht-lock.c
new file mode 100644
index 00000000000..638821ccee5
--- /dev/null
+++ b/xlators/cluster/dht/src/dht-lock.c
@@ -0,0 +1,1392 @@
+/*
+  Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "dht-lock.h"
+
+static char *
+dht_lock_asprintf(dht_lock_t *lock)
+{
+    char *lk_buf = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {
+        0,
+    };
+
+    if (lock == NULL)
+        goto out;
+
+    uuid_utoa_r(lock->loc.gfid, gfid);
+
+    gf_asprintf(&lk_buf, "%s:%s", lock->xl->name, gfid);
+
+out:
+    return lk_buf;
+}
+
+static void
+dht_log_lk_array(char *name, gf_loglevel_t log_level, dht_lock_t **lk_array,
+                 int count)
+{
+    int i = 0;
+    char *lk_buf = NULL;
+
+    if ((lk_array == NULL) || (count == 0))
+        goto out;
+
+    for (i = 0; i < count; i++) {
+        lk_buf = dht_lock_asprintf(lk_array[i]);
+        if (!lk_buf)
+            goto out;
+
+        gf_smsg(name, log_level, 0, DHT_MSG_LK_ARRAY_INFO, "index=%d", i,
+                "lk_buf=%s", lk_buf, NULL);
+        GF_FREE(lk_buf);
+    }
+
+out:
+    return;
+}
+
+static void
+dht_lock_stack_destroy(call_frame_t *lock_frame, dht_lock_type_t lk)
+{
+    dht_local_t *local = NULL;
+
+    local = lock_frame->local;
+
+    if (lk == DHT_INODELK) {
+        local->lock[0].layout.my_layout.locks = NULL;
+        local->lock[0].layout.my_layout.lk_count = 0;
+    } else {
+        local->lock[0].ns.directory_ns.locks = NULL;
+        local->lock[0].ns.directory_ns.lk_count = 0;
+    }
+
+    DHT_STACK_DESTROY(lock_frame);
+    return;
+}
+
+static void
+dht_lock_free(dht_lock_t *lock)
+{
+    if (lock == NULL)
+        goto out;
+
+    loc_wipe(&lock->loc);
+    GF_FREE(lock->domain);
+    GF_FREE(lock->basename);
+    mem_put(lock);
+
+out:
+    return;
+}
+
+static void
+dht_set_lkowner(dht_lock_t **lk_array, int count, gf_lkowner_t *lkowner)
+{
+    int i = 0;
+
+    if (!lk_array || !lkowner)
+        goto out;
+
+    for (i = 0; i < count; i++) {
+        lk_array[i]->lk_owner = *lkowner;
+    }
+
+out:
+    return;
+}
+
+static int
+dht_lock_request_cmp(const void *val1, const void *val2)
+{
+    dht_lock_t *lock1 = NULL;
+    dht_lock_t *lock2 = NULL;
+    int ret = -1;
+
+    lock1 = *(dht_lock_t **)val1;
+    lock2 = *(dht_lock_t **)val2;
+
+    GF_VALIDATE_OR_GOTO("dht-locks", lock1, out);
+    GF_VALIDATE_OR_GOTO("dht-locks", lock2, out);
+
+    ret = strcmp(lock1->xl->name, lock2->xl->name);
+
+    if (ret == 0) {
+        ret = gf_uuid_compare(lock1->loc.gfid, lock2->loc.gfid);
+    }
+
+out:
+    return ret;
+}
+
+static int
+dht_lock_order_requests(dht_lock_t **locks, int count)
+{
+    int ret = -1;
+
+    if (!locks || !count)
+        goto out;
+
+    qsort(locks, count, sizeof(*locks), dht_lock_request_cmp);
+    ret = 0;
+
+out:
+    return ret;
+}
+
+void
+dht_lock_array_free(dht_lock_t **lk_array, int count)
+{
+    int i = 0;
+    dht_lock_t *lock = NULL;
+
+    if (lk_array == NULL)
+        goto out;
+
+    for (i = 0; i < count; i++) {
+        lock = lk_array[i];
+        lk_array[i] = NULL;
+        dht_lock_free(lock);
+    }
+
+out:
+    return;
+}
+
+int32_t
+dht_lock_count(dht_lock_t **lk_array, int lk_count)
+{
+    int i = 0, locked = 0;
+
+    if ((lk_array == NULL) || (lk_count == 0))
+        goto out;
+
+    for (i = 0; i < lk_count; i++) {
+        if (lk_array[i]->locked)
+            locked++;
+    }
+out:
+    return locked;
+}
+
+static call_frame_t *
+dht_lock_frame(call_frame_t *parent_frame)
+{
+    call_frame_t *lock_frame = NULL;
+
+    lock_frame = copy_frame(parent_frame);
+    if (lock_frame == NULL)
+        goto out;
+
+    set_lk_owner_from_ptr(&lock_frame->root->lk_owner, parent_frame->root);
+
+out:
+    return lock_frame;
+}
+
+dht_lock_t *
+dht_lock_new(xlator_t *this, xlator_t *xl, loc_t *loc, short type,
+             const char *domain, const char *basename,
+             dht_reaction_type_t do_on_failure)
+{
+    dht_conf_t *conf = NULL;
+    dht_lock_t *lock = NULL;
+
+    conf = this->private;
+
+    lock = mem_get0(conf->lock_pool);
+    if (lock == NULL)
+        goto out;
+
+    lock->xl = xl;
+    lock->type = type;
+    lock->do_on_failure = do_on_failure;
+
+    lock->domain = gf_strdup(domain);
+    if (lock->domain == NULL) {
+        dht_lock_free(lock);
+        lock = NULL;
+        goto out;
+    }
+
+    if (basename) {
+        lock->basename = gf_strdup(basename);
+        if (lock->basename == NULL) {
+            dht_lock_free(lock);
+            lock = NULL;
+            goto out;
+        }
+    }
+
+    /* Fill only inode and gfid.
+       posix and protocol/server give preference to pargfid/basename over
+       gfid/inode for resolution if all the three parameters of loc_t are
+       present. I want to avoid the following hypothetical situation:
+
+       1. rebalance did a lookup on a dentry and got a gfid.
+       2. rebalance acquires lock on loc_t which was filled with gfid and
+          path (pargfid/bname) from step 1.
+       3. somebody deleted and recreated the same file
+       4. rename on the same path acquires lock on loc_t which now points
+          to a different inode (and hence gets the lock).
+       5. rebalance continues to migrate file (note that not all fops done
+          by rebalance during migration are inode/gfid based Eg., unlink)
+       6. rename continues.
+    */
+    lock->loc.inode = inode_ref(loc->inode);
+    loc_gfid(loc, lock->loc.gfid);
+
+out:
+    return lock;
+}
+
+static int
+dht_local_entrylk_init(call_frame_t *frame, dht_lock_t **lk_array, int lk_count,
+                       fop_entrylk_cbk_t entrylk_cbk)
+{
+    int ret = -1;
+    dht_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local == NULL) {
+        local = dht_local_init(frame, NULL, NULL, 0);
+    }
+
+    if (local == NULL) {
+        goto out;
+    }
+
+    local->lock[0].ns.directory_ns.entrylk_cbk = entrylk_cbk;
+    local->lock[0].ns.directory_ns.locks = lk_array;
+    local->lock[0].ns.directory_ns.lk_count = lk_count;
+
+    ret = dht_lock_order_requests(local->lock[0].ns.directory_ns.locks,
+                                  local->lock[0].ns.directory_ns.lk_count);
+    if (ret < 0)
+        goto out;
+
+    ret = 0;
+out:
+    return ret;
+}
+
+static void
+dht_entrylk_done(call_frame_t *lock_frame)
+{
+    fop_entrylk_cbk_t entrylk_cbk = NULL;
+    call_frame_t *main_frame = NULL;
+    dht_local_t *local = NULL;
+
+    local = lock_frame->local;
+    main_frame = local->main_frame;
+
+    local->lock[0].ns.directory_ns.locks = NULL;
+    local->lock[0].ns.directory_ns.lk_count = 0;
+
+    entrylk_cbk = local->lock[0].ns.directory_ns.entrylk_cbk;
+    local->lock[0].ns.directory_ns.entrylk_cbk = NULL;
+
+    entrylk_cbk(main_frame, NULL, main_frame->this,
+                local->lock[0].ns.directory_ns.op_ret,
+                local->lock[0].ns.directory_ns.op_errno, NULL);
+
+    dht_lock_stack_destroy(lock_frame, DHT_ENTRYLK);
+    return;
+}
+
+static int32_t
+dht_unlock_entrylk_done(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    gf_uuid_unparse(local->lock[0].ns.directory_ns.locks[0]->loc.inode->gfid,
+                    gfid);
+
+    if (op_ret < 0) {
+        gf_smsg(this->name, GF_LOG_WARNING, op_errno,
+                DHT_MSG_UNLOCK_GFID_FAILED, "gfid=%s", gfid,
+                "DHT_LAYOUT_HEAL_DOMAIN", NULL);
+    }
+
+    DHT_STACK_DESTROY(frame);
+    return 0;
+}
+
+static int32_t
+dht_unlock_entrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int lk_index = 0, call_cnt = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    lk_index = (long)cookie;
+
+    local = frame->local;
+
+    uuid_utoa_r(local->lock[0].ns.directory_ns.locks[lk_index]->loc.gfid, gfid);
+
+    if (op_ret < 0) {
+        gf_smsg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_UNLOCKING_FAILED,
+                "name=%s",
+                local->lock[0].ns.directory_ns.locks[lk_index]->xl->name,
+                "gfid=%s", gfid, NULL);
+    } else {
+        local->lock[0].ns.directory_ns.locks[lk_index]->locked = 0;
+    }
+
+    call_cnt = dht_frame_return(frame);
+    if (is_last_call(call_cnt)) {
+        dht_entrylk_done(frame);
+    }
+
+    return 0;
+}
+
+static int32_t
+dht_unlock_entrylk(call_frame_t *frame, dht_lock_t **lk_array, int lk_count,
+                   fop_entrylk_cbk_t entrylk_cbk)
+{
+    dht_local_t *local = NULL;
+    int ret = -1, i = 0;
+    call_frame_t *lock_frame = NULL;
+    int call_cnt = 0;
+
+    GF_VALIDATE_OR_GOTO("dht-locks", frame, done);
+    GF_VALIDATE_OR_GOTO(frame->this->name, lk_array, done);
+    GF_VALIDATE_OR_GOTO(frame->this->name, entrylk_cbk, done);
+
+    call_cnt = dht_lock_count(lk_array, lk_count);
+    if (call_cnt == 0) {
+        ret = 0;
+        goto done;
+    }
+
+    lock_frame = dht_lock_frame(frame);
+    if (lock_frame == NULL) {
+        gf_smsg(frame->this->name, GF_LOG_WARNING, 0,
+                DHT_MSG_ALLOC_FRAME_FAILED_NOT_UNLOCKING_FOLLOWING_ENTRYLKS,
+                NULL);
+
+        dht_log_lk_array(frame->this->name, GF_LOG_WARNING, lk_array, lk_count);
+        goto done;
+    }
+
+    ret = dht_local_entrylk_init(lock_frame, lk_array, lk_count, entrylk_cbk);
+    if (ret < 0) {
+        gf_smsg(frame->this->name, GF_LOG_WARNING, 0,
+                DHT_MSG_LOCAL_LOCKS_STORE_FAILED_UNLOCKING_FOLLOWING_ENTRYLK,
+                NULL);
+
+        dht_log_lk_array(frame->this->name, GF_LOG_WARNING, lk_array, lk_count);
+
+        goto done;
+    }
+
+    local = lock_frame->local;
+    local->main_frame = frame;
+    local->call_cnt = call_cnt;
+
+    for (i = 0; i < local->lock[0].ns.directory_ns.lk_count; i++) {
+        if (!local->lock[0].ns.directory_ns.locks[i]->locked)
+            continue;
+
+        lock_frame->root
+            ->lk_owner = local->lock[0].ns.directory_ns.locks[i]->lk_owner;
+        STACK_WIND_COOKIE(
+            lock_frame, dht_unlock_entrylk_cbk, (void *)(long)i,
+            local->lock[0].ns.directory_ns.locks[i]->xl,
+            local->lock[0].ns.directory_ns.locks[i]->xl->fops->entrylk,
+            local->lock[0].ns.directory_ns.locks[i]->domain,
+            &local->lock[0].ns.directory_ns.locks[i]->loc,
+            local->lock[0].ns.directory_ns.locks[i]->basename, ENTRYLK_UNLOCK,
+            ENTRYLK_WRLCK, NULL);
+        if (!--call_cnt)
+            break;
+    }
+
+    return 0;
+
+done:
+    if (lock_frame)
+        dht_lock_stack_destroy(lock_frame, DHT_ENTRYLK);
+
+    /* no locks acquired, invoke entrylk_cbk */
+    if (ret == 0)
+        entrylk_cbk(frame, NULL, frame->this, 0, 0, NULL);
+
+    return ret;
+}
+
+int32_t
+dht_unlock_entrylk_wrapper(call_frame_t *frame, dht_elock_wrap_t *entrylk)
+{
+    dht_local_t *local = NULL, *lock_local = NULL;
+    call_frame_t *lock_frame = NULL;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    int ret = 0;
+
+    local = frame->local;
+
+    if (!entrylk || !entrylk->locks)
+        goto out;
+
+    gf_uuid_unparse(local->loc.parent->gfid, pgfid);
+
+    lock_frame = copy_frame(frame);
+    if (lock_frame == NULL) {
+        gf_smsg(frame->this->name, GF_LOG_WARNING, ENOMEM,
+                DHT_MSG_COPY_FRAME_FAILED, "pgfid=%s", pgfid, "name=%s",
+                local->loc.name, "path=%s", local->loc.path, NULL);
+        goto done;
+    }
+
+    lock_local = dht_local_init(lock_frame, NULL, NULL, 0);
+    if (lock_local == NULL) {
+        gf_smsg(frame->this->name, GF_LOG_WARNING, ENOMEM,
+                DHT_MSG_CREATE_FAILED, "local", "pgfid=%s", pgfid, "name=%s",
+                local->loc.name, "path=%s", local->loc.path, NULL);
+        goto done;
+    }
+
+    lock_frame->local = lock_local;
+
+    lock_local->lock[0].ns.directory_ns.locks = entrylk->locks;
+    lock_local->lock[0].ns.directory_ns.lk_count = entrylk->lk_count;
+    entrylk->locks = NULL;
+    entrylk->lk_count = 0;
+
+    ret = dht_unlock_entrylk(
+        lock_frame, lock_local->lock[0].ns.directory_ns.locks,
+        lock_local->lock[0].ns.directory_ns.lk_count, dht_unlock_entrylk_done);
+    if (ret)
+        goto done;
+
+    lock_frame = NULL;
+
+done:
+    if (lock_frame != NULL) {
+        DHT_STACK_DESTROY(lock_frame);
+    }
+
+out:
+    return 0;
+}
+
+static int
+dht_entrylk_cleanup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_entrylk_done(frame);
+    return 0;
+}
+
+static void
+dht_entrylk_cleanup(call_frame_t *lock_frame)
+{
+    dht_lock_t **lk_array = NULL;
+    int lk_count = 0, lk_acquired = 0;
+    dht_local_t *local = NULL;
+
+    local = lock_frame->local;
+
+    lk_array = local->lock[0].ns.directory_ns.locks;
+    lk_count = local->lock[0].ns.directory_ns.lk_count;
+
+    lk_acquired = dht_lock_count(lk_array, lk_count);
+    if (lk_acquired != 0) {
+        dht_unlock_entrylk(lock_frame, lk_array, lk_count,
+                           dht_entrylk_cleanup_cbk);
+    } else {
+        dht_entrylk_done(lock_frame);
+    }
+
+    return;
+}
+
+static int32_t
+dht_blocking_entrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    int lk_index = 0;
+    int i = 0;
+    dht_local_t *local = NULL;
+
+    lk_index = (long)cookie;
+
+    local = frame->local;
+    if (op_ret == 0) {
+        local->lock[0].ns.directory_ns.locks[lk_index]->locked = _gf_true;
+    } else {
+        switch (op_errno) {
+            case ESTALE:
+            case ENOENT:
+                if (local->lock[0]
+                        .ns.directory_ns.locks[lk_index]
+                        ->do_on_failure != IGNORE_ENOENT_ESTALE) {
+                    local->lock[0].ns.directory_ns.op_ret = -1;
+                    local->lock[0].ns.directory_ns.op_errno = op_errno;
+                    goto cleanup;
+                }
+                break;
+            default:
+                local->lock[0].ns.directory_ns.op_ret = -1;
+                local->lock[0].ns.directory_ns.op_errno = op_errno;
+                goto cleanup;
+        }
+    }
+
+    if (lk_index == (local->lock[0].ns.directory_ns.lk_count - 1)) {
+        for (i = 0; (i < local->lock[0].ns.directory_ns.lk_count) &&
+                    (!local->lock[0].ns.directory_ns.locks[i]->locked);
+             i++)
+            ;
+
+        if (i == local->lock[0].ns.directory_ns.lk_count) {
+            local->lock[0].ns.directory_ns.op_ret = -1;
+            local->lock[0].ns.directory_ns.op_errno = op_errno;
+        }
+
+        dht_entrylk_done(frame);
+    } else {
+        dht_blocking_entrylk_rec(frame, ++lk_index);
+    }
+
+    return 0;
+
+cleanup:
+    dht_entrylk_cleanup(frame);
+
+    return 0;
+}
+
+void
+dht_blocking_entrylk_rec(call_frame_t *frame, int i)
+{
+    dht_local_t *local = NULL;
+
+    local = frame->local;
+
+    STACK_WIND_COOKIE(
+        frame, dht_blocking_entrylk_cbk, (void *)(long)i,
+        local->lock[0].ns.directory_ns.locks[i]->xl,
+        local->lock[0].ns.directory_ns.locks[i]->xl->fops->entrylk,
+        local->lock[0].ns.directory_ns.locks[i]->domain,
+        &local->lock[0].ns.directory_ns.locks[i]->loc,
+        local->lock[0].ns.directory_ns.locks[i]->basename, ENTRYLK_LOCK,
+        ENTRYLK_WRLCK, NULL);
+
+    return;
+}
+
+int
+dht_blocking_entrylk(call_frame_t *frame, dht_lock_t **lk_array, int lk_count,
+                     fop_entrylk_cbk_t entrylk_cbk)
+{
+    int ret = -1;
+    call_frame_t *lock_frame = NULL;
+    dht_local_t *local = NULL;
+
+    GF_VALIDATE_OR_GOTO("dht-locks", frame, out);
+    GF_VALIDATE_OR_GOTO(frame->this->name, lk_array, out);
+    GF_VALIDATE_OR_GOTO(frame->this->name, entrylk_cbk, out);
+
+    lock_frame = dht_lock_frame(frame);
+    if (lock_frame == NULL)
+        goto out;
+
+    ret = dht_local_entrylk_init(lock_frame, lk_array, lk_count, entrylk_cbk);
+    if (ret < 0) {
+        goto out;
+    }
+
+    dht_set_lkowner(lk_array, lk_count, &lock_frame->root->lk_owner);
+
+    local = lock_frame->local;
+    local->main_frame = frame;
+
+    dht_blocking_entrylk_rec(lock_frame, 0);
+
+    return 0;
+out:
+    if (lock_frame)
+        dht_lock_stack_destroy(lock_frame, DHT_ENTRYLK);
+
+    return -1;
+}
+
+static int
+dht_local_inodelk_init(call_frame_t *frame, dht_lock_t **lk_array, int lk_count,
+                       fop_inodelk_cbk_t inodelk_cbk)
+{
+    int ret = -1;
+    dht_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local == NULL) {
+        local = dht_local_init(frame, NULL, NULL, 0);
+    }
+
+    if (local == NULL) {
+        goto out;
+    }
+
+    local->lock[0].layout.my_layout.inodelk_cbk = inodelk_cbk;
+    local->lock[0].layout.my_layout.locks = lk_array;
+    local->lock[0].layout.my_layout.lk_count = lk_count;
+
+    ret = dht_lock_order_requests(local->lock[0].layout.my_layout.locks,
+                                  local->lock[0].layout.my_layout.lk_count);
+    if (ret < 0)
+        goto out;
+
+    ret = 0;
+out:
+    return ret;
+}
+
+static void
+dht_inodelk_done(call_frame_t *lock_frame)
+{
+    fop_inodelk_cbk_t inodelk_cbk = NULL;
+    call_frame_t *main_frame = NULL;
+    dht_local_t *local = NULL;
+
+    local = lock_frame->local;
+    main_frame = local->main_frame;
+
+    local->lock[0].layout.my_layout.locks = NULL;
+    local->lock[0].layout.my_layout.lk_count = 0;
+
+    inodelk_cbk = local->lock[0].layout.my_layout.inodelk_cbk;
+    local->lock[0].layout.my_layout.inodelk_cbk = NULL;
+
+    inodelk_cbk(main_frame, NULL, main_frame->this,
+                local->lock[0].layout.my_layout.op_ret,
+                local->lock[0].layout.my_layout.op_errno, NULL);
+
+    dht_lock_stack_destroy(lock_frame, DHT_INODELK);
+    return;
+}
+
+static int32_t
+dht_unlock_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int lk_index = 0, call_cnt = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    lk_index = (long)cookie;
+
+    local = frame->local;
+    if (op_ret < 0) {
+        uuid_utoa_r(local->lock[0].layout.my_layout.locks[lk_index]->loc.gfid,
+                    gfid);
+
+        gf_smsg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_UNLOCKING_FAILED,
+                "name=%s",
+                local->lock[0].layout.my_layout.locks[lk_index]->xl->name,
+                "gfid=%s", gfid, NULL);
+    } else {
+        local->lock[0].layout.my_layout.locks[lk_index]->locked = 0;
+    }
+
+    call_cnt = dht_frame_return(frame);
+    if (is_last_call(call_cnt)) {
+        dht_inodelk_done(frame);
+    }
+
+    return 0;
+}
+
+static int32_t
+dht_unlock_inodelk_done(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    gf_uuid_unparse(local->lock[0].layout.my_layout.locks[0]->loc.inode->gfid,
+                    gfid);
+
+    if (op_ret < 0) {
+        gf_smsg(this->name, GF_LOG_WARNING, op_errno,
+                DHT_MSG_UNLOCK_GFID_FAILED, "DHT_LAYOUT_HEAL_DOMAIN gfid=%s",
+                gfid, NULL);
+    }
+
+    DHT_STACK_DESTROY(frame);
+    return 0;
+}
+
+int32_t
+dht_unlock_inodelk(call_frame_t *frame, dht_lock_t **lk_array, int lk_count,
+                   fop_inodelk_cbk_t inodelk_cbk)
+{
+    dht_local_t *local = NULL;
+    struct gf_flock flock = {
+        0,
+    };
+    int ret = -1, i = 0;
+    call_frame_t *lock_frame = NULL;
+    int call_cnt = 0;
+
+    GF_VALIDATE_OR_GOTO("dht-locks", frame, done);
+    GF_VALIDATE_OR_GOTO(frame->this->name, lk_array, done);
+    GF_VALIDATE_OR_GOTO(frame->this->name, inodelk_cbk, done);
+
+    call_cnt = dht_lock_count(lk_array, lk_count);
+    if (call_cnt == 0) {
+        ret = 0;
+        goto done;
+    }
+
+    lock_frame = dht_lock_frame(frame);
+    if (lock_frame == NULL) {
+        gf_smsg(frame->this->name, GF_LOG_WARNING, 0,
+                DHT_MSG_ALLOC_FRAME_FAILED_NOT_UNLOCKING_FOLLOWING_ENTRYLKS,
+                NULL);
+
+        dht_log_lk_array(frame->this->name, GF_LOG_WARNING, lk_array, lk_count);
+        goto done;
+    }
+
+    ret = dht_local_inodelk_init(lock_frame, lk_array, lk_count, inodelk_cbk);
+    if (ret < 0) {
+        gf_smsg(frame->this->name, GF_LOG_WARNING, 0,
+                DHT_MSG_LOCAL_LOCKS_STORE_FAILED_UNLOCKING_FOLLOWING_ENTRYLK,
+                NULL);
+
+        dht_log_lk_array(frame->this->name, GF_LOG_WARNING, lk_array, lk_count);
+
+        goto done;
+    }
+
+    local = lock_frame->local;
+    local->main_frame = frame;
+    local->call_cnt = call_cnt;
+
+    flock.l_type = F_UNLCK;
+
+    for (i = 0; i < local->lock[0].layout.my_layout.lk_count; i++) {
+        if (!local->lock[0].layout.my_layout.locks[i]->locked)
+            continue;
+
+        lock_frame->root
+            ->lk_owner = local->lock[0].layout.my_layout.locks[i]->lk_owner;
+        STACK_WIND_COOKIE(
+            lock_frame, dht_unlock_inodelk_cbk, (void *)(long)i,
+            local->lock[0].layout.my_layout.locks[i]->xl,
+            local->lock[0].layout.my_layout.locks[i]->xl->fops->inodelk,
+            local->lock[0].layout.my_layout.locks[i]->domain,
+            &local->lock[0].layout.my_layout.locks[i]->loc, F_SETLK, &flock,
+            NULL);
+        if (!--call_cnt)
+            break;
+    }
+
+    return 0;
+
+done:
+    if (lock_frame)
+        dht_lock_stack_destroy(lock_frame, DHT_INODELK);
+
+    /* no locks acquired, invoke inodelk_cbk */
+    if (ret == 0)
+        inodelk_cbk(frame, NULL, frame->this, 0, 0, NULL);
+
+    return ret;
+}
+
+int32_t
+dht_unlock_inodelk_wrapper(call_frame_t *frame, dht_ilock_wrap_t *inodelk)
+{
+    dht_local_t *local = NULL, *lock_local = NULL;
+    call_frame_t *lock_frame = NULL;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    int ret = 0;
+
+    local = frame->local;
+
+    if (!inodelk || !inodelk->locks)
+        goto out;
+
+    gf_uuid_unparse(local->loc.parent->gfid, pgfid);
+
+    lock_frame = copy_frame(frame);
+    if (lock_frame == NULL) {
+        gf_smsg(frame->this->name, GF_LOG_WARNING, ENOMEM,
+                DHT_MSG_COPY_FRAME_FAILED, "pgfid=%s", pgfid, "name=%s",
+                local->loc.name, "path=%s", local->loc.path, NULL);
+        goto done;
+    }
+
+    lock_local = dht_local_init(lock_frame, NULL, NULL, 0);
+    if (lock_local == NULL) {
+        gf_smsg(frame->this->name, GF_LOG_WARNING, ENOMEM,
+                DHT_MSG_CREATE_FAILED, "local", "gfid=%s", pgfid, "name=%s",
+                local->loc.name, "path=%s", local->loc.path, NULL);
+        goto done;
+    }
+
+    lock_frame->local = lock_local;
+
+    lock_local->lock[0].layout.my_layout.locks = inodelk->locks;
+    lock_local->lock[0].layout.my_layout.lk_count = inodelk->lk_count;
+    inodelk->locks = NULL;
+    inodelk->lk_count = 0;
+
+    ret = dht_unlock_inodelk(
+        lock_frame, lock_local->lock[0].layout.my_layout.locks,
+        lock_local->lock[0].layout.my_layout.lk_count, dht_unlock_inodelk_done);
+
+    if (ret)
+        goto done;
+
+    lock_frame = NULL;
+
+done:
+    if (lock_frame != NULL) {
+        DHT_STACK_DESTROY(lock_frame);
+    }
+out:
+    return 0;
+}
+
+static int
+dht_inodelk_cleanup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_inodelk_done(frame);
+    return 0;
+}
+
+static void
+dht_inodelk_cleanup(call_frame_t *lock_frame)
+{
+    dht_lock_t **lk_array = NULL;
+    int lk_count = 0, lk_acquired = 0;
+    dht_local_t *local = NULL;
+
+    local = lock_frame->local;
+
+    lk_array = local->lock[0].layout.my_layout.locks;
+    lk_count = local->lock[0].layout.my_layout.lk_count;
+
+    lk_acquired = dht_lock_count(lk_array, lk_count);
+    if (lk_acquired != 0) {
+        dht_unlock_inodelk(lock_frame, lk_array, lk_count,
+                           dht_inodelk_cleanup_cbk);
+    } else {
+        dht_inodelk_done(lock_frame);
+    }
+
+    return;
+}
+
+static int32_t
+dht_nonblocking_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int lk_index = 0, call_cnt = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    lk_index = (long)cookie;
+
+    if (op_ret == -1) {
+        local->lock[0].layout.my_layout.op_ret = -1;
+        local->lock[0].layout.my_layout.op_errno = op_errno;
+
+        if (local && local->lock[0].layout.my_layout.locks[lk_index]) {
+            uuid_utoa_r(local->lock[0]
+                            .layout.my_layout.locks[lk_index]
+                            ->loc.inode->gfid,
+                        gfid);
+
+            gf_msg_debug(
+                this->name, op_errno,
+                "inodelk failed on gfid: %s "
+                "subvolume: %s",
+                gfid,
+                local->lock[0].layout.my_layout.locks[lk_index]->xl->name);
+        }
+
+        goto out;
+    }
+
+    local->lock[0].layout.my_layout.locks[lk_index]->locked = _gf_true;
+
+out:
+    call_cnt = dht_frame_return(frame);
+    if (is_last_call(call_cnt)) {
+        if (local->lock[0].layout.my_layout.op_ret < 0) {
+            dht_inodelk_cleanup(frame);
+            return 0;
+        }
+
+        dht_inodelk_done(frame);
+    }
+
+    return 0;
+}
+
+int
+dht_nonblocking_inodelk(call_frame_t *frame, dht_lock_t **lk_array,
+                        int lk_count, fop_inodelk_cbk_t inodelk_cbk)
+{
+    struct gf_flock flock = {
+        0,
+    };
+    int i = 0, ret = 0;
+    dht_local_t *local = NULL;
+    call_frame_t *lock_frame = NULL;
+
+    GF_VALIDATE_OR_GOTO("dht-locks", frame, out);
+    GF_VALIDATE_OR_GOTO(frame->this->name, lk_array, out);
+    GF_VALIDATE_OR_GOTO(frame->this->name, inodelk_cbk, out);
+
+    lock_frame = dht_lock_frame(frame);
+    if (lock_frame == NULL)
+        goto out;
+
+    ret = dht_local_inodelk_init(lock_frame, lk_array, lk_count, inodelk_cbk);
+    if (ret < 0) {
+        goto out;
+    }
+
+    dht_set_lkowner(lk_array, lk_count, &lock_frame->root->lk_owner);
+
+    local = lock_frame->local;
+    local->main_frame = frame;
+
+    local->call_cnt = lk_count;
+
+    for (i = 0; i < lk_count; i++) {
+        flock.l_type = local->lock[0].layout.my_layout.locks[i]->type;
+
+        STACK_WIND_COOKIE(
+            lock_frame, dht_nonblocking_inodelk_cbk, (void *)(long)i,
+            local->lock[0].layout.my_layout.locks[i]->xl,
+            local->lock[0].layout.my_layout.locks[i]->xl->fops->inodelk,
+            local->lock[0].layout.my_layout.locks[i]->domain,
+            &local->lock[0].layout.my_layout.locks[i]->loc, F_SETLK, &flock,
+            NULL);
+    }
+
+    return 0;
+
+out:
+    if (lock_frame)
+        dht_lock_stack_destroy(lock_frame, DHT_INODELK);
+
+    return -1;
+}
+
+static int32_t
+dht_blocking_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    int lk_index = 0;
+    int i = 0;
+    dht_local_t *local = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {
+        0,
+    };
+    dht_reaction_type_t reaction = 0;
+
+    lk_index = (long)cookie;
+
+    local = frame->local;
+    if (op_ret == 0) {
+        local->lock[0].layout.my_layout.locks[lk_index]->locked = _gf_true;
+    } else {
+        switch (op_errno) {
+            case ESTALE:
+            case ENOENT:
+                reaction = local->lock[0]
+                               .layout.my_layout.locks[lk_index]
+                               ->do_on_failure;
+                if ((reaction != IGNORE_ENOENT_ESTALE) &&
+                    (reaction != IGNORE_ENOENT_ESTALE_EIO)) {
+                    gf_uuid_unparse(local->lock[0]
+                                        .layout.my_layout.locks[lk_index]
+                                        ->loc.gfid,
+                                    gfid);
+                    local->lock[0].layout.my_layout.op_ret = -1;
+                    local->lock[0].layout.my_layout.op_errno = op_errno;
+                    gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                            DHT_MSG_INODELK_FAILED, "subvol=%s",
+                            local->lock[0]
+                                .layout.my_layout.locks[lk_index]
+                                ->xl->name,
+                            "gfid=%s", gfid, NULL);
+                    goto cleanup;
+                }
+                break;
+            case EIO:
+                reaction = local->lock[0]
+                               .layout.my_layout.locks[lk_index]
+                               ->do_on_failure;
+                if (reaction != IGNORE_ENOENT_ESTALE_EIO) {
+                    gf_uuid_unparse(local->lock[0]
+                                        .layout.my_layout.locks[lk_index]
+                                        ->loc.gfid,
+                                    gfid);
+                    local->lock[0].layout.my_layout.op_ret = -1;
+                    local->lock[0].layout.my_layout.op_errno = op_errno;
+                    gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                            DHT_MSG_INODELK_FAILED, "subvol=%s",
+                            local->lock[0]
+                                .layout.my_layout.locks[lk_index]
+                                ->xl->name,
+                            "gfid=%s", gfid, NULL);
+                    goto cleanup;
+                }
+                break;
+
+            default:
+                gf_uuid_unparse(
+                    local->lock[0].layout.my_layout.locks[lk_index]->loc.gfid,
+                    gfid);
+                local->lock[0].layout.my_layout.op_ret = -1;
+                local->lock[0].layout.my_layout.op_errno = op_errno;
+                gf_smsg(
+                    this->name, GF_LOG_ERROR, op_errno, DHT_MSG_INODELK_FAILED,
+                    "subvol=%s",
+                    local->lock[0].layout.my_layout.locks[lk_index]->xl->name,
+                    "gfid=%s", gfid, NULL);
+                goto cleanup;
+        }
+    }
+
+    if (lk_index == (local->lock[0].layout.my_layout.lk_count - 1)) {
+        for (i = 0; (i < local->lock[0].layout.my_layout.lk_count) &&
+                    (!local->lock[0].layout.my_layout.locks[i]->locked);
+             i++)
+            ;
+
+        if (i == local->lock[0].layout.my_layout.lk_count) {
+            local->lock[0].layout.my_layout.op_ret = -1;
+            local->lock[0].layout.my_layout.op_errno = op_errno;
+        }
+
+        dht_inodelk_done(frame);
+    } else {
+        dht_blocking_inodelk_rec(frame, ++lk_index);
+    }
+
+    return 0;
+
+cleanup:
+    dht_inodelk_cleanup(frame);
+
+    return 0;
+}
+
+void
+dht_blocking_inodelk_rec(call_frame_t *frame, int i)
+{
+    dht_local_t *local = NULL;
+    struct gf_flock flock = {
+        0,
+    };
+
+    local = frame->local;
+
+    flock.l_type = local->lock[0].layout.my_layout.locks[i]->type;
+
+    STACK_WIND_COOKIE(
+        frame, dht_blocking_inodelk_cbk, (void *)(long)i,
+        local->lock[0].layout.my_layout.locks[i]->xl,
+        local->lock[0].layout.my_layout.locks[i]->xl->fops->inodelk,
+        local->lock[0].layout.my_layout.locks[i]->domain,
+        &local->lock[0].layout.my_layout.locks[i]->loc, F_SETLKW, &flock, NULL);
+
+    return;
+}
+
+int
+dht_blocking_inodelk(call_frame_t *frame, dht_lock_t **lk_array, int lk_count,
+                     fop_inodelk_cbk_t inodelk_cbk)
+{
+    int ret = -1;
+    call_frame_t *lock_frame = NULL;
+    dht_local_t *local = NULL;
+    dht_local_t *tmp_local = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("dht-locks", frame, out);
+    GF_VALIDATE_OR_GOTO(frame->this->name, lk_array, out);
+    GF_VALIDATE_OR_GOTO(frame->this->name, inodelk_cbk, out);
+
+    tmp_local = frame->local;
+
+    lock_frame = dht_lock_frame(frame);
+    if (lock_frame == NULL) {
+        gf_uuid_unparse(tmp_local->loc.gfid, gfid);
+        gf_smsg("dht", GF_LOG_ERROR, ENOMEM, DHT_MSG_LOCK_FRAME_FAILED,
+                "gfid=%s", gfid, "path=%s", tmp_local->loc.path, NULL);
+        goto out;
+    }
+
+    ret = dht_local_inodelk_init(lock_frame, lk_array, lk_count, inodelk_cbk);
+    if (ret < 0) {
+        gf_uuid_unparse(tmp_local->loc.gfid, gfid);
+        gf_smsg("dht", GF_LOG_ERROR, ENOMEM, DHT_MSG_LOCAL_LOCK_INIT_FAILED,
+                "gfid=%s", gfid, "path=%s", tmp_local->loc.path, NULL);
+        goto out;
+    }
+
+    dht_set_lkowner(lk_array, lk_count, &lock_frame->root->lk_owner);
+
+    local = lock_frame->local;
+    local->main_frame = frame;
+
+    dht_blocking_inodelk_rec(lock_frame, 0);
+
+    return 0;
+out:
+    if (lock_frame)
+        dht_lock_stack_destroy(lock_frame, DHT_INODELK);
+
+    return -1;
+}
+
+void
+dht_unlock_namespace(call_frame_t *frame, dht_dir_transaction_t *lock)
+{
+    GF_VALIDATE_OR_GOTO("dht-locks", frame, out);
+    GF_VALIDATE_OR_GOTO(frame->this->name, lock, out);
+
+    dht_unlock_entrylk_wrapper(frame, &lock->ns.directory_ns);
+    dht_unlock_inodelk_wrapper(frame, &lock->ns.parent_layout);
+
+out:
+    return;
+}
+
+static int32_t
+dht_protect_namespace_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+
+    local = frame->local;
+    if (op_ret != 0)
+        dht_unlock_inodelk_wrapper(frame, &local->current->ns.parent_layout);
+
+    local->current->ns.ns_cbk(frame, cookie, this, op_ret, op_errno, xdata);
+    return 0;
+}
+
+int32_t
+dht_blocking_entrylk_after_inodelk(call_frame_t *frame, void *cookie,
+                                   xlator_t *this, int32_t op_ret,
+                                   int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int ret = -1;
+    loc_t *loc = NULL;
+    dht_lock_t **lk_array = NULL;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    int count = 0;
+    dht_elock_wrap_t *entrylk = NULL;
+
+    local = frame->local;
+    entrylk = &local->current->ns.directory_ns;
+
+    if (op_ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+        goto err;
+    }
+
+    loc = &entrylk->locks[0]->loc;
+    gf_uuid_unparse(loc->gfid, pgfid);
+
+    local->op_ret = 0;
+    lk_array = entrylk->locks;
+    count = entrylk->lk_count;
+
+    ret = dht_blocking_entrylk(frame, lk_array, count,
+                               dht_protect_namespace_cbk);
+
+    if (ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = EIO;
+        gf_smsg(this->name, GF_LOG_WARNING, local->op_errno,
+                DHT_MSG_ENTRYLK_FAILED_AFT_INODELK, "fop=%s",
+                gf_fop_list[local->fop], "pgfid=%s", pgfid, "basename=%s",
+                entrylk->locks[0]->basename, NULL);
+        goto err;
+    }
+
+    return 0;
+
+err:
+    if (lk_array != NULL) {
+        dht_lock_array_free(lk_array, count);
+        GF_FREE(lk_array);
+        entrylk->locks = NULL;
+        entrylk->lk_count = 0;
+    }
+
+    /* Unlock inodelk. No harm calling unlock twice */
+    dht_unlock_inodelk_wrapper(frame, &local->current->ns.parent_layout);
+    /* Call ns_cbk. It will take care of unwinding */
+    local->current->ns.ns_cbk(frame, NULL, this, local->op_ret, local->op_errno,
+                              NULL);
+    return 0;
+}
+
+/* Given the loc and the subvol, this routine takes the inodelk on
+ * the parent inode and entrylk on (parent, loc->name). This routine
+ * is specific as it supports only one subvol on which it takes inodelk
+ * and then entrylk serially.
+ */
+int
+dht_protect_namespace(call_frame_t *frame, loc_t *loc, xlator_t *subvol,
+                      struct dht_namespace *ns, fop_entrylk_cbk_t ns_cbk)
+{
+    dht_ilock_wrap_t *inodelk = NULL;
+    dht_elock_wrap_t *entrylk = NULL;
+    dht_lock_t **lk_array = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *this = NULL;
+    loc_t parent = {
+        0,
+    };
+    int ret = -1;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    int32_t op_errno = 0;
+    int count = 1;
+
+    GF_VALIDATE_OR_GOTO("dht-locks", frame, out);
+    GF_VALIDATE_OR_GOTO(frame->this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(frame->this->name, loc->parent, out);
+    GF_VALIDATE_OR_GOTO(frame->this->name, subvol, out);
+
+    local = frame->local;
+    this = frame->this;
+
+    inodelk = &ns->parent_layout;
+    entrylk = &ns->directory_ns;
+
+    /* Initialize entrylk_cbk and parent loc */
+    ns->ns_cbk = ns_cbk;
+
+    ret = dht_build_parent_loc(this, &parent, loc, &op_errno);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno, DHT_MSG_LOC_FAILED,
+                "gfid=%s", loc->gfid, "name=%s", loc->name, "path=%s",
+                loc->path, NULL);
+        goto out;
+    }
+    gf_uuid_unparse(parent.gfid, pgfid);
+
+    /* Alloc inodelk */
+    inodelk->locks = GF_CALLOC(count, sizeof(*lk_array), gf_common_mt_pointer);
+    if (inodelk->locks == NULL) {
+        local->op_errno = ENOMEM;
+        gf_smsg(this->name, GF_LOG_WARNING, local->op_errno,
+                DHT_MSG_CALLOC_FAILED, "fop=%s", gf_fop_list[local->fop],
+                "pgfid=%s", pgfid, "name=%s", loc->name, "path=%s", loc->path,
+                NULL);
+        goto out;
+    }
+
+    inodelk->locks[0] = dht_lock_new(this, subvol, &parent, F_RDLCK,
+                                     DHT_LAYOUT_HEAL_DOMAIN, NULL,
+                                     FAIL_ON_ANY_ERROR);
+    if (inodelk->locks[0] == NULL) {
+        local->op_errno = ENOMEM;
+        gf_smsg(this->name, GF_LOG_WARNING, local->op_errno,
+                DHT_MSG_LOCK_ALLOC_FAILED, "inodelk-fop=%s",
+                gf_fop_list[local->fop], "pgfid=%s", pgfid, "name=%s",
+                loc->name, "path=%s", loc->path, NULL);
+        goto err;
+    }
+    inodelk->lk_count = count;
+
+    /* Allock entrylk */
+    entrylk->locks = GF_CALLOC(count, sizeof(*lk_array), gf_common_mt_pointer);
+    if (entrylk->locks == NULL) {
+        local->op_errno = ENOMEM;
+        gf_smsg(this->name, GF_LOG_WARNING, local->op_errno,
+                DHT_MSG_CALLOC_FAILED, "entrylk-fop=%s",
+                gf_fop_list[local->fop], "pgfid=%s", pgfid, "name=%s",
+                loc->name, "path=%s", loc->path, NULL);
+
+        goto err;
+    }
+
+    entrylk->locks[0] = dht_lock_new(this, subvol, &parent, F_WRLCK,
+                                     DHT_ENTRY_SYNC_DOMAIN, loc->name,
+                                     FAIL_ON_ANY_ERROR);
+    if (entrylk->locks[0] == NULL) {
+        local->op_errno = ENOMEM;
+        gf_smsg(this->name, GF_LOG_WARNING, local->op_errno,
+                DHT_MSG_LOCK_ALLOC_FAILED, "entrylk-fop=%s",
+                gf_fop_list[local->fop], "pgfid=%s", pgfid, "name=%s",
+                loc->name, "path=%s", loc->path, NULL);
+
+        goto err;
+    }
+    entrylk->lk_count = count;
+
+    /* Take read inodelk on parent. If it is successful, take write entrylk
+     * on name in cbk.
+     */
+    lk_array = inodelk->locks;
+    ret = dht_blocking_inodelk(frame, lk_array, count,
+                               dht_blocking_entrylk_after_inodelk);
+    if (ret < 0) {
+        local->op_errno = EIO;
+        gf_smsg(this->name, GF_LOG_WARNING, local->op_errno,
+                DHT_MSG_BLOCK_INODELK_FAILED, "fop=%s", gf_fop_list[local->fop],
+                "pgfid=%s", pgfid, "name=%s", loc->name, "path=%s", loc->path,
+                NULL);
+
+        goto err;
+    }
+
+    loc_wipe(&parent);
+
+    return 0;
+err:
+    if (entrylk->locks != NULL) {
+        dht_lock_array_free(entrylk->locks, count);
+        GF_FREE(entrylk->locks);
+        entrylk->locks = NULL;
+        entrylk->lk_count = 0;
+    }
+
+    if (inodelk->locks != NULL) {
+        dht_lock_array_free(inodelk->locks, count);
+        GF_FREE(inodelk->locks);
+        inodelk->locks = NULL;
+        inodelk->lk_count = 0;
+    }
+
+    loc_wipe(&parent);
+out:
+    return -1;
+}
diff --git a/xlators/cluster/dht/src/dht-lock.h b/xlators/cluster/dht/src/dht-lock.h
new file mode 100644
index 00000000000..6485c03fb6e
--- /dev/null
+++ b/xlators/cluster/dht/src/dht-lock.h
@@ -0,0 +1,91 @@
+/*
+  Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _DHT_LOCK_H
+#define _DHT_LOCK_H
+
+#include "dht-common.h"
+
+void
+dht_lock_array_free(dht_lock_t **lk_array, int count);
+
+int32_t
+dht_lock_count(dht_lock_t **lk_array, int lk_count);
+
+dht_lock_t *
+dht_lock_new(xlator_t *this, xlator_t *xl, loc_t *loc, short type,
+             const char *domain, const char *basename,
+             dht_reaction_type_t do_on_failure);
+
+int32_t
+dht_unlock_entrylk_wrapper(call_frame_t *, dht_elock_wrap_t *);
+
+void
+dht_blocking_entrylk_rec(call_frame_t *frame, int i);
+
+int
+dht_blocking_entrylk(call_frame_t *frame, dht_lock_t **lk_array, int lk_count,
+                     fop_inodelk_cbk_t entrylk_cbk);
+
+int32_t
+dht_unlock_inodelk(call_frame_t *frame, dht_lock_t **lk_array, int lk_count,
+                   fop_inodelk_cbk_t inodelk_cbk);
+
+int32_t
+dht_unlock_inodelk_wrapper(call_frame_t *, dht_ilock_wrap_t *);
+
+/* Acquire non-blocking inodelk on a list of xlators.
+ *
+ * @lk_array: array of lock requests lock on.
+ *
+ * @lk_count: number of locks in @lk_array
+ *
+ * @inodelk_cbk: will be called after inodelk replies are received
+ *
+ * @retval: -1 if stack_winding inodelk fails. 0 otherwise.
+ *          inodelk_cbk is called with appropriate error on errors.
+ *          On failure to acquire lock on all members of list, successful
+ *          locks are unlocked before invoking cbk.
+ */
+
+int
+dht_nonblocking_inodelk(call_frame_t *frame, dht_lock_t **lk_array,
+                        int lk_count, fop_inodelk_cbk_t inodelk_cbk);
+
+void
+dht_blocking_inodelk_rec(call_frame_t *frame, int i);
+
+/* same as dht_nonblocking_inodelk, but issues sequential blocking locks on
+ * @lk_array directly. locks are issued on some order which remains same
+ * for a list of xlators (irrespective of order of xlators within list).
+ */
+
+int
+dht_blocking_inodelk(call_frame_t *frame, dht_lock_t **lk_array, int lk_count,
+                     fop_inodelk_cbk_t inodelk_cbk);
+
+int32_t
+dht_blocking_entrylk_after_inodelk(call_frame_t *frame, void *cookie,
+                                   xlator_t *this, int32_t op_ret,
+                                   int32_t op_errno, dict_t *xdata);
+
+int32_t
+dht_blocking_entrylk_after_inodelk_rename(call_frame_t *frame, void *cookie,
+                                          xlator_t *this, int32_t op_ret,
+                                          int32_t op_errno, dict_t *xdata);
+
+void
+dht_unlock_namespace(call_frame_t *, dht_dir_transaction_t *);
+
+int
+dht_protect_namespace(call_frame_t *frame, loc_t *loc, xlator_t *subvol,
+                      struct dht_namespace *ns, fop_entrylk_cbk_t ns_cbk);
+
+#endif /* _DHT_LOCK_H */
diff --git a/xlators/cluster/dht/src/dht-mem-types.h b/xlators/cluster/dht/src/dht-mem-types.h
index e893eb48fd8..e3c4471334a 100644
--- a/xlators/cluster/dht/src/dht-mem-types.h
+++ b/xlators/cluster/dht/src/dht-mem-types.h
@@ -8,28 +8,31 @@
   cases as published by the Free Software Foundation.
 */
 
-
 #ifndef __DHT_MEM_TYPES_H__
 #define __DHT_MEM_TYPES_H__
 
-#include "mem-types.h"
+#include <glusterfs/mem-types.h>
 
 enum gf_dht_mem_types_ {
-        gf_dht_mt_dht_du_t = gf_common_mt_end + 1,
-        gf_dht_mt_dht_conf_t,
-        gf_dht_mt_char,
-        gf_dht_mt_int32_t,
-        gf_dht_mt_xlator_t,
-        gf_dht_mt_dht_layout_t,
-        gf_switch_mt_dht_conf_t,
-        gf_switch_mt_dht_du_t,
-        gf_switch_mt_switch_sched_array,
-        gf_switch_mt_switch_struct,
-        gf_dht_mt_subvol_time,
-        gf_dht_mt_loc_t,
-        gf_defrag_info_mt,
-        gf_dht_mt_inode_ctx_t,
-        gf_dht_mt_ctx_stat_time_t,
-        gf_dht_mt_end
+    gf_dht_mt_dht_du_t = gf_common_mt_end + 1,
+    gf_dht_mt_dht_conf_t,
+    gf_dht_mt_char,
+    gf_dht_mt_int32_t,
+    gf_dht_mt_xlator_t,
+    gf_dht_mt_dht_layout_t,
+    gf_switch_mt_switch_sched_array,
+    gf_switch_mt_switch_struct,
+    gf_dht_mt_subvol_time,
+    gf_dht_mt_loc_t,
+    gf_defrag_info_mt,
+    gf_dht_mt_inode_ctx_t,
+    gf_dht_mt_dirent_t,
+    gf_dht_mt_container_t,
+    gf_dht_mt_octx_t,
+    gf_dht_mt_miginfo_t,
+    gf_dht_mt_fd_ctx_t,
+    gf_dht_ret_cache_t,
+    gf_dht_nodeuuids_t,
+    gf_dht_mt_end
 };
 #endif
diff --git a/xlators/cluster/dht/src/dht-messages.h b/xlators/cluster/dht/src/dht-messages.h
new file mode 100644
index 00000000000..601f8dad78b
--- /dev/null
+++ b/xlators/cluster/dht/src/dht-messages.h
@@ -0,0 +1,386 @@
+/*Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _DHT_MESSAGES_H_
+#define _DHT_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(
+    DHT, DHT_MSG_CACHED_SUBVOL_GET_FAILED, DHT_MSG_CREATE_LINK_FAILED,
+    DHT_MSG_DICT_SET_FAILED, DHT_MSG_DIR_ATTR_HEAL_FAILED,
+    DHT_MSG_DIR_SELFHEAL_FAILED, DHT_MSG_DIR_SELFHEAL_XATTR_FAILED,
+    DHT_MSG_FILE_ON_MULT_SUBVOL, DHT_MSG_FILE_TYPE_MISMATCH,
+    DHT_MSG_GFID_MISMATCH, DHT_MSG_GFID_NULL, DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+    DHT_MSG_INIT_FAILED, DHT_MSG_INVALID_CONFIGURATION,
+    DHT_MSG_INVALID_DISK_LAYOUT, DHT_MSG_INVALID_OPTION,
+    DHT_MSG_LAYOUT_FIX_FAILED, DHT_MSG_LAYOUT_MERGE_FAILED,
+    DHT_MSG_LAYOUT_MISMATCH, DHT_MSG_LAYOUT_NULL, DHT_MSG_MIGRATE_DATA_COMPLETE,
+    DHT_MSG_MIGRATE_DATA_FAILED, DHT_MSG_MIGRATE_FILE_COMPLETE,
+    DHT_MSG_MIGRATE_FILE_FAILED, DHT_MSG_NO_MEMORY, DHT_MSG_OPENDIR_FAILED,
+    DHT_MSG_REBALANCE_FAILED, DHT_MSG_REBALANCE_START_FAILED,
+    DHT_MSG_REBALANCE_STATUS, DHT_MSG_REBALANCE_STOPPED, DHT_MSG_RENAME_FAILED,
+    DHT_MSG_SETATTR_FAILED, DHT_MSG_SUBVOL_INSUFF_INODES,
+    DHT_MSG_SUBVOL_INSUFF_SPACE, DHT_MSG_UNLINK_FAILED,
+    DHT_MSG_LAYOUT_SET_FAILED, DHT_MSG_LOG_FIXED_LAYOUT,
+    DHT_MSG_GET_XATTR_FAILED, DHT_MSG_FILE_LOOKUP_FAILED,
+    DHT_MSG_OPEN_FD_FAILED, DHT_MSG_SET_INODE_CTX_FAILED,
+    DHT_MSG_UNLOCKING_FAILED, DHT_MSG_DISK_LAYOUT_NULL, DHT_MSG_SUBVOL_INFO,
+    DHT_MSG_CHUNK_SIZE_INFO, DHT_MSG_LAYOUT_FORM_FAILED, DHT_MSG_SUBVOL_ERROR,
+    DHT_MSG_LAYOUT_SORT_FAILED, DHT_MSG_REGEX_INFO, DHT_MSG_FOPEN_FAILED,
+    DHT_MSG_SET_HOSTNAME_FAILED, DHT_MSG_BRICK_ERROR, DHT_MSG_SYNCOP_FAILED,
+    DHT_MSG_MIGRATE_INFO, DHT_MSG_SOCKET_ERROR, DHT_MSG_CREATE_FD_FAILED,
+    DHT_MSG_READDIR_ERROR, DHT_MSG_CHILD_LOC_BUILD_FAILED,
+    DHT_MSG_SET_SWITCH_PATTERN_ERROR, DHT_MSG_COMPUTE_HASH_FAILED,
+    DHT_MSG_FIND_LAYOUT_ANOMALIES_ERROR, DHT_MSG_ANOMALIES_INFO,
+    DHT_MSG_LAYOUT_INFO, DHT_MSG_INODE_LK_ERROR, DHT_MSG_RENAME_INFO,
+    DHT_MSG_DATA_NULL, DHT_MSG_AGGREGATE_QUOTA_XATTR_FAILED,
+    DHT_MSG_UNLINK_LOOKUP_INFO, DHT_MSG_LINK_FILE_LOOKUP_INFO,
+    DHT_MSG_OPERATION_NOT_SUP, DHT_MSG_NOT_LINK_FILE_ERROR, DHT_MSG_CHILD_DOWN,
+    DHT_MSG_UUID_PARSE_ERROR, DHT_MSG_GET_DISK_INFO_ERROR,
+    DHT_MSG_INVALID_VALUE, DHT_MSG_SWITCH_PATTERN_INFO,
+    DHT_MSG_SUBVOL_OP_FAILED, DHT_MSG_LAYOUT_PRESET_FAILED,
+    DHT_MSG_INVALID_LINKFILE, DHT_MSG_FIX_LAYOUT_INFO,
+    DHT_MSG_GET_HOSTNAME_FAILED, DHT_MSG_WRITE_FAILED,
+    DHT_MSG_MIGRATE_HARDLINK_FILE_FAILED, DHT_MSG_FSYNC_FAILED,
+    DHT_MSG_SUBVOL_DECOMMISSION_INFO, DHT_MSG_BRICK_QUERY_FAILED,
+    DHT_MSG_SUBVOL_NO_LAYOUT_INFO, DHT_MSG_OPEN_FD_ON_DST_FAILED,
+    DHT_MSG_SUBVOL_NOT_FOUND, DHT_MSG_FILE_LOOKUP_ON_DST_FAILED,
+    DHT_MSG_DISK_LAYOUT_MISSING, DHT_MSG_DICT_GET_FAILED,
+    DHT_MSG_REVALIDATE_CBK_INFO, DHT_MSG_UPGRADE_BRICKS, DHT_MSG_LK_ARRAY_INFO,
+    DHT_MSG_RENAME_NOT_LOCAL, DHT_MSG_RECONFIGURE_INFO,
+    DHT_MSG_INIT_LOCAL_SUBVOL_FAILED, DHT_MSG_SYS_CALL_GET_TIME_FAILED,
+    DHT_MSG_NO_DISK_USAGE_STATUS, DHT_MSG_SUBVOL_DOWN_ERROR,
+    DHT_MSG_REBAL_THROTTLE_INFO, DHT_MSG_COMMIT_HASH_INFO,
+    DHT_MSG_REBAL_STRUCT_SET, DHT_MSG_HAS_MIGINFO, DHT_MSG_SETTLE_HASH_FAILED,
+    DHT_MSG_DEFRAG_PROCESS_DIR_FAILED, DHT_MSG_FD_CTX_SET_FAILED,
+    DHT_MSG_STALE_LOOKUP, DHT_MSG_PARENT_LAYOUT_CHANGED,
+    DHT_MSG_LOCK_MIGRATION_FAILED, DHT_MSG_LOCK_INODE_UNREF_FAILED,
+    DHT_MSG_ASPRINTF_FAILED, DHT_MSG_DIR_LOOKUP_FAILED, DHT_MSG_INODELK_FAILED,
+    DHT_MSG_LOCK_FRAME_FAILED, DHT_MSG_LOCAL_LOCK_INIT_FAILED,
+    DHT_MSG_ENTRYLK_ERROR, DHT_MSG_INODELK_ERROR, DHT_MSG_LOC_FAILED,
+    DHT_MSG_UNKNOWN_FOP, DHT_MSG_MIGRATE_FILE_SKIPPED,
+    DHT_MSG_DIR_XATTR_HEAL_FAILED, DHT_MSG_HASHED_SUBVOL_DOWN,
+    DHT_MSG_NON_HASHED_SUBVOL_DOWN, DHT_MSG_SYNCTASK_CREATE_FAILED,
+    DHT_MSG_DIR_HEAL_ABORT, DHT_MSG_MIGRATE_SKIP, DHT_MSG_FD_CREATE_FAILED,
+    DHT_MSG_DICT_NEW_FAILED, DHT_MSG_FAILED_TO_OPEN, DHT_MSG_CREATE_FAILED,
+    DHT_MSG_FILE_NOT_EXIST, DHT_MSG_CHOWN_FAILED, DHT_MSG_FALLOCATE_FAILED,
+    DHT_MSG_FTRUNCATE_FAILED, DHT_MSG_STATFS_FAILED, DHT_MSG_WRITE_CROSS,
+    DHT_MSG_NEW_TARGET_FOUND, DHT_MSG_INSUFF_MEMORY, DHT_MSG_SET_XATTR_FAILED,
+    DHT_MSG_SET_MODE_FAILED, DHT_MSG_FILE_EXISTS_IN_DEST,
+    DHT_MSG_SYMLINK_FAILED, DHT_MSG_LINKFILE_DEL_FAILED, DHT_MSG_MKNOD_FAILED,
+    DHT_MSG_MIGRATE_CLEANUP_FAILED, DHT_MSG_LOCK_MIGRATE,
+    DHT_MSG_PARENT_BUILD_FAILED, DHT_MSG_HASHED_SUBVOL_NOT_FOUND,
+    DHT_MSG_ACQUIRE_ENTRYLK_FAILED, DHT_MSG_CREATE_DST_FAILED,
+    DHT_MSG_MIGRATION_EXIT, DHT_MSG_CHANGED_DST, DHT_MSG_TRACE_FAILED,
+    DHT_MSG_WRITE_LOCK_FAILED, DHT_MSG_GETACTIVELK_FAILED, DHT_MSG_STAT_FAILED,
+    DHT_MSG_UNLINK_PERFORM_FAILED, DHT_MSG_CLANUP_SOURCE_FILE_FAILED,
+    DHT_MSG_UNLOCK_FILE_FAILED, DHT_MSG_REMOVE_XATTR_FAILED,
+    DHT_MSG_DATA_MIGRATE_ABORT, DHT_MSG_DEFRAG_NULL, DHT_MSG_PARENT_NULL,
+    DHT_MSG_GFID_NOT_PRESENT, DHT_MSG_CHILD_LOC_FAILED,
+    DHT_MSG_SET_LOOKUP_FAILED, DHT_MSG_DIR_REMOVED, DHT_MSG_FIX_NOT_COMP,
+    DHT_MSG_SUBVOL_DETER_FAILED, DHT_MSG_LOCAL_SUBVOL, DHT_MSG_NODE_UUID,
+    DHT_MSG_SIZE_FILE, DHT_MSG_GET_DATA_SIZE_FAILED,
+    DHT_MSG_PTHREAD_JOIN_FAILED, DHT_MSG_COUNTER_THREAD_CREATE_FAILED,
+    DHT_MSG_MIGRATION_INIT_QUEUE_FAILED, DHT_MSG_PAUSED_TIMEOUT, DHT_MSG_WOKE,
+    DHT_MSG_ABORT_REBALANCE, DHT_MSG_CREATE_TASK_REBAL_FAILED,
+    DHT_MSG_REBAL_ESTIMATE_NOT_AVAIL, DHT_MSG_ADD_CHOICES_ERROR,
+    DHT_MSG_GET_CHOICES_ERROR, DHT_MSG_PREPARE_STATUS_ERROR,
+    DHT_MSG_SET_CHOICE_FAILED, DHT_MSG_SET_HASHED_SUBVOL_FAILED,
+    DHT_MSG_XATTR_HEAL_NOT_POSS, DHT_MSG_LINKTO_FILE_FAILED,
+    DHT_MSG_STALE_LINKFILE_DELETE, DHT_MSG_NO_SUBVOL_FOR_LINKTO,
+    DHT_MSG_SUBVOL_RETURNED, DHT_MSG_UNKNOWN_LOCAL_XSEL, DHT_MSG_GET_XATTR_ERR,
+    DHT_MSG_ALLOC_OR_FILL_FAILED, DHT_MSG_GET_REAL_NAME_FAILED,
+    DHT_MSG_COPY_UUID_FAILED, DHT_MSG_MDS_DETER_FAILED,
+    DHT_MSG_CREATE_REBAL_FAILED, DHT_MSG_LINK_LAYOUT_FAILED,
+    DHT_MSG_NO_SUBVOL_IN_LAYOUT, DHT_MSG_MEM_ALLOC_FAILED,
+    DHT_MSG_SET_IN_PARAMS_DICT_FAILED, DHT_MSG_LOC_COPY_FAILED,
+    DHT_MSG_PARENT_LOC_FAILED, DHT_MSG_CREATE_LOCK_FAILED,
+    DHT_MSG_PREV_ATTEMPT_FAILED, DHT_MSG_REFRESH_ATTEMPT,
+    DHT_MSG_ACQUIRE_LOCK_FAILED, DHT_MSG_CREATE_STUB_FAILED,
+    DHT_MSG_WIND_LOCK_REQ_FAILED, DHT_MSG_REFRESH_FAILED,
+    DHT_MSG_CACHED_SUBVOL_ERROR, DHT_MSG_NO_LINK_SUBVOL, DHT_MSG_SET_KEY_FAILED,
+    DHT_MSG_REMOVE_LINKTO_FAILED, DHT_MSG_LAYOUT_DICT_SET_FAILED,
+    DHT_MSG_XATTR_DICT_NULL, DHT_MSG_DUMMY_ALLOC_FAILED, DHT_MSG_DICT_IS_NULL,
+    DHT_MSG_LINK_INODE_FAILED, DHT_MSG_SELFHEAL_FAILED, DHT_MSG_NO_MDS_SUBVOL,
+    DHT_MSG_LIST_XATTRS_FAILED, DHT_MSG_RESET_INTER_XATTR_FAILED,
+    DHT_MSG_MDS_DOWN_UNABLE_TO_SET, DHT_MSG_WIND_UNLOCK_FAILED,
+    DHT_MSG_COMMIT_HASH_FAILED, DHT_MSG_UNLOCK_GFID_FAILED,
+    DHT_MSG_UNLOCK_FOLLOW_ENTRYLK, DHT_MSG_COPY_FRAME_FAILED,
+    DHT_MSG_UNLOCK_FOLLOW_LOCKS, DHT_MSG_ENTRYLK_FAILED_AFT_INODELK,
+    DHT_MSG_CALLOC_FAILED, DHT_MSG_LOCK_ALLOC_FAILED,
+    DHT_MSG_BLOCK_INODELK_FAILED,
+    DHT_MSG_LOCAL_LOCKS_STORE_FAILED_UNLOCKING_FOLLOWING_ENTRYLK,
+    DHT_MSG_ALLOC_FRAME_FAILED_NOT_UNLOCKING_FOLLOWING_ENTRYLKS,
+    DHT_MSG_DST_NULL_SET_FAILED);
+
+#define DHT_MSG_FD_CTX_SET_FAILED_STR "Failed to set fd ctx"
+#define DHT_MSG_INVALID_VALUE_STR "Different dst found in the fd ctx"
+#define DHT_MSG_UNKNOWN_FOP_STR "Unknown FOP on file"
+#define DHT_MSG_OPEN_FD_ON_DST_FAILED_STR "Failed to open the fd on file"
+#define DHT_MSG_SYNCTASK_CREATE_FAILED_STR "Failed to create synctask"
+#define DHT_MSG_ASPRINTF_FAILED_STR                                            \
+    "asprintf failed while fetching subvol from the id"
+#define DHT_MSG_HAS_MIGINFO_STR "Found miginfo in the inode ctx"
+#define DHT_MSG_FILE_LOOKUP_FAILED_STR "failed to lookup the file"
+#define DHT_MSG_INVALID_LINKFILE_STR                                           \
+    "linkto target is different from cached-subvol. treating as destination "  \
+    "subvol"
+#define DHT_MSG_GFID_MISMATCH_STR "gfid different on the target file"
+#define DHT_MSG_GET_XATTR_FAILED_STR "failed to get 'linkto' xattr"
+#define DHT_MSG_SET_INODE_CTX_FAILED_STR "failed to set inode-ctx target file"
+#define DHT_MSG_DIR_SELFHEAL_FAILED_STR "Healing of path failed"
+#define DHT_MSG_DIR_HEAL_ABORT_STR                                             \
+    "Failed to get path from subvol. Aborting directory healing"
+#define DHT_MSG_DIR_XATTR_HEAL_FAILED_STR "xattr heal failed for directory"
+#define DHT_MSG_LOCK_INODE_UNREF_FAILED_STR                                    \
+    "Found a NULL inode. Failed to unref the inode"
+#define DHT_MSG_DICT_SET_FAILED_STR "Failed to set dictionary value"
+#define DHT_MSG_NOT_LINK_FILE_ERROR_STR "got non-linkfile"
+#define DHT_MSG_CREATE_LINK_FAILED_STR "failed to initialize linkfile data"
+#define DHT_MSG_UNLINK_FAILED_STR "Unlinking linkfile on subvolume failed"
+#define DHT_MSG_MIGRATE_FILE_FAILED_STR "Migrate file failed"
+#define DHT_MSG_NO_MEMORY_STR "could not allocate memory for dict"
+#define DHT_MSG_SUBVOL_ERROR_STR "Failed to get linkto subvol"
+#define DHT_MSG_MIGRATE_HARDLINK_FILE_FAILED_STR "link failed on subvol"
+#define DHT_MSG_MIGRATE_FILE_SKIPPED_STR "Migration skipped"
+#define DHT_MSG_FD_CREATE_FAILED_STR "fd create failed"
+#define DHT_MSG_DICT_NEW_FAILED_STR "dict_new failed"
+#define DHT_MSG_FAILED_TO_OPEN_STR "failed to open"
+#define DHT_MSG_CREATE_FAILED_STR "failed to create"
+#define DHT_MSG_FILE_NOT_EXIST_STR "file does not exist"
+#define DHT_MSG_CHOWN_FAILED_STR "chown failed"
+#define DHT_MSG_FALLOCATE_FAILED_STR "fallocate failed"
+#define DHT_MSG_FTRUNCATE_FAILED_STR "ftruncate failed"
+#define DHT_MSG_STATFS_FAILED_STR "failed to get statfs"
+#define DHT_MSG_WRITE_CROSS_STR                                                \
+    "write will cross min-fre-disk for file on subvol. looking for new subvol"
+#define DHT_MSG_SUBVOL_INSUFF_SPACE_STR                                        \
+    "Could not find any subvol with space accommodating the file. Cosider "    \
+    "adding bricks"
+#define DHT_MSG_NEW_TARGET_FOUND_STR "New target found for file"
+#define DHT_MSG_INSUFF_MEMORY_STR "insufficient memory"
+#define DHT_MSG_SET_XATTR_FAILED_STR "failed to set xattr"
+#define DHT_MSG_SET_MODE_FAILED_STR "failed to set mode"
+#define DHT_MSG_FILE_EXISTS_IN_DEST_STR "file exists in destination"
+#define DHT_MSG_LINKFILE_DEL_FAILED_STR "failed to delete the linkfile"
+#define DHT_MSG_SYMLINK_FAILED_STR "symlink failed"
+#define DHT_MSG_MKNOD_FAILED_STR "mknod failed"
+#define DHT_MSG_SETATTR_FAILED_STR "failed to perform setattr"
+#define DHT_MSG_MIGRATE_CLEANUP_FAILED_STR                                     \
+    "Migrate file cleanup failed: failed to fstat file"
+#define DHT_MSG_LOCK_MIGRATE_STR "locks will be migrated for file"
+#define DHT_MSG_PARENT_BUILD_FAILED_STR                                        \
+    "failed to build parent loc, which is needed to acquire entrylk to "       \
+    "synchronize with renames on this path. Skipping migration"
+#define DHT_MSG_HASHED_SUBVOL_NOT_FOUND_STR                                    \
+    "cannot find hashed subvol which is needed to synchronize with renames "   \
+    "on this path. Skipping migration"
+#define DHT_MSG_ACQUIRE_ENTRYLK_FAILED_STR "failed to acquire entrylk on subvol"
+#define DHT_MSG_CREATE_DST_FAILED_STR "create dst failed for file"
+#define DHT_MSG_MIGRATION_EXIT_STR "Exiting migration"
+#define DHT_MSG_CHANGED_DST_STR "destination changed fo file"
+#define DHT_MSG_TRACE_FAILED_STR "Trace failed"
+#define DHT_MSG_WRITE_LOCK_FAILED_STR "write lock failed"
+#define DHT_MSG_GETACTIVELK_FAILED_STR "getactivelk failed for file"
+#define DHT_MSG_STAT_FAILED_STR "failed to do a stat"
+#define DHT_MSG_UNLINK_PERFORM_FAILED_STR "failed to perform unlink"
+#define DHT_MSG_MIGRATE_FILE_COMPLETE_STR "completed migration"
+#define DHT_MSG_CLANUP_SOURCE_FILE_FAILED_STR "failed to cleanup source file"
+#define DHT_MSG_UNLOCK_FILE_FAILED_STR "failed to unlock file"
+#define DHT_MSG_REMOVE_XATTR_FAILED_STR "remove xattr failed"
+#define DHT_MSG_SOCKET_ERROR_STR "Failed to unlink listener socket"
+#define DHT_MSG_HASHED_SUBVOL_GET_FAILED_STR "Failed to get hashed subvolume"
+#define DHT_MSG_CACHED_SUBVOL_GET_FAILED_STR "Failed to get cached subvolume"
+#define DHT_MSG_MIGRATE_DATA_FAILED_STR "migrate-data failed"
+#define DHT_MSG_DEFRAG_NULL_STR "defrag is NULL"
+#define DHT_MSG_DATA_MIGRATE_ABORT_STR                                         \
+    "Readdirp failed. Aborting data migration for dict"
+#define DHT_MSG_LAYOUT_FIX_FAILED_STR "fix layout failed"
+#define DHT_MSG_PARENT_NULL_STR "parent is NULL"
+#define DHT_MSG_GFID_NOT_PRESENT_STR "gfid not present"
+#define DHT_MSG_CHILD_LOC_FAILED_STR "Child loc build failed"
+#define DHT_MSG_SET_LOOKUP_FAILED_STR "Failed to set lookup"
+#define DHT_MSG_DIR_LOOKUP_FAILED_STR "lookup failed"
+#define DHT_MSG_DIR_REMOVED_STR "Dir renamed or removed. Skipping"
+#define DHT_MSG_READDIR_ERROR_STR "readdir failed, Aborting fix-layout"
+#define DHT_MSG_SETTLE_HASH_FAILED_STR "Settle hash failed"
+#define DHT_MSG_DEFRAG_PROCESS_DIR_FAILED_STR "gf_defrag_process_dir failed"
+#define DHT_MSG_FIX_NOT_COMP_STR                                               \
+    "Unable to retrieve fixlayout xattr. Assume background fix layout not "    \
+    "complete"
+#define DHT_MSG_SUBVOL_DETER_FAILED_STR                                        \
+    "local subvolume determination failed with error"
+#define DHT_MSG_LOCAL_SUBVOL_STR "local subvol"
+#define DHT_MSG_NODE_UUID_STR "node uuid"
+#define DHT_MSG_SIZE_FILE_STR "Total size files"
+#define DHT_MSG_GET_DATA_SIZE_FAILED_STR                                       \
+    "Failed to get the total data size. Unable to estimate time to complete "  \
+    "rebalance"
+#define DHT_MSG_PTHREAD_JOIN_FAILED_STR                                        \
+    "file_counter_thread: pthread_join failed"
+#define DHT_MSG_COUNTER_THREAD_CREATE_FAILED_STR                               \
+    "Failed to create the file counter thread"
+#define DHT_MSG_MIGRATION_INIT_QUEUE_FAILED_STR                                \
+    "Failed to initialise migration queue"
+#define DHT_MSG_REBALANCE_STOPPED_STR "Received stop command on rebalance"
+#define DHT_MSG_PAUSED_TIMEOUT_STR "Request pause timer timeout"
+#define DHT_MSG_WOKE_STR "woken"
+#define DHT_MSG_ABORT_REBALANCE_STR "Aborting rebalance"
+#define DHT_MSG_REBALANCE_START_FAILED_STR                                     \
+    "Failed to start rebalance: look up on / failed"
+#define DHT_MSG_CREATE_TASK_REBAL_FAILED_STR                                   \
+    "Could not create task for rebalance"
+#define DHT_MSG_REBAL_ESTIMATE_NOT_AVAIL_STR                                   \
+    "Rebalance estimates will not be available"
+#define DHT_MSG_REBALANCE_STATUS_STR "Rebalance status"
+#define DHT_MSG_DATA_NULL_STR "data value is NULL"
+#define DHT_MSG_ADD_CHOICES_ERROR_STR "Error to add choices in buffer"
+#define DHT_MSG_GET_CHOICES_ERROR_STR "Error to get choices"
+#define DHT_MSG_PREPARE_STATUS_ERROR_STR "Error to prepare status"
+#define DHT_MSG_SET_CHOICE_FAILED_STR "Failed to set full choice"
+#define DHT_MSG_AGGREGATE_QUOTA_XATTR_FAILED_STR                               \
+    "Failed to aggregate quota xattr"
+#define DHT_MSG_FILE_TYPE_MISMATCH_STR                                         \
+    "path exists as a file on one subvolume and directory on another. Please " \
+    "fix it manually"
+#define DHT_MSG_LAYOUT_SET_FAILED_STR "failed to set layout for subvolume"
+#define DHT_MSG_LAYOUT_MERGE_FAILED_STR "failed to merge layouts for subvolume"
+#define DHT_MSG_SET_HASHED_SUBVOL_FAILED_STR "Failed to set hashed subvolume"
+#define DHT_MSG_XATTR_HEAL_NOT_POSS_STR                                        \
+    "No gfid exists for path. so healing xattr is not possible"
+#define DHT_MSG_REVALIDATE_CBK_INFO_STR "Revalidate: subvolume returned -1"
+#define DHT_MSG_LAYOUT_MISMATCH_STR "Mismatching layouts"
+#define DHT_MSG_UNLINK_LOOKUP_INFO_STR "lookup_unlink retuened"
+#define DHT_MSG_LINKTO_FILE_FAILED_STR                                         \
+    "Could not unlink the linkto file as either fd is open and/or linkto "     \
+    "xattr is set"
+#define DHT_MSG_LAYOUT_PRESET_FAILED_STR                                       \
+    "Could not set pre-set layout for subvolume"
+#define DHT_MSG_FILE_ON_MULT_SUBVOL_STR                                        \
+    "multiple subvolumes have file (preferably rename the file in the "        \
+    "backend, and do a fresh lookup"
+#define DHT_MSG_STALE_LINKFILE_DELETE_STR                                      \
+    "attempting deletion of stale linkfile"
+#define DHT_MSG_LINK_FILE_LOOKUP_INFO_STR "Lookup on following linkfile"
+#define DHT_MSG_NO_SUBVOL_FOR_LINKTO_STR "No link subvolume for linkto"
+#define DHT_MSG_SUBVOL_RETURNED_STR "Subvolume returned -1"
+#define DHT_MSG_UNKNOWN_LOCAL_XSEL_STR "Unknown local->xsel"
+#define DHT_MSG_DICT_GET_FAILED_STR "Failed to get"
+#define DHT_MSG_UUID_PARSE_ERROR_STR "Failed to parse uuid"
+#define DHT_MSG_GET_XATTR_ERR_STR "getxattr err for dir"
+#define DHT_MSG_ALLOC_OR_FILL_FAILED_STR "alloc or fill failed"
+#define DHT_MSG_UPGRADE_BRICKS_STR                                             \
+    "At least one of the bricks does not support this operation. Please "      \
+    "upgrade all bricks"
+#define DHT_MSG_GET_REAL_NAME_FAILED_STR "Failed to get real filename"
+#define DHT_MSG_LAYOUT_NULL_STR "Layout is NULL"
+#define DHT_MSG_COPY_UUID_FAILED_STR "Failed to copy node uuid key"
+#define DHT_MSG_MDS_DETER_FAILED_STR                                           \
+    "Cannot determine MDS, fetching xattr randomly from a subvol"
+#define DHT_MSG_HASHED_SUBVOL_DOWN_STR                                         \
+    "MDS is down for path, so fetching xattr randomly from subvol"
+#define DHT_MSG_CREATE_REBAL_FAILED_STR                                        \
+    "failed to create a new rebalance synctask"
+#define DHT_MSG_FIX_LAYOUT_INFO_STR "fixing the layout"
+#define DHT_MSG_OPERATION_NOT_SUP_STR "wrong directory-spread-count value"
+#define DHT_MSG_LINK_LAYOUT_FAILED_STR "failed to link the layout in inode"
+#define DHT_MSG_NO_SUBVOL_IN_LAYOUT_STR "no subvolume in layout for path"
+#define DHT_MSG_INODE_LK_ERROR_STR "mknod lock failed for file"
+#define DHT_MSG_MEM_ALLOC_FAILED_STR "mem allocation failed"
+#define DHT_MSG_PARENT_LAYOUT_CHANGED_STR                                      \
+    "extracting in-memory layout of parent failed"
+#define DHT_MSG_SET_IN_PARAMS_DICT_FAILED_STR                                  \
+    "setting in params dictionary failed"
+#define DHT_MSG_LOC_COPY_FAILED_STR "loc_copy failed"
+#define DHT_MSG_LOC_FAILED_STR "parent loc build failed"
+#define DHT_MSG_PARENT_LOC_FAILED_STR "locking parent failed"
+#define DHT_MSG_CREATE_LOCK_FAILED_STR "Create lock failed"
+#define DHT_MSG_PREV_ATTEMPT_FAILED_STR                                        \
+    "mkdir loop detected. parent layout didn't change even though previous "   \
+    "attempt of mkdir failed because of in-memory layout not matching with "   \
+    "that on disk."
+#define DHT_MSG_REFRESH_ATTEMPT_STR                                            \
+    "mkdir parent layout changed. Attempting a refresh and then a retry"
+#define DHT_MSG_ACQUIRE_LOCK_FAILED_STR                                        \
+    "Acquiring lock on parent to guard against layout-change failed"
+#define DHT_MSG_CREATE_STUB_FAILED_STR "creating stub failed"
+#define DHT_MSG_WIND_LOCK_REQ_FAILED_STR                                       \
+    "cannot wind lock request to guard parent layout"
+#define DHT_MSG_REFRESH_FAILED_STR "refreshing parent layout failed."
+#define DHT_MSG_CACHED_SUBVOL_ERROR_STR "On cached subvol"
+#define DHT_MSG_NO_LINK_SUBVOL_STR "Linkfile does not have link subvolume"
+#define DHT_MSG_SET_KEY_FAILED_STR "failed to set key"
+#define DHT_MSG_CHILD_DOWN_STR "Received CHILD_DOWN. Exiting"
+#define DHT_MSG_LOG_FIXED_LAYOUT_STR "log layout fixed"
+#define DHT_MSG_REBAL_STRUCT_SET_STR "local->rebalance already set"
+#define DHT_MSG_REMOVE_LINKTO_FAILED_STR "Removal of linkto failed at subvol"
+#define DHT_MSG_LAYOUT_DICT_SET_FAILED_STR "dht layout dict set failed"
+#define DHT_MSG_SUBVOL_INFO_STR "creating subvolume"
+#define DHT_MSG_COMPUTE_HASH_FAILED_STR "hash computation failed"
+#define DHT_MSG_INVALID_DISK_LAYOUT_STR                                        \
+    "Invalid disk layout: Catastrophic error layout with unknown type found"
+#define DHT_MSG_LAYOUT_SORT_FAILED_STR "layout sort failed"
+#define DHT_MSG_ANOMALIES_INFO_STR "Found anomalies"
+#define DHT_MSG_XATTR_DICT_NULL_STR "xattr dictionary is NULL"
+#define DHT_MSG_DISK_LAYOUT_MISSING_STR "Disk layout missing"
+#define DHT_MSG_LAYOUT_INFO_STR "layout info"
+#define DHT_MSG_SUBVOL_NO_LAYOUT_INFO_STR "no pre-set layout for subvol"
+#define DHT_MSG_SELFHEAL_XATTR_FAILED_STR "layout setxattr failed"
+#define DHT_MSG_DIR_SELFHEAL_XATTR_FAILED_STR "Directory self heal xattr failed"
+#define DHT_MSG_DUMMY_ALLOC_FAILED_STR "failed to allocate dummy layout"
+#define DHT_MSG_DICT_IS_NULL_STR                                               \
+    "dict is NULL, need to make sure gfids are same"
+#define DHT_MSG_ENTRYLK_ERROR_STR "acquiring entrylk after inodelk failed"
+#define DHT_MSG_NO_DISK_USAGE_STATUS_STR "no du stats"
+#define DHT_MSG_LINK_INODE_FAILED_STR "linking inode failed"
+#define DHT_MSG_SELFHEAL_FAILED_STR "Directory selfheal failed"
+#define DHT_MSG_NO_MDS_SUBVOL_STR "No mds subvol"
+#define DHT_MSG_LIST_XATTRS_FAILED_STR "failed to list xattrs"
+#define DHT_MSG_RESET_INTER_XATTR_FAILED_STR "Failed to reset internal xattr"
+#define DHT_MSG_MDS_DOWN_UNABLE_TO_SET_STR                                     \
+    "mds subvol is down, unable to set xattr"
+#define DHT_MSG_DIR_ATTR_HEAL_FAILED_STR                                       \
+    "Directory attr heal failed. Failed to set uid/gid"
+#define DHT_MSG_WIND_UNLOCK_FAILED_STR                                         \
+    "Winding unlock failed: stale locks left on brick"
+#define DHT_MSG_COMMIT_HASH_FAILED_STR "Directory commit hash updaten failed"
+#define DHT_MSG_LK_ARRAY_INFO_STR "lk info"
+#define DHT_MSG_UNLOCK_GFID_FAILED_STR                                         \
+    "unlock failed on gfid: stale lock might be left"
+#define DHT_MSG_UNLOCKING_FAILED_STR "unlocking failed"
+#define DHT_MSG_UNLOCK_FOLLOW_ENTRYLK_STR "not unlocking following entrylks"
+#define DHT_MSG_COPY_FRAME_FAILED_STR "copy frame failed"
+#define DHT_MSG_UNLOCK_FOLLOW_LOCKS_STR "not unlocking following locks"
+#define DHT_MSG_INODELK_FAILED_STR "inodelk failed on subvol"
+#define DHT_MSG_LOCK_FRAME_FAILED_STR "memory allocation failed for lock_frame"
+#define DHT_MSG_LOCAL_LOCK_INIT_FAILED_STR "dht_local_lock_init failed"
+#define DHT_MSG_ENTRYLK_FAILED_AFT_INODELK_STR                                 \
+    "dht_blocking_entrylk failed after taking inodelk"
+#define DHT_MSG_BLOCK_INODELK_FAILED_STR "dht_blocking_inodelk failed"
+#define DHT_MSG_CALLOC_FAILED_STR "calloc failed"
+#define DHT_MSG_LOCK_ALLOC_FAILED_STR "lock allocation failed"
+#define DHT_MSG_ALLOC_FRAME_FAILED_NOT_UNLOCKING_FOLLOWING_ENTRYLKS_STR        \
+    "cannot allocate a frame, not unlocking following entrylks"
+#define DHT_MSG_LOCAL_LOCKS_STORE_FAILED_UNLOCKING_FOLLOWING_ENTRYLK_STR       \
+    "storing locks in local failed, not unlocking following entrylks"
+#define DHT_MSG_DST_NULL_SET_FAILED_STR                                        \
+    "src or dst is NULL, Failed to set dictionary value"
+
+#endif /* _DHT_MESSAGES_H_ */
diff --git a/xlators/cluster/dht/src/dht-rebalance.c b/xlators/cluster/dht/src/dht-rebalance.c
index 56e90f69866..8ba8082bd86 100644
--- a/xlators/cluster/dht/src/dht-rebalance.c
+++ b/xlators/cluster/dht/src/dht-rebalance.c
@@ -8,655 +8,1457 @@
   cases as published by the Free Software Foundation.
 */
 
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
 #include "dht-common.h"
-#include "xlator.h"
+#include <glusterfs/syscall.h>
 #include <fnmatch.h>
+#include <signal.h>
+#include <glusterfs/events.h>
+#include "glusterfs/compat-errno.h"  // for ENODATA on BSD
+
+#define GF_DISK_SECTOR_SIZE 512
+#define DHT_REBALANCE_PID 4242        /* Change it if required */
+#define DHT_REBALANCE_BLKSIZE 1048576 /* 1 MB */
+#define MAX_MIGRATE_QUEUE_COUNT 500
+#define MIN_MIGRATE_QUEUE_COUNT 200
+#define MAX_REBAL_TYPE_SIZE 16
+#define FILE_CNT_INTERVAL 600       /* 10 mins */
+#define ESTIMATE_START_INTERVAL 600 /* 10 mins */
+#define HARDLINK_MIG_INPROGRESS -2
+#define SKIP_MIGRATION_FD_POSITIVE -3
+#ifndef MAX
+#define MAX(a, b) (((a) > (b)) ? (a) : (b))
+#endif
 
-#define GF_DISK_SECTOR_SIZE             512
-#define DHT_REBALANCE_PID               4242 /* Change it if required */
-#define DHT_REBALANCE_BLKSIZE           (128 * 1024)
+#define GF_CRAWL_INDEX_MOVE(idx, sv_cnt)                                       \
+    {                                                                          \
+        idx++;                                                                 \
+        idx %= sv_cnt;                                                         \
+    }
 
-static int
-dht_write_with_holes (xlator_t *to, fd_t *fd, struct iovec *vec, int count,
-                      int32_t size, off_t offset, struct iobref *iobref)
+uint64_t g_totalfiles = 0;
+uint64_t g_totalsize = 0;
+
+void
+gf_defrag_free_dir_dfmeta(struct dir_dfmeta *meta, int local_subvols_cnt)
 {
-        int i            = 0;
-        int ret          = -1;
-        int start_idx    = 0;
-        int tmp_offset   = 0;
-        int write_needed = 0;
-        int buf_len      = 0;
-        int size_pending = 0;
-        char *buf        = NULL;
-
-        /* loop through each vector */
-        for (i = 0; i < count; i++) {
-                buf = vec[i].iov_base;
-                buf_len = vec[i].iov_len;
-
-                for (start_idx = 0; (start_idx + GF_DISK_SECTOR_SIZE) <= buf_len;
-                     start_idx += GF_DISK_SECTOR_SIZE) {
-
-                        if (mem_0filled (buf + start_idx, GF_DISK_SECTOR_SIZE) != 0) {
-                                write_needed = 1;
-                                continue;
-                        }
+    int i = 0;
+
+    if (meta) {
+        for (i = 0; i < local_subvols_cnt; i++) {
+            if (meta->equeue)
+                gf_dirent_free(&meta->equeue[i]);
+            if (meta->lfd && meta->lfd[i])
+                fd_unref(meta->lfd[i]);
+        }
 
-                        if (write_needed) {
-                                ret = syncop_write (to, fd, (buf + tmp_offset),
-                                                    (start_idx - tmp_offset),
-                                                    (offset + tmp_offset),
-                                                    iobref, 0);
-                                /* 'path' will be logged in calling function */
-                                if (ret < 0) {
-                                        gf_log (THIS->name, GF_LOG_WARNING,
-                                                "failed to write (%s)",
-                                                strerror (errno));
-                                        goto out;
-                                }
-
-                                write_needed = 0;
-                        }
-                        tmp_offset = start_idx + GF_DISK_SECTOR_SIZE;
-                }
+        GF_FREE(meta->equeue);
+        GF_FREE(meta->head);
+        GF_FREE(meta->iterator);
+        GF_FREE(meta->offset_var);
+        GF_FREE(meta->fetch_entries);
+        GF_FREE(meta->lfd);
+        GF_FREE(meta);
+    }
+}
 
-                if ((start_idx < buf_len) || write_needed) {
-                        /* This means, last chunk is not yet written.. write it */
-                        ret = syncop_write (to, fd, (buf + tmp_offset),
-                                            (buf_len - tmp_offset),
-                                            (offset + tmp_offset), iobref, 0);
-                        if (ret < 0) {
-                                /* 'path' will be logged in calling function */
-                                gf_log (THIS->name, GF_LOG_WARNING,
-                                        "failed to write (%s)",
-                                        strerror (errno));
-                                goto out;
-                        }
-                }
+void
+gf_defrag_free_container(struct dht_container *container)
+{
+    if (container) {
+        gf_dirent_entry_free(container->df_entry);
 
-                size_pending = (size - buf_len);
-                if (!size_pending)
-                        break;
+        if (container->parent_loc) {
+            loc_wipe(container->parent_loc);
         }
 
-        ret = size;
-out:
-        return ret;
+        GF_FREE(container->parent_loc);
 
+        GF_FREE(container);
+    }
 }
 
-int32_t
-gf_defrag_handle_hardlink (xlator_t *this, loc_t *loc, dict_t  *xattrs,
-                           struct iatt *stbuf)
+void
+dht_set_global_defrag_error(gf_defrag_info_t *defrag, int ret)
 {
-        int32_t                 ret             = -1;
-        xlator_t               *cached_subvol   = NULL;
-        xlator_t               *hashed_subvol   = NULL;
-        xlator_t               *linkto_subvol   = NULL;
-        data_t                 *data            = NULL;
-        struct iatt             iatt            = {0,};
-        int32_t                 op_errno        = 0;
-        dht_conf_t             *conf            = NULL;
-
-        GF_VALIDATE_OR_GOTO ("defrag", loc, out);
-        GF_VALIDATE_OR_GOTO ("defrag", loc->name, out);
-        GF_VALIDATE_OR_GOTO ("defrag", stbuf, out);
-        GF_VALIDATE_OR_GOTO ("defrag", this, out);
-        GF_VALIDATE_OR_GOTO ("defrag", xattrs, out);
-        GF_VALIDATE_OR_GOTO ("defrag", this->private, out);
-
-        conf = this->private;
-
-        if (uuid_is_null (loc->pargfid)) {
-                gf_log ("", GF_LOG_ERROR, "loc->pargfid is NULL for "
-                        "%s", loc->path);
-                goto out;
-        }
+    LOCK(&defrag->lock);
+    {
+        defrag->global_error = ret;
+    }
+    UNLOCK(&defrag->lock);
+    return;
+}
 
-        if (uuid_is_null (loc->gfid)) {
-                gf_log ("", GF_LOG_ERROR, "loc->gfid is NULL for "
-                        "%s", loc->path);
-                goto out;
-        }
+static int
+dht_send_rebalance_event(xlator_t *this, int cmd, gf_defrag_status_t status)
+{
+    int ret = -1;
+    char *volname = NULL;
+    char *tmpstr = NULL;
+    char *ptr = NULL;
+    char *suffix = "-dht";
+    int len = 0;
 
-        cached_subvol = dht_subvol_get_cached (this, loc->inode);
-        if (!cached_subvol) {
-                gf_log (this->name, GF_LOG_ERROR, "Failed to get cached subvol"
-                        " for %s on %s", loc->name, this->name);
-                goto out;
-        }
+    eventtypes_t event = EVENT_LAST;
 
-        hashed_subvol = dht_subvol_get_hashed (this, loc);
-        if (!hashed_subvol) {
-                gf_log (this->name, GF_LOG_ERROR, "Failed to get hashed subvol"
-                        " for %s on %s", loc->name, this->name);
-                goto out;
+    switch (status) {
+        case GF_DEFRAG_STATUS_COMPLETE:
+            event = EVENT_VOLUME_REBALANCE_COMPLETE;
+            break;
+        case GF_DEFRAG_STATUS_FAILED:
+            event = EVENT_VOLUME_REBALANCE_FAILED;
+            break;
+        case GF_DEFRAG_STATUS_STOPPED:
+            event = EVENT_VOLUME_REBALANCE_STOP;
+            break;
+        default:
+            break;
+    }
+
+    /* DHT volume */
+    len = strlen(this->name) - strlen(suffix);
+    tmpstr = gf_strdup(this->name);
+    if (tmpstr) {
+        ptr = tmpstr + len;
+        if (!strcmp(ptr, suffix)) {
+            tmpstr[len] = '\0';
+            volname = tmpstr;
         }
+    }
 
-        gf_log (this->name, GF_LOG_INFO, "Attempting to migrate hardlink %s "
-                "with gfid %s from %s -> %s", loc->name, uuid_utoa (loc->gfid),
-                cached_subvol->name, hashed_subvol->name);
-        data = dict_get (xattrs, conf->link_xattr_name);
-        /* set linkto on cached -> hashed if not present, else link it */
-        if (!data) {
-                ret = dict_set_str (xattrs, conf->link_xattr_name,
-                                    hashed_subvol->name);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_ERROR, "Failed to set "
-                                "linkto xattr in dict for %s", loc->name);
-                        goto out;
-                }
+    if (!volname) {
+        /* Better than nothing */
+        volname = this->name;
+    }
 
-                ret = syncop_setxattr (cached_subvol, loc, xattrs, 0);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_ERROR, "Linkto setxattr "
-                                "failed %s -> %s (%s)", cached_subvol->name,
-                                loc->name, strerror (errno));
-                        goto out;
-                }
-                goto out;
-        } else {
-                linkto_subvol = dht_linkfile_subvol (this, NULL, NULL, xattrs);
-                if (!linkto_subvol) {
-                        gf_log (this->name, GF_LOG_ERROR, "Failed to get "
-                                "linkto subvol for %s", loc->name);
-                } else {
-                        hashed_subvol = linkto_subvol;
-                }
+    if (event != EVENT_LAST) {
+        gf_event(event, "volume=%s", volname);
+    }
 
-                ret = syncop_link (hashed_subvol, loc, loc);
-                if  (ret) {
-                        op_errno = errno;
-                        gf_log (this->name, GF_LOG_ERROR, "link of %s -> %s"
-                                " failed on  subvol %s (%s)", loc->name,
-                                uuid_utoa(loc->gfid),
-                                hashed_subvol->name, strerror (op_errno));
-                        if (op_errno != EEXIST)
-                                goto out;
-                }
-        }
-        ret = syncop_lookup (hashed_subvol, loc, NULL, &iatt, NULL, NULL);
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "Failed lookup %s on %s (%s)"
-                        , loc->name, hashed_subvol->name, strerror (errno));
-                goto out;
-        }
+    GF_FREE(tmpstr);
+    return ret;
+}
 
-        if (iatt.ia_nlink == stbuf->ia_nlink) {
-                ret = dht_migrate_file (this, loc, cached_subvol, hashed_subvol,
-                                        GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS);
-                if (ret)
-                        goto out;
-        }
-        ret = 0;
-out:
-        return ret;
+static void
+dht_strip_out_acls(dict_t *dict)
+{
+    if (dict) {
+        dict_del(dict, "trusted.SGI_ACL_FILE");
+        dict_del(dict, POSIX_ACL_ACCESS_XATTR);
+    }
 }
 
+/*
+   return values:
+   -1 : failure
+   -2 : success
 
-static inline int
-__is_file_migratable (xlator_t *this, loc_t *loc,
-                      struct iatt *stbuf, dict_t *xattrs, int flags)
-{
-        int ret = -1;
+Hard link migration is carried out in three stages.
 
-        if (IA_ISDIR (stbuf->ia_type)) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: migrate-file called on directory", loc->path);
-                ret = -1;
-                goto out;
-        }
+(Say there are n hardlinks)
+Stage 1: Setting the new hashed subvol information on the 1st hardlink
+         encountered (linkto setxattr)
 
-        if (flags == GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS) {
-                ret = 0;
-                goto out;
-        }
-        if (stbuf->ia_nlink > 1) {
-                /* support for decomission */
-                if (flags == GF_DHT_MIGRATE_HARDLINK) {
-                        ret = gf_defrag_handle_hardlink (this, loc,
-                                                         xattrs, stbuf);
-                        if (ret) {
-                                gf_log (this->name, GF_LOG_WARNING,
-                                        "%s: failed to migrate file with link",
-                                        loc->path);
-                        }
-                } else {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "%s: file has hardlinks", loc->path);
-                }
-                ret = ENOTSUP;
-                goto out;
-        }
+Stage 2: Creating hardlinks on new hashed subvol for the 2nd to (n-1)th
+         hardlink
 
-        ret = 0;
+Stage 3: Physical migration of the data file for nth hardlink
 
-out:
-        return ret;
-}
+Why to deem "-2" as success and not "0":
 
-static inline int
-__dht_rebalance_create_dst_file (xlator_t *to, xlator_t *from, loc_t *loc, struct iatt *stbuf,
-                                 dict_t *dict, fd_t **dst_fd)
-{
-        xlator_t    *this = NULL;
-        int          ret  = -1;
-        fd_t        *fd   = NULL;
-        struct iatt  new_stbuf = {0,};
-        dht_conf_t  *conf = NULL;
+   dht_migrate_file expects return value "0" from _is_file_migratable if
+the file has to be migrated.
+
+   _is_file_migratable returns zero only when it is called with the
+flag "GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS".
 
-        this = THIS;
-        conf = this->private;
+   gf_defrag_handle_hardlink calls dht_migrate_file for physical migration
+of the data file with the flag "GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS"
 
-        ret = dict_set_static_bin (dict, "gfid-req", stbuf->ia_gfid, 16);
+Hence, gf_defrag_handle_hardlink returning "0" for success will force
+"dht_migrate_file" to migrate each of the hardlink which is not intended.
+
+For each of the three stage mentioned above "-2" will be returned and will
+be converted to "0" in dht_migrate_file.
+
+*/
+
+int32_t
+gf_defrag_handle_hardlink(xlator_t *this, loc_t *loc, int *fop_errno)
+{
+    int32_t ret = -1;
+    xlator_t *cached_subvol = NULL;
+    xlator_t *hashed_subvol = NULL;
+    xlator_t *linkto_subvol = NULL;
+    data_t *data = NULL;
+    struct iatt iatt = {
+        0,
+    };
+    int32_t op_errno = 0;
+    dht_conf_t *conf = NULL;
+    gf_loglevel_t loglevel = 0;
+    dict_t *link_xattr = NULL;
+    dict_t *dict = NULL;
+    dict_t *xattr_rsp = NULL;
+    struct iatt stbuf = {
+        0,
+    };
+
+    *fop_errno = EINVAL;
+
+    GF_VALIDATE_OR_GOTO("defrag", loc, out);
+    GF_VALIDATE_OR_GOTO("defrag", loc->name, out);
+    GF_VALIDATE_OR_GOTO("defrag", this, out);
+    GF_VALIDATE_OR_GOTO("defrag", this->private, out);
+
+    conf = this->private;
+
+    if (gf_uuid_is_null(loc->pargfid)) {
+        gf_msg("", GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed :"
+               "loc->pargfid is NULL for %s",
+               loc->path);
+        *fop_errno = EINVAL;
+        ret = -1;
+        goto out;
+    }
+
+    if (gf_uuid_is_null(loc->gfid)) {
+        gf_msg("", GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed :"
+               "loc->gfid is NULL for %s",
+               loc->path);
+        *fop_errno = EINVAL;
+        ret = -1;
+        goto out;
+    }
+
+    link_xattr = dict_new();
+    if (!link_xattr) {
+        ret = -1;
+        *fop_errno = ENOMEM;
+        goto out;
+    }
+
+    /*
+      Parallel migration can lead to migration of the hard link multiple
+      times which can lead to data loss. Hence, adding a fresh lookup to
+      decide whether migration is required or not.
+
+      Elaborating the scenario for let say 10 hardlinks [link{1..10}]:
+          Let say the first hard link "link1"  does the setxattr of the
+      new hashed subvolume info on the cached file. As there are multiple
+      threads working, we might have already all the links created on the
+      new hashed by the time we reach hardlink let say link5. Now the
+      number of links on hashed is equal to that of cached. Hence, file
+      migration will happen for link6.
+
+             Cached                                 Hashed
+      --------T link6                        rwxrwxrwx   link6
+
+      Now post above state all the link file on the cached will be zero
+      byte linkto files. Hence, if we still do migration for the following
+      files link{7..10}, we will end up migrating 0 data leading to data
+      loss.
+            Hence, a lookup can make sure whether we need to migrate the
+      file or not.
+    */
+
+    dict = dict_new();
+    if (!dict) {
+        ret = -1;
+        *fop_errno = ENOMEM;
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+               "could not allocate memory for dict");
+        goto out;
+    }
+
+    ret = dict_set_int32(dict, conf->link_xattr_name, 256);
+    if (ret) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed:"
+               "%s: failed to set 'linkto' key in dict",
+               loc->path);
+        goto out;
+    }
+
+    ret = syncop_lookup(this, loc, &stbuf, NULL, dict, &xattr_rsp);
+    if (ret) {
+        /*Ignore ENOENT and ESTALE as file might have been
+          migrated already*/
+        if (-ret == ENOENT || -ret == ESTALE) {
+            ret = -2;
+            goto out;
+        }
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed:%s lookup failed with ret = %d", loc->path,
+               ret);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    cached_subvol = dht_subvol_get_cached(this, loc->inode);
+    if (!cached_subvol) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed :"
+               "Failed to get cached subvol"
+               " for %s on %s",
+               loc->name, this->name);
+        *fop_errno = EINVAL;
+        ret = -1;
+        goto out;
+    }
+
+    hashed_subvol = dht_subvol_get_hashed(this, loc);
+    if (!hashed_subvol) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed :"
+               "Failed to get hashed subvol"
+               " for %s on %s",
+               loc->name, this->name);
+        *fop_errno = EINVAL;
+        ret = -1;
+        goto out;
+    }
+
+    /* Hardlink migration happens only with remove-brick. So this condition will
+     * be true only when the migration has happened. In case hardlinks are
+     * migrated for rebalance case, remove this check. Having this check here
+     * avoid redundant calls below*/
+    if (hashed_subvol == cached_subvol) {
+        ret = -2;
+        goto out;
+    }
+
+    gf_log(this->name, GF_LOG_INFO,
+           "Attempting to migrate hardlink %s "
+           "with gfid %s from %s -> %s",
+           loc->name, uuid_utoa(loc->gfid), cached_subvol->name,
+           hashed_subvol->name);
+
+    data = dict_get(xattr_rsp, conf->link_xattr_name);
+    /* set linkto on cached -> hashed if not present, else link it */
+    if (!data) {
+        ret = dict_set_str(link_xattr, conf->link_xattr_name,
+                           hashed_subvol->name);
         if (ret) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s: failed to set gfid in dict for create", loc->path);
-                goto out;
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Migrate file failed :"
+                   "Failed to set dictionary value:"
+                   " key = %s for %s",
+                   conf->link_xattr_name, loc->name);
+            *fop_errno = ENOMEM;
+            ret = -1;
+            goto out;
         }
 
-        ret = dict_set_str (dict, conf->link_xattr_name, from->name);
+        ret = syncop_setxattr(cached_subvol, loc, link_xattr, 0, NULL, NULL);
         if (ret) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s: failed to set gfid in dict for create", loc->path);
-                goto out;
+            gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Migrate file failed :"
+                   "Linkto setxattr failed %s -> %s",
+                   cached_subvol->name, loc->name);
+            *fop_errno = -ret;
+            ret = -1;
+            goto out;
         }
 
-        fd = fd_create (loc->inode, DHT_REBALANCE_PID);
-        if (!fd) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s: fd create failed (destination) (%s)",
-                        loc->path, strerror (errno));
-                ret = -1;
-                goto out;
+        gf_msg_debug(this->name, 0,
+                     "hardlink target subvol created on %s "
+                     ",cached %s, file %s",
+                     hashed_subvol->name, cached_subvol->name, loc->path);
+
+        ret = -2;
+        goto out;
+    } else {
+        linkto_subvol = dht_linkfile_subvol(this, NULL, NULL, xattr_rsp);
+        if (!linkto_subvol) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SUBVOL_ERROR,
+                   "Failed to get "
+                   "linkto subvol for %s",
+                   loc->name);
+        } else {
+            hashed_subvol = linkto_subvol;
         }
 
-        ret = syncop_lookup (to, loc, NULL, &new_stbuf, NULL, NULL);
-        if (!ret) {
-                /* File exits in the destination, check if gfid matches */
-                if (uuid_compare (stbuf->ia_gfid, new_stbuf.ia_gfid) != 0) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "file %s exits in %s with different gfid",
-                                loc->path, to->name);
-                        fd_unref (fd);
-                        goto out;
-                }
-        }
-        if ((ret == -1) && (errno != ENOENT)) {
-                /* File exists in destination, but not accessible */
-                gf_log (THIS->name, GF_LOG_WARNING,
-                        "%s: failed to lookup file (%s)",
-                        loc->path, strerror (errno));
+        ret = syncop_link(hashed_subvol, loc, loc, &iatt, NULL, NULL);
+        if (ret) {
+            op_errno = -ret;
+            ret = -1;
+
+            loglevel = (op_errno == EEXIST) ? GF_LOG_DEBUG : GF_LOG_ERROR;
+            gf_msg(this->name, loglevel, op_errno,
+                   DHT_MSG_MIGRATE_HARDLINK_FILE_FAILED,
+                   "link of %s -> %s"
+                   " failed on  subvol %s",
+                   loc->name, uuid_utoa(loc->gfid), hashed_subvol->name);
+            if (op_errno != EEXIST) {
+                *fop_errno = op_errno;
                 goto out;
+            }
+        } else {
+            gf_msg_debug(this->name, 0,
+                         "syncop_link successful for"
+                         " hardlink %s on subvol %s, cached %s",
+                         loc->path, hashed_subvol->name, cached_subvol->name);
         }
+    }
 
-        /* Create the destination with LINKFILE mode, and linkto xattr,
-           if the linkfile already exists, it will just open the file */
-        ret = syncop_create (to, loc, O_RDWR, DHT_LINKFILE_MODE, fd,
-                             dict, &new_stbuf);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "failed to create %s on %s (%s)",
-                        loc->path, to->name, strerror (errno));
-                goto out;
+    ret = syncop_lookup(hashed_subvol, loc, &iatt, NULL, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed :Failed lookup %s on %s ", loc->name,
+               hashed_subvol->name);
+
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    /* There is a race where on the target subvol for the hardlink
+     * (note: hash subvol for the hardlink might differ from this), some
+     * other client(non-rebalance) would have created a linkto file for that
+     * hardlink as part of lookup. So let say there are 10 hardlinks, on the
+     * 5th hardlink it self the hardlinks might have migrated. Now for
+     * (6..10th) hardlinks the cached and target would be same as the file
+     * has already migrated. Hence this check is needed  */
+    if (cached_subvol == hashed_subvol) {
+        gf_msg_debug(this->name, 0,
+                     "source %s and destination %s "
+                     "for hardlink %s are same",
+                     cached_subvol->name, hashed_subvol->name, loc->path);
+        ret = -2;
+        goto out;
+    }
+
+    if (iatt.ia_nlink == stbuf.ia_nlink) {
+        ret = dht_migrate_file(this, loc, cached_subvol, hashed_subvol,
+                               GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS, fop_errno);
+        if (ret) {
+            goto out;
         }
+    }
+    ret = -2;
+out:
+    if (link_xattr)
+        dict_unref(link_xattr);
 
-        ret = syncop_ftruncate (to, fd, stbuf->ia_size);
-        if (ret < 0)
-                gf_log (this->name, GF_LOG_ERROR,
-                        "ftruncate failed for %s on %s (%s)",
-                        loc->path, to->name, strerror (errno));
+    if (xattr_rsp)
+        dict_unref(xattr_rsp);
 
-        ret = syncop_fsetattr (to, fd, stbuf,
-                               (GF_SET_ATTR_UID | GF_SET_ATTR_GID),
-                                NULL, NULL);
-        if (ret < 0)
-                gf_log (this->name, GF_LOG_ERROR,
-                        "chown failed for %s on %s (%s)",
-                        loc->path, to->name, strerror (errno));
+    if (dict)
+        dict_unref(dict);
 
-        if (dst_fd)
-                *dst_fd = fd;
+    return ret;
+}
 
-        /* success */
-        ret = 0;
+static int
+__check_file_has_hardlink(xlator_t *this, loc_t *loc, struct iatt *stbuf,
+                          dict_t *xattrs, int flags, gf_defrag_info_t *defrag,
+                          dht_conf_t *conf, int *fop_errno)
+{
+    int ret = 0;
 
-out:
+    if (flags == GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS) {
+        ret = 0;
         return ret;
+    }
+    if (stbuf->ia_nlink > 1) {
+        /* support for decomission */
+        if (flags == GF_DHT_MIGRATE_HARDLINK) {
+            synclock_lock(&conf->link_lock);
+            ret = gf_defrag_handle_hardlink(this, loc, fop_errno);
+            synclock_unlock(&conf->link_lock);
+            /*
+            Returning zero will force the file to be remigrated.
+            Checkout gf_defrag_handle_hardlink for more information.
+            */
+            if (ret && ret != -2) {
+                gf_msg(this->name, GF_LOG_WARNING, 0,
+                       DHT_MSG_MIGRATE_FILE_FAILED,
+                       "Migrate file failed:"
+                       "%s: failed to migrate file with link",
+                       loc->path);
+            }
+        } else {
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Migration skipped for:"
+                   "%s: file has hardlinks",
+                   loc->path);
+            *fop_errno = ENOTSUP;
+            ret = 1;
+        }
+    }
+
+    return ret;
 }
 
-static inline int
-__dht_check_free_space (xlator_t *to, xlator_t *from, loc_t *loc,
-                        struct iatt *stbuf, int flag)
+/*
+     return values
+     0 : File will be migrated
+    -2 : File will not be migrated
+         (This is the return value from gf_defrag_handle_hardlink. Checkout
+         gf_defrag_handle_hardlink for description of "returning -2")
+    -1 : failure
+*/
+static int
+__is_file_migratable(xlator_t *this, loc_t *loc, struct iatt *stbuf,
+                     dict_t *xattrs, int flags, gf_defrag_info_t *defrag,
+                     dht_conf_t *conf, int *fop_errno)
 {
-        struct statvfs  src_statfs = {0,};
-        struct statvfs  dst_statfs = {0,};
-        int             ret        = -1;
-        xlator_t       *this       = NULL;
-
-        uint64_t        src_statfs_blocks = 1;
-        uint64_t        dst_statfs_blocks = 1;
-
-        this = THIS;
+    int ret = -1;
+    int lock_count = 0;
+
+    if (IA_ISDIR(stbuf->ia_type)) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed:"
+               "%s: migrate-file called on directory",
+               loc->path);
+        *fop_errno = EISDIR;
+        ret = -1;
+        goto out;
+    }
 
-        ret = syncop_statfs (from, loc, &src_statfs);
+    if (!conf->lock_migration_enabled) {
+        ret = dict_get_int32(xattrs, GLUSTERFS_POSIXLK_COUNT, &lock_count);
         if (ret) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "failed to get statfs of %s on %s (%s)",
-                        loc->path, from->name, strerror (errno));
-                goto out;
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Migrate file failed:"
+                   "%s: Unable to get lock count for file",
+                   loc->path);
+            *fop_errno = EINVAL;
+            ret = -1;
+            goto out;
         }
 
-        ret = syncop_statfs (to, loc, &dst_statfs);
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "failed to get statfs of %s on %s (%s)",
-                        loc->path, to->name, strerror (errno));
-                goto out;
+        if (lock_count) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Migrate file failed: %s: File has locks."
+                   " Skipping file migration",
+                   loc->path);
+            *fop_errno = ENOTSUP;
+            ret = 1;
+            goto out;
         }
+    }
 
-        /* if force option is given, do not check for space @ dst.
-         * Check only if space is avail for the file */
-        if (flag != GF_DHT_MIGRATE_DATA)
-                goto check_avail_space;
+    /* Check if file has hardlink*/
+    ret = __check_file_has_hardlink(this, loc, stbuf, xattrs, flags, defrag,
+                                    conf, fop_errno);
+out:
+    return ret;
+}
 
-        /* Check:
-           During rebalance `migrate-data` - Destination subvol experiences
-           a `reduction` in 'blocks' of free space, at the same time source
-           subvol gains certain 'blocks' of free space. A valid check is
-           necessary here to avoid errorneous move to destination where
-           the space could be scantily available.
+static int
+__dht_rebalance_create_dst_file(xlator_t *this, xlator_t *to, xlator_t *from,
+                                loc_t *loc, struct iatt *stbuf, fd_t **dst_fd,
+                                int *fop_errno, int file_has_holes)
+{
+    int ret = -1;
+    int ret2 = -1;
+    fd_t *fd = NULL;
+    struct iatt new_stbuf = {
+        0,
+    };
+    struct iatt check_stbuf = {
+        0,
+    };
+    dht_conf_t *conf = NULL;
+    dict_t *dict = NULL;
+    dict_t *xdata = NULL;
+
+    conf = this->private;
+
+    dict = dict_new();
+    if (!dict) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+               "dictionary allocation failed for"
+               "path:%s",
+               loc->path);
+        goto out;
+    }
+    ret = dict_set_gfuuid(dict, "gfid-req", stbuf->ia_gfid, true);
+    if (ret) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+               "%s: failed to set dictionary value: key = gfid-req", loc->path);
+        goto out;
+    }
+
+    ret = dict_set_str(dict, conf->link_xattr_name, from->name);
+    if (ret) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+               "%s: failed to set dictionary value: key = %s ", loc->path,
+               conf->link_xattr_name);
+        goto out;
+    }
+
+    fd = fd_create(loc->inode, DHT_REBALANCE_PID);
+    if (!fd) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: fd create failed (destination)", loc->path);
+        goto out;
+    }
+
+    xdata = dict_new();
+    if (!xdata) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: dict_new failed)", loc->path);
+        goto out;
+    }
+
+    ret = dict_set_int32_sizen(xdata, GF_CLEAN_WRITE_PROTECTION, 1);
+    if (ret) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+               "%s: failed to set dictionary value: key = %s ", loc->path,
+               GF_CLEAN_WRITE_PROTECTION);
+        goto out;
+    }
+
+    ret = syncop_lookup(to, loc, &new_stbuf, NULL, xdata, NULL);
+    if (!ret) {
+        /* File exits in the destination, check if gfid matches */
+        if (gf_uuid_compare(stbuf->ia_gfid, new_stbuf.ia_gfid) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_GFID_MISMATCH,
+                   "file %s exists in %s with different gfid", loc->path,
+                   to->name);
+            *fop_errno = EINVAL;
+            ret = -1;
+            goto out;
+        }
+    }
+    if ((ret < 0) && (-ret != ENOENT)) {
+        /* File exists in destination, but not accessible */
+        gf_msg(THIS->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: failed to lookup file", loc->path);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    /* Create the destination with LINKFILE mode, and linkto xattr,
+       if the linkfile already exists, just open the file */
+    if (!ret) {
+        /*
+         * File already present, just open the file.
          */
-        if (stbuf) {
-                dst_statfs_blocks = ((dst_statfs.f_bavail *
-                                      dst_statfs.f_bsize) /
-                                     GF_DISK_SECTOR_SIZE);
-                src_statfs_blocks = ((src_statfs.f_bavail *
-                                      src_statfs.f_bsize) /
-                                     GF_DISK_SECTOR_SIZE);
-                if ((dst_statfs_blocks - stbuf->ia_blocks) <
-                    (src_statfs_blocks + stbuf->ia_blocks)) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "data movement attempted from node (%s) with"
-                                " higher disk space to a node (%s) with "
-                                "lesser disk space (%s)", from->name,
-                                to->name, loc->path);
-
-                        /* this is not a 'failure', but we don't want to
-                           consider this as 'success' too :-/ */
-                        ret = 1;
-                        goto out;
+        ret = syncop_open(to, loc, O_RDWR, fd, NULL, NULL);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "failed to open %s on %s", loc->path, to->name);
+            *fop_errno = -ret;
+            ret = -1;
+            goto out;
+        }
+    } else {
+        ret = syncop_create(to, loc, O_RDWR, DHT_LINKFILE_MODE, fd, &new_stbuf,
+                            dict, NULL);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "failed to create %s on %s", loc->path, to->name);
+            *fop_errno = -ret;
+            ret = -1;
+            goto out;
+        }
+    }
+
+    fd_bind(fd);
+
+    /*Reason of doing lookup after create again:
+     *In the create, there is some time-gap between opening fd at the
+     *server (posix_layer) and binding it in server (incrementing fd count),
+     *so if in that time-gap, if other process sends unlink considering it
+     *as a linkto file, because inode->fd count will be 0, so file will be
+     *unlinked at the backend. And because further operations are performed
+     *on fd, so though migration will be done but will end with no file
+     *at  the backend.
+     */
+
+    ret = syncop_lookup(to, loc, &check_stbuf, NULL, NULL, NULL);
+    if (!ret) {
+        if (gf_uuid_compare(stbuf->ia_gfid, check_stbuf.ia_gfid) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_GFID_MISMATCH,
+                   "file %s exists in %s with different gfid,"
+                   "found in lookup after create",
+                   loc->path, to->name);
+            *fop_errno = EINVAL;
+            ret = -1;
+            goto out;
+        }
+    }
+
+    if (-ret == ENOENT) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: file does not exist"
+               "on %s",
+               loc->path, to->name);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    ret = syncop_fsetattr(to, fd, stbuf, (GF_SET_ATTR_UID | GF_SET_ATTR_GID),
+                          NULL, NULL, NULL, NULL);
+    if (ret < 0) {
+        *fop_errno = -ret;
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "chown failed for %s on %s", loc->path, to->name);
+    }
+
+    /* No need to bother about 0 byte size files */
+    if (stbuf->ia_size > 0) {
+        if (conf->use_fallocate && !file_has_holes) {
+            ret = syncop_fallocate(to, fd, 0, 0, stbuf->ia_size, NULL, NULL);
+            if (ret < 0) {
+                if (ret == -EOPNOTSUPP || ret == -EINVAL || ret == -ENOSYS) {
+                    conf->use_fallocate = _gf_false;
+                } else {
+                    gf_msg(this->name, GF_LOG_ERROR, -ret,
+                           DHT_MSG_MIGRATE_FILE_FAILED,
+                           "fallocate failed for %s on %s", loc->path,
+                           to->name);
+
+                    *fop_errno = -ret;
+
+                    /* fallocate does not release the space
+                     * in some cases
+                     */
+                    ret2 = syncop_ftruncate(to, fd, 0, NULL, NULL, NULL, NULL);
+                    if (ret2 < 0) {
+                        gf_msg(this->name, GF_LOG_WARNING, -ret2,
+                               DHT_MSG_MIGRATE_FILE_FAILED,
+                               "ftruncate failed for "
+                               "%s on %s",
+                               loc->path, to->name);
+                    }
+                    goto out;
                 }
+            }
+        } else {
+            ret = syncop_ftruncate(to, fd, stbuf->ia_size, NULL, NULL, NULL,
+                                   NULL);
+            if (ret < 0) {
+                *fop_errno = -ret;
+                gf_msg(this->name, GF_LOG_WARNING, -ret,
+                       DHT_MSG_MIGRATE_FILE_FAILED,
+                       "ftruncate failed for %s on %s", loc->path, to->name);
+            }
+        }
+    }
+
+    /* success */
+    ret = 0;
+
+    if (dst_fd)
+        *dst_fd = fd;
+
+out:
+    if (ret) {
+        if (fd) {
+            fd_unref(fd);
+        }
+    }
+    if (dict)
+        dict_unref(dict);
+
+    if (xdata)
+        dict_unref(xdata);
+
+    return ret;
+}
+
+static int
+__dht_check_free_space(xlator_t *this, xlator_t *to, xlator_t *from, loc_t *loc,
+                       struct iatt *stbuf, int flag, dht_conf_t *conf,
+                       gf_boolean_t *target_changed, xlator_t **new_subvol,
+                       int *fop_errno)
+{
+    struct statvfs src_statfs = {
+        0,
+    };
+    struct statvfs dst_statfs = {
+        0,
+    };
+    int ret = -1;
+    dict_t *xdata = NULL;
+    dht_layout_t *layout = NULL;
+    uint64_t src_statfs_blocks = 1;
+    uint64_t dst_statfs_blocks = 1;
+    double dst_post_availspacepercent = 0;
+    double src_post_availspacepercent = 0;
+    uint64_t file_blocks = 0;
+    uint64_t src_total_blocks = 0;
+    uint64_t dst_total_blocks = 0;
+
+    xdata = dict_new();
+    if (!xdata) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+               "failed to allocate dictionary");
+        goto out;
+    }
+
+    ret = dict_set_int8(xdata, GF_INTERNAL_IGNORE_DEEM_STATFS, 1);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Failed to set " GF_INTERNAL_IGNORE_DEEM_STATFS " in dict");
+        ret = -1;
+        *fop_errno = ENOMEM;
+        goto out;
+    }
+
+    ret = syncop_statfs(from, loc, &src_statfs, xdata, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "failed to get statfs of %s on %s", loc->path, from->name);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    ret = syncop_statfs(to, loc, &dst_statfs, xdata, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "failed to get statfs of %s on %s", loc->path, to->name);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    gf_msg_debug(this->name, 0,
+                 "min_free_disk - %f , block available - %" PRId64
+                 ", block size - %lu",
+                 conf->min_free_disk, dst_statfs.f_bavail, dst_statfs.f_bsize);
+
+    dst_statfs_blocks = dst_statfs.f_bavail *
+                        (dst_statfs.f_frsize / GF_DISK_SECTOR_SIZE);
+
+    src_statfs_blocks = src_statfs.f_bavail *
+                        (src_statfs.f_frsize / GF_DISK_SECTOR_SIZE);
+
+    dst_total_blocks = dst_statfs.f_blocks *
+                       (dst_statfs.f_frsize / GF_DISK_SECTOR_SIZE);
+
+    src_total_blocks = src_statfs.f_blocks *
+                       (src_statfs.f_frsize / GF_DISK_SECTOR_SIZE);
+
+    /* if force option is given, do not check for space @ dst.
+     * Check only if space is avail for the file */
+    if (flag != GF_DHT_MIGRATE_DATA)
+        goto check_avail_space;
+
+    /* Check:
+       During rebalance `migrate-data` - Destination subvol experiences
+       a `reduction` in 'blocks' of free space, at the same time source
+       subvol gains certain 'blocks' of free space. A valid check is
+       necessary here to avoid erroneous move to destination where
+       the space could be scantily available.
+       With heterogeneous brick support, an actual space comparison could
+       prevent any files being migrated to newly added bricks if they are
+       smaller then the free space available on the existing bricks.
+     */
+    if (!conf->use_fallocate) {
+        file_blocks = stbuf->ia_size + GF_DISK_SECTOR_SIZE - 1;
+        file_blocks /= GF_DISK_SECTOR_SIZE;
+
+        if (file_blocks >= dst_statfs_blocks) {
+            dst_statfs_blocks = 0;
+        } else {
+            dst_statfs_blocks -= file_blocks;
         }
+    }
+
+    src_post_availspacepercent = ((src_statfs_blocks + file_blocks) * 100) /
+                                 src_total_blocks;
+
+    dst_post_availspacepercent = (dst_statfs_blocks * 100) / dst_total_blocks;
+
+    if (dst_post_availspacepercent < src_post_availspacepercent) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "data movement of file "
+               "{blocks:%" PRIu64
+               " name:(%s)} would result in "
+               "dst node (%s:%" PRIu64
+               ") having lower disk "
+               "space than the source node (%s:%" PRIu64
+               ")"
+               ".Skipping file.",
+               stbuf->ia_blocks, loc->path, to->name, dst_statfs_blocks,
+               from->name, src_statfs_blocks);
+
+        /* this is not a 'failure', but we don't want to
+           consider this as 'success' too :-/ */
+        *fop_errno = ENOSPC;
+        ret = 1;
+        goto out;
+    }
+
 check_avail_space:
-        if (((dst_statfs.f_bavail * dst_statfs.f_bsize) /
-              GF_DISK_SECTOR_SIZE) < stbuf->ia_blocks) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "data movement attempted from node (%s) with "
-                        "to node (%s) which does not have required free space"
-                        " for %s", from->name, to->name, loc->path);
-                ret = 1;
-                goto out;
+    if (conf->disk_unit == 'p' && dst_statfs.f_blocks) {
+        dst_post_availspacepercent = (dst_statfs_blocks * 100) /
+                                     dst_total_blocks;
+
+        gf_msg_debug(this->name, 0,
+                     "file : %s, post_availspacepercent"
+                     " : %lf f_bavail : %" PRIu64 " min-free-disk: %lf",
+                     loc->path, dst_post_availspacepercent, dst_statfs.f_bavail,
+                     conf->min_free_disk);
+
+        if (dst_post_availspacepercent < conf->min_free_disk) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, 0,
+                   "Write will cross min-free-disk for "
+                   "file - %s on subvol - %s. Looking "
+                   "for new subvol",
+                   loc->path, to->name);
+
+            goto find_new_subvol;
+        } else {
+            ret = 0;
+            goto out;
         }
+    }
+
+    if (conf->disk_unit != 'p') {
+        if ((dst_statfs_blocks * GF_DISK_SECTOR_SIZE) < conf->min_free_disk) {
+            gf_msg_debug(this->name, 0,
+                         "file : %s,  destination frsize: %lu "
+                         "f_bavail : %" PRIu64 " min-free-disk: %lf",
+                         loc->path, dst_statfs.f_frsize, dst_statfs.f_bavail,
+                         conf->min_free_disk);
+
+            gf_msg(this->name, GF_LOG_WARNING, 0, 0,
+                   "write will"
+                   " cross min-free-disk for file - %s on subvol -"
+                   " %s. looking for new subvol",
+                   loc->path, to->name);
 
+            goto find_new_subvol;
+
+        } else {
+            ret = 0;
+            goto out;
+        }
+    }
+
+find_new_subvol:
+    layout = dht_layout_get(this, loc->parent);
+    if (!layout) {
+        gf_log(this->name, GF_LOG_ERROR, "Layout is NULL");
+        *fop_errno = EINVAL;
+        ret = -1;
+        goto out;
+    }
+
+    *new_subvol = dht_subvol_with_free_space_inodes(this, to, from, layout,
+                                                    stbuf->ia_size);
+    if ((!(*new_subvol)) || (*new_subvol == from)) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_SUBVOL_INSUFF_SPACE,
+               "Could not find any subvol"
+               " with space accommodating the file - %s. Consider "
+               "adding bricks",
+               loc->path);
+
+        *target_changed = _gf_false;
+        *fop_errno = ENOSPC;
+        ret = -1;
+    } else {
+        gf_msg(this->name, GF_LOG_INFO, 0, 0,
+               "new target found - %s"
+               " for file - %s",
+               (*new_subvol)->name, loc->path);
+        *target_changed = _gf_true;
         ret = 0;
+    }
+
 out:
-        return ret;
+    if (xdata)
+        dict_unref(xdata);
+    return ret;
 }
 
-static inline int
-__dht_rebalance_migrate_data (xlator_t *from, xlator_t *to, fd_t *src, fd_t *dst,
-                             uint64_t ia_size, int hole_exists)
+static int
+__dht_rebalance_migrate_data(xlator_t *this, gf_defrag_info_t *defrag,
+                             xlator_t *from, xlator_t *to, fd_t *src, fd_t *dst,
+                             uint64_t ia_size, int hole_exists, int *fop_errno)
 {
-        int            ret    = 0;
-        int            count  = 0;
-        off_t          offset = 0;
-        struct iovec  *vector = NULL;
-        struct iobref *iobref = NULL;
-        uint64_t       total  = 0;
-        size_t         read_size = 0;
-
-        /* if file size is '0', no need to enter this loop */
-        while (total < ia_size) {
-                read_size = (((ia_size - total) > DHT_REBALANCE_BLKSIZE) ?
-                             DHT_REBALANCE_BLKSIZE : (ia_size - total));
-                ret = syncop_readv (from, src, read_size,
-                                    offset, 0, &vector, &count, &iobref);
-                if (!ret || (ret < 0)) {
-                        break;
+    int ret = 0;
+    int count = 0;
+    off_t offset = 0;
+    off_t data_offset = 0;
+    off_t hole_offset = 0;
+    struct iovec *vector = NULL;
+    struct iobref *iobref = NULL;
+    uint64_t total = 0;
+    size_t read_size = 0;
+    size_t data_block_size = 0;
+    dict_t *xdata = NULL;
+    dht_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    /* if file size is '0', no need to enter this loop */
+    while (total < ia_size) {
+        /* This is a regular file - read it sequentially */
+        if (!hole_exists) {
+            read_size = (((ia_size - total) > DHT_REBALANCE_BLKSIZE)
+                             ? DHT_REBALANCE_BLKSIZE
+                             : (ia_size - total));
+        } else {
+            /* This is a sparse file - read only the data segments in the file
+             */
+
+            /* If the previous data block is fully copied, find the next data
+             * segment
+             * starting at the offset of the last read and written byte,  */
+            if (data_block_size <= 0) {
+                ret = syncop_seek(from, src, offset, GF_SEEK_DATA, NULL,
+                                  &data_offset);
+                if (ret) {
+                    if (ret == -ENXIO)
+                        ret = 0; /* No more data segments */
+                    else
+                        *fop_errno = -ret; /* Error occurred */
+
+                    break;
                 }
 
-                if (hole_exists)
-                        ret = dht_write_with_holes (to, dst, vector, count,
-                                                    ret, offset, iobref);
-                else
-                        ret = syncop_writev (to, dst, vector, count,
-                                             offset, iobref, 0);
-                if (ret < 0) {
+                /* If the position of the current data segment is greater than
+                 * the position of the next hole, find the next hole in order to
+                 * calculate the length of the new data segment */
+                if (data_offset > hole_offset) {
+                    /* Starting at the offset of the last data segment, find the
+                     * next hole */
+                    ret = syncop_seek(from, src, data_offset, GF_SEEK_HOLE,
+                                      NULL, &hole_offset);
+                    if (ret) {
+                        /* If an error occurred here it's a real error because
+                         * if the seek for a data segment was successful then
+                         * necessarily another hole must exist (EOF is a hole)
+                         */
+                        *fop_errno = -ret;
                         break;
-                }
-                offset += ret;
-                total += ret;
+                    }
 
-                GF_FREE (vector);
-                if (iobref)
-                        iobref_unref (iobref);
-                iobref = NULL;
-                vector = NULL;
+                    /* Calculate the total size of the current data block */
+                    data_block_size = hole_offset - data_offset;
+                }
+            } else {
+                /* There is still data in the current segment, move the
+                 * data_offset to the position of the last written byte */
+                data_offset = offset;
+            }
+
+            /* Calculate how much data needs to be read and written. If the data
+             * segment's length is bigger than DHT_REBALANCE_BLKSIZE, read and
+             * write DHT_REBALANCE_BLKSIZE data length and the rest in the
+             * next iteration(s) */
+            read_size = ((data_block_size > DHT_REBALANCE_BLKSIZE)
+                             ? DHT_REBALANCE_BLKSIZE
+                             : data_block_size);
+
+            /* Calculate the remaining size of the data block - maybe there's no
+             * need to seek for data in the next iteration */
+            data_block_size -= read_size;
+
+            /* Set offset to the offset of the data segment so read and write
+             * will have the correct position */
+            offset = data_offset;
         }
-        if (iobref)
-                iobref_unref (iobref);
-        GF_FREE (vector);
 
-        if (ret >= 0)
-                ret = 0;
+        ret = syncop_readv(from, src, read_size, offset, 0, &vector, &count,
+                           &iobref, NULL, NULL, NULL);
 
-        return ret;
-}
-
-
-static inline int
-__dht_rebalance_open_src_file (xlator_t *from, xlator_t *to, loc_t *loc,
-                               struct iatt *stbuf, fd_t **src_fd)
-{
-        int          ret  = 0;
-        fd_t        *fd   = NULL;
-        dict_t      *dict = NULL;
-        xlator_t    *this = NULL;
-        struct iatt  iatt = {0,};
-        dht_conf_t  *conf = NULL;
+        if (!ret || (ret < 0)) {
+            if (!ret) {
+                /* File was probably truncated*/
+                ret = -1;
+                *fop_errno = ENOSPC;
+            } else {
+                *fop_errno = -ret;
+            }
+            break;
+        }
 
-        this = THIS;
-        conf = this->private;
+        if (!conf->force_migration) {
+            if (!xdata) {
+                xdata = dict_new();
+                if (!xdata) {
+                    gf_msg("dht", GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                           "insufficient memory");
+                    ret = -1;
+                    *fop_errno = ENOMEM;
+                    break;
+                }
 
-        fd = fd_create (loc->inode, DHT_REBALANCE_PID);
-        if (!fd) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s: fd create failed (source)", loc->path);
-                ret = -1;
-                goto out;
+                /* Fail this write and abort rebalance if we
+                 * detect a write from client since migration of
+                 * this file started. This is done to avoid
+                 * potential data corruption due to out of order
+                 * writes from rebalance and client to the same
+                 * region (as compared between src and dst
+                 * files). See
+                 * https://github.com/gluster/glusterfs/issues/308
+                 * for more details.
+                 */
+                ret = dict_set_int32_sizen(xdata, GF_AVOID_OVERWRITE, 1);
+                if (ret) {
+                    gf_msg("dht", GF_LOG_ERROR, 0, ENOMEM,
+                           "failed to set dict");
+                    ret = -1;
+                    *fop_errno = ENOMEM;
+                    break;
+                }
+            }
         }
 
-        ret = syncop_open (from, loc, O_RDWR, fd);
-        if (ret == -1) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "failed to open file %s on %s (%s)",
-                        loc->path, from->name, strerror (errno));
-                goto out;
+        ret = syncop_writev(to, dst, vector, count, offset, iobref, 0, NULL,
+                            NULL, xdata, NULL);
+        if (ret < 0) {
+            *fop_errno = -ret;
+            break;
         }
 
+        offset += ret;
+        total += ret;
+
+        GF_FREE(vector);
+        if (iobref)
+            iobref_unref(iobref);
+        iobref = NULL;
+        vector = NULL;
+    }
+    if (iobref)
+        iobref_unref(iobref);
+    GF_FREE(vector);
+
+    if (ret >= 0)
+        ret = 0;
+    else
         ret = -1;
-        dict = dict_new ();
-        if (!dict)
-                goto out;
 
-        ret = dict_set_str (dict, conf->link_xattr_name, to->name);
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "failed to set xattr in dict for %s (linkto:%s)",
-                        loc->path, to->name);
-                goto out;
-        }
+    if (xdata) {
+        dict_unref(xdata);
+    }
 
-        /* Once the migration starts, the source should have 'linkto' key set
-           to show which is the target, so other clients can work around it */
-        ret = syncop_setxattr (from, loc, dict, 0);
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "failed to set xattr on %s in %s (%s)",
-                        loc->path, from->name, strerror (errno));
-                goto out;
-        }
+    return ret;
+}
 
-        /* mode should be (+S+T) to indicate migration is in progress */
-        iatt.ia_prot = stbuf->ia_prot;
-        iatt.ia_type = stbuf->ia_type;
-        iatt.ia_prot.sticky = 1;
-        iatt.ia_prot.sgid = 1;
+static int
+__dht_rebalance_open_src_file(xlator_t *this, xlator_t *from, xlator_t *to,
+                              loc_t *loc, struct iatt *stbuf, fd_t **src_fd,
+                              gf_boolean_t *clean_src, int *fop_errno)
+{
+    int ret = 0;
+    fd_t *fd = NULL;
+    dict_t *dict = NULL;
+    struct iatt iatt = {
+        0,
+    };
+    dht_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    *clean_src = _gf_false;
+
+    fd = fd_create(loc->inode, DHT_REBALANCE_PID);
+    if (!fd) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: fd create failed (source)", loc->path);
+        *fop_errno = ENOMEM;
+        ret = -1;
+        goto out;
+    }
+
+    ret = syncop_open(from, loc, O_RDWR, fd, NULL, NULL);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "failed to open file %s on %s", loc->path, from->name);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
 
-        ret = syncop_setattr (from, loc, &iatt, GF_SET_ATTR_MODE, NULL, NULL);
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "failed to set mode on %s in %s (%s)",
-                        loc->path, from->name, strerror (errno));
-                goto out;
-        }
+    fd_bind(fd);
 
-        if (src_fd)
-                *src_fd = fd;
+    if (src_fd)
+        *src_fd = fd;
 
-        /* success */
-        ret = 0;
+    ret = -1;
+    dict = dict_new();
+    if (!dict) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: Could not allocate memory for dict", loc->path);
+        *fop_errno = ENOMEM;
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_set_str(dict, conf->link_xattr_name, to->name);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "failed to set xattr in dict for %s (linkto:%s)", loc->path,
+               to->name);
+        *fop_errno = ENOMEM;
+        ret = -1;
+        goto out;
+    }
+
+    /* Once the migration starts, the source should have 'linkto' key set
+       to show which is the target, so other clients can work around it */
+    ret = syncop_setxattr(from, loc, dict, 0, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "failed to set xattr on %s in %s", loc->path, from->name);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    /* Reset source mode/xattr if migration fails*/
+    *clean_src = _gf_true;
+
+    /* mode should be (+S+T) to indicate migration is in progress */
+    iatt.ia_prot = stbuf->ia_prot;
+    iatt.ia_type = stbuf->ia_type;
+    iatt.ia_prot.sticky = 1;
+    iatt.ia_prot.sgid = 1;
+
+    ret = syncop_setattr(from, loc, &iatt, GF_SET_ATTR_MODE, NULL, NULL, NULL,
+                         NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "failed to set mode on %s in %s", loc->path, from->name);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    /* success */
+    ret = 0;
 out:
-        if (dict)
-                dict_unref (dict);
+    if (dict)
+        dict_unref(dict);
 
-        return ret;
+    return ret;
 }
 
 int
-migrate_special_files (xlator_t *this, xlator_t *from, xlator_t *to, loc_t *loc,
-                       struct iatt *buf)
+migrate_special_files(xlator_t *this, xlator_t *from, xlator_t *to, loc_t *loc,
+                      struct iatt *buf, int *fop_errno)
 {
-        int          ret      = -1;
-        dict_t      *rsp_dict = NULL;
-        dict_t      *dict     = NULL;
-        char        *link     = NULL;
-        struct iatt  stbuf    = {0,};
-        dht_conf_t  *conf     = this->private;
-
-        dict = dict_new ();
-        if (!dict)
-                goto out;
+    int ret = -1;
+    dict_t *rsp_dict = NULL;
+    dict_t *dict = NULL;
+    char *link = NULL;
+    struct iatt stbuf = {
+        0,
+    };
+    dht_conf_t *conf = this->private;
+
+    dict = dict_new();
+    if (!dict) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        goto out;
+    }
+    ret = dict_set_int32(dict, conf->link_xattr_name, 256);
+    if (ret) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        gf_log(this->name, GF_LOG_ERROR,
+               "%s: failed to set 'linkto' key in dict", loc->path);
+        goto out;
+    }
+
+    /* check in the destination if the file is link file */
+    ret = syncop_lookup(to, loc, &stbuf, NULL, dict, &rsp_dict);
+    if ((ret < 0) && (-ret != ENOENT)) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: lookup failed", loc->path);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    /* we no more require this key */
+    dict_del(dict, conf->link_xattr_name);
+
+    /* file exists in target node, only if it is 'linkfile' its valid,
+       otherwise, error out */
+    if (!ret) {
+        if (!check_is_linkfile(loc->inode, &stbuf, rsp_dict,
+                               conf->link_xattr_name)) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "%s: file exists in destination", loc->path);
+            *fop_errno = EINVAL;
+            ret = -1;
+            goto out;
+        }
 
-        ret = dict_set_int32 (dict, conf->link_xattr_name, 256);
+        /* as file is linkfile, delete it */
+        ret = syncop_unlink(to, loc, NULL, NULL);
         if (ret) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s: failed to set 'linkto' key in dict", loc->path);
-                goto out;
+            gf_msg(this->name, GF_LOG_WARNING, -ret,
+                   DHT_MSG_MIGRATE_FILE_FAILED,
+                   "%s: failed to delete the linkfile", loc->path);
+            *fop_errno = -ret;
+            ret = -1;
+            goto out;
         }
+    }
 
-        /* check in the destination if the file is link file */
-        ret = syncop_lookup (to, loc, dict, &stbuf, &rsp_dict, NULL);
-        if ((ret == -1) && (errno != ENOENT)) {
-                gf_log (this->name, GF_LOG_WARNING, "%s: lookup failed (%s)",
-                        loc->path, strerror (errno));
-                goto out;
+    /* Set the gfid of the source file in dict */
+    ret = dict_set_gfuuid(dict, "gfid-req", buf->ia_gfid, true);
+    if (ret) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        gf_log(this->name, GF_LOG_ERROR,
+               "%s: failed to set gfid in dict for create", loc->path);
+        goto out;
+    }
+
+    /* Create the file in target */
+    if (IA_ISLNK(buf->ia_type)) {
+        /* Handle symlinks separately */
+        ret = syncop_readlink(from, loc, &link, buf->ia_size, NULL, NULL);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_WARNING, -ret,
+                   DHT_MSG_MIGRATE_FILE_FAILED,
+                   "%s: readlink on symlink failed", loc->path);
+            *fop_errno = -ret;
+            ret = -1;
+            goto out;
         }
 
-        /* we no more require this key */
-        dict_del (dict, conf->link_xattr_name);
+        ret = syncop_symlink(to, loc, link, 0, dict, NULL);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_WARNING, -ret,
+                   DHT_MSG_MIGRATE_FILE_FAILED, "%s: creating symlink failed",
+                   loc->path);
+            *fop_errno = -ret;
+            ret = -1;
+            goto out;
+        }
 
-        /* file exists in target node, only if it is 'linkfile' its valid,
-           otherwise, error out */
-        if (!ret) {
-                if (!check_is_linkfile (loc->inode, &stbuf, rsp_dict,
-                                        conf->link_xattr_name)) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "%s: file exists in destination", loc->path);
-                        ret = -1;
-                        goto out;
-                }
+        goto done;
+    }
 
-                /* as file is linkfile, delete it */
-                ret = syncop_unlink (to, loc);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "%s: failed to delete the linkfile (%s)",
-                                loc->path, strerror (errno));
-                        goto out;
-                }
-        }
+    ret = syncop_mknod(to, loc, st_mode_from_ia(buf->ia_prot, buf->ia_type),
+                       makedev(ia_major(buf->ia_rdev), ia_minor(buf->ia_rdev)),
+                       0, dict, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: mknod failed", loc->path);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
 
-        /* Set the gfid of the source file in dict */
-        ret = dict_set_static_bin (dict, "gfid-req", buf->ia_gfid, 16);
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s: failed to set gfid in dict for create", loc->path);
-                goto out;
-        }
+done:
+    ret = syncop_setattr(to, loc, buf,
+                         (GF_SET_ATTR_MTIME | GF_SET_ATTR_UID |
+                          GF_SET_ATTR_GID | GF_SET_ATTR_MODE),
+                         NULL, NULL, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: failed to perform setattr on %s", loc->path, to->name);
+        *fop_errno = -ret;
+    }
+
+    ret = syncop_unlink(from, loc, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: unlink failed", loc->path);
+        *fop_errno = -ret;
+        ret = -1;
+    }
 
-        /* Create the file in target */
-        if (IA_ISLNK (buf->ia_type)) {
-                /* Handle symlinks separately */
-                ret = syncop_readlink (from, loc, &link, buf->ia_size);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "%s: readlink on symlink failed (%s)",
-                                loc->path, strerror (errno));
-                        goto out;
-                }
+out:
+    GF_FREE(link);
+    if (dict)
+        dict_unref(dict);
 
-                ret = syncop_symlink (to, loc, link, dict, 0);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "%s: creating symlink failed (%s)",
-                                loc->path, strerror (errno));
-                        goto out;
-                }
+    if (rsp_dict)
+        dict_unref(rsp_dict);
 
-                goto done;
-        }
+    return ret;
+}
 
-        ret = syncop_mknod (to, loc, st_mode_from_ia (buf->ia_prot,
-                                                      buf->ia_type),
-                            makedev (ia_major (buf->ia_rdev),
-                                     ia_minor (buf->ia_rdev)), dict, 0);
-        if (ret) {
-                gf_log (this->name, GF_LOG_WARNING, "%s: mknod failed (%s)",
-                        loc->path, strerror (errno));
-                goto out;
-        }
+static int
+__dht_migration_cleanup_src_file(xlator_t *this, loc_t *loc, fd_t *fd,
+                                 xlator_t *from, ia_prot_t *src_ia_prot)
+{
+    int ret = -1;
+    dht_conf_t *conf = NULL;
+    struct iatt new_stbuf = {
+        0,
+    };
+
+    if (!this || !fd || !from || !src_ia_prot) {
+        goto out;
+    }
+
+    conf = this->private;
+
+    /*Revert source mode and xattr changes*/
+    ret = syncop_fstat(from, fd, &new_stbuf, NULL, NULL);
+    if (ret < 0) {
+        /* Failed to get the stat info */
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file cleanup failed: failed to fstat "
+               "file %s on %s ",
+               loc->path, from->name);
+        ret = -1;
+        goto out;
+    }
 
-done:
-        ret = syncop_setattr (to, loc, buf,
-                              (GF_SET_ATTR_UID | GF_SET_ATTR_GID |
-                               GF_SET_ATTR_MODE), NULL, NULL);
-        if (ret) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to perform setattr on %s (%s)",
-                        loc->path, to->name, strerror (errno));
-        }
+    /* Remove the sticky bit and sgid bit set, reset it to 0*/
+    if (!src_ia_prot->sticky)
+        new_stbuf.ia_prot.sticky = 0;
 
-        ret = syncop_unlink (from, loc);
-        if (ret)
-                gf_log (this->name, GF_LOG_WARNING, "%s: unlink failed (%s)",
-                        loc->path, strerror (errno));
+    if (!src_ia_prot->sgid)
+        new_stbuf.ia_prot.sgid = 0;
 
-out:
-        if (dict)
-                dict_unref (dict);
+    ret = syncop_fsetattr(from, fd, &new_stbuf,
+                          (GF_SET_ATTR_GID | GF_SET_ATTR_MODE), NULL, NULL,
+                          NULL, NULL);
+
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file cleanup failed:"
+               "%s: failed to perform fsetattr on %s ",
+               loc->path, from->name);
+        ret = -1;
+        goto out;
+    }
+
+    ret = syncop_fremovexattr(from, fd, conf->link_xattr_name, 0, NULL);
+    if (ret) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "%s: failed to remove linkto xattr on %s (%s)", loc->path,
+               from->name, strerror(-ret));
+        ret = -1;
+        goto out;
+    }
 
-        if (rsp_dict)
-                dict_unref (rsp_dict);
+    ret = 0;
 
-        return ret;
+out:
+    return ret;
 }
 
 /*
@@ -667,1148 +1469,3234 @@ out:
     1 : not a failure, but we can't migrate data as of now
 */
 int
-dht_migrate_file (xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to,
-                  int flag)
+dht_migrate_file(xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to,
+                 int flag, int *fop_errno)
 {
-        int             ret            = -1;
-        struct iatt     new_stbuf      = {0,};
-        struct iatt     stbuf          = {0,};
-        struct iatt     empty_iatt     = {0,};
-        ia_prot_t       src_ia_prot    = {0,};
-        fd_t           *src_fd         = NULL;
-        fd_t           *dst_fd         = NULL;
-        dict_t         *dict           = NULL;
-        dict_t         *xattr          = NULL;
-        dict_t         *xattr_rsp      = NULL;
-        int             file_has_holes = 0;
-        dht_conf_t     *conf           = this->private;
-
-        gf_log (this->name, GF_LOG_INFO, "%s: attempting to move from %s to %s",
-                loc->path, from->name, to->name);
-
-        dict = dict_new ();
-        if (!dict)
-                goto out;
+    int ret = -1;
+    struct iatt new_stbuf = {
+        0,
+    };
+    struct iatt stbuf = {
+        0,
+    };
+    struct iatt empty_iatt = {
+        0,
+    };
+    ia_prot_t src_ia_prot = {
+        0,
+    };
+    fd_t *src_fd = NULL;
+    fd_t *dst_fd = NULL;
+    dict_t *dict = NULL;
+    dict_t *xattr = NULL;
+    dict_t *xattr_rsp = NULL;
+    int file_has_holes = 0;
+    dht_conf_t *conf = this->private;
+    int rcvd_enoent_from_src = 0;
+    struct gf_flock flock = {
+        0,
+    };
+    struct gf_flock plock = {
+        0,
+    };
+    loc_t tmp_loc = {
+        0,
+    };
+    loc_t parent_loc = {
+        0,
+    };
+    gf_boolean_t inodelk_locked = _gf_false;
+    gf_boolean_t entrylk_locked = _gf_false;
+    gf_boolean_t p_locked = _gf_false;
+    int lk_ret = -1;
+    gf_defrag_info_t *defrag = NULL;
+    gf_boolean_t clean_src = _gf_false;
+    gf_boolean_t clean_dst = _gf_false;
+    int log_level = GF_LOG_INFO;
+    gf_boolean_t delete_src_linkto = _gf_true;
+    lock_migration_info_t locklist;
+    dict_t *meta_dict = NULL;
+    gf_boolean_t meta_locked = _gf_false;
+    gf_boolean_t target_changed = _gf_false;
+    xlator_t *new_target = NULL;
+    xlator_t *old_target = NULL;
+    xlator_t *hashed_subvol = NULL;
+    fd_t *linkto_fd = NULL;
+    dict_t *xdata = NULL;
+
+    if (from == to) {
+        gf_msg_debug(this->name, 0,
+                     "destination and source are same. file %s"
+                     " might have migrated already",
+                     loc->path);
+        ret = 0;
+        goto out;
+    }
+
+    gf_log(this->name, log_level, "%s: attempting to move from %s to %s",
+           loc->path, from->name, to->name);
 
-        ret = dict_set_int32 (dict, conf->link_xattr_name, 256);
+    dict = dict_new();
+    if (!dict) {
+        ret = -1;
+        *fop_errno = ENOMEM;
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+               "Could not allocate memory for dict");
+        goto out;
+    }
+    ret = dict_set_int32(dict, conf->link_xattr_name, 256);
+    if (ret) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed:"
+               "%s: failed to set 'linkto' key in dict",
+               loc->path);
+        goto out;
+    }
+
+    /* Do not migrate file in case lock migration is not enabled on the
+     * volume*/
+    if (!conf->lock_migration_enabled) {
+        ret = dict_set_int32(dict, GLUSTERFS_POSIXLK_COUNT, sizeof(int32_t));
         if (ret) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "%s: failed to set 'linkto' key in dict", loc->path);
-                goto out;
+            *fop_errno = ENOMEM;
+            ret = -1;
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Migrate file failed: %s: failed to "
+                   "set " GLUSTERFS_POSIXLK_COUNT " key in dict",
+                   loc->path);
+            goto out;
+        }
+    } else {
+        gf_msg(this->name, GF_LOG_INFO, 0, 0,
+               "locks will be migrated"
+               " for file: %s",
+               loc->path);
+    }
+
+    /* The file is locked to prevent a rename during a migration. Renames
+     * and migrations on the file at the same time can lead to data loss.
+     */
+
+    ret = dht_build_parent_loc(this, &parent_loc, loc, fop_errno);
+    if (ret < 0) {
+        ret = -1;
+        gf_msg(this->name, GF_LOG_WARNING, *fop_errno,
+               DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: failed to build parent loc, which is needed to "
+               "acquire entrylk to synchronize with renames on this "
+               "path. Skipping migration",
+               loc->path);
+        goto out;
+    }
+
+    hashed_subvol = dht_subvol_get_hashed(this, loc);
+    if (hashed_subvol == NULL) {
+        ret = -1;
+        gf_msg(this->name, GF_LOG_WARNING, EINVAL, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: cannot find hashed subvol which is needed to "
+               "synchronize with renames on this path. "
+               "Skipping migration",
+               loc->path);
+        goto out;
+    }
+
+    flock.l_type = F_WRLCK;
+
+    tmp_loc.inode = inode_ref(loc->inode);
+    gf_uuid_copy(tmp_loc.gfid, loc->gfid);
+    tmp_loc.path = gf_strdup(loc->path);
+
+    /* this inodelk happens with flock.owner being zero. But to synchronize
+     * hardlink migration we need to have different lkowner for each migration
+     * Filed a bug here: https://bugzilla.redhat.com/show_bug.cgi?id=1468202 to
+     * track the fix for this. Currently synclock takes care of synchronizing
+     * hardlink migration. Once this bug is fixed we can avoid taking synclock
+     */
+    ret = syncop_inodelk(from, DHT_FILE_MIGRATE_DOMAIN, &tmp_loc, F_SETLKW,
+                         &flock, NULL, NULL);
+    if (ret < 0) {
+        *fop_errno = -ret;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_WARNING, *fop_errno,
+               DHT_MSG_MIGRATE_FILE_FAILED,
+               "migrate file failed: "
+               "%s: failed to lock file on %s",
+               loc->path, from->name);
+        goto out;
+    }
+
+    inodelk_locked = _gf_true;
+
+    /* dht_rename has changed to use entrylk on hashed subvol for
+     * synchronization. So, rebalance too has to acquire an entrylk on
+     * hashed subvol.
+     */
+    ret = syncop_entrylk(hashed_subvol, DHT_ENTRY_SYNC_DOMAIN, &parent_loc,
+                         loc->name, ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL, NULL);
+    if (ret < 0) {
+        *fop_errno = -ret;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_WARNING, *fop_errno,
+               DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: failed to acquire entrylk on subvol %s", loc->path,
+               hashed_subvol->name);
+        goto out;
+    }
+
+    entrylk_locked = _gf_true;
+
+    /* Phase 1 - Data migration is in progress from now on */
+    ret = syncop_lookup(from, loc, &stbuf, NULL, dict, &xattr_rsp);
+    if (ret) {
+        *fop_errno = -ret;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, *fop_errno,
+               DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed:"
+               "%s: lookup failed on %s",
+               loc->path, from->name);
+        goto out;
+    }
+
+    /* preserve source mode, so set the same to the destination */
+    src_ia_prot = stbuf.ia_prot;
+
+    /* Check if file can be migrated */
+    ret = __is_file_migratable(this, loc, &stbuf, xattr_rsp, flag, defrag, conf,
+                               fop_errno);
+    if (ret) {
+        if (ret == HARDLINK_MIG_INPROGRESS)
+            ret = 0;
+        goto out;
+    }
+
+    /* Take care of the special files */
+    if (!IA_ISREG(stbuf.ia_type)) {
+        /* Special files */
+        ret = migrate_special_files(this, from, to, loc, &stbuf, fop_errno);
+        goto out;
+    }
+
+    /* Try to preserve 'holes' while migrating data */
+    if (stbuf.ia_size > (stbuf.ia_blocks * GF_DISK_SECTOR_SIZE))
+        file_has_holes = 1;
+
+    /* create the destination, with required modes/xattr */
+    ret = __dht_rebalance_create_dst_file(this, to, from, loc, &stbuf, &dst_fd,
+                                          fop_errno, file_has_holes);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+               "Create dst failed"
+               " on - %s for file - %s",
+               to->name, loc->path);
+        goto out;
+    }
+
+    clean_dst = _gf_true;
+
+    ret = __dht_check_free_space(this, to, from, loc, &stbuf, flag, conf,
+                                 &target_changed, &new_target, fop_errno);
+    if (target_changed) {
+        /* Can't handle for hardlinks. Marking this as failure */
+        if (flag == GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS || stbuf.ia_nlink > 1) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SUBVOL_INSUFF_SPACE,
+                   "Exiting migration for"
+                   " file - %s. flag - %d, stbuf.ia_nlink - %d",
+                   loc->path, flag, stbuf.ia_nlink);
+            ret = -1;
+            goto out;
         }
 
-        /* Phase 1 - Data migration is in progress from now on */
-        ret = syncop_lookup (from, loc, dict, &stbuf, &xattr_rsp, NULL);
+        ret = syncop_ftruncate(to, dst_fd, 0, NULL, NULL, NULL, NULL);
         if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "%s: lookup failed on %s (%s)",
-                        loc->path, from->name, strerror (errno));
-                goto out;
+            gf_log(this->name, GF_LOG_WARNING,
+                   "%s: failed to perform truncate on %s (%s)", loc->path,
+                   to->name, strerror(-ret));
         }
 
-        /* we no more require this key */
-        dict_del (dict, conf->link_xattr_name);
+        syncop_close(dst_fd);
+        dst_fd = NULL;
 
-        /* preserve source mode, so set the same to the destination */
-        src_ia_prot = stbuf.ia_prot;
+        old_target = to;
+        to = new_target;
 
-        /* Check if file can be migrated */
-        ret = __is_file_migratable (this, loc, &stbuf, xattr_rsp, flag);
-        if (ret)
-                goto out;
+        clean_dst = _gf_false;
 
-        /* Take care of the special files */
-        if (!IA_ISREG (stbuf.ia_type)) {
-                /* Special files */
-                ret = migrate_special_files (this, from, to, loc, &stbuf);
-                goto out;
+        /* if the file migration is successful to this new target, then
+         * update the xattr on the old destination to point the new
+         * destination. We need to do update this only post migration
+         * as in case of failure the linkto needs to point to the source
+         * subvol */
+        ret = __dht_rebalance_create_dst_file(
+            this, to, from, loc, &stbuf, &dst_fd, fop_errno, file_has_holes);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "Create dst failed"
+                   " on - %s for file - %s",
+                   to->name, loc->path);
+            goto out;
+        } else {
+            gf_msg(this->name, GF_LOG_INFO, 0, 0,
+                   "destination for file "
+                   "- %s is changed to - %s",
+                   loc->path, to->name);
+            clean_dst = _gf_true;
         }
+    }
+
+    if (ret) {
+        goto out;
+    }
+
+    /* Open the source, and also update mode/xattr */
+    ret = __dht_rebalance_open_src_file(this, from, to, loc, &stbuf, &src_fd,
+                                        &clean_src, fop_errno);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed: failed to open %s on %s", loc->path,
+               from->name);
+        goto out;
+    }
+
+    /* TODO: move all xattr related operations to fd based operations */
+    ret = syncop_listxattr(from, loc, &xattr, NULL, NULL);
+    if (ret < 0) {
+        *fop_errno = -ret;
+        gf_msg(this->name, GF_LOG_WARNING, *fop_errno,
+               DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed:"
+               "%s: failed to get xattr from %s",
+               loc->path, from->name);
+        ret = -1;
+        goto out;
+    }
+
+    /* Copying posix acls to the linkto file messes up the permissions*/
+    dht_strip_out_acls(xattr);
+
+    /* Remove the linkto xattr as we don't want to overwrite the value
+     * set on the dst.
+     */
+    dict_del(xattr, conf->link_xattr_name);
+
+    /* We need to error out if this fails as having the wrong shard xattrs
+     * set on the dst could cause data corruption
+     */
+    ret = syncop_fsetxattr(to, dst_fd, xattr, 0, NULL, NULL);
+    if (ret < 0) {
+        *fop_errno = -ret;
+        gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: failed to set xattr on %s", loc->path, to->name);
+        ret = -1;
+        goto out;
+    }
 
-        /* create the destination, with required modes/xattr */
-        ret = __dht_rebalance_create_dst_file (to, from, loc, &stbuf,
-                                               dict, &dst_fd);
-        if (ret)
-                goto out;
+    if (xattr_rsp) {
+        /* we no more require this key */
+        dict_del(dict, conf->link_xattr_name);
+        dict_unref(xattr_rsp);
+    }
+
+    ret = syncop_fstat(from, src_fd, &stbuf, dict, &xattr_rsp);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed:failed to lookup %s on %s ", loc->path,
+               from->name);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    /* Check again if file has hardlink */
+    ret = __check_file_has_hardlink(this, loc, &stbuf, xattr_rsp, flag, defrag,
+                                    conf, fop_errno);
+    if (ret) {
+        if (ret == HARDLINK_MIG_INPROGRESS)
+            ret = 0;
+        goto out;
+    }
+
+    ret = __dht_rebalance_migrate_data(this, defrag, from, to, src_fd, dst_fd,
+                                       stbuf.ia_size, file_has_holes,
+                                       fop_errno);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed: %s: failed to migrate data", loc->path);
 
-        ret = __dht_check_free_space (to, from, loc, &stbuf, flag);
-        if (ret) {
-                goto out;
+        ret = -1;
+        goto out;
+    }
+
+    /* TODO: Sync the locks */
+
+    xdata = dict_new();
+    if (!xdata || dict_set_int8(xdata, "last-fsync", 1)) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "%s: failed to set last-fsync flag on "
+               "%s (%s)",
+               loc->path, to->name, strerror(ENOMEM));
+    }
+
+    ret = syncop_fsync(to, dst_fd, 0, NULL, NULL, xdata, NULL);
+    if (ret) {
+        gf_log(this->name, GF_LOG_WARNING, "%s: failed to fsync on %s (%s)",
+               loc->path, to->name, strerror(-ret));
+        *fop_errno = -ret;
+    }
+
+    /* Phase 2 - Data-Migration Complete, Housekeeping updates pending */
+
+    ret = syncop_fstat(from, src_fd, &new_stbuf, NULL, NULL);
+    if (ret < 0) {
+        /* Failed to get the stat info */
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed: failed to fstat file %s on %s ", loc->path,
+               from->name);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    /* Lock the entire source file to prevent clients from taking a
+       lock on it as dht_lk does not handle file migration.
+
+       This still leaves a small window where conflicting locks can
+       be granted to different clients. If client1 requests a blocking
+       lock on the src file, it will be granted after the migrating
+       process releases its lock. If client2 requests a lock on the dst
+       data file, it will also be granted, but all FOPs will be redirected
+       to the dst data file.
+    */
+
+    /* Take meta lock  */
+
+    if (conf->lock_migration_enabled) {
+        meta_dict = dict_new();
+        if (!meta_dict) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "dict_new failed");
+
+            *fop_errno = ENOMEM;
+            ret = -1;
+            goto out;
         }
 
-        /* Open the source, and also update mode/xattr */
-        ret = __dht_rebalance_open_src_file (from, to, loc, &stbuf, &src_fd);
+        ret = dict_set_str(meta_dict, GLUSTERFS_INTERNAL_FOP_KEY, "yes");
         if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "failed to open %s on %s",
-                        loc->path, from->name);
-                goto out;
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                   "Failed to set dictionary value: key = %s,"
+                   " path = %s",
+                   GLUSTERFS_INTERNAL_FOP_KEY, loc->path);
+            *fop_errno = ENOMEM;
+            ret = -1;
+            goto out;
         }
 
-        ret = syncop_fstat (from, src_fd, &stbuf);
+        ret = dict_set_int32(meta_dict, GF_META_LOCK_KEY, 1);
         if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "failed to lookup %s on %s (%s)",
-                        loc->path, from->name, strerror (errno));
-                goto out;
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Trace dict_set failed");
+            *fop_errno = ENOMEM;
+            ret = -1;
+            goto out;
         }
 
-        /* Try to preserve 'holes' while migrating data */
-        if (stbuf.ia_size > (stbuf.ia_blocks * GF_DISK_SECTOR_SIZE))
-                file_has_holes = 1;
-
-        /* All I/O happens in this function */
-        ret = __dht_rebalance_migrate_data (from, to, src_fd, dst_fd,
-					    stbuf.ia_size, file_has_holes);
+        ret = syncop_setxattr(from, loc, meta_dict, 0, NULL, NULL);
         if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "%s: failed to migrate data",
-                        loc->path);
-                /* reset the destination back to 0 */
-                ret = syncop_ftruncate (to, dst_fd, 0);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "%s: failed to reset target size back to 0 (%s)",
-                                loc->path, strerror (errno));
-                }
+            gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Trace syncop_setxattr metalock failed");
 
-                ret = -1;
-                goto out;
+            *fop_errno = -ret;
+            ret = -1;
+            goto out;
+        } else {
+            meta_locked = _gf_true;
         }
+    }
 
-        /* TODO: move all xattr related operations to fd based operations */
-        ret = syncop_listxattr (from, loc, &xattr);
-        if (ret == -1)
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to get xattr from %s (%s)",
-                        loc->path, from->name, strerror (errno));
+    if (!conf->lock_migration_enabled) {
+        plock.l_type = F_WRLCK;
+        plock.l_start = 0;
+        plock.l_len = 0;
+        plock.l_whence = SEEK_SET;
 
-        ret = syncop_setxattr (to, loc, xattr, 0);
-        if (ret == -1)
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to set xattr on %s (%s)",
-                        loc->path, to->name, strerror (errno));
+        ret = syncop_lk(from, src_fd, F_SETLK, &plock, NULL, NULL);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Migrate file failed:"
+                   "%s: Failed to lock on %s",
+                   loc->path, from->name);
+            *fop_errno = -ret;
+            ret = -1;
+            goto out;
+        }
 
-        /* TODO: Sync the locks */
+        p_locked = _gf_true;
 
-        ret = syncop_fsync (to, dst_fd, 0);
-        if (ret)
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to fsync on %s (%s)",
-                        loc->path, to->name, strerror (errno));
+    } else {
+        INIT_LIST_HEAD(&locklist.list);
 
+        ret = syncop_getactivelk(from, loc, &locklist, NULL, NULL);
+        if (ret == 0) {
+            gf_log(this->name, GF_LOG_INFO, "No active locks on:%s", loc->path);
 
-        /* Phase 2 - Data-Migration Complete, Housekeeping updates pending */
+        } else if (ret > 0) {
+            ret = syncop_setactivelk(to, loc, &locklist, NULL, NULL);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, -ret,
+                       DHT_MSG_LOCK_MIGRATION_FAILED, "write lock failed on:%s",
+                       loc->path);
 
-        ret = syncop_fstat (from, src_fd, &new_stbuf);
-        if (ret < 0) {
-                /* Failed to get the stat info */
-                gf_log (this->name, GF_LOG_ERROR,
-                        "failed to fstat file %s on %s (%s)",
-                        loc->path, from->name, strerror (errno));
-                goto out;
+                *fop_errno = -ret;
+                ret = -1;
+                goto metaunlock;
+            }
+        } else {
+            gf_msg(this->name, GF_LOG_ERROR, -ret,
+                   DHT_MSG_LOCK_MIGRATION_FAILED,
+                   "getactivelk failed for file: %s", loc->path);
+            *fop_errno = -ret;
         }
-
-        /* source would have both sticky bit and sgid bit set, reset it to 0,
-           and set the source permission on destination, if it was not set
-           prior to setting rebalance-modes in source  */
-        if (!src_ia_prot.sticky)
-                new_stbuf.ia_prot.sticky = 0;
-
-        if (!src_ia_prot.sgid)
-                new_stbuf.ia_prot.sgid = 0;
-
-        /* TODO: if the source actually had sticky bit, or sgid bit set,
-           we are not handling it */
-
-        ret = syncop_fsetattr (to, dst_fd, &new_stbuf,
-                               (GF_SET_ATTR_UID | GF_SET_ATTR_GID |
-                                GF_SET_ATTR_MODE), NULL, NULL);
+    }
+
+    /* source would have both sticky bit and sgid bit set, reset it to 0,
+       and set the source permission on destination, if it was not set
+       prior to setting rebalance-modes in source  */
+    if (!src_ia_prot.sticky)
+        new_stbuf.ia_prot.sticky = 0;
+
+    if (!src_ia_prot.sgid)
+        new_stbuf.ia_prot.sgid = 0;
+
+    /* TODO: if the source actually had sticky bit, or sgid bit set,
+       we are not handling it */
+
+    ret = syncop_fsetattr(
+        to, dst_fd, &new_stbuf,
+        (GF_SET_ATTR_UID | GF_SET_ATTR_GID | GF_SET_ATTR_MODE), NULL, NULL,
+        NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed:"
+               "%s: failed to perform setattr on %s ",
+               loc->path, to->name);
+        *fop_errno = -ret;
+        ret = -1;
+        goto metaunlock;
+    }
+
+    /* Because 'futimes' is not portable */
+    ret = syncop_setattr(to, loc, &new_stbuf,
+                         (GF_SET_ATTR_MTIME | GF_SET_ATTR_ATIME), NULL, NULL,
+                         NULL, NULL);
+    if (ret) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "%s: failed to perform setattr on %s ", loc->path, to->name);
+        *fop_errno = -ret;
+    }
+
+    if (target_changed) {
+        dict_del(dict, GLUSTERFS_POSIXLK_COUNT);
+        ret = dict_set_str(dict, conf->link_xattr_name, to->name);
         if (ret) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to perform setattr on %s (%s)",
-                        loc->path, to->name, strerror (errno));
-                goto out;
+            gf_log(this->name, GF_LOG_ERROR,
+                   "failed to set xattr in dict for %s (linkto:%s)", loc->path,
+                   to->name);
+            *fop_errno = ENOMEM;
+            ret = -1;
+            goto out;
         }
 
-        /* Because 'futimes' is not portable */
-        ret = syncop_setattr (to, loc, &new_stbuf,
-                              (GF_SET_ATTR_MTIME | GF_SET_ATTR_ATIME),
-                              NULL, NULL);
-        if (ret) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to perform setattr on %s (%s)",
-                        loc->path, to->name, strerror (errno));
+        ret = syncop_setxattr(old_target, loc, dict, 0, NULL, NULL);
+        if (ret && -ret != ESTALE && -ret != ENOENT) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "failed to set xattr on %s in %s", loc->path,
+                   old_target->name);
+            *fop_errno = -ret;
+            ret = -1;
+            goto out;
+        } else if (-ret == ESTALE || -ret == ENOENT) {
+            /* The failure ESTALE indicates that the linkto
+             * file on the hashed subvol might have been deleted.
+             * In this case will create a linkto file with new target
+             * as linkto xattr value*/
+            linkto_fd = fd_create(loc->inode, DHT_REBALANCE_PID);
+            if (!linkto_fd) {
+                gf_msg(this->name, GF_LOG_ERROR, errno,
+                       DHT_MSG_MIGRATE_FILE_FAILED, "%s: fd create failed",
+                       loc->path);
+                *fop_errno = ENOMEM;
+                ret = -1;
+                goto out;
+            }
+            ret = syncop_create(old_target, loc, O_RDWR, DHT_LINKFILE_MODE,
+                                linkto_fd, NULL, dict, NULL);
+            if (ret != 0 && -ret != EEXIST && -ret != ESTALE) {
+                *fop_errno = -ret;
+                ret = -1;
+                gf_msg(this->name, GF_LOG_ERROR, -ret,
+                       DHT_MSG_MIGRATE_FILE_FAILED,
+                       "failed to create linkto file on %s in %s", loc->path,
+                       old_target->name);
+                goto out;
+            } else if (ret == 0) {
+                ret = syncop_fsetattr(old_target, linkto_fd, &stbuf,
+                                      (GF_SET_ATTR_UID | GF_SET_ATTR_GID), NULL,
+                                      NULL, NULL, NULL);
+                if (ret < 0) {
+                    *fop_errno = -ret;
+                    gf_msg(this->name, GF_LOG_ERROR, -ret,
+                           DHT_MSG_MIGRATE_FILE_FAILED,
+                           "chown failed for %s on %s", loc->path,
+                           old_target->name);
+                }
+            }
+        }
+    }
+
+    clean_dst = _gf_false;
+
+    /* Posix acls are not set on DHT linkto files as part of the initial
+     * initial xattrs set on the dst file, so these need
+     * to be set on the dst file after the linkto attrs are removed.
+     * TODO: Optimize this.
+     */
+    if (xattr) {
+        dict_unref(xattr);
+        xattr = NULL;
+    }
+
+    /* Set only the Posix ACLs this time */
+    ret = syncop_getxattr(from, loc, &xattr, POSIX_ACL_ACCESS_XATTR, NULL,
+                          NULL);
+    if (ret < 0) {
+        if ((-ret != ENODATA) && (-ret != ENOATTR)) {
+            gf_msg(this->name, GF_LOG_WARNING, -ret,
+                   DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Migrate file failed:"
+                   "%s: failed to get xattr from %s",
+                   loc->path, from->name);
+            *fop_errno = -ret;
+        }
+    } else {
+        ret = syncop_setxattr(to, loc, xattr, 0, NULL, NULL);
+        if (ret < 0) {
+            /* Potential problem here where Posix ACLs will
+             * not be set on the target file */
+
+            gf_msg(this->name, GF_LOG_WARNING, -ret,
+                   DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Migrate file failed:"
+                   "%s: failed to set xattr on %s",
+                   loc->path, to->name);
+            *fop_errno = -ret;
         }
+    }
+
+    /* The src file is being unlinked after this so we don't need
+       to clean it up */
+    clean_src = _gf_false;
+
+    /* Make the source as a linkfile first before deleting it */
+    empty_iatt.ia_prot.sticky = 1;
+    ret = syncop_fsetattr(from, src_fd, &empty_iatt, GF_SET_ATTR_MODE, NULL,
+                          NULL, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed:"
+               "%s: failed to perform setattr on %s ",
+               loc->path, from->name);
+        *fop_errno = -ret;
+        ret = -1;
+        goto metaunlock;
+    }
+
+    /* Free up the data blocks on the source node, as the whole
+        file is migrated */
+    ret = syncop_ftruncate(from, src_fd, 0, NULL, NULL, NULL, NULL);
+    if (ret) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "%s: failed to perform truncate on %s (%s)", loc->path,
+               from->name, strerror(-ret));
+        *fop_errno = -ret;
+    }
+
+    /* remove the 'linkto' xattr from the destination */
+    ret = syncop_fremovexattr(to, dst_fd, conf->link_xattr_name, 0, NULL);
+    if (ret) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "%s: failed to perform removexattr on %s (%s)", loc->path,
+               to->name, strerror(-ret));
+        *fop_errno = -ret;
+    }
+
+    /* Do a stat and check the gfid before unlink */
+
+    /*
+     * Cached file changes its state from non-linkto to linkto file after
+     * migrating data. If lookup from any other mount-point is performed,
+     * converted-linkto-cached file will be treated as a stale and will be
+     * unlinked. But by this time, file is already migrated. So further
+     * failure because of ENOENT should  not be treated as error
+     */
+
+    ret = syncop_stat(from, loc, &empty_iatt, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: failed to do a stat on %s", loc->path, from->name);
+
+        if (-ret != ENOENT) {
+            *fop_errno = -ret;
+            ret = -1;
+            goto metaunlock;
+        }
+
+        rcvd_enoent_from_src = 1;
+    }
 
-        /* Make the source as a linkfile first before deleting it */
-        empty_iatt.ia_prot.sticky = 1;
-        ret = syncop_fsetattr (from, src_fd, &empty_iatt,
-                               GF_SET_ATTR_MODE, NULL, NULL);
+    if ((gf_uuid_compare(empty_iatt.ia_gfid, loc->gfid) == 0) &&
+        (!rcvd_enoent_from_src) && delete_src_linkto) {
+        /* take out the source from namespace */
+        ret = syncop_unlink(from, loc, NULL, NULL);
         if (ret) {
-                gf_log (this->name, GF_LOG_WARNING,             \
-                        "%s: failed to perform setattr on %s (%s)",
-                        loc->path, from->name, strerror (errno));
-                goto out;
+            gf_msg(this->name, GF_LOG_WARNING, -ret,
+                   DHT_MSG_MIGRATE_FILE_FAILED,
+                   "%s: failed to perform unlink on %s", loc->path, from->name);
+            *fop_errno = -ret;
+            ret = -1;
+            goto metaunlock;
         }
+    }
+
+    ret = syncop_lookup(this, loc, NULL, NULL, NULL, NULL);
+    if (ret) {
+        gf_msg_debug(this->name, -ret,
+                     "%s: failed to lookup the file on subvolumes", loc->path);
+        *fop_errno = -ret;
+    }
+
+    gf_msg(this->name, log_level, 0, DHT_MSG_MIGRATE_FILE_COMPLETE,
+           "completed migration of %s from subvolume %s to %s", loc->path,
+           from->name, to->name);
 
-       /* Free up the data blocks on the source node, as the whole
-           file is migrated */
-        ret = syncop_ftruncate (from, src_fd, 0);
+    ret = 0;
+
+metaunlock:
+
+    if (conf->lock_migration_enabled && meta_locked) {
+        dict_del(meta_dict, GF_META_LOCK_KEY);
+
+        ret = dict_set_int32(meta_dict, GF_META_UNLOCK_KEY, 1);
         if (ret) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to perform truncate on %s (%s)",
-                        loc->path, from->name, strerror (errno));
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Trace dict_set failed");
+
+            *fop_errno = ENOMEM;
+            ret = -1;
+            goto out;
         }
 
-        /* remove the 'linkto' xattr from the destination */
-        ret = syncop_fremovexattr (to, dst_fd, conf->link_xattr_name);
+        if (clean_dst == _gf_false)
+            ret = dict_set_int32(meta_dict, "status", 1);
+        else
+            ret = dict_set_int32(meta_dict, "status", 0);
+
         if (ret) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to perform removexattr on %s (%s)",
-                        loc->path, to->name, strerror (errno));
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Trace dict_set failed");
+
+            *fop_errno = ENOMEM;
+            ret = -1;
+            goto out;
         }
 
-        /* Do a stat and check the gfid before unlink */
-        ret = syncop_stat (from, loc, &empty_iatt);
+        ret = syncop_setxattr(from, loc, meta_dict, 0, NULL, NULL);
         if (ret) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to do a stat on %s (%s)",
-                        loc->path, from->name, strerror (errno));
-                goto out;
+            gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Trace syncop_setxattr meta unlock failed");
+
+            *fop_errno = -ret;
+            ret = -1;
+            goto out;
         }
+    }
 
-        if (uuid_compare (empty_iatt.ia_gfid, loc->gfid) == 0) {
-                /* take out the source from namespace */
-                ret = syncop_unlink (from, loc);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "%s: failed to perform unlink on %s (%s)",
-                                loc->path, from->name, strerror (errno));
-                        goto out;
-                }
+out:
+    if (clean_src) {
+        /* Revert source mode and xattr changes*/
+        lk_ret = __dht_migration_cleanup_src_file(this, loc, src_fd, from,
+                                                  &src_ia_prot);
+        if (lk_ret) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "%s: failed to cleanup source file on %s", loc->path,
+                   from->name);
         }
+    }
+
+    /* reset the destination back to 0 */
+    if (clean_dst) {
+        lk_ret = syncop_ftruncate(to, dst_fd, 0, NULL, NULL, NULL, NULL);
+        if (lk_ret) {
+            gf_msg(this->name, GF_LOG_ERROR, -lk_ret,
+                   DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Migrate file failed: "
+                   "%s: failed to reset target size back to 0",
+                   loc->path);
+        }
+    }
 
-        ret = syncop_lookup (this, loc, NULL, NULL, NULL, NULL);
-        if (ret) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "%s: failed to lookup the file on subvolumes (%s)",
-                        loc->path, strerror (errno));
+    if (inodelk_locked) {
+        flock.l_type = F_UNLCK;
+
+        lk_ret = syncop_inodelk(from, DHT_FILE_MIGRATE_DOMAIN, &tmp_loc,
+                                F_SETLK, &flock, NULL, NULL);
+        if (lk_ret < 0) {
+            gf_msg(this->name, GF_LOG_WARNING, -lk_ret,
+                   DHT_MSG_MIGRATE_FILE_FAILED,
+                   "%s: failed to unlock file on %s", loc->path, from->name);
+        }
+    }
+
+    if (entrylk_locked) {
+        lk_ret = syncop_entrylk(hashed_subvol, DHT_ENTRY_SYNC_DOMAIN,
+                                &parent_loc, loc->name, ENTRYLK_UNLOCK,
+                                ENTRYLK_UNLOCK, NULL, NULL);
+        if (lk_ret < 0) {
+            gf_msg(this->name, GF_LOG_WARNING, -lk_ret,
+                   DHT_MSG_MIGRATE_FILE_FAILED,
+                   "%s: failed to unlock entrylk on %s", loc->path,
+                   hashed_subvol->name);
         }
+    }
 
-        gf_log (this->name, GF_LOG_INFO,
-                "completed migration of %s from subvolume %s to %s",
-                loc->path, from->name, to->name);
+    if (p_locked) {
+        plock.l_type = F_UNLCK;
+        lk_ret = syncop_lk(from, src_fd, F_SETLK, &plock, NULL, NULL);
 
-        ret = 0;
-out:
-        if (dict)
-                dict_unref (dict);
+        if (lk_ret < 0) {
+            gf_msg(this->name, GF_LOG_WARNING, -lk_ret,
+                   DHT_MSG_MIGRATE_FILE_FAILED,
+                   "%s: failed to unlock file on %s", loc->path, from->name);
+        }
+    }
 
-        if (xattr)
-                dict_unref (xattr);
-        if (xattr_rsp)
-                dict_unref (xattr_rsp);
+    lk_ret = syncop_removexattr(to, loc, GF_PROTECT_FROM_EXTERNAL_WRITES, NULL,
+                                NULL);
+    if (lk_ret && (lk_ret != -ENODATA) && (lk_ret != -ENOATTR)) {
+        gf_msg(this->name, GF_LOG_WARNING, -lk_ret, 0,
+               "%s: removexattr failed key %s", loc->path,
+               GF_PROTECT_FROM_EXTERNAL_WRITES);
+    }
 
-        if (dst_fd)
-                syncop_close (dst_fd);
-        if (src_fd)
-                syncop_close (src_fd);
+    if (dict)
+        dict_unref(dict);
 
-        return ret;
-}
+    if (xattr)
+        dict_unref(xattr);
+    if (xattr_rsp)
+        dict_unref(xattr_rsp);
 
-static int
-rebalance_task (void *data)
-{
-        int           ret   = -1;
-        dht_local_t  *local = NULL;
-        call_frame_t *frame = NULL;
+    if (dst_fd)
+        syncop_close(dst_fd);
 
-        frame = data;
+    if (src_fd)
+        syncop_close(src_fd);
+    if (linkto_fd)
+        syncop_close(linkto_fd);
 
-        local = frame->local;
+    if (xdata)
+        dict_unref(xdata);
 
-        /* This function is 'synchrounous', hence if it returns,
-           we are done with the task */
-        ret = dht_migrate_file (THIS, &local->loc, local->rebalance.from_subvol,
-                                local->rebalance.target_node, local->flags);
+    loc_wipe(&tmp_loc);
+    loc_wipe(&parent_loc);
 
-        return ret;
+    return ret;
 }
 
 static int
-rebalance_task_completion (int op_ret, call_frame_t *sync_frame, void *data)
+rebalance_task(void *data)
 {
-        int           ret        = -1;
-        uint64_t      layout_int = 0;
-        dht_layout_t *layout     = 0;
-        xlator_t     *this       = NULL;
-        dht_local_t  *local      = NULL;
-        int32_t       op_errno   = EINVAL;
-
-        this = THIS;
-        local = sync_frame->local;
-
-        if (!op_ret) {
-                /* Make sure we have valid 'layout' in inode ctx
-                   after the operation */
-                ret = inode_ctx_del (local->loc.inode, this, &layout_int);
-                if (!ret && layout_int) {
-                        layout = (dht_layout_t *)(long)layout_int;
-                        dht_layout_unref (this, layout);
-                }
+    int ret = -1;
+    dht_local_t *local = NULL;
+    call_frame_t *frame = NULL;
+    int fop_errno = 0;
 
-                ret = dht_layout_preset (this, local->rebalance.target_node,
-                                         local->loc.inode);
-                if (ret)
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "%s: failed to set inode ctx", local->loc.path);
-        }
+    frame = data;
 
-        if (op_ret == -1) {
-                /* Failure of migration process, mostly due to write process.
-                   as we can't preserve the exact errno, lets say there was
-                   no space to migrate-data
-                */
-                op_errno = ENOSPC;
-        }
+    local = frame->local;
 
-        if (op_ret == 1) {
-                /* migration didn't happen, but is not a failure, let the user
-                   understand that he doesn't have permission to migrate the
-                   file.
-                */
-                op_ret = -1;
-                op_errno = EPERM;
-        }
+    /* This function is 'synchrounous', hence if it returns,
+       we are done with the task */
+    ret = dht_migrate_file(THIS, &local->loc, local->rebalance.from_subvol,
+                           local->rebalance.target_node, local->flags,
+                           &fop_errno);
 
-        DHT_STACK_UNWIND (setxattr, sync_frame, op_ret, op_errno, NULL);
-        return 0;
+    return ret;
+}
+
+static int
+rebalance_task_completion(int op_ret, call_frame_t *sync_frame, void *data)
+{
+    int32_t op_errno = EINVAL;
+
+    if (op_ret == -1) {
+        /* Failure of migration process, mostly due to write process.
+           as we can't preserve the exact errno, lets say there was
+           no space to migrate-data
+        */
+        op_errno = ENOSPC;
+    } else if (op_ret == 1) {
+        /* migration didn't happen, but is not a failure, let the user
+           understand that he doesn't have permission to migrate the
+           file.
+        */
+        op_ret = -1;
+        op_errno = EPERM;
+    } else if (op_ret != 0) {
+        op_errno = -op_ret;
+        op_ret = -1;
+    }
+
+    DHT_STACK_UNWIND(setxattr, sync_frame, op_ret, op_errno, NULL);
+    return 0;
 }
 
 int
-dht_start_rebalance_task (xlator_t *this, call_frame_t *frame)
+dht_start_rebalance_task(xlator_t *this, call_frame_t *frame)
 {
-        int         ret     = -1;
+    int ret = -1;
 
-        ret = synctask_new (this->ctx->env, rebalance_task,
-                            rebalance_task_completion,
-                            frame, frame);
-        return ret;
+    ret = synctask_new(this->ctx->env, rebalance_task,
+                       rebalance_task_completion, frame, frame);
+    return ret;
 }
 
 int
-gf_listener_stop (xlator_t *this)
+gf_listener_stop(xlator_t *this)
 {
-        glusterfs_ctx_t  *ctx = NULL;
-        cmd_args_t       *cmd_args = NULL;
-        int              ret = 0;
-
-        ctx = this->ctx;
-        GF_ASSERT (ctx);
-        cmd_args = &ctx->cmd_args;
-        if (cmd_args->sock_file) {
-                ret = unlink (cmd_args->sock_file);
-                if (ret && (ENOENT == errno)) {
-                        ret = 0;
-                }
-        }
-
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "Failed to unlink listener "
-                        "socket %s, error: %s", cmd_args->sock_file,
-                        strerror (errno));
+    glusterfs_ctx_t *ctx = NULL;
+    cmd_args_t *cmd_args = NULL;
+    int ret = 0;
+
+    ctx = this->ctx;
+    GF_ASSERT(ctx);
+    cmd_args = &ctx->cmd_args;
+    if (cmd_args->sock_file) {
+        ret = sys_unlink(cmd_args->sock_file);
+        if (ret && (ENOENT == errno)) {
+            ret = 0;
         }
-        return ret;
+    }
+
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, DHT_MSG_SOCKET_ERROR,
+               "Failed to unlink listener "
+               "socket %s",
+               cmd_args->sock_file);
+    }
+    return ret;
 }
 
 void
-dht_build_root_inode (xlator_t *this, inode_t **inode)
+dht_build_root_inode(xlator_t *this, inode_t **inode)
 {
-        inode_table_t    *itable        = NULL;
-        uuid_t            root_gfid     = {0, };
+    inode_table_t *itable = NULL;
+    static uuid_t root_gfid = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1};
 
-        itable = inode_table_new (0, this);
-        if (!itable)
-                return;
+    itable = inode_table_new(0, this);
+    if (!itable)
+        return;
 
-        root_gfid[15] = 1;
-        *inode = inode_find (itable, root_gfid);
+    *inode = inode_find(itable, root_gfid);
 }
 
 void
-dht_build_root_loc (inode_t *inode, loc_t *loc)
+dht_build_root_loc(inode_t *inode, loc_t *loc)
 {
-        loc->path = "/";
-        loc->inode = inode;
-        loc->inode->ia_type = IA_IFDIR;
-        memset (loc->gfid, 0, 16);
-        loc->gfid[15] = 1;
+    loc->path = "/";
+    loc->inode = inode;
+    loc->inode->ia_type = IA_IFDIR;
+    memset(loc->gfid, 0, 16);
+    loc->gfid[15] = 1;
 }
 
-
 /* return values: 1 -> error, bug ignore and continue
                   0 -> proceed
                  -1 -> error, handle it */
 int32_t
-gf_defrag_handle_migrate_error (int32_t op_errno, gf_defrag_info_t *defrag)
+gf_defrag_handle_migrate_error(int32_t op_errno, gf_defrag_info_t *defrag)
 {
-        /* if errno is not ENOSPC or ENOTCONN, we can still continue
-           with rebalance process */
-        if ((errno != ENOSPC) || (errno != ENOTCONN))
-                return 1;
-
-        if (errno == ENOTCONN) {
-                /* Most probably mount point went missing (mostly due
-                   to a brick down), say rebalance failure to user,
-                   let him restart it if everything is fine */
-                defrag->defrag_status = GF_DEFRAG_STATUS_FAILED;
-                return -1;
-        }
-
-        if (errno == ENOSPC) {
-                /* rebalance process itself failed, may be
-                   remote brick went down, or write failed due to
-                   disk full etc etc.. */
-                defrag->defrag_status = GF_DEFRAG_STATUS_FAILED;
-                return -1;
-        }
+    int ret = 0;
+    /* if errno is not ENOTCONN, we can still continue
+       with rebalance process */
+    if (op_errno != ENOTCONN) {
+        ret = 1;
+        goto out;
+    }
+
+    if (op_errno == ENOTCONN) {
+        /* Most probably mount point went missing (mostly due
+           to a brick down), say rebalance failure to user,
+           let him restart it if everything is fine */
+        defrag->defrag_status = GF_DEFRAG_STATUS_FAILED;
+        ret = -1;
+        goto out;
+    }
 
-        return 0;
+out:
+    return ret;
 }
 
 static gf_boolean_t
-gf_defrag_pattern_match (gf_defrag_info_t *defrag, char *name, uint64_t size)
+gf_defrag_pattern_match(gf_defrag_info_t *defrag, char *name, uint64_t size)
 {
-        gf_defrag_pattern_list_t *trav = NULL;
-        gf_boolean_t               match = _gf_false;
-        gf_boolean_t               ret = _gf_false;
+    gf_defrag_pattern_list_t *trav = NULL;
+    gf_boolean_t match = _gf_false;
+    gf_boolean_t ret = _gf_false;
 
-        GF_VALIDATE_OR_GOTO ("dht", defrag, out);
+    GF_VALIDATE_OR_GOTO("dht", defrag, out);
 
-        trav = defrag->defrag_pattern;
-        while (trav) {
-                if (!fnmatch (trav->path_pattern, name, FNM_NOESCAPE)) {
-                        match = _gf_true;
-                        break;
-                }
-                trav = trav->next;
+    trav = defrag->defrag_pattern;
+    while (trav) {
+        if (!fnmatch(trav->path_pattern, name, FNM_NOESCAPE)) {
+            match = _gf_true;
+            break;
         }
+        trav = trav->next;
+    }
 
-        if ((match == _gf_true) && (size >= trav->size))
-                ret = _gf_true;
+    if ((match == _gf_true) && (size >= trav->size))
+        ret = _gf_true;
 
- out:
-        return ret;
+out:
+    return ret;
+}
+
+int
+dht_dfreaddirp_done(dht_dfoffset_ctx_t *offset_var, int cnt)
+{
+    int i;
+    int result = 1;
+
+    for (i = 0; i < cnt; i++) {
+        if (offset_var[i].readdir_done == 0) {
+            result = 0;
+            break;
+        }
+    }
+    return result;
 }
 
-/* We do a depth first traversal of directories. But before we move into
- * subdirs, we complete the data migration of those directories whose layouts
- * have been fixed
+int static gf_defrag_ctx_subvols_init(dht_dfoffset_ctx_t *offset_var,
+                                      xlator_t *this)
+{
+    int i;
+    dht_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf)
+        return -1;
+
+    for (i = 0; i < conf->local_subvols_cnt; i++) {
+        offset_var[i].this = conf->local_subvols[i];
+        offset_var[i].offset = (off_t)0;
+        offset_var[i].readdir_done = 0;
+    }
+
+    return 0;
+}
+
+static int
+dht_get_first_non_null_index(subvol_nodeuuids_info_t *entry)
+{
+    int i = 0;
+    int index = 0;
+
+    for (i = 0; i < entry->count; i++) {
+        if (!gf_uuid_is_null(entry->elements[i].uuid)) {
+            index = i;
+            goto out;
+        }
+    }
+
+    if (i == entry->count) {
+        index = -1;
+    }
+out:
+    return index;
+}
+
+/* Return value
+ * 0 : this node does not migrate the file
+ * 1 : this node migrates the file
+ *
+ * Use the hash value of the gfid to determine which node will migrate files.
+ * Using the gfid instead of the name also ensures that the same node handles
+ * all hardlinks.
  */
 
-int
-gf_defrag_migrate_data (xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
-                        dict_t *migrate_data)
+gf_boolean_t
+gf_defrag_should_i_migrate(xlator_t *this, int local_subvol_index, uuid_t gfid)
 {
-        int                      ret            = -1;
-        loc_t                    entry_loc      = {0,};
-        fd_t                    *fd             = NULL;
-        gf_dirent_t              entries;
-        gf_dirent_t             *tmp            = NULL;
-        gf_dirent_t             *entry          = NULL;
-        gf_boolean_t             free_entries   = _gf_false;
-        off_t                    offset         = 0;
-        dict_t                  *dict           = NULL;
-        struct iatt              iatt           = {0,};
-        int32_t                  op_errno       = 0;
-        char                    *uuid_str       = NULL;
-        uuid_t                   node_uuid      = {0,};
-        int                      readdir_operrno = 0;
-        struct timeval           dir_start      = {0,};
-        struct timeval           end            = {0,};
-        double                   elapsed        = {0,};
-        struct timeval           start          = {0,};
-        int32_t                  err            = 0;
-
-        gf_log (this->name, GF_LOG_INFO, "migrate data called on %s",
-                loc->path);
-        gettimeofday (&dir_start, NULL);
-
-        fd = fd_create (loc->inode, defrag->pid);
-        if (!fd) {
-                gf_log (this->name, GF_LOG_ERROR, "Failed to create fd");
-                goto out;
+    gf_boolean_t ret = _gf_false;
+    int i = local_subvol_index;
+    char *str = NULL;
+    uint32_t hashval = 0;
+    int32_t index = 0;
+    dht_conf_t *conf = NULL;
+    char buf[UUID_CANONICAL_FORM_LEN + 1] = {
+        0,
+    };
+    subvol_nodeuuids_info_t *entry = NULL;
+
+    conf = this->private;
+
+    /* Pure distribute. A subvol in this case
+        will be handled by only one node */
+
+    entry = &(conf->local_nodeuuids[i]);
+    if (entry->count == 1) {
+        return 1;
+    }
+
+    str = uuid_utoa_r(gfid, buf);
+    if (dht_hash_compute(this, 0, str, &hashval) == 0) {
+        index = (hashval % entry->count);
+        if (entry->elements[index].info == REBAL_NODEUUID_MINE) {
+            /* Index matches this node's nodeuuid.*/
+            ret = _gf_true;
+            goto out;
         }
 
-        ret = syncop_opendir (this, loc, fd);
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "Failed to open dir %s",
-                        loc->path);
+        /* Brick down - some other node has to migrate these files*/
+        if (gf_uuid_is_null(entry->elements[index].uuid)) {
+            /* Fall back to the first non-null index */
+            index = dht_get_first_non_null_index(entry);
+
+            if (index == -1) {
+                /* None of the bricks in the subvol are up.
+                 * CHILD_DOWN will kill the process soon */
+
+                return _gf_false;
+            }
+
+            if (entry->elements[index].info == REBAL_NODEUUID_MINE) {
+                /* Index matches this node's nodeuuid.*/
+                ret = _gf_true;
                 goto out;
+            }
         }
+    }
+out:
+    return ret;
+}
 
-        INIT_LIST_HEAD (&entries.list);
+int
+gf_defrag_migrate_single_file(void *opaque)
+{
+    xlator_t *this = NULL;
+    dht_conf_t *conf = NULL;
+    gf_defrag_info_t *defrag = NULL;
+    int ret = 0;
+    gf_dirent_t *entry = NULL;
+    struct timeval start = {
+        0,
+    };
+    loc_t entry_loc = {
+        0,
+    };
+    loc_t *loc = NULL;
+    struct iatt iatt = {
+        0,
+    };
+    dict_t *migrate_data = NULL;
+    struct timeval end = {
+        0,
+    };
+    double elapsed = {
+        0,
+    };
+    struct dht_container *rebal_entry = NULL;
+    inode_t *inode = NULL;
+    xlator_t *hashed_subvol = NULL;
+    xlator_t *cached_subvol = NULL;
+    call_frame_t *statfs_frame = NULL;
+    xlator_t *old_THIS = NULL;
+    data_t *tmp = NULL;
+    int fop_errno = 0;
+    gf_dht_migrate_data_type_t rebal_type = GF_DHT_MIGRATE_DATA;
+    char value[MAX_REBAL_TYPE_SIZE] = {
+        0,
+    };
+    struct iatt *iatt_ptr = NULL;
+    gf_boolean_t update_skippedcount = _gf_true;
+    int i = 0;
+    gf_boolean_t should_i_migrate = 0;
+
+    rebal_entry = (struct dht_container *)opaque;
+    if (!rebal_entry) {
+        gf_log("DHT", GF_LOG_ERROR, "rebal_entry is NULL");
+        ret = -1;
+        goto out;
+    }
 
-        while ((ret = syncop_readdirp (this, fd, 131072, offset, NULL,
-                                       &entries)) != 0) {
+    this = rebal_entry->this;
 
-                if (ret < 0) {
+    conf = this->private;
 
-                        gf_log (this->name, GF_LOG_ERROR, "Readdir returned %s."
-                                " Aborting migrate-data",
-                                strerror(readdir_operrno));
-                        goto out;
-                }
+    defrag = conf->defrag;
 
-                /* Need to keep track of ENOENT errno, that means, there is no
-                   need to send more readdirp() */
-                readdir_operrno = errno;
+    loc = rebal_entry->parent_loc;
 
-                if (list_empty (&entries.list))
-                        break;
+    migrate_data = rebal_entry->migrate_data;
 
-                free_entries = _gf_true;
+    entry = rebal_entry->df_entry;
+    iatt_ptr = &entry->d_stat;
 
-                list_for_each_entry_safe (entry, tmp, &entries.list, list) {
-                        if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) {
-                                ret = 1;
-                                goto out;
-                        }
+    if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) {
+        ret = -1;
+        goto out;
+    }
 
-                        offset = entry->d_off;
+    if (defrag->stats == _gf_true) {
+        gettimeofday(&start, NULL);
+    }
 
-                        if (!strcmp (entry->d_name, ".") ||
-                            !strcmp (entry->d_name, ".."))
-                                continue;
+    if (defrag->defrag_pattern &&
+        (gf_defrag_pattern_match(defrag, entry->d_name,
+                                 entry->d_stat.ia_size) == _gf_false)) {
+        gf_log(this->name, GF_LOG_ERROR, "pattern_match failed");
+        goto out;
+    }
 
-                        if (IA_ISDIR (entry->d_stat.ia_type))
-                                continue;
+    memset(&entry_loc, 0, sizeof(entry_loc));
 
-                        defrag->num_files_lookedup++;
-                        if (defrag->stats == _gf_true) {
-                                gettimeofday (&start, NULL);
-                        }
-                        if (defrag->defrag_pattern &&
-                            (gf_defrag_pattern_match (defrag, entry->d_name,
-                                                      entry->d_stat.ia_size)
-                             == _gf_false)) {
-                                continue;
-                        }
-                        loc_wipe (&entry_loc);
-                        ret =dht_build_child_loc (this, &entry_loc, loc,
-                                                  entry->d_name);
-                        if (ret) {
-                                gf_log (this->name, GF_LOG_ERROR, "Child loc"
-                                        " build failed");
-                                goto out;
-                        }
+    ret = dht_build_child_loc(this, &entry_loc, loc, entry->d_name);
+    if (ret) {
+        LOCK(&defrag->lock);
+        {
+            defrag->total_failures += 1;
+        }
+        UNLOCK(&defrag->lock);
 
-                        if (uuid_is_null (entry->d_stat.ia_gfid)) {
-                                gf_log (this->name, GF_LOG_ERROR, "%s/%s"
-                                        " gfid not present", loc->path,
-                                         entry->d_name);
-                                continue;
-                        }
+        ret = 0;
 
-                        uuid_copy (entry_loc.gfid, entry->d_stat.ia_gfid);
+        gf_log(this->name, GF_LOG_ERROR, "Child loc build failed");
 
-                        if (uuid_is_null (loc->gfid)) {
-                                gf_log (this->name, GF_LOG_ERROR, "%s/%s"
-                                        " gfid not present", loc->path,
-                                         entry->d_name);
-                                continue;
-                        }
+        goto out;
+    }
 
-                        uuid_copy (entry_loc.pargfid, loc->gfid);
+    should_i_migrate = gf_defrag_should_i_migrate(
+        this, rebal_entry->local_subvol_index, entry->d_stat.ia_gfid);
 
-                        entry_loc.inode->ia_type = entry->d_stat.ia_type;
+    gf_uuid_copy(entry_loc.gfid, entry->d_stat.ia_gfid);
 
-                        ret = syncop_lookup (this, &entry_loc, NULL, &iatt,
-                                             NULL, NULL);
-                        if (ret) {
-                                gf_log (this->name, GF_LOG_ERROR, "%s"
-                                        " lookup failed", entry_loc.path);
-                                continue;
-                        }
+    gf_uuid_copy(entry_loc.pargfid, loc->gfid);
 
-                        ret = syncop_getxattr (this, &entry_loc, &dict,
-                                               GF_XATTR_NODE_UUID_KEY);
-                        if(ret < 0) {
-                                gf_log (this->name, GF_LOG_ERROR, "Failed to "
-                                        "get node-uuid for %s", entry_loc.path);
-                                continue;
-                        }
+    ret = syncop_lookup(this, &entry_loc, &iatt, NULL, NULL, NULL);
 
-                        ret = dict_get_str (dict, GF_XATTR_NODE_UUID_KEY,
-                                            &uuid_str);
-                        if(ret < 0) {
-                                gf_log (this->name, GF_LOG_ERROR, "Failed to "
-                                        "get node-uuid from dict for %s",
-                                        entry_loc.path);
-                                continue;
-                        }
+    if (!should_i_migrate) {
+        /* this node isn't supposed to migrate the file. suppressing any
+         * potential error from lookup as this file is under migration by
+         * another node */
+        if (ret) {
+            gf_msg_debug(this->name, -ret,
+                         "Ignoring lookup failure: node isn't migrating %s",
+                         entry_loc.path);
+            ret = 0;
+        }
+        gf_msg_debug(this->name, 0, "Don't migrate %s ", entry_loc.path);
+        goto out;
+    }
 
-                        if (uuid_parse (uuid_str, node_uuid)) {
-                                gf_log (this->name, GF_LOG_ERROR, "uuid_parse "
-                                        "failed for %s", entry_loc.path);
-                                continue;
-                        }
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed: %s lookup failed", entry_loc.path);
 
-                        /* if file belongs to different node, skip migration
-                         * the other node will take responsibility of migration
-                         */
-                        if (uuid_compare (node_uuid, defrag->node_uuid)) {
-                                gf_log (this->name, GF_LOG_TRACE, "%s does not"
-                                        "belong to this node", entry_loc.path);
-                                continue;
-                        }
+        /* Increase failure count only for remove-brick op, so that
+         * user is warned to check the removed-brick for any files left
+         * unmigrated
+         */
+        if (conf->decommission_subvols_cnt) {
+            LOCK(&defrag->lock);
+            {
+                defrag->total_failures += 1;
+            }
+            UNLOCK(&defrag->lock);
+        }
 
-                        uuid_str = NULL;
+        ret = 0;
+        goto out;
+    }
 
-                        dict_del (dict, GF_XATTR_NODE_UUID_KEY);
+    iatt_ptr = &iatt;
 
+    hashed_subvol = dht_subvol_get_hashed(this, &entry_loc);
+    if (!hashed_subvol) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+               "Failed to get hashed subvol for %s", entry_loc.path);
+        ret = 0;
+        goto out;
+    }
 
-                        /* if distribute is present, it will honor this key.
-                         * -1 is returned if distribute is not present or file
-                         * doesn't have a link-file. If file has link-file, the
-                         * path of link-file will be the value, and also that
-                         * guarantees that file has to be mostly migrated */
+    cached_subvol = dht_subvol_get_cached(this, entry_loc.inode);
+    if (!cached_subvol) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_CACHED_SUBVOL_GET_FAILED,
+               "Failed to get cached subvol for %s", entry_loc.path);
 
-                        ret = syncop_getxattr (this, &entry_loc, &dict,
-                                               GF_XATTR_LINKINFO_KEY);
-                        if (ret < 0) {
-                                gf_log (this->name, GF_LOG_TRACE, "failed to "
-                                        "get link-to key for %s",
-                                        entry_loc.path);
-                                continue;
-                        }
+        ret = 0;
+        goto out;
+    }
 
-                        ret = syncop_setxattr (this, &entry_loc, migrate_data,
-                                               0);
-                        if (ret) {
-                                err = op_errno;
-                                /* errno is overloaded. See
-                                 * rebalance_task_completion () */
-                                if (err != ENOSPC) {
-                                        gf_log (this->name, GF_LOG_DEBUG,
-                                                "migrate-data skipped for %s"
-                                                " due to space constraints",
-                                                entry_loc.path);
-                                        defrag->skipped +=1;
-                                } else{
-                                        gf_log (this->name, GF_LOG_ERROR,
-                                                "migrate-data failed for %s",
-                                                entry_loc.path);
-                                        defrag->total_failures +=1;
-                                }
+    if (hashed_subvol == cached_subvol) {
+        ret = 0;
+        goto out;
+    }
+
+    inode = inode_link(entry_loc.inode, entry_loc.parent, entry->d_name, &iatt);
+    inode_unref(entry_loc.inode);
+    /* use the inode returned by inode_link */
+    entry_loc.inode = inode;
+
+    old_THIS = THIS;
+    THIS = this;
+    statfs_frame = create_frame(this, this->ctx->pool);
+    if (!statfs_frame) {
+        gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_NO_MEMORY, ENOMEM,
+               "Insufficient memory. Frame creation failed");
+        ret = -1;
+        goto out;
+    }
+
+    /* async statfs information for honoring min-free-disk */
+    dht_get_du_info(statfs_frame, this, loc);
+    THIS = old_THIS;
+
+    tmp = dict_get(migrate_data, GF_XATTR_FILE_MIGRATE_KEY);
+    if (tmp) {
+        memcpy(value, tmp->data, tmp->len);
+        if (strcmp(value, "force") == 0)
+            rebal_type = GF_DHT_MIGRATE_DATA_EVEN_IF_LINK_EXISTS;
+
+        if (conf->decommission_in_progress)
+            rebal_type = GF_DHT_MIGRATE_HARDLINK;
+    }
+
+    ret = dht_migrate_file(this, &entry_loc, cached_subvol, hashed_subvol,
+                           rebal_type, &fop_errno);
+    if (ret == 1) {
+        if (fop_errno == ENOSPC) {
+            gf_msg_debug(this->name, 0,
+                         "migrate-data skipped for"
+                         " %s due to space constraints",
+                         entry_loc.path);
+
+            /* For remove-brick case if the source is not one of the
+             * removed-brick, do not mark the error as failure */
+            if (conf->decommission_subvols_cnt) {
+                for (i = 0; i < conf->subvolume_cnt; i++) {
+                    if (conf->decommissioned_bricks[i] == cached_subvol) {
+                        LOCK(&defrag->lock);
+                        {
+                            defrag->total_failures += 1;
+                            update_skippedcount = _gf_false;
                         }
+                        UNLOCK(&defrag->lock);
 
-                        if (ret == -1) {
-                                op_errno = errno;
-                                ret = gf_defrag_handle_migrate_error (op_errno,
-                                                                      defrag);
-
-                                if (!ret)
-                                        gf_log (this->name, GF_LOG_DEBUG,
-                                                "migrate-data on %s failed: %s",
-                                                entry_loc.path,
-                                                strerror (op_errno));
-                                else if (ret == 1)
-                                        continue;
-                                else if (ret == -1)
-                                        goto out;
-                        }
+                        break;
+                    }
+                }
+            }
 
-                        LOCK (&defrag->lock);
-                        {
-                                defrag->total_files += 1;
-                                defrag->total_data += iatt.ia_size;
-                        }
-                        UNLOCK (&defrag->lock);
-                        if (defrag->stats == _gf_true) {
-                                gettimeofday (&end, NULL);
-                                elapsed = (end.tv_sec - start.tv_sec) * 1e6 +
-                                          (end.tv_usec - start.tv_usec);
-                                gf_log (this->name, GF_LOG_INFO, "Migration of "
-                                        "file:%s size:%"PRIu64" bytes took %.2f"
-                                        "secs", entry_loc.path, iatt.ia_size,
-                                         elapsed/1e6);
-                        }
+            if (update_skippedcount) {
+                LOCK(&defrag->lock);
+                {
+                    defrag->skipped += 1;
                 }
+                UNLOCK(&defrag->lock);
+
+                gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_MIGRATE_FILE_SKIPPED,
+                       "File migration skipped for %s.", entry_loc.path);
+            }
+
+        } else if (fop_errno == ENOTSUP) {
+            gf_msg_debug(this->name, 0,
+                         "migrate-data skipped for"
+                         " hardlink %s ",
+                         entry_loc.path);
+            LOCK(&defrag->lock);
+            {
+                defrag->skipped += 1;
+            }
+            UNLOCK(&defrag->lock);
+
+            gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_MIGRATE_FILE_SKIPPED,
+                   "File migration skipped for %s.", entry_loc.path);
+        }
+
+        ret = 0;
+        goto out;
+    } else if (ret < 0) {
+        if (fop_errno != EEXIST) {
+            gf_msg(this->name, GF_LOG_ERROR, fop_errno,
+                   DHT_MSG_MIGRATE_FILE_FAILED, "migrate-data failed for %s",
+                   entry_loc.path);
+
+            LOCK(&defrag->lock);
+            {
+                defrag->total_failures += 1;
+            }
+            UNLOCK(&defrag->lock);
+        }
 
-                gf_dirent_free (&entries);
-                free_entries = _gf_false;
-                INIT_LIST_HEAD (&entries.list);
+        ret = gf_defrag_handle_migrate_error(fop_errno, defrag);
 
-                if (readdir_operrno == ENOENT)
-                        break;
+        if (!ret) {
+            gf_msg(this->name, GF_LOG_ERROR, fop_errno,
+                   DHT_MSG_MIGRATE_FILE_FAILED,
+                   "migrate-data on %s failed:", entry_loc.path);
+        } else if (ret == 1) {
+            ret = 0;
         }
 
-        gettimeofday (&end, NULL);
-        elapsed = (end.tv_sec - dir_start.tv_sec) * 1e6 +
-                  (end.tv_usec - dir_start.tv_usec);
-        gf_log (this->name, GF_LOG_INFO, "Migration operation on dir %s took "
-                "%.2f secs", loc->path, elapsed/1e6);
-        ret = 0;
+        goto out;
+    }
+
+    LOCK(&defrag->lock);
+    {
+        defrag->total_files += 1;
+        defrag->total_data += iatt.ia_size;
+    }
+    UNLOCK(&defrag->lock);
+
+    if (defrag->stats == _gf_true) {
+        gettimeofday(&end, NULL);
+        elapsed = gf_tvdiff(&start, &end);
+        gf_log(this->name, GF_LOG_INFO,
+               "Migration of "
+               "file:%s size:%" PRIu64
+               " bytes took %.2f"
+               "secs and ret: %d",
+               entry_loc.name, iatt.ia_size, elapsed / 1e6, ret);
+    }
+
 out:
-        if (free_entries)
-                gf_dirent_free (&entries);
+    if (statfs_frame) {
+        STACK_DESTROY(statfs_frame->root);
+    }
 
-        loc_wipe (&entry_loc);
+    if (iatt_ptr) {
+        LOCK(&defrag->lock);
+        {
+            defrag->size_processed += iatt_ptr->ia_size;
+        }
+        UNLOCK(&defrag->lock);
+    }
+    loc_wipe(&entry_loc);
 
-        if (dict)
-                dict_unref(dict);
+    return ret;
+}
 
-        if (fd)
-                fd_unref (fd);
-        return ret;
+void *
+gf_defrag_task(void *opaque)
+{
+    struct list_head *q_head = NULL;
+    struct dht_container *iterator = NULL;
+    gf_defrag_info_t *defrag = NULL;
+    int ret = 0;
+    pid_t pid = GF_CLIENT_PID_DEFRAG;
+
+    defrag = (gf_defrag_info_t *)opaque;
+    if (!defrag) {
+        gf_msg("dht", GF_LOG_ERROR, 0, 0, "defrag is NULL");
+        goto out;
+    }
+
+    syncopctx_setfspid(&pid);
+
+    q_head = &(defrag->queue[0].list);
+
+    /* The following while loop will dequeue one entry from the defrag->queue
+       under lock. We will update the defrag->global_error only when there
+       is an error which is critical to stop the rebalance process. The stop
+       message will be intimated to other migrator threads by setting the
+       defrag->defrag_status to GF_DEFRAG_STATUS_FAILED.
+
+       In defrag->queue, a low watermark (MIN_MIGRATE_QUEUE_COUNT) is
+       maintained so that crawler does not starve the file migration
+       workers and a high watermark (MAX_MIGRATE_QUEUE_COUNT) so that
+       crawler does not go far ahead in filling up the queue.
+     */
+
+    while (_gf_true) {
+        if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) {
+            pthread_cond_broadcast(&defrag->rebalance_crawler_alarm);
+            pthread_cond_broadcast(&defrag->parallel_migration_cond);
+            goto out;
+        }
 
+        pthread_mutex_lock(&defrag->dfq_mutex);
+        {
+            /*Throttle down:
+              If the reconfigured count is less than current thread
+              count, then the current thread will sleep */
+
+            /*TODO: Need to refactor the following block to work
+             *under defrag->lock. For now access
+             * defrag->current_thread_count and rthcount under
+             * dfq_mutex lock */
+            while (!defrag->crawl_done && (defrag->recon_thread_count <
+                                           defrag->current_thread_count)) {
+                defrag->current_thread_count--;
+                gf_msg_debug("DHT", 0,
+                             "Thread sleeping. "
+                             "current thread count: %d",
+                             defrag->current_thread_count);
+
+                pthread_cond_wait(&defrag->df_wakeup_thread,
+                                  &defrag->dfq_mutex);
+
+                defrag->current_thread_count++;
+                gf_msg_debug("DHT", 0,
+                             "Thread wokeup. "
+                             "current thread count: %d",
+                             defrag->current_thread_count);
+            }
+
+            if (defrag->q_entry_count) {
+                iterator = list_entry(q_head->next, typeof(*iterator), list);
+
+                gf_msg_debug("DHT", 0,
+                             "picking entry "
+                             "%s",
+                             iterator->df_entry->d_name);
+
+                list_del_init(&(iterator->list));
+
+                defrag->q_entry_count--;
+
+                if ((defrag->q_entry_count < MIN_MIGRATE_QUEUE_COUNT) &&
+                    defrag->wakeup_crawler) {
+                    pthread_cond_broadcast(&defrag->rebalance_crawler_alarm);
+                }
+                pthread_mutex_unlock(&defrag->dfq_mutex);
+                ret = gf_defrag_migrate_single_file((void *)iterator);
+
+                /*Critical errors: ENOTCONN and ENOSPACE*/
+                if (ret) {
+                    dht_set_global_defrag_error(defrag, ret);
+
+                    defrag->defrag_status = GF_DEFRAG_STATUS_FAILED;
+
+                    pthread_cond_broadcast(&defrag->rebalance_crawler_alarm);
+
+                    pthread_cond_broadcast(&defrag->parallel_migration_cond);
+
+                    goto out;
+                }
+
+                gf_defrag_free_container(iterator);
+
+                continue;
+            } else {
+                /* defrag->crawl_done flag is set means crawling
+                 file system is done and hence a list_empty when
+                 the above flag is set indicates there are no more
+                 entries to be added to the queue and rebalance is
+                 finished */
+
+                if (!defrag->crawl_done) {
+                    defrag->current_thread_count--;
+                    gf_msg_debug("DHT", 0,
+                                 "Thread "
+                                 "sleeping while  waiting "
+                                 "for migration entries. "
+                                 "current thread  count:%d",
+                                 defrag->current_thread_count);
+
+                    pthread_cond_wait(&defrag->parallel_migration_cond,
+                                      &defrag->dfq_mutex);
+                }
+
+                if (defrag->crawl_done && !defrag->q_entry_count) {
+                    defrag->current_thread_count++;
+                    gf_msg_debug("DHT", 0, "Exiting thread");
+
+                    pthread_cond_broadcast(&defrag->parallel_migration_cond);
+                    goto unlock;
+                } else {
+                    defrag->current_thread_count++;
+                    gf_msg_debug("DHT", 0,
+                                 "Thread woke up"
+                                 " as found migrating entries. "
+                                 "current thread count:%d",
+                                 defrag->current_thread_count);
+
+                    pthread_mutex_unlock(&defrag->dfq_mutex);
+                    continue;
+                }
+            }
+        }
+    unlock:
+        pthread_mutex_unlock(&defrag->dfq_mutex);
+        break;
+    }
+out:
+    return NULL;
 }
 
+int static gf_defrag_get_entry(xlator_t *this, int i,
+                               struct dht_container **container, loc_t *loc,
+                               dht_conf_t *conf, gf_defrag_info_t *defrag,
+                               fd_t *fd, dict_t *migrate_data,
+                               struct dir_dfmeta *dir_dfmeta, dict_t *xattr_req,
+                               int *perrno)
+{
+    int ret = 0;
+    char is_linkfile = 0;
+    gf_dirent_t *df_entry = NULL;
+    struct dht_container *tmp_container = NULL;
+
+    if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) {
+        ret = -1;
+        goto out;
+    }
+
+    if (dir_dfmeta->offset_var[i].readdir_done == 1) {
+        ret = 0;
+        goto out;
+    }
+
+    if (dir_dfmeta->fetch_entries[i] == 1) {
+        if (!fd) {
+            dir_dfmeta->fetch_entries[i] = 0;
+            dir_dfmeta->offset_var[i].readdir_done = 1;
+            ret = 0;
+            goto out;
+        }
+
+        ret = syncop_readdirp(conf->local_subvols[i], fd, 131072,
+                              dir_dfmeta->offset_var[i].offset,
+                              &(dir_dfmeta->equeue[i]), xattr_req, NULL);
+        if (ret == 0) {
+            dir_dfmeta->offset_var[i].readdir_done = 1;
+            ret = 0;
+            goto out;
+        }
+
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_WARNING, -ret,
+                   DHT_MSG_MIGRATE_DATA_FAILED,
+                   "Readdirp failed. Aborting data migration for "
+                   "directory: %s",
+                   loc->path);
+            *perrno = -ret;
+            ret = -1;
+            goto out;
+        }
+
+        if (list_empty(&(dir_dfmeta->equeue[i].list))) {
+            dir_dfmeta->offset_var[i].readdir_done = 1;
+            ret = 0;
+            goto out;
+        }
+
+        dir_dfmeta->fetch_entries[i] = 0;
+    }
+
+    while (1) {
+        if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) {
+            ret = -1;
+            goto out;
+        }
+
+        df_entry = list_entry(dir_dfmeta->iterator[i]->next, typeof(*df_entry),
+                              list);
+
+        if (&df_entry->list == dir_dfmeta->head[i]) {
+            gf_dirent_free(&(dir_dfmeta->equeue[i]));
+            INIT_LIST_HEAD(&(dir_dfmeta->equeue[i].list));
+            dir_dfmeta->fetch_entries[i] = 1;
+            dir_dfmeta->iterator[i] = dir_dfmeta->head[i];
+            ret = 0;
+            goto out;
+        }
+
+        dir_dfmeta->iterator[i] = dir_dfmeta->iterator[i]->next;
+
+        dir_dfmeta->offset_var[i].offset = df_entry->d_off;
+        if (!strcmp(df_entry->d_name, ".") || !strcmp(df_entry->d_name, ".."))
+            continue;
+
+        if (IA_ISDIR(df_entry->d_stat.ia_type)) {
+            defrag->size_processed += df_entry->d_stat.ia_size;
+            continue;
+        }
+
+        defrag->num_files_lookedup++;
+
+        if (defrag->defrag_pattern &&
+            (gf_defrag_pattern_match(defrag, df_entry->d_name,
+                                     df_entry->d_stat.ia_size) == _gf_false)) {
+            defrag->size_processed += df_entry->d_stat.ia_size;
+            continue;
+        }
+
+        is_linkfile = check_is_linkfile(NULL, &df_entry->d_stat, df_entry->dict,
+                                        conf->link_xattr_name);
+
+        if (is_linkfile) {
+            /* No need to add linkto file to the queue for
+               migration. Only the actual data file need to
+               be checked for migration criteria.
+            */
+
+            gf_msg_debug(this->name, 0,
+                         "Skipping linkfile"
+                         " %s on subvol: %s",
+                         df_entry->d_name, conf->local_subvols[i]->name);
+            continue;
+        }
+
+        /*Build Container Structure */
+
+        tmp_container = GF_CALLOC(1, sizeof(struct dht_container),
+                                  gf_dht_mt_container_t);
+        if (!tmp_container) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "Failed to allocate "
+                   "memory for container");
+            ret = -1;
+            goto out;
+        }
+        tmp_container->df_entry = gf_dirent_for_name(df_entry->d_name);
+        if (!tmp_container->df_entry) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "Failed to allocate "
+                   "memory for df_entry");
+            ret = -1;
+            goto out;
+        }
+
+        tmp_container->local_subvol_index = i;
+
+        tmp_container->df_entry->d_stat = df_entry->d_stat;
+
+        tmp_container->df_entry->d_ino = df_entry->d_ino;
+
+        tmp_container->df_entry->d_type = df_entry->d_type;
+
+        tmp_container->df_entry->d_len = df_entry->d_len;
+
+        tmp_container->parent_loc = GF_CALLOC(1, sizeof(*loc), gf_dht_mt_loc_t);
+        if (!tmp_container->parent_loc) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "Failed to allocate "
+                   "memory for loc");
+            ret = -1;
+            goto out;
+        }
+
+        ret = loc_copy(tmp_container->parent_loc, loc);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR, "loc_copy failed");
+            ret = -1;
+            goto out;
+        }
+
+        tmp_container->migrate_data = migrate_data;
+
+        tmp_container->this = this;
+
+        if (df_entry->dict)
+            tmp_container->df_entry->dict = dict_ref(df_entry->dict);
+
+        /*Build Container Structure >> END*/
+
+        ret = 0;
+        goto out;
+    }
+
+out:
+    if (ret == 0) {
+        *container = tmp_container;
+    } else {
+        if (tmp_container) {
+            gf_defrag_free_container(tmp_container);
+        }
+    }
+
+    return ret;
+}
 
 int
-gf_defrag_fix_layout (xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
-                  dict_t *fix_layout, dict_t *migrate_data)
+gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
+                      dict_t *migrate_data, int *perrno)
 {
-        int                      ret            = -1;
-        loc_t                    entry_loc      = {0,};
-        fd_t                    *fd             = NULL;
-        gf_dirent_t              entries;
-        gf_dirent_t             *tmp            = NULL;
-        gf_dirent_t             *entry          = NULL;
-        gf_boolean_t             free_entries   = _gf_false;
-        dict_t                  *dict           = NULL;
-        off_t                    offset         = 0;
-        struct iatt              iatt           = {0,};
-        int                      readdirp_errno = 0;
-
-        ret = syncop_lookup (this, loc, NULL, &iatt, NULL, NULL);
+    int ret = -1;
+    dht_conf_t *conf = NULL;
+    gf_dirent_t entries;
+    dict_t *xattr_req = NULL;
+    struct timeval dir_start = {
+        0,
+    };
+    struct timeval end = {
+        0,
+    };
+    double elapsed = {
+        0,
+    };
+    int local_subvols_cnt = 0;
+    int i = 0;
+    int j = 0;
+    struct dht_container *container = NULL;
+    int ldfq_count = 0;
+    int dfc_index = 0;
+    int throttle_up = 0;
+    struct dir_dfmeta *dir_dfmeta = NULL;
+    xlator_t *old_THIS = NULL;
+
+    gf_log(this->name, GF_LOG_INFO, "migrate data called on %s", loc->path);
+    gettimeofday(&dir_start, NULL);
+
+    conf = this->private;
+    local_subvols_cnt = conf->local_subvols_cnt;
+
+    if (!local_subvols_cnt) {
+        ret = 0;
+        goto out;
+    }
+
+    old_THIS = THIS;
+    THIS = this;
+
+    dir_dfmeta = GF_CALLOC(1, sizeof(*dir_dfmeta), gf_common_mt_pointer);
+    if (!dir_dfmeta) {
+        gf_log(this->name, GF_LOG_ERROR, "dir_dfmeta is NULL");
+        ret = -1;
+        goto out;
+    }
+
+    dir_dfmeta->lfd = GF_CALLOC(local_subvols_cnt, sizeof(fd_t *),
+                                gf_common_mt_pointer);
+    if (!dir_dfmeta->lfd) {
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_INSUFF_MEMORY,
+                "for dir_dfmeta", NULL);
+        ret = -1;
+        *perrno = ENOMEM;
+        goto out;
+    }
+
+    for (i = 0; i < local_subvols_cnt; i++) {
+        dir_dfmeta->lfd[i] = fd_create(loc->inode, defrag->pid);
+        if (!dir_dfmeta->lfd[i]) {
+            gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_FD_CREATE_FAILED,
+                    NULL);
+            *perrno = ENOMEM;
+            ret = -1;
+            goto out;
+        }
+
+        ret = syncop_opendir(conf->local_subvols[i], loc, dir_dfmeta->lfd[i],
+                             NULL, NULL);
         if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "Lookup failed on %s",
-                        loc->path);
+            fd_unref(dir_dfmeta->lfd[i]);
+            dir_dfmeta->lfd[i] = NULL;
+            gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_FAILED_TO_OPEN,
+                    "dir: %s", loc->path, "subvol: %s",
+                    conf->local_subvols[i]->name, NULL);
+
+            if (conf->decommission_in_progress) {
+                *perrno = -ret;
+                ret = -1;
                 goto out;
+            }
+        } else {
+            fd_bind(dir_dfmeta->lfd[i]);
         }
+    }
 
-        if (defrag->cmd != GF_DEFRAG_CMD_START_LAYOUT_FIX) {
-                ret = gf_defrag_migrate_data (this, defrag, loc, migrate_data);
-                if (ret)
-                        goto out;
+    dir_dfmeta->head = GF_CALLOC(local_subvols_cnt, sizeof(*(dir_dfmeta->head)),
+                                 gf_common_mt_pointer);
+    if (!dir_dfmeta->head) {
+        gf_log(this->name, GF_LOG_ERROR, "dir_dfmeta->head is NULL");
+        ret = -1;
+        goto out;
+    }
+
+    dir_dfmeta->iterator = GF_CALLOC(local_subvols_cnt,
+                                     sizeof(*(dir_dfmeta->iterator)),
+                                     gf_common_mt_pointer);
+    if (!dir_dfmeta->iterator) {
+        gf_log(this->name, GF_LOG_ERROR, "dir_dfmeta->iterator is NULL");
+        ret = -1;
+        goto out;
+    }
+
+    dir_dfmeta->equeue = GF_CALLOC(local_subvols_cnt, sizeof(entries),
+                                   gf_dht_mt_dirent_t);
+    if (!dir_dfmeta->equeue) {
+        gf_log(this->name, GF_LOG_ERROR, "dir_dfmeta->equeue is NULL");
+        ret = -1;
+        goto out;
+    }
+
+    dir_dfmeta->offset_var = GF_CALLOC(
+        local_subvols_cnt, sizeof(dht_dfoffset_ctx_t), gf_dht_mt_octx_t);
+    if (!dir_dfmeta->offset_var) {
+        gf_log(this->name, GF_LOG_ERROR, "dir_dfmeta->offset_var is NULL");
+        ret = -1;
+        goto out;
+    }
+
+    ret = gf_defrag_ctx_subvols_init(dir_dfmeta->offset_var, this);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "dht_dfoffset_ctx_t"
+               "initialization failed");
+        ret = -1;
+        goto out;
+    }
+
+    dir_dfmeta->fetch_entries = GF_CALLOC(local_subvols_cnt, sizeof(int),
+                                          gf_common_mt_int);
+    if (!dir_dfmeta->fetch_entries) {
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_INSUFF_MEMORY,
+                "for dir_dfmeta->fetch_entries", NULL);
+        ret = -1;
+        goto out;
+    }
+
+    for (i = 0; i < local_subvols_cnt; i++) {
+        INIT_LIST_HEAD(&(dir_dfmeta->equeue[i].list));
+        dir_dfmeta->head[i] = &(dir_dfmeta->equeue[i].list);
+        dir_dfmeta->iterator[i] = dir_dfmeta->head[i];
+        dir_dfmeta->fetch_entries[i] = 1;
+    }
+
+    xattr_req = dict_new();
+    if (!xattr_req) {
+        gf_log(this->name, GF_LOG_ERROR, "dict_new failed");
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_set_uint32(xattr_req, conf->link_xattr_name, 256);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "failed to set dict for "
+               "key: %s",
+               conf->link_xattr_name);
+        ret = -1;
+        goto out;
+    }
+
+    /*
+     Job: Read entries from each local subvol and store the entries
+          in equeue array of linked list. Now pick one entry from the
+          equeue array in a round robin basis and add them to defrag Queue.
+    */
+
+    while (!dht_dfreaddirp_done(dir_dfmeta->offset_var, local_subvols_cnt)) {
+        pthread_mutex_lock(&defrag->dfq_mutex);
+        {
+            /*Throttle up: If reconfigured count is higher than
+              current thread count, wake up the sleeping threads
+              TODO: Need to refactor this. Instead of making the
+              thread sleep and wake, we should terminate and spawn
+              threads on-demand*/
+
+            if (defrag->recon_thread_count > defrag->current_thread_count) {
+                throttle_up = (defrag->recon_thread_count -
+                               defrag->current_thread_count);
+                for (j = 0; j < throttle_up; j++) {
+                    pthread_cond_signal(&defrag->df_wakeup_thread);
+                }
+            }
+
+            while (defrag->q_entry_count > MAX_MIGRATE_QUEUE_COUNT) {
+                defrag->wakeup_crawler = 1;
+                pthread_cond_wait(&defrag->rebalance_crawler_alarm,
+                                  &defrag->dfq_mutex);
+            }
+
+            ldfq_count = defrag->q_entry_count;
+
+            if (defrag->wakeup_crawler) {
+                defrag->wakeup_crawler = 0;
+            }
         }
+        pthread_mutex_unlock(&defrag->dfq_mutex);
 
-        gf_log (this->name, GF_LOG_TRACE, "fix layout called on %s", loc->path);
+        while (
+            ldfq_count <= MAX_MIGRATE_QUEUE_COUNT &&
+            !dht_dfreaddirp_done(dir_dfmeta->offset_var, local_subvols_cnt)) {
+            ret = gf_defrag_get_entry(this, dfc_index, &container, loc, conf,
+                                      defrag, dir_dfmeta->lfd[dfc_index],
+                                      migrate_data, dir_dfmeta, xattr_req,
+                                      perrno);
+
+            if (defrag->defrag_status == GF_DEFRAG_STATUS_STOPPED) {
+                goto out;
+            }
+
+            if (ret) {
+                gf_log(this->name, GF_LOG_WARNING,
+                       "Found "
+                       "error from gf_defrag_get_entry");
 
-        fd = fd_create (loc->inode, defrag->pid);
-        if (!fd) {
-                gf_log (this->name, GF_LOG_ERROR, "Failed to create fd");
                 ret = -1;
                 goto out;
+            }
+
+            /* Check if we got an entry, else we need to move the
+               index to the next subvol */
+            if (!container) {
+                GF_CRAWL_INDEX_MOVE(dfc_index, local_subvols_cnt);
+                continue;
+            }
+
+            /* Q this entry in the dfq */
+            pthread_mutex_lock(&defrag->dfq_mutex);
+            {
+                list_add_tail(&container->list, &(defrag->queue[0].list));
+                defrag->q_entry_count++;
+                ldfq_count = defrag->q_entry_count;
+
+                gf_msg_debug(this->name, 0,
+                             "added "
+                             "file:%s parent:%s to the queue ",
+                             container->df_entry->d_name,
+                             container->parent_loc->path);
+
+                pthread_cond_signal(&defrag->parallel_migration_cond);
+            }
+            pthread_mutex_unlock(&defrag->dfq_mutex);
+
+            GF_CRAWL_INDEX_MOVE(dfc_index, local_subvols_cnt);
         }
+    }
+
+    gettimeofday(&end, NULL);
+    elapsed = gf_tvdiff(&dir_start, &end);
+    gf_log(this->name, GF_LOG_INFO,
+           "Migration operation on dir %s took "
+           "%.2f secs",
+           loc->path, elapsed / 1e6);
+    ret = 0;
+out:
+    THIS = old_THIS;
+    gf_defrag_free_dir_dfmeta(dir_dfmeta, local_subvols_cnt);
+
+    if (xattr_req)
+        dict_unref(xattr_req);
+
+    /* It does not matter if it errored out - this number is
+     * used to calculate rebalance estimated time to complete.
+     * No locking required as dirs are processed by a single thread.
+     */
+    defrag->num_dirs_processed++;
+    return ret;
+}
 
-        ret = syncop_opendir (this, loc, fd);
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "Failed to open dir %s",
-                        loc->path);
-                ret = -1;
+int
+gf_defrag_settle_hash(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
+                      dict_t *fix_layout)
+{
+    int ret;
+    dht_conf_t *conf = NULL;
+    /*
+     * Now we're ready to update the directory commit hash for the volume
+     * root, so that hash miscompares and broadcast lookups can stop.
+     * However, we want to skip that if fix-layout is all we did.  In
+     * that case, we want the miscompares etc. to continue until a real
+     * rebalance is complete.
+     */
+    if (defrag->cmd == GF_DEFRAG_CMD_START_LAYOUT_FIX ||
+        defrag->cmd == GF_DEFRAG_CMD_DETACH_START) {
+        return 0;
+    }
+
+    conf = this->private;
+    if (!conf) {
+        /*Uh oh
+         */
+        return -1;
+    }
+
+    if (conf->local_subvols_cnt == 0 || !conf->lookup_optimize) {
+        /* Commit hash updates are only done on local subvolumes and
+         * only when lookup optimization is needed (for older client
+         * support)
+         */
+        return 0;
+    }
+
+    ret = dict_set_uint32(fix_layout, "new-commit-hash",
+                          defrag->new_commit_hash);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR, "Failed to set new-commit-hash");
+        return -1;
+    }
+
+    ret = syncop_setxattr(this, loc, fix_layout, 0, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_LAYOUT_FIX_FAILED,
+               "fix layout on %s failed", loc->path);
+
+        if (-ret == ENOENT || -ret == ESTALE) {
+            /* Dir most likely is deleted */
+            return 0;
+        }
+
+        return -1;
+    }
+
+    /* TBD: find more efficient solution than adding/deleting every time */
+    dict_del(fix_layout, "new-commit-hash");
+
+    return 0;
+}
+
+int
+gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
+                     dict_t *fix_layout, dict_t *migrate_data)
+{
+    int ret = -1;
+    loc_t entry_loc = {
+        0,
+    };
+    fd_t *fd = NULL;
+    gf_dirent_t entries;
+    gf_dirent_t *tmp = NULL;
+    gf_dirent_t *entry = NULL;
+    gf_boolean_t free_entries = _gf_false;
+    off_t offset = 0;
+    struct iatt iatt = {
+        0,
+    };
+    inode_t *linked_inode = NULL, *inode = NULL;
+    dht_conf_t *conf = NULL;
+    int perrno = 0;
+
+    conf = this->private;
+    if (!conf) {
+        ret = -1;
+        goto out;
+    }
+
+    ret = syncop_lookup(this, loc, &iatt, NULL, NULL, NULL);
+    if (ret) {
+        if (strcmp(loc->path, "/") == 0) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_DIR_LOOKUP_FAILED,
+                   "lookup failed for:%s", loc->path);
+
+            defrag->total_failures++;
+            ret = -1;
+            goto out;
+        }
+
+        if (-ret == ENOENT || -ret == ESTALE) {
+            gf_msg(this->name, GF_LOG_INFO, -ret, DHT_MSG_DIR_LOOKUP_FAILED,
+                   "Dir:%s renamed or removed. Skipping", loc->path);
+            if (conf->decommission_subvols_cnt) {
+                defrag->total_failures++;
+            }
+            ret = 0;
+            goto out;
+        } else {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_DIR_LOOKUP_FAILED,
+                   "lookup failed for:%s", loc->path);
+
+            defrag->total_failures++;
+            goto out;
+        }
+    }
+
+    fd = fd_create(loc->inode, defrag->pid);
+    if (!fd) {
+        gf_log(this->name, GF_LOG_ERROR, "Failed to create fd");
+        ret = -1;
+        goto out;
+    }
+
+    ret = syncop_opendir(this, loc, fd, NULL, NULL);
+    if (ret) {
+        if (-ret == ENOENT || -ret == ESTALE) {
+            if (conf->decommission_subvols_cnt) {
+                defrag->total_failures++;
+            }
+            ret = 0;
+            goto out;
+        }
+
+        gf_log(this->name, GF_LOG_ERROR,
+               "Failed to open dir %s, "
+               "err:%d",
+               loc->path, -ret);
+
+        ret = -1;
+        goto out;
+    }
+
+    fd_bind(fd);
+    INIT_LIST_HEAD(&entries.list);
+
+    while ((ret = syncop_readdirp(this, fd, 131072, offset, &entries, NULL,
+                                  NULL)) != 0) {
+        if (ret < 0) {
+            if (-ret == ENOENT || -ret == ESTALE) {
+                if (conf->decommission_subvols_cnt) {
+                    defrag->total_failures++;
+                }
+                ret = 0;
                 goto out;
+            }
+
+            gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_READDIR_ERROR,
+                   "readdirp failed for "
+                   "path %s. Aborting fix-layout",
+                   loc->path);
+
+            ret = -1;
+            goto out;
         }
 
-        INIT_LIST_HEAD (&entries.list);
-        while ((ret = syncop_readdirp (this, fd, 131072, offset, NULL,
-                &entries)) != 0)
+        if (list_empty(&entries.list))
+            break;
+
+        free_entries = _gf_true;
+
+        list_for_each_entry_safe(entry, tmp, &entries.list, list)
         {
+            if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) {
+                ret = 1;
+                goto out;
+            }
 
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_ERROR, "Readdir returned %s"
-                                ". Aborting fix-layout",strerror(errno));
-                        goto out;
+            offset = entry->d_off;
+
+            if (!strcmp(entry->d_name, ".") || !strcmp(entry->d_name, ".."))
+                continue;
+            if (!IA_ISDIR(entry->d_stat.ia_type)) {
+                continue;
+            }
+            loc_wipe(&entry_loc);
+
+            ret = dht_build_child_loc(this, &entry_loc, loc, entry->d_name);
+            if (ret) {
+                gf_log(this->name, GF_LOG_ERROR,
+                       "Child loc"
+                       " build failed for entry: %s",
+                       entry->d_name);
+
+                if (conf->decommission_in_progress) {
+                    defrag->defrag_status = GF_DEFRAG_STATUS_FAILED;
+
+                    goto out;
+                } else {
+                    continue;
                 }
+            }
+
+            if (gf_uuid_is_null(entry->d_stat.ia_gfid)) {
+                gf_log(this->name, GF_LOG_ERROR,
+                       "%s/%s"
+                       " gfid not present",
+                       loc->path, entry->d_name);
+                continue;
+            }
+
+            gf_uuid_copy(entry_loc.gfid, entry->d_stat.ia_gfid);
+
+            /*In case the gfid stored in the inode by inode_link
+             * and the gfid obtained in the lookup differs, then
+             * client3_3_lookup_cbk will return ESTALE and proper
+             * error will be captured
+             */
+
+            linked_inode = inode_link(entry_loc.inode, loc->inode,
+                                      entry->d_name, &entry->d_stat);
+
+            inode = entry_loc.inode;
+            entry_loc.inode = linked_inode;
+            inode_unref(inode);
+
+            if (gf_uuid_is_null(loc->gfid)) {
+                gf_log(this->name, GF_LOG_ERROR,
+                       "%s/%s"
+                       " gfid not present",
+                       loc->path, entry->d_name);
+                continue;
+            }
+
+            gf_uuid_copy(entry_loc.pargfid, loc->gfid);
+
+            ret = syncop_lookup(this, &entry_loc, &iatt, NULL, NULL, NULL);
+            if (ret) {
+                if (-ret == ENOENT || -ret == ESTALE) {
+                    gf_msg(this->name, GF_LOG_INFO, -ret,
+                           DHT_MSG_DIR_LOOKUP_FAILED,
+                           "Dir:%s renamed or removed. "
+                           "Skipping",
+                           loc->path);
+                    ret = 0;
+                    if (conf->decommission_subvols_cnt) {
+                        defrag->total_failures++;
+                    }
+                    continue;
+                } else {
+                    gf_msg(this->name, GF_LOG_ERROR, -ret,
+                           DHT_MSG_DIR_LOOKUP_FAILED, "lookup failed for:%s",
+                           entry_loc.path);
 
-                /* Need to keep track of ENOENT errno, that means, there is no
-                   need to send more readdirp() */
-                readdirp_errno = errno;
+                    defrag->total_failures++;
 
-                if (list_empty (&entries.list))
-                        break;
+                    if (conf->decommission_in_progress) {
+                        defrag->defrag_status = GF_DEFRAG_STATUS_FAILED;
+                        ret = -1;
+                        goto out;
+                    } else {
+                        continue;
+                    }
+                }
+            }
 
-                free_entries = _gf_true;
+            /* A return value of 2 means, either process_dir or
+             * lookup of a dir failed. Hence, don't commit hash
+             * for the current directory*/
 
-                list_for_each_entry_safe (entry, tmp, &entries.list, list) {
-                        if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) {
-                                ret = 1;
-                                goto out;
-                        }
+            ret = gf_defrag_fix_layout(this, defrag, &entry_loc, fix_layout,
+                                       migrate_data);
 
-                        offset = entry->d_off;
+            if (defrag->defrag_status == GF_DEFRAG_STATUS_STOPPED ||
+                defrag->defrag_status == GF_DEFRAG_STATUS_FAILED) {
+                goto out;
+            }
 
-                        if (!strcmp (entry->d_name, ".") ||
-                            !strcmp (entry->d_name, ".."))
-                                continue;
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LAYOUT_FIX_FAILED,
+                       "Fix layout failed for %s", entry_loc.path);
 
-                        if (!IA_ISDIR (entry->d_stat.ia_type))
-                                continue;
+                defrag->total_failures++;
 
-                        loc_wipe (&entry_loc);
-                        ret =dht_build_child_loc (this, &entry_loc, loc,
-                                                  entry->d_name);
-                        if (ret) {
-                                gf_log (this->name, GF_LOG_ERROR, "Child loc"
-                                        " build failed");
-                                goto out;
-                        }
+                if (conf->decommission_in_progress) {
+                    defrag->defrag_status = GF_DEFRAG_STATUS_FAILED;
 
-                        if (uuid_is_null (entry->d_stat.ia_gfid)) {
-                                gf_log (this->name, GF_LOG_ERROR, "%s/%s"
-                                        " gfid not present", loc->path,
-                                         entry->d_name);
-                                continue;
-                        }
+                    goto out;
+                } else {
+                    /* Let's not commit-hash if
+                     * gf_defrag_fix_layout failed*/
+                    continue;
+                }
+            }
+        }
 
-                        entry_loc.inode->ia_type = entry->d_stat.ia_type;
+        gf_dirent_free(&entries);
+        free_entries = _gf_false;
+        INIT_LIST_HEAD(&entries.list);
+    }
+
+    /* A directory layout is fixed only after its subdirs are healed to
+     * any newly added bricks. If the layout is fixed before subdirs are
+     * healed, the newly added brick will get a non-null layout.
+     * Any subdirs which hash to that layout will no longer show up
+     * in a directory listing until they are healed.
+     */
+
+    ret = syncop_setxattr(this, loc, fix_layout, 0, NULL, NULL);
+
+    /* In case of a race where the directory is deleted just before
+     * layout setxattr, the errors are updated in the layout structure.
+     * We can use this information to make a decision whether the directory
+     * is deleted entirely.
+     */
+    if (ret == 0) {
+        ret = dht_dir_layout_error_check(this, loc->inode);
+        ret = -ret;
+    }
+
+    if (ret) {
+        if (-ret == ENOENT || -ret == ESTALE) {
+            gf_msg(this->name, GF_LOG_INFO, -ret, DHT_MSG_LAYOUT_FIX_FAILED,
+                   "Setxattr failed. Dir %s "
+                   "renamed or removed",
+                   loc->path);
+            if (conf->decommission_subvols_cnt) {
+                defrag->total_failures++;
+            }
+            ret = 0;
+            goto out;
+        } else {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_LAYOUT_FIX_FAILED,
+                   "Setxattr failed for %s", loc->path);
 
-                        uuid_copy (entry_loc.gfid, entry->d_stat.ia_gfid);
-                        if (uuid_is_null (loc->gfid)) {
-                                gf_log (this->name, GF_LOG_ERROR, "%s/%s"
-                                        " gfid not present", loc->path,
-                                         entry->d_name);
-                                continue;
-                        }
+            defrag->total_failures++;
 
-                        uuid_copy (entry_loc.pargfid, loc->gfid);
+            if (conf->decommission_in_progress) {
+                defrag->defrag_status = GF_DEFRAG_STATUS_FAILED;
+                ret = -1;
+                goto out;
+            }
+        }
+    }
 
-                        ret = syncop_lookup (this, &entry_loc, NULL, &iatt,
-                                             NULL, NULL);
-                        if (ret) {
-                                gf_log (this->name, GF_LOG_ERROR, "%s"
-                                        " lookup failed", entry_loc.path);
-                                continue;
-                        }
+    if (defrag->cmd != GF_DEFRAG_CMD_START_LAYOUT_FIX) {
+        ret = gf_defrag_process_dir(this, defrag, loc, migrate_data, &perrno);
 
-                        ret = syncop_setxattr (this, &entry_loc, fix_layout,
-                                               0);
-                        if (ret) {
-                                gf_log (this->name, GF_LOG_ERROR, "Setxattr "
-                                        "failed for %s", entry_loc.path);
-                                defrag->defrag_status =
-                                GF_DEFRAG_STATUS_FAILED;
-                                defrag->total_failures ++;
-                                goto out;
-                        }
-                        ret = gf_defrag_fix_layout (this, defrag, &entry_loc,
-                                                    fix_layout, migrate_data);
-
-                        if (ret) {
-                                gf_log (this->name, GF_LOG_ERROR, "Fix layout "
-                                        "failed for %s", entry_loc.path);
-                                defrag->total_failures++;
-                                goto out;
-                        }
+        if (ret) {
+            if (perrno == ENOENT || perrno == ESTALE) {
+                ret = 0;
+                goto out;
+            } else {
+                defrag->total_failures++;
+
+                gf_msg(this->name, GF_LOG_ERROR, 0,
+                       DHT_MSG_DEFRAG_PROCESS_DIR_FAILED,
+                       "gf_defrag_process_dir failed for "
+                       "directory: %s",
+                       loc->path);
 
+                if (conf->decommission_in_progress) {
+                    goto out;
                 }
-                gf_dirent_free (&entries);
-                free_entries = _gf_false;
-                INIT_LIST_HEAD (&entries.list);
-                if (readdirp_errno == ENOENT)
-                        break;
+            }
         }
+    }
 
-        ret = 0;
+    gf_msg_trace(this->name, 0, "fix layout called on %s", loc->path);
+
+    if (gf_defrag_settle_hash(this, defrag, loc, fix_layout) != 0) {
+        defrag->total_failures++;
+
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SETTLE_HASH_FAILED,
+               "Settle hash failed for %s", loc->path);
+
+        ret = -1;
+
+        if (conf->decommission_in_progress) {
+            defrag->defrag_status = GF_DEFRAG_STATUS_FAILED;
+            goto out;
+        }
+    }
+
+    ret = 0;
 out:
-        if (free_entries)
-                gf_dirent_free (&entries);
+    if (free_entries)
+        gf_dirent_free(&entries);
 
-        loc_wipe (&entry_loc);
+    loc_wipe(&entry_loc);
 
-        if (dict)
-                dict_unref(dict);
+    if (fd)
+        fd_unref(fd);
 
-        if (fd)
-                fd_unref (fd);
+    return ret;
+}
 
+int
+dht_init_local_subvols_and_nodeuuids(xlator_t *this, dht_conf_t *conf,
+                                     loc_t *loc)
+{
+    dict_t *dict = NULL;
+    uuid_t *uuid_ptr = NULL;
+    int ret = -1;
+    int i = 0;
+    int j = 0;
+
+    /* Find local subvolumes */
+    ret = syncop_getxattr(this, loc, &dict, GF_REBAL_FIND_LOCAL_SUBVOL, NULL,
+                          NULL);
+    if (ret && (ret != -ENODATA)) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, 0,
+               "local "
+               "subvolume determination failed with error: %d",
+               -ret);
+        ret = -1;
+        goto out;
+    }
+
+    if (!ret)
+        goto out;
+
+    ret = syncop_getxattr(this, loc, &dict, GF_REBAL_OLD_FIND_LOCAL_SUBVOL,
+                          NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, 0,
+               "local "
+               "subvolume determination failed with error: %d",
+               -ret);
+        ret = -1;
+        goto out;
+    }
+    ret = 0;
+
+out:
+    if (ret) {
         return ret;
+    }
+
+    for (i = 0; i < conf->local_subvols_cnt; i++) {
+        gf_msg(this->name, GF_LOG_INFO, 0, 0,
+               "local subvol: "
+               "%s",
+               conf->local_subvols[i]->name);
+
+        for (j = 0; j < conf->local_nodeuuids[i].count; j++) {
+            uuid_ptr = &(conf->local_nodeuuids[i].elements[j].uuid);
+            gf_msg(this->name, GF_LOG_INFO, 0, 0, "node uuid : %s",
+                   uuid_utoa(*uuid_ptr));
+        }
+    }
 
+    return ret;
 }
 
+/* Functions for the rebalance estimates feature */
 
-int
-gf_defrag_start_crawl (void *data)
+uint64_t
+gf_defrag_subvol_file_size(xlator_t *this, loc_t *root_loc)
 {
-        xlator_t                *this   = NULL;
-        dht_conf_t              *conf   = NULL;
-        gf_defrag_info_t        *defrag = NULL;
-        int                      ret    = -1;
-        loc_t                    loc    = {0,};
-        struct iatt              iatt   = {0,};
-        struct iatt              parent = {0,};
-        dict_t                  *fix_layout = NULL;
-        dict_t                  *migrate_data = NULL;
-        dict_t                  *status = NULL;
-        glusterfs_ctx_t         *ctx = NULL;
-
-        this = data;
-        if (!this)
-                goto out;
+    int ret = -1;
+    struct statvfs buf = {
+        0,
+    };
+
+    ret = syncop_statfs(this, root_loc, &buf, NULL, NULL);
+    if (ret) {
+        /* Aargh! */
+        return 0;
+    }
+    return ((buf.f_blocks - buf.f_bfree) * buf.f_frsize);
+}
 
-        ctx = this->ctx;
-        if (!ctx)
-                goto out;
+uint64_t
+gf_defrag_total_file_size(xlator_t *this, loc_t *root_loc)
+{
+    dht_conf_t *conf = NULL;
+    int i = 0;
+    uint64_t size_files = 0;
+    uint64_t total_size = 0;
 
-        conf = this->private;
-        if (!conf)
-                goto out;
+    conf = this->private;
+    if (!conf) {
+        return 0;
+    }
+
+    for (i = 0; i < conf->local_subvols_cnt; i++) {
+        size_files = gf_defrag_subvol_file_size(conf->local_subvols[i],
+                                                root_loc);
+        total_size += size_files;
+        gf_msg(this->name, GF_LOG_INFO, 0, 0,
+               "local subvol: %s,"
+               "cnt = %" PRIu64,
+               conf->local_subvols[i]->name, size_files);
+    }
+
+    gf_msg(this->name, GF_LOG_INFO, 0, 0, "Total size files = %" PRIu64,
+           total_size);
+
+    return total_size;
+}
 
-        defrag = conf->defrag;
-        if (!defrag)
-                goto out;
+static void *
+dht_file_counter_thread(void *args)
+{
+    gf_defrag_info_t *defrag = NULL;
+    loc_t root_loc = {
+        0,
+    };
+    struct timespec time_to_wait = {
+        0,
+    };
+    uint64_t tmp_size = 0;
+
+    if (!args)
+        return NULL;
 
-        gettimeofday (&defrag->start_time, NULL);
-        dht_build_root_inode (this, &defrag->root_inode);
-        if (!defrag->root_inode)
-                goto out;
+    defrag = (gf_defrag_info_t *)args;
+    dht_build_root_loc(defrag->root_inode, &root_loc);
 
-        dht_build_root_loc (defrag->root_inode, &loc);
+    while (defrag->defrag_status == GF_DEFRAG_STATUS_STARTED) {
+        timespec_now(&time_to_wait);
+        time_to_wait.tv_sec += 600;
 
-        /* fix-layout on '/' first */
+        pthread_mutex_lock(&defrag->fc_mutex);
+        pthread_cond_timedwait(&defrag->fc_wakeup_cond, &defrag->fc_mutex,
+                               &time_to_wait);
 
-        ret = syncop_lookup (this, &loc, NULL, &iatt, NULL, &parent);
+        pthread_mutex_unlock(&defrag->fc_mutex);
 
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "look up on / failed");
-                goto out;
+        if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED)
+            break;
+
+        tmp_size = gf_defrag_total_file_size(defrag->this, &root_loc);
+
+        gf_log("dht", GF_LOG_INFO, "tmp data size =%" PRIu64, tmp_size);
+
+        if (!tmp_size) {
+            gf_msg("dht", GF_LOG_ERROR, 0, 0,
+                   "Failed to get "
+                   "the total data size. Unable to estimate "
+                   "time to complete rebalance.");
+        } else {
+            g_totalsize = tmp_size;
+            gf_msg_debug("dht", 0, "total data size =%" PRIu64, g_totalsize);
         }
+    }
 
-        fix_layout = dict_new ();
-        if (!fix_layout) {
-                ret = -1;
-                goto out;
+    return NULL;
+}
+
+int
+gf_defrag_estimates_cleanup(xlator_t *this, gf_defrag_info_t *defrag,
+                            pthread_t filecnt_thread)
+{
+    int ret = -1;
+
+    /* Wake up the filecounter thread.
+     * By now the defrag status will no longer be
+     * GF_DEFRAG_STATUS_STARTED so the thread will exit the loop.
+     */
+    pthread_mutex_lock(&defrag->fc_mutex);
+    {
+        pthread_cond_broadcast(&defrag->fc_wakeup_cond);
+    }
+    pthread_mutex_unlock(&defrag->fc_mutex);
+
+    ret = pthread_join(filecnt_thread, NULL);
+    if (ret) {
+        gf_msg("dht", GF_LOG_ERROR, ret, 0,
+               "file_counter_thread: pthread_join failed.");
+        ret = -1;
+    }
+    return ret;
+}
+
+int
+gf_defrag_estimates_init(xlator_t *this, loc_t *loc, pthread_t *filecnt_thread)
+{
+    int ret = -1;
+    dht_conf_t *conf = NULL;
+    gf_defrag_info_t *defrag = NULL;
+
+    conf = this->private;
+    defrag = conf->defrag;
+
+    g_totalsize = gf_defrag_total_file_size(this, loc);
+    if (!g_totalsize) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+               "Failed to get "
+               "the total data size. Unable to estimate "
+               "time to complete rebalance.");
+        goto out;
+    }
+
+    ret = gf_thread_create(filecnt_thread, NULL, dht_file_counter_thread,
+                           (void *)defrag, "dhtfcnt");
+
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, ret, 0,
+               "Failed to "
+               "create the file counter thread ");
+        ret = -1;
+        goto out;
+    }
+    ret = 0;
+out:
+    return ret;
+}
+
+/* Init and cleanup functions for parallel file migration*/
+int
+gf_defrag_parallel_migration_init(xlator_t *this, gf_defrag_info_t *defrag,
+                                  pthread_t **tid_array, int *thread_index)
+{
+    int ret = -1;
+    int thread_spawn_count = 0;
+    int index = 0;
+    pthread_t *tid = NULL;
+
+    if (!defrag)
+        goto out;
+
+    /* Initialize global entry queue */
+    defrag->queue = GF_CALLOC(1, sizeof(struct dht_container),
+                              gf_dht_mt_container_t);
+
+    if (!defrag->queue) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, 0,
+               "Failed to initialise migration queue");
+        ret = -1;
+        goto out;
+    }
+
+    INIT_LIST_HEAD(&(defrag->queue[0].list));
+
+    thread_spawn_count = MAX(MAX_REBAL_THREADS, 4);
+
+    gf_msg_debug(this->name, 0, "thread_spawn_count: %d", thread_spawn_count);
+
+    tid = GF_CALLOC(thread_spawn_count, sizeof(pthread_t),
+                    gf_common_mt_pthread_t);
+    if (!tid) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, 0,
+               "Failed to create migration threads");
+        ret = -1;
+        goto out;
+    }
+    defrag->current_thread_count = thread_spawn_count;
+
+    /*Spawn Threads Here*/
+    while (index < thread_spawn_count) {
+        ret = gf_thread_create(&(tid[index]), NULL, gf_defrag_task,
+                               (void *)defrag, "dhtmig%d", (index + 1) & 0x3ff);
+        if (ret != 0) {
+            gf_msg("DHT", GF_LOG_ERROR, ret, 0, "Thread[%d] creation failed. ",
+                   index);
+            ret = -1;
+            goto out;
+        } else {
+            gf_log("DHT", GF_LOG_INFO,
+                   "Thread[%d] "
+                   "creation successful",
+                   index);
         }
+        index++;
+    }
+
+    ret = 0;
+out:
+    *thread_index = index;
+    *tid_array = tid;
+
+    return ret;
+}
+
+int
+gf_defrag_parallel_migration_cleanup(gf_defrag_info_t *defrag,
+                                     pthread_t *tid_array, int thread_index)
+{
+    int ret = -1;
+    int i = 0;
 
-        ret = dict_set_str (fix_layout, GF_XATTR_FIX_LAYOUT_KEY, "yes");
+    if (!defrag)
+        goto out;
+
+    /* Wake up all migration threads */
+    pthread_mutex_lock(&defrag->dfq_mutex);
+    {
+        defrag->crawl_done = 1;
+
+        pthread_cond_broadcast(&defrag->parallel_migration_cond);
+        pthread_cond_broadcast(&defrag->df_wakeup_thread);
+    }
+    pthread_mutex_unlock(&defrag->dfq_mutex);
+
+    /*Wait for all the threads to complete their task*/
+    for (i = 0; i < thread_index; i++) {
+        pthread_join(tid_array[i], NULL);
+    }
+
+    GF_FREE(tid_array);
+
+    /* Cleanup the migration queue */
+    if (defrag->queue) {
+        gf_dirent_free(defrag->queue[0].df_entry);
+        INIT_LIST_HEAD(&(defrag->queue[0].list));
+    }
+
+    GF_FREE(defrag->queue);
+
+    ret = 0;
+out:
+    return ret;
+}
+
+int
+gf_defrag_start_crawl(void *data)
+{
+    xlator_t *this = NULL;
+    dht_conf_t *conf = NULL;
+    gf_defrag_info_t *defrag = NULL;
+    dict_t *fix_layout = NULL;
+    dict_t *migrate_data = NULL;
+    dict_t *status = NULL;
+    glusterfs_ctx_t *ctx = NULL;
+    call_frame_t *statfs_frame = NULL;
+    xlator_t *old_THIS = NULL;
+    int ret = -1;
+    loc_t loc = {
+        0,
+    };
+    struct iatt iatt = {
+        0,
+    };
+    struct iatt parent = {
+        0,
+    };
+    int thread_index = 0;
+    pthread_t *tid = NULL;
+    pthread_t filecnt_thread;
+    gf_boolean_t fc_thread_started = _gf_false;
+
+    this = data;
+    if (!this)
+        goto exit;
+
+    ctx = this->ctx;
+    if (!ctx)
+        goto exit;
+
+    conf = this->private;
+    if (!conf)
+        goto exit;
+
+    defrag = conf->defrag;
+    if (!defrag)
+        goto exit;
+
+    defrag->start_time = gf_time();
+
+    dht_build_root_inode(this, &defrag->root_inode);
+    if (!defrag->root_inode)
+        goto out;
+
+    dht_build_root_loc(defrag->root_inode, &loc);
+
+    /* fix-layout on '/' first */
+
+    ret = syncop_lookup(this, &loc, &iatt, &parent, NULL, NULL);
+
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_REBALANCE_START_FAILED,
+               "Failed to start rebalance: look up on / failed");
+        ret = -1;
+        goto out;
+    }
+
+    old_THIS = THIS;
+    THIS = this;
+
+    statfs_frame = create_frame(this, this->ctx->pool);
+    if (!statfs_frame) {
+        gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_NO_MEMORY, ENOMEM,
+               "Insufficient memory. Frame creation failed");
+        ret = -1;
+        goto out;
+    }
+
+    /* async statfs update for honoring min-free-disk */
+    dht_get_du_info(statfs_frame, this, &loc);
+    THIS = old_THIS;
+
+    fix_layout = dict_new();
+    if (!fix_layout) {
+        ret = -1;
+        goto out;
+    }
+
+    /*
+     * Unfortunately, we can't do special xattrs (like fix.layout) and
+     * real ones in the same call currently, and changing it seems
+     * riskier than just doing two calls.
+     */
+
+    gf_log(this->name, GF_LOG_INFO, "%s using commit hash %u", __func__,
+           conf->vol_commit_hash);
+
+    ret = dict_set_uint32(fix_layout, conf->commithash_xattr_name,
+                          conf->vol_commit_hash);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR, "Failed to set %s",
+               conf->commithash_xattr_name);
+        defrag->total_failures++;
+        ret = -1;
+        goto out;
+    }
+
+    ret = syncop_setxattr(this, &loc, fix_layout, 0, NULL, NULL);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Failed to set commit hash on %s. "
+               "Rebalance cannot proceed.",
+               loc.path);
+        defrag->total_failures++;
+        ret = -1;
+        goto out;
+    }
+
+    /* We now return to our regularly scheduled program. */
+
+    ret = dict_set_str(fix_layout, GF_XATTR_FIX_LAYOUT_KEY, "yes");
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_REBALANCE_START_FAILED,
+               "Failed to start rebalance:"
+               "Failed to set dictionary value: key = %s",
+               GF_XATTR_FIX_LAYOUT_KEY);
+        defrag->total_failures++;
+        ret = -1;
+        goto out;
+    }
+
+    defrag->new_commit_hash = conf->vol_commit_hash;
+
+    ret = syncop_setxattr(this, &loc, fix_layout, 0, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_REBALANCE_FAILED,
+               "fix layout on %s failed", loc.path);
+        defrag->total_failures++;
+        ret = -1;
+        goto out;
+    }
+
+    if (defrag->cmd != GF_DEFRAG_CMD_START_LAYOUT_FIX) {
+        /* We need to migrate files */
+
+        migrate_data = dict_new();
+        if (!migrate_data) {
+            defrag->total_failures++;
+            ret = -1;
+            goto out;
+        }
+        ret = dict_set_str(
+            migrate_data, GF_XATTR_FILE_MIGRATE_KEY,
+            (defrag->cmd == GF_DEFRAG_CMD_START_FORCE) ? "force" : "non-force");
         if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "Failed to set dict str");
-                goto out;
+            defrag->total_failures++;
+            ret = -1;
+            goto out;
         }
 
-        ret = syncop_setxattr (this, &loc, fix_layout, 0);
+        ret = dht_init_local_subvols_and_nodeuuids(this, conf, &loc);
         if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "fix layout on %s failed",
-                        loc.path);
-                defrag->total_failures++;
-                goto out;
+            ret = -1;
+            goto out;
         }
 
-        if (defrag->cmd != GF_DEFRAG_CMD_START_LAYOUT_FIX) {
-                migrate_data = dict_new ();
-                if (!migrate_data) {
-                        ret = -1;
-                        goto out;
-                }
-                if (defrag->cmd == GF_DEFRAG_CMD_START_FORCE)
-                        ret = dict_set_str (migrate_data,
-                                            "distribute.migrate-data", "force");
-                else
-                        ret = dict_set_str (migrate_data,
-                                            "distribute.migrate-data",
-                                            "non-force");
-                if (ret)
-                        goto out;
+        /* Initialise the structures required for parallel migration */
+        ret = gf_defrag_parallel_migration_init(this, defrag, &tid,
+                                                &thread_index);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, 0, "Aborting rebalance.");
+            goto out;
         }
-        ret = gf_defrag_fix_layout (this, defrag, &loc, fix_layout,
-                                    migrate_data);
-        if ((defrag->defrag_status != GF_DEFRAG_STATUS_STOPPED) &&
-            (defrag->defrag_status != GF_DEFRAG_STATUS_FAILED)) {
-                defrag->defrag_status = GF_DEFRAG_STATUS_COMPLETE;
+
+        ret = gf_defrag_estimates_init(this, &loc, &filecnt_thread);
+        if (ret) {
+            /* Not a fatal error. Allow the rebalance to proceed*/
+            ret = 0;
+        } else {
+            fc_thread_started = _gf_true;
         }
+    }
 
+    ret = gf_defrag_fix_layout(this, defrag, &loc, fix_layout, migrate_data);
+    if (ret) {
+        defrag->total_failures++;
+        ret = -1;
+        goto out;
+    }
 
+    if (gf_defrag_settle_hash(this, defrag, &loc, fix_layout) != 0) {
+        defrag->total_failures++;
+        ret = -1;
+        goto out;
+    }
 
+    gf_log("DHT", GF_LOG_INFO, "crawling file-system completed");
 out:
-        LOCK (&defrag->lock);
-        {
-                status = dict_new ();
-                gf_defrag_status_get (defrag, status);
-                if (ctx->notify)
-                        ctx->notify (GF_EN_DEFRAG_STATUS, status);
-                if (status)
-                        dict_unref (status);
-                defrag->is_exiting = 1;
-        }
-        UNLOCK (&defrag->lock);
-
-        if (defrag) {
-                GF_FREE (defrag);
-                conf->defrag = NULL;
-        }
 
-        return ret;
+    /* We are here means crawling the entire file system is done
+       or something failed. Set defrag->crawl_done flag to intimate
+       the migrator threads to exhaust the defrag->queue and terminate*/
+
+    if (ret) {
+        defrag->defrag_status = GF_DEFRAG_STATUS_FAILED;
+    }
+
+    gf_defrag_parallel_migration_cleanup(defrag, tid, thread_index);
+
+    if ((defrag->defrag_status != GF_DEFRAG_STATUS_STOPPED) &&
+        (defrag->defrag_status != GF_DEFRAG_STATUS_FAILED)) {
+        defrag->defrag_status = GF_DEFRAG_STATUS_COMPLETE;
+    }
+
+    if (fc_thread_started) {
+        gf_defrag_estimates_cleanup(this, defrag, filecnt_thread);
+    }
+
+    dht_send_rebalance_event(this, defrag->cmd, defrag->defrag_status);
+
+    status = dict_new();
+    LOCK(&defrag->lock);
+    {
+        gf_defrag_status_get(conf, status);
+        if (ctx && ctx->notify)
+            ctx->notify(GF_EN_DEFRAG_STATUS, status);
+        if (status)
+            dict_unref(status);
+        defrag->is_exiting = 1;
+    }
+    UNLOCK(&defrag->lock);
+
+    GF_FREE(defrag);
+    conf->defrag = NULL;
+
+    if (migrate_data)
+        dict_unref(migrate_data);
+
+    if (statfs_frame) {
+        STACK_DESTROY(statfs_frame->root);
+    }
+exit:
+    return ret;
 }
 
-
 static int
-gf_defrag_done  (int ret, call_frame_t *sync_frame, void *data)
+gf_defrag_done(int ret, call_frame_t *sync_frame, void *data)
 {
-        gf_listener_stop (sync_frame->this);
+    gf_listener_stop(sync_frame->this);
 
-        STACK_DESTROY (sync_frame->root);
-        kill (getpid(), SIGTERM);
-        return 0;
+    STACK_DESTROY(sync_frame->root);
+    kill(getpid(), SIGTERM);
+    return 0;
 }
 
 void *
-gf_defrag_start (void *data)
+gf_defrag_start(void *data)
 {
-        int                      ret    = -1;
-        call_frame_t            *frame  = NULL;
-        dht_conf_t              *conf   = NULL;
-        gf_defrag_info_t        *defrag = NULL;
-        xlator_t                *this  = NULL;
-
-        this = data;
-        conf = this->private;
-        if (!conf)
-                goto out;
+    int ret = -1;
+    call_frame_t *frame = NULL;
+    dht_conf_t *conf = NULL;
+    gf_defrag_info_t *defrag = NULL;
+    xlator_t *this = NULL;
+    xlator_t *old_THIS = NULL;
 
-        defrag = conf->defrag;
-        if (!defrag)
-                goto out;
+    this = data;
+    conf = this->private;
+    if (!conf)
+        goto out;
 
-        frame = create_frame (this, this->ctx->pool);
-        if (!frame)
-                goto out;
+    defrag = conf->defrag;
+    if (!defrag)
+        goto out;
+
+    frame = create_frame(this, this->ctx->pool);
+    if (!frame)
+        goto out;
 
-        frame->root->pid = GF_CLIENT_PID_DEFRAG;
+    frame->root->pid = GF_CLIENT_PID_DEFRAG;
 
-        defrag->pid = frame->root->pid;
+    defrag->pid = frame->root->pid;
 
-        defrag->defrag_status = GF_DEFRAG_STATUS_STARTED;
+    defrag->defrag_status = GF_DEFRAG_STATUS_STARTED;
 
-        ret = synctask_new (this->ctx->env, gf_defrag_start_crawl,
-                            gf_defrag_done, frame, this);
+    old_THIS = THIS;
+    THIS = this;
+    ret = synctask_new(this->ctx->env, gf_defrag_start_crawl, gf_defrag_done,
+                       frame, this);
 
-        if (ret)
-                gf_log (this->name, GF_LOG_ERROR, "Could not create"
-                        " task for rebalance");
+    if (ret)
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_REBALANCE_START_FAILED,
+               "Could not create task for rebalance");
+    THIS = old_THIS;
 out:
-        return NULL;
+    return NULL;
 }
 
-int
-gf_defrag_status_get (gf_defrag_info_t *defrag, dict_t *dict)
+uint64_t
+gf_defrag_get_estimates_based_on_size(dht_conf_t *conf)
 {
-        int      ret    = 0;
-        uint64_t files  = 0;
-        uint64_t size   = 0;
-        uint64_t lookup = 0;
-        uint64_t failures = 0;
-        uint64_t skipped = 0;
-        char     *status = "";
-        double   elapsed = 0;
-        struct timeval end = {0,};
-
-
-        if (!defrag)
-                goto out;
+    gf_defrag_info_t *defrag = NULL;
+    double rate_processed = 0;
+    uint64_t total_processed = 0;
+    uint64_t tmp_count = 0;
+    uint64_t time_to_complete = 0;
+    double elapsed = 0;
 
-        ret = 0;
-        if (defrag->defrag_status == GF_DEFRAG_STATUS_NOT_STARTED)
-                goto out;
+    defrag = conf->defrag;
 
-        files  = defrag->total_files;
-        size   = defrag->total_data;
-        lookup = defrag->num_files_lookedup;
-        failures = defrag->total_failures;
-        skipped = defrag->skipped;
+    if (!g_totalsize)
+        goto out;
 
-        gettimeofday (&end, NULL);
+    elapsed = gf_time() - defrag->start_time;
 
-        elapsed = end.tv_sec - defrag->start_time.tv_sec;
+    /* Don't calculate the estimates for the first 10 minutes.
+     * It is unlikely to be accurate and estimates are not required
+     * if the process finishes in less than 10 mins.
+     */
 
-        if (!dict)
-                goto log;
+    if (elapsed < ESTIMATE_START_INTERVAL) {
+        gf_msg(THIS->name, GF_LOG_INFO, 0, 0,
+               "Rebalance estimates will not be available for the "
+               "first %d seconds.",
+               ESTIMATE_START_INTERVAL);
 
-        ret = dict_set_uint64 (dict, "files", files);
-        if (ret)
-                gf_log (THIS->name, GF_LOG_WARNING,
-                        "failed to set file count");
+        goto out;
+    }
 
-        ret = dict_set_uint64 (dict, "size", size);
-        if (ret)
-                gf_log (THIS->name, GF_LOG_WARNING,
-                        "failed to set size of xfer");
+    total_processed = defrag->size_processed;
 
-        ret = dict_set_uint64 (dict, "lookups", lookup);
-        if (ret)
-                gf_log (THIS->name, GF_LOG_WARNING,
-                        "failed to set lookedup file count");
+    /* rate at which files processed */
+    rate_processed = (total_processed) / elapsed;
 
+    tmp_count = g_totalsize;
 
-        ret = dict_set_int32 (dict, "status", defrag->defrag_status);
-        if (ret)
-                gf_log (THIS->name, GF_LOG_WARNING,
-                        "failed to set status");
-        if (elapsed) {
-                ret = dict_set_double (dict, "run-time", elapsed);
-                if (ret)
-                        gf_log (THIS->name, GF_LOG_WARNING,
-                                "failed to set run-time");
-        }
+    if (rate_processed) {
+        time_to_complete = (tmp_count) / rate_processed;
+
+    } else {
+        gf_msg(THIS->name, GF_LOG_ERROR, 0, 0,
+               "Unable to calculate estimated time for rebalance");
+    }
+
+    gf_log(THIS->name, GF_LOG_INFO,
+           "TIME: (size) total_processed=%" PRIu64 " tmp_cnt = %" PRIu64
+           ","
+           "rate_processed=%f, elapsed = %f",
+           total_processed, tmp_count, rate_processed, elapsed);
 
-        ret = dict_set_uint64 (dict, "failures", failures);
-        if (ret)
-                gf_log (THIS->name, GF_LOG_WARNING,
-                        "failed to set failure count");
+out:
+    return time_to_complete;
+}
+
+int
+gf_defrag_status_get(dht_conf_t *conf, dict_t *dict)
+{
+    int ret = 0;
+    uint64_t files = 0;
+    uint64_t size = 0;
+    uint64_t lookup = 0;
+    uint64_t failures = 0;
+    uint64_t skipped = 0;
+    char *status = "";
+    double elapsed = 0;
+    uint64_t time_to_complete = 0;
+    uint64_t time_left = 0;
+    gf_defrag_info_t *defrag = conf->defrag;
+
+    if (!defrag)
+        goto out;
+
+    ret = 0;
+    if (defrag->defrag_status == GF_DEFRAG_STATUS_NOT_STARTED)
+        goto out;
+
+    files = defrag->total_files;
+    size = defrag->total_data;
+    lookup = defrag->num_files_lookedup;
+    failures = defrag->total_failures;
+    skipped = defrag->skipped;
+
+    elapsed = gf_time() - defrag->start_time;
+
+    /* The rebalance is still in progress */
+
+    if (defrag->defrag_status == GF_DEFRAG_STATUS_STARTED) {
+        time_to_complete = gf_defrag_get_estimates_based_on_size(conf);
+
+        if (time_to_complete && (time_to_complete > elapsed))
+            time_left = time_to_complete - elapsed;
+
+        gf_log(THIS->name, GF_LOG_INFO,
+               "TIME: Estimated total time to complete (size)= %" PRIu64
+               " seconds, seconds left = %" PRIu64 "",
+               time_to_complete, time_left);
+    }
+
+    if (!dict)
+        goto log;
+
+    ret = dict_set_uint64(dict, "files", files);
+    if (ret)
+        gf_log(THIS->name, GF_LOG_WARNING, "failed to set file count");
+
+    ret = dict_set_uint64(dict, "size", size);
+    if (ret)
+        gf_log(THIS->name, GF_LOG_WARNING, "failed to set size of xfer");
+
+    ret = dict_set_uint64(dict, "lookups", lookup);
+    if (ret)
+        gf_log(THIS->name, GF_LOG_WARNING, "failed to set lookedup file count");
+
+    ret = dict_set_int32(dict, "status", defrag->defrag_status);
+    if (ret)
+        gf_log(THIS->name, GF_LOG_WARNING, "failed to set status");
+
+    ret = dict_set_double(dict, "run-time", elapsed);
+    if (ret)
+        gf_log(THIS->name, GF_LOG_WARNING, "failed to set run-time");
+
+    ret = dict_set_uint64(dict, "failures", failures);
+    if (ret)
+        gf_log(THIS->name, GF_LOG_WARNING, "failed to set failure count");
+
+    ret = dict_set_uint64(dict, "skipped", skipped);
+    if (ret)
+        gf_log(THIS->name, GF_LOG_WARNING, "failed to set skipped file count");
+
+    ret = dict_set_uint64(dict, "time-left", time_left);
+    if (ret)
+        gf_log(THIS->name, GF_LOG_WARNING, "failed to set time-left");
 
-        ret = dict_set_uint64 (dict, "skipped", skipped);
-        if (ret)
-                gf_log (THIS->name, GF_LOG_WARNING,
-                        "failed to set skipped file count");
 log:
-        switch (defrag->defrag_status) {
+    switch (defrag->defrag_status) {
         case GF_DEFRAG_STATUS_NOT_STARTED:
-                status = "not started";
-                break;
+            status = "not started";
+            break;
         case GF_DEFRAG_STATUS_STARTED:
-                status = "in progress";
-                break;
+            status = "in progress";
+            break;
         case GF_DEFRAG_STATUS_STOPPED:
-                status = "stopped";
-                break;
+            status = "stopped";
+            break;
         case GF_DEFRAG_STATUS_COMPLETE:
-                status = "completed";
-                break;
+            status = "completed";
+            break;
         case GF_DEFRAG_STATUS_FAILED:
-                status = "failed";
-                break;
+            status = "failed";
+            break;
         default:
-                break;
-        }
-
-        gf_log (THIS->name, GF_LOG_INFO, "Rebalance is %s. Time taken is %.2f "
-                "secs", status, elapsed);
-        gf_log (THIS->name, GF_LOG_INFO, "Files migrated: %"PRIu64", size: %"
-                PRIu64", lookups: %"PRIu64", failures: %"PRIu64", skipped: "
-                "%"PRIu64, files, size, lookup, failures, skipped);
-
-
+            break;
+    }
+
+    gf_msg(THIS->name, GF_LOG_INFO, 0, DHT_MSG_REBALANCE_STATUS,
+           "Rebalance is %s. Time taken is %.2f secs", status, elapsed);
+    gf_msg(THIS->name, GF_LOG_INFO, 0, DHT_MSG_REBALANCE_STATUS,
+           "Files migrated: %" PRIu64 ", size: %" PRIu64 ", lookups: %" PRIu64
+           ", failures: %" PRIu64
+           ", skipped: "
+           "%" PRIu64,
+           files, size, lookup, failures, skipped);
 out:
-        return 0;
+    return 0;
 }
 
 int
-gf_defrag_stop (gf_defrag_info_t *defrag, dict_t *output)
+gf_defrag_stop(dht_conf_t *conf, gf_defrag_status_t status, dict_t *output)
 {
-        /* TODO: set a variable 'stop_defrag' here, it should be checked
-           in defrag loop */
-        int     ret = -1;
-        GF_ASSERT (defrag);
+    /* TODO: set a variable 'stop_defrag' here, it should be checked
+       in defrag loop */
+    int ret = -1;
+    gf_defrag_info_t *defrag = conf->defrag;
 
-        if (defrag->defrag_status == GF_DEFRAG_STATUS_NOT_STARTED) {
-                goto out;
-        }
+    GF_ASSERT(defrag);
 
-        gf_log ("", GF_LOG_INFO, "Received stop command on rebalance");
-        defrag->defrag_status = GF_DEFRAG_STATUS_STOPPED;
+    if (defrag->defrag_status == GF_DEFRAG_STATUS_NOT_STARTED) {
+        goto out;
+    }
 
-        if (output)
-                gf_defrag_status_get (defrag, output);
-        ret = 0;
+    gf_msg("", GF_LOG_INFO, 0, DHT_MSG_REBALANCE_STOPPED,
+           "Received stop command on rebalance");
+    defrag->defrag_status = status;
+
+    if (output)
+        gf_defrag_status_get(conf, output);
+    ret = 0;
 out:
-        gf_log ("", GF_LOG_DEBUG, "Returning %d", ret);
-        return ret;
+    gf_msg_debug("", 0, "Returning %d", ret);
+    return ret;
 }
diff --git a/xlators/cluster/dht/src/dht-rename.c b/xlators/cluster/dht/src/dht-rename.c
index 5d6f4f2322e..d9dbf50492f 100644
--- a/xlators/cluster/dht/src/dht-rename.c
+++ b/xlators/cluster/dht/src/dht-rename.c
@@ -11,914 +11,1987 @@
 /* TODO: link(oldpath, newpath) fails if newpath already exists. DHT should
  *       delete the newpath if it gets EEXISTS from link() call.
  */
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "xlator.h"
 #include "dht-common.h"
-#include "defaults.h"
+#include "dht-lock.h"
+#include <glusterfs/defaults.h>
 
+int
+dht_rename_unlock(call_frame_t *frame, xlator_t *this);
+int32_t
+dht_rename_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *xdata);
 
 int
-dht_rename_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, struct iatt *stbuf,
-                    struct iatt *preoldparent, struct iatt *postoldparent,
-                    struct iatt *prenewparent, struct iatt *postnewparent,
-                    dict_t *xdata)
-{
-        dht_local_t  *local = NULL;
-        int           this_call_cnt = 0;
-        call_frame_t *prev = NULL;
-
-        local = frame->local;
-        prev = cookie;
-
-        if (op_ret == -1) {
-                /* TODO: undo the damage */
-
-                gf_log (this->name, GF_LOG_INFO,
-                        "rename %s -> %s on %s failed (%s)",
-                        local->loc.path, local->loc2.path,
-                        prev->this->name, strerror (op_errno));
-
-                local->op_ret   = op_ret;
-                local->op_errno = op_errno;
-                goto unwind;
-        }
-        /* TODO: construct proper stbuf for dir */
-        /*
-         * FIXME: is this the correct way to build stbuf and
-         * parent bufs?
-         */
-        dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
-        dht_iatt_merge (this, &local->preoldparent, preoldparent,
-                        prev->this);
-        dht_iatt_merge (this, &local->postoldparent, postoldparent,
-                        prev->this);
-        dht_iatt_merge (this, &local->preparent, prenewparent,
-                        prev->this);
-        dht_iatt_merge (this, &local->postparent, postnewparent,
-                        prev->this);
+dht_rename_unlock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
 
+    local = frame->local;
 
-unwind:
-        this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt)) {
-                WIPE (&local->preoldparent);
-                WIPE (&local->postoldparent);
-                WIPE (&local->preparent);
-                WIPE (&local->postparent);
-
-                DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
-                DHT_STACK_UNWIND (rename, frame, local->op_ret, local->op_errno,
-                                  &local->stbuf, &local->preoldparent,
-                                  &local->postoldparent,
-                                  &local->preparent, &local->postparent, xdata);
-        }
+    dht_set_fixed_dir_stat(&local->preoldparent);
+    dht_set_fixed_dir_stat(&local->postoldparent);
+    dht_set_fixed_dir_stat(&local->preparent);
+    dht_set_fixed_dir_stat(&local->postparent);
 
-        return 0;
+    if (IA_ISREG(local->stbuf.ia_type))
+        DHT_STRIP_PHASE1_FLAGS(&local->stbuf);
+
+    DHT_STACK_UNWIND(rename, frame, local->op_ret, local->op_errno,
+                     &local->stbuf, &local->preoldparent, &local->postoldparent,
+                     &local->preparent, &local->postparent, local->xattr);
+    return 0;
 }
 
+static void
+dht_rename_dir_unlock_src(call_frame_t *frame, xlator_t *this)
+{
+    dht_local_t *local = NULL;
+
+    local = frame->local;
+    dht_unlock_namespace(frame, &local->lock[0]);
+    return;
+}
+
+static void
+dht_rename_dir_unlock_dst(call_frame_t *frame, xlator_t *this)
+{
+    dht_local_t *local = NULL;
+    int op_ret = -1;
+    char src_gfid[GF_UUID_BUF_SIZE] = {0};
+    char dst_gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+
+    /* Unlock entrylk */
+    dht_unlock_entrylk_wrapper(frame, &local->lock[1].ns.directory_ns);
+
+    /* Unlock inodelk */
+    op_ret = dht_unlock_inodelk(frame, local->lock[1].ns.parent_layout.locks,
+                                local->lock[1].ns.parent_layout.lk_count,
+                                dht_rename_unlock_cbk);
+    if (op_ret < 0) {
+        uuid_utoa_r(local->loc.inode->gfid, src_gfid);
+
+        if (local->loc2.inode)
+            uuid_utoa_r(local->loc2.inode->gfid, dst_gfid);
+
+        if (IA_ISREG(local->stbuf.ia_type))
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_UNLOCKING_FAILED,
+                   "winding unlock inodelk failed "
+                   "rename (%s:%s:%s %s:%s:%s), "
+                   "stale locks left on bricks",
+                   local->loc.path, src_gfid, local->src_cached->name,
+                   local->loc2.path, dst_gfid,
+                   local->dst_cached ? local->dst_cached->name : NULL);
+        else
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_UNLOCKING_FAILED,
+                   "winding unlock inodelk failed "
+                   "rename (%s:%s %s:%s), "
+                   "stale locks left on bricks",
+                   local->loc.path, src_gfid, local->loc2.path, dst_gfid);
+
+        dht_rename_unlock_cbk(frame, NULL, this, 0, 0, NULL);
+    }
+
+    return;
+}
 
+static int
+dht_rename_dir_unlock(call_frame_t *frame, xlator_t *this)
+{
+    dht_rename_dir_unlock_src(frame, this);
+    dht_rename_dir_unlock_dst(frame, this);
+    return 0;
+}
 int
-dht_rename_hashed_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                           int32_t op_ret, int32_t op_errno, struct iatt *stbuf,
-                           struct iatt *preoldparent,
-                           struct iatt *postoldparent,
-                           struct iatt *prenewparent,
-                           struct iatt *postnewparent, dict_t *xdata)
-{
-        dht_conf_t   *conf = NULL;
-        dht_local_t  *local = NULL;
-        int           call_cnt = 0;
-        call_frame_t *prev = NULL;
-        int           i = 0;
-
-        conf = this->private;
-        local = frame->local;
-        prev = cookie;
-
-        if (op_ret == -1) {
-                /* TODO: undo the damage */
-
-                gf_log (this->name, GF_LOG_INFO,
-                        "rename %s -> %s on %s failed (%s)",
-                        local->loc.path, local->loc2.path,
-                        prev->this->name, strerror (op_errno));
-
-                local->op_ret   = op_ret;
-                local->op_errno = op_errno;
-                goto unwind;
-        }
-        /* TODO: construct proper stbuf for dir */
-        /*
-         * FIXME: is this the correct way to build stbuf and
-         * parent bufs?
+dht_rename_dir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *stbuf,
+                   struct iatt *preoldparent, struct iatt *postoldparent,
+                   struct iatt *prenewparent, struct iatt *postnewparent,
+                   dict_t *xdata)
+{
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+    int i = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    int subvol_cnt = -1;
+
+    conf = this->private;
+    local = frame->local;
+    prev = cookie;
+    subvol_cnt = dht_subvol_cnt(this, prev);
+    local->ret_cache[subvol_cnt] = op_ret;
+
+    if (op_ret == -1) {
+        gf_uuid_unparse(local->loc.inode->gfid, gfid);
+
+        gf_msg(this->name, GF_LOG_INFO, op_errno, DHT_MSG_RENAME_FAILED,
+               "Rename %s -> %s on %s failed, (gfid = %s)", local->loc.path,
+               local->loc2.path, prev->name, gfid);
+
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        goto unwind;
+    }
+    /* TODO: construct proper stbuf for dir */
+    /*
+     * FIXME: is this the correct way to build stbuf and
+     * parent bufs?
+     */
+    dht_iatt_merge(this, &local->stbuf, stbuf);
+    dht_iatt_merge(this, &local->preoldparent, preoldparent);
+    dht_iatt_merge(this, &local->postoldparent, postoldparent);
+    dht_iatt_merge(this, &local->preparent, prenewparent);
+    dht_iatt_merge(this, &local->postparent, postnewparent);
+
+unwind:
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        /* We get here with local->call_cnt == 0. Which means
+         * we are the only one executing this code, there is
+         * no contention. Therefore it's safe to manipulate or
+         * deref local->call_cnt directly (without locking).
          */
-        dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
-        dht_iatt_merge (this, &local->preoldparent, preoldparent,
-                        prev->this);
-        dht_iatt_merge (this, &local->postoldparent, postoldparent,
-                        prev->this);
-        dht_iatt_merge (this, &local->preparent, prenewparent,
-                        prev->this);
-        dht_iatt_merge (this, &local->postparent, postnewparent,
-                        prev->this);
-
-        call_cnt = local->call_cnt = conf->subvolume_cnt - 1;
-
-        if (!local->call_cnt)
-                goto unwind;
-
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                if (conf->subvolumes[i] == local->dst_hashed)
+        if (local->ret_cache[conf->subvolume_cnt] == 0) {
+            /* count errant subvols in last field of ret_cache */
+            for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (local->ret_cache[i] != 0)
+                    ++local->ret_cache[conf->subvolume_cnt];
+            }
+            if (local->ret_cache[conf->subvolume_cnt]) {
+                /* undoing the damage:
+                 * for all subvolumes, where rename
+                 * succeeded, we perform the reverse operation
+                 */
+                for (i = 0; i < conf->subvolume_cnt; i++) {
+                    if (local->ret_cache[i] == 0)
+                        ++local->call_cnt;
+                }
+                for (i = 0; i < conf->subvolume_cnt; i++) {
+                    if (local->ret_cache[i])
                         continue;
-                STACK_WIND (frame, dht_rename_dir_cbk,
-                            conf->subvolumes[i],
-                            conf->subvolumes[i]->fops->rename,
-                            &local->loc, &local->loc2, NULL);
-                if (!--call_cnt)
-                        break;
-        }
 
+                    STACK_WIND(frame, dht_rename_dir_cbk, conf->subvolumes[i],
+                               conf->subvolumes[i]->fops->rename, &local->loc2,
+                               &local->loc, NULL);
+                }
 
-        return 0;
-unwind:
-        WIPE (&local->preoldparent);
-        WIPE (&local->postoldparent);
-        WIPE (&local->preparent);
-        WIPE (&local->postparent);
+                return 0;
+            }
+        }
 
-        DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
-        DHT_STACK_UNWIND (rename, frame, local->op_ret, local->op_errno,
-                          &local->stbuf, &local->preoldparent,
-                          &local->postoldparent,
-                          &local->preparent, &local->postparent, NULL);
+        WIPE(&local->preoldparent);
+        WIPE(&local->postoldparent);
+        WIPE(&local->preparent);
+        WIPE(&local->postparent);
 
-        return 0;
+        dht_rename_dir_unlock(frame, this);
+    }
+
+    return 0;
 }
 
+int
+dht_rename_hashed_dir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, struct iatt *stbuf,
+                          struct iatt *preoldparent, struct iatt *postoldparent,
+                          struct iatt *prenewparent, struct iatt *postnewparent,
+                          dict_t *xdata)
+{
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    int call_cnt = 0;
+    xlator_t *prev = NULL;
+    int i = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    conf = this->private;
+    local = frame->local;
+    prev = cookie;
+
+    if (op_ret == -1) {
+        gf_uuid_unparse(local->loc.inode->gfid, gfid);
+
+        gf_msg(this->name, GF_LOG_INFO, op_errno, DHT_MSG_RENAME_FAILED,
+               "rename %s -> %s on %s failed, (gfid = %s) ", local->loc.path,
+               local->loc2.path, prev->name, gfid);
+
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        goto unwind;
+    }
+    /* TODO: construct proper stbuf for dir */
+    /*
+     * FIXME: is this the correct way to build stbuf and
+     * parent bufs?
+     */
+    dht_iatt_merge(this, &local->stbuf, stbuf);
+    dht_iatt_merge(this, &local->preoldparent, preoldparent);
+    dht_iatt_merge(this, &local->postoldparent, postoldparent);
+    dht_iatt_merge(this, &local->preparent, prenewparent);
+    dht_iatt_merge(this, &local->postparent, postnewparent);
+
+    call_cnt = local->call_cnt = conf->subvolume_cnt - 1;
+
+    if (!local->call_cnt)
+        goto unwind;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (conf->subvolumes[i] == local->dst_hashed)
+            continue;
+        STACK_WIND_COOKIE(
+            frame, dht_rename_dir_cbk, conf->subvolumes[i], conf->subvolumes[i],
+            conf->subvolumes[i]->fops->rename, &local->loc, &local->loc2, NULL);
+        if (!--call_cnt)
+            break;
+    }
+
+    return 0;
+unwind:
+    WIPE(&local->preoldparent);
+    WIPE(&local->postoldparent);
+    WIPE(&local->preparent);
+    WIPE(&local->postparent);
+
+    dht_rename_dir_unlock(frame, this);
+    return 0;
+}
 
 int
-dht_rename_dir_do (call_frame_t *frame, xlator_t *this)
+dht_rename_dir_do(call_frame_t *frame, xlator_t *this)
 {
-        dht_local_t  *local = NULL;
+    dht_local_t *local = NULL;
 
-        local = frame->local;
+    local = frame->local;
 
-        if (local->op_ret == -1)
-                goto err;
+    if (local->op_ret == -1)
+        goto err;
 
-        local->op_ret = 0;
+    local->op_ret = 0;
 
-        STACK_WIND (frame, dht_rename_hashed_dir_cbk,
-                    local->dst_hashed,
-                    local->dst_hashed->fops->rename,
-                    &local->loc, &local->loc2, NULL);
-        return 0;
+    STACK_WIND_COOKIE(frame, dht_rename_hashed_dir_cbk, local->dst_hashed,
+                      local->dst_hashed, local->dst_hashed->fops->rename,
+                      &local->loc, &local->loc2, NULL);
+    return 0;
 
 err:
-        DHT_STACK_UNWIND (rename, frame, local->op_ret, local->op_errno, NULL, NULL,
-                          NULL, NULL, NULL, NULL);
-        return 0;
+    dht_rename_dir_unlock(frame, this);
+    return 0;
 }
 
-
 int
-dht_rename_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int op_ret, int op_errno, gf_dirent_t *entries,
-                        dict_t *xdata)
-{
-        dht_local_t  *local = NULL;
-        int           this_call_cnt = -1;
-        call_frame_t *prev = NULL;
-
-        local = frame->local;
-        prev  = cookie;
-
-        if (op_ret > 2) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "readdir on %s for %s returned %d entries",
-                        prev->this->name, local->loc.path, op_ret);
-                local->op_ret = -1;
-                local->op_errno = ENOTEMPTY;
-        }
+dht_rename_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno, gf_dirent_t *entries,
+                       dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = -1;
+    xlator_t *prev = NULL;
 
-        this_call_cnt = dht_frame_return (frame);
+    local = frame->local;
+    prev = cookie;
 
-        if (is_last_call (this_call_cnt)) {
-                dht_rename_dir_do (frame, this);
-        }
+    if (op_ret > 2) {
+        gf_msg_trace(this->name, 0, "readdir on %s for %s returned %d entries",
+                     prev->name, local->loc.path, op_ret);
+        local->op_ret = -1;
+        local->op_errno = ENOTEMPTY;
+    }
 
-        return 0;
-}
+    this_call_cnt = dht_frame_return(frame);
 
+    if (is_last_call(this_call_cnt)) {
+        dht_rename_dir_do(frame, this);
+    }
+
+    return 0;
+}
 
 int
-dht_rename_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int op_ret, int op_errno, fd_t *fd, dict_t *xdata)
+dht_rename_opendir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno, fd_t *fd, dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
-        int           this_call_cnt = -1;
-        call_frame_t *prev = NULL;
+    dht_local_t *local = NULL;
+    int this_call_cnt = -1;
+    xlator_t *prev = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
 
+    local = frame->local;
+    prev = cookie;
 
-        local = frame->local;
-        prev  = cookie;
+    if (op_ret == -1) {
+        gf_uuid_unparse(local->loc.inode->gfid, gfid);
+        gf_msg(this->name, GF_LOG_INFO, op_errno, DHT_MSG_OPENDIR_FAILED,
+               "opendir on %s for %s failed,(gfid = %s) ", prev->name,
+               local->loc.path, gfid);
+        goto err;
+    }
 
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "opendir on %s for %s failed (%s)",
-                        prev->this->name, local->loc.path,
-                        strerror (op_errno));
-                goto err;
-        }
-
-        STACK_WIND (frame, dht_rename_readdir_cbk,
-                    prev->this, prev->this->fops->readdir,
-                    local->fd, 4096, 0, NULL);
+    fd_bind(fd);
+    STACK_WIND_COOKIE(frame, dht_rename_readdir_cbk, prev, prev,
+                      prev->fops->readdir, local->fd, 4096, 0, NULL);
 
-        return 0;
+    return 0;
 
 err:
-        this_call_cnt = dht_frame_return (frame);
+    this_call_cnt = dht_frame_return(frame);
 
-        if (is_last_call (this_call_cnt)) {
-                dht_rename_dir_do (frame, this);
-        }
+    if (is_last_call(this_call_cnt)) {
+        dht_rename_dir_do(frame, this);
+    }
 
-        return 0;
+    return 0;
 }
 
-
 int
-dht_rename_dir (call_frame_t *frame, xlator_t *this)
+dht_rename_dir_lock2_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        dht_conf_t  *conf = NULL;
-        dht_local_t *local = NULL;
-        int          i = 0;
-        int          op_errno = -1;
+    dht_local_t *local = NULL;
+    char src_gfid[GF_UUID_BUF_SIZE] = {0};
+    char dst_gfid[GF_UUID_BUF_SIZE] = {0};
+    dht_conf_t *conf = NULL;
+    int i = 0;
+
+    local = frame->local;
+    conf = this->private;
+
+    if (op_ret < 0) {
+        uuid_utoa_r(local->loc.inode->gfid, src_gfid);
+
+        if (local->loc2.inode)
+            uuid_utoa_r(local->loc2.inode->gfid, dst_gfid);
+
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_INODE_LK_ERROR,
+               "acquiring entrylk after inodelk failed"
+               "rename (%s:%s:%s %s:%s:%s)",
+               local->loc.path, src_gfid, local->src_cached->name,
+               local->loc2.path, dst_gfid,
+               local->dst_cached ? local->dst_cached->name : NULL);
+
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+        goto err;
+    }
+
+    local->fd = fd_create(local->loc.inode, frame->root->pid);
+    if (!local->fd) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->op_ret = 0;
+
+    if (!local->dst_cached) {
+        dht_rename_dir_do(frame, this);
+        return 0;
+    }
 
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        STACK_WIND_COOKIE(frame, dht_rename_opendir_cbk, conf->subvolumes[i],
+                          conf->subvolumes[i],
+                          conf->subvolumes[i]->fops->opendir, &local->loc2,
+                          local->fd, NULL);
+    }
 
-        conf = frame->this->private;
-        local = frame->local;
+    return 0;
 
-        local->call_cnt = conf->subvolume_cnt;
+err:
+    /* No harm in calling an extra unlock */
+    dht_rename_dir_unlock(frame, this);
+    return 0;
+}
 
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                if (!conf->subvolume_status[i]) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "one of the subvolumes down (%s)",
-                                conf->subvolumes[i]->name);
-                        op_errno = ENOTCONN;
-                        goto err;
-                }
-        }
+int
+dht_rename_dir_lock1_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    char src_gfid[GF_UUID_BUF_SIZE] = {0};
+    char dst_gfid[GF_UUID_BUF_SIZE] = {0};
+    int ret = 0;
+    loc_t *loc = NULL;
+    xlator_t *subvol = NULL;
+
+    local = frame->local;
+
+    if (op_ret < 0) {
+        uuid_utoa_r(local->loc.inode->gfid, src_gfid);
+
+        if (local->loc2.inode)
+            uuid_utoa_r(local->loc2.inode->gfid, dst_gfid);
+
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_INODE_LK_ERROR,
+               "acquiring entrylk after inodelk failed"
+               "rename (%s:%s:%s %s:%s:%s)",
+               local->loc.path, src_gfid, local->src_cached->name,
+               local->loc2.path, dst_gfid,
+               local->dst_cached ? local->dst_cached->name : NULL);
+
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+        goto err;
+    }
+
+    if (local->current == &local->lock[0]) {
+        loc = &local->loc2;
+        subvol = local->dst_hashed;
+        local->current = &local->lock[1];
+    } else {
+        loc = &local->loc;
+        subvol = local->src_hashed;
+        local->current = &local->lock[0];
+    }
+    ret = dht_protect_namespace(frame, loc, subvol, &local->current->ns,
+                                dht_rename_dir_lock2_cbk);
+    if (ret < 0) {
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    return 0;
+err:
+    /* No harm in calling an extra unlock */
+    dht_rename_dir_unlock(frame, this);
+    return 0;
+}
 
-        local->fd = fd_create (local->loc.inode, frame->root->pid);
-        if (!local->fd) {
-                op_errno = ENOMEM;
-                goto err;
+/*
+ * If the hashed subvolumes of both source and dst are the different,
+ * lock in dictionary order of hashed subvol->name. This is important
+ * in case the parent directory is the same for both src and dst to
+ * prevent inodelk deadlocks when racing with a fix-layout op on the parent.
+ *
+ * If the hashed subvols are the same, use the gfid/name to determine
+ * the order of taking locks to prevent entrylk deadlocks when the parent
+ * dirs are the same.
+ *
+ */
+static int
+dht_order_rename_lock(call_frame_t *frame, loc_t **loc, xlator_t **subvol)
+{
+    int ret = 0;
+    int op_ret = 0;
+    dht_local_t *local = NULL;
+    char *src = NULL;
+    char *dst = NULL;
+
+    local = frame->local;
+
+    if (local->src_hashed->name == local->dst_hashed->name) {
+        ret = 0;
+    } else {
+        ret = strcmp(local->src_hashed->name, local->dst_hashed->name);
+    }
+
+    if (ret == 0) {
+        /* hashed subvols are the same for src and dst */
+        /* Entrylks need to be ordered*/
+
+        src = alloca(GF_UUID_BNAME_BUF_SIZE + strlen(local->loc.name) + 1);
+        if (!src) {
+            gf_msg(frame->this->name, GF_LOG_ERROR, ENOMEM, 0,
+                   "Insufficient memory for src");
+            op_ret = -1;
+            goto out;
         }
 
-        local->op_ret = 0;
+        if (!gf_uuid_is_null(local->loc.pargfid))
+            uuid_utoa_r(local->loc.pargfid, src);
+        else if (local->loc.parent)
+            uuid_utoa_r(local->loc.parent->gfid, src);
+        else
+            src[0] = '\0';
 
-        if (!local->dst_cached) {
-                dht_rename_dir_do (frame, this);
-                return 0;
-        }
+        strcat(src, local->loc.name);
 
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                STACK_WIND (frame, dht_rename_opendir_cbk,
-                            conf->subvolumes[i],
-                            conf->subvolumes[i]->fops->opendir,
-                            &local->loc2, local->fd, NULL);
+        dst = alloca(GF_UUID_BNAME_BUF_SIZE + strlen(local->loc2.name) + 1);
+        if (!dst) {
+            gf_msg(frame->this->name, GF_LOG_ERROR, ENOMEM, 0,
+                   "Insufficient memory for dst");
+            op_ret = -1;
+            goto out;
         }
 
-        return 0;
+        if (!gf_uuid_is_null(local->loc2.pargfid))
+            uuid_utoa_r(local->loc2.pargfid, dst);
+        else if (local->loc2.parent)
+            uuid_utoa_r(local->loc2.parent->gfid, dst);
+        else
+            dst[0] = '\0';
+
+        strcat(dst, local->loc2.name);
+        ret = strcmp(src, dst);
+    }
+
+    if (ret <= 0) {
+        /*inodelk in dictionary order of hashed subvol names*/
+        /*entrylk in dictionary order of gfid/basename */
+        local->current = &local->lock[0];
+        *loc = &local->loc;
+        *subvol = local->src_hashed;
+
+    } else {
+        local->current = &local->lock[1];
+        *loc = &local->loc2;
+        *subvol = local->dst_hashed;
+    }
+
+    op_ret = 0;
+
+out:
+    return op_ret;
+}
+
+int
+dht_rename_dir(call_frame_t *frame, xlator_t *this)
+{
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    loc_t *loc = NULL;
+    xlator_t *subvol = NULL;
+    int i = 0;
+    int ret = 0;
+    int op_errno = -1;
+
+    conf = frame->this->private;
+    local = frame->local;
+
+    local->ret_cache = GF_CALLOC(conf->subvolume_cnt + 1, sizeof(int),
+                                 gf_dht_ret_cache_t);
+
+    if (local->ret_cache == NULL) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->call_cnt = conf->subvolume_cnt;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (!conf->subvolume_status[i]) {
+            gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_RENAME_FAILED,
+                   "Rename dir failed: subvolume down (%s)",
+                   conf->subvolumes[i]->name);
+            op_errno = ENOTCONN;
+            goto err;
+        }
+    }
+
+    /* Locks on src and dst needs to ordered which otherwise might cause
+     * deadlocks when rename (src, dst) and rename (dst, src) is done from
+     * two different clients
+     */
+    ret = dht_order_rename_lock(frame, &loc, &subvol);
+    if (ret) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    /* Rename must take locks on src to avoid lookup selfheal from
+     * recreating src on those subvols where the rename was successful.
+     * The locks can't be issued parallel as two different clients might
+     * attempt same rename command and be in dead lock.
+     */
+    ret = dht_protect_namespace(frame, loc, subvol, &local->current->ns,
+                                dht_rename_dir_lock1_cbk);
+    if (ret < 0) {
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (rename, frame, -1, op_errno, NULL, NULL, NULL, NULL,
-                          NULL, NULL);
-        return 0;
+    DHT_STACK_UNWIND(rename, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL,
+                     NULL);
+    return 0;
 }
-#define DHT_MARK_FOP_INTERNAL(xattr) do {                                      \
-                int tmp = -1;                                                  \
-                if (!xattr) {                                                  \
-                        xattr = dict_new ();                                   \
-                        if (!xattr)                                            \
-                                break;                                         \
-                }                                                              \
-                tmp = dict_set_str (xattr, GLUSTERFS_INTERNAL_FOP_KEY, "yes"); \
-                if (tmp) {                                                     \
-                        gf_log (this->name, GF_LOG_ERROR, "Failed to set"      \
-                                " internal dict key for %s", local->loc.path); \
-                }                                                              \
-        }while (0)
+
+static int
+dht_rename_track_for_changelog(xlator_t *this, dict_t *xattr, loc_t *oldloc,
+                               loc_t *newloc)
+{
+    int ret = -1;
+    dht_changelog_rename_info_t *info = NULL;
+    char *name = NULL;
+    int len1 = 0;
+    int len2 = 0;
+    int size = 0;
+
+    if (!xattr || !oldloc || !newloc || !this)
+        return ret;
+
+    len1 = strlen(oldloc->name) + 1;
+    len2 = strlen(newloc->name) + 1;
+    size = sizeof(dht_changelog_rename_info_t) + len1 + len2;
+
+    info = GF_CALLOC(size, sizeof(char), gf_common_mt_char);
+    if (!info) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+               "Failed to calloc memory");
+        return ret;
+    }
+
+    gf_uuid_copy(info->old_pargfid, oldloc->pargfid);
+    gf_uuid_copy(info->new_pargfid, newloc->pargfid);
+
+    info->oldname_len = len1;
+    info->newname_len = len2;
+    strncpy(info->buffer, oldloc->name, len1);
+    name = info->buffer + len1;
+    strncpy(name, newloc->name, len2);
+
+    ret = dict_set_bin(xattr, DHT_CHANGELOG_RENAME_OP_KEY, info, size);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+               "Failed to set dictionary value: key = %s,"
+               " path = %s",
+               DHT_CHANGELOG_RENAME_OP_KEY, oldloc->name);
+        GF_FREE(info);
+    }
+
+    return ret;
+}
+
+#define DHT_MARKER_DONT_ACCOUNT(xattr)                                         \
+    do {                                                                       \
+        int tmp = -1;                                                          \
+        if (!xattr) {                                                          \
+            xattr = dict_new();                                                \
+            if (!xattr)                                                        \
+                break;                                                         \
+        }                                                                      \
+        tmp = dict_set_str(xattr, GLUSTERFS_MARKER_DONT_ACCOUNT_KEY, "yes");   \
+        if (tmp) {                                                             \
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,       \
+                   "Failed to set dictionary value: key = %s,"                 \
+                   " path = %s",                                               \
+                   GLUSTERFS_MARKER_DONT_ACCOUNT_KEY, local->loc.path);        \
+        }                                                                      \
+    } while (0)
+
+#define DHT_CHANGELOG_TRACK_AS_RENAME(xattr, oldloc, newloc)                   \
+    do {                                                                       \
+        int tmp = -1;                                                          \
+        if (!xattr) {                                                          \
+            xattr = dict_new();                                                \
+            if (!xattr) {                                                      \
+                gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,   \
+                       "Failed to create dictionary to "                       \
+                       "track rename");                                        \
+                break;                                                         \
+            }                                                                  \
+        }                                                                      \
+                                                                               \
+        tmp = dht_rename_track_for_changelog(this, xattr, oldloc, newloc);     \
+                                                                               \
+        if (tmp) {                                                             \
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,       \
+                   "Failed to set dictionary value: key = %s,"                 \
+                   " path = %s",                                               \
+                   DHT_CHANGELOG_RENAME_OP_KEY, (oldloc)->path);               \
+        }                                                                      \
+    } while (0)
+
+int
+dht_rename_unlock(call_frame_t *frame, xlator_t *this)
+{
+    dht_local_t *local = NULL;
+    int op_ret = -1;
+    char src_gfid[GF_UUID_BUF_SIZE] = {0};
+    char dst_gfid[GF_UUID_BUF_SIZE] = {0};
+    dht_ilock_wrap_t inodelk_wrapper = {
+        0,
+    };
+
+    local = frame->local;
+    inodelk_wrapper.locks = local->rename_inodelk_backward_compatible;
+    inodelk_wrapper.lk_count = local->rename_inodelk_bc_count;
+
+    op_ret = dht_unlock_inodelk_wrapper(frame, &inodelk_wrapper);
+    if (op_ret < 0) {
+        uuid_utoa_r(local->loc.inode->gfid, src_gfid);
+
+        if (local->loc2.inode)
+            uuid_utoa_r(local->loc2.inode->gfid, dst_gfid);
+
+        if (IA_ISREG(local->stbuf.ia_type))
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_UNLOCKING_FAILED,
+                   "winding unlock inodelk failed "
+                   "rename (%s:%s:%s %s:%s:%s), "
+                   "stale locks left on bricks",
+                   local->loc.path, src_gfid, local->src_cached->name,
+                   local->loc2.path, dst_gfid,
+                   local->dst_cached ? local->dst_cached->name : NULL);
+        else
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_UNLOCKING_FAILED,
+                   "winding unlock inodelk failed "
+                   "rename (%s:%s %s:%s), "
+                   "stale locks left on bricks",
+                   local->loc.path, src_gfid, local->loc2.path, dst_gfid);
+    }
+
+    dht_unlock_namespace(frame, &local->lock[0]);
+    dht_unlock_namespace(frame, &local->lock[1]);
+
+    dht_rename_unlock_cbk(frame, NULL, this, local->op_ret, local->op_errno,
+                          NULL);
+    return 0;
+}
+
 int
-dht_rename_done (call_frame_t *frame, xlator_t *this)
+dht_rename_done(call_frame_t *frame, xlator_t *this)
 {
-        dht_local_t     *local = NULL;
+    dht_local_t *local = NULL;
 
-        local = frame->local;
+    local = frame->local;
 
-        if (local->linked == _gf_true) {
-                local->linked = _gf_false;
-                dht_linkfile_attr_heal (frame, this);
-        }
-        DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
-        DHT_STACK_UNWIND (rename, frame, local->op_ret, local->op_errno,
-                          &local->stbuf, &local->preoldparent,
-                          &local->postoldparent, &local->preparent,
-                          &local->postparent, NULL);
+    if (local->linked == _gf_true) {
+        local->linked = _gf_false;
+        dht_linkfile_attr_heal(frame, this);
+    }
 
-        return 0;
+    dht_rename_unlock(frame, this);
+    return 0;
 }
 
 int
-dht_rename_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, struct iatt *preparent,
-                       struct iatt *postparent, dict_t *xdata)
+dht_rename_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                      struct iatt *postparent, dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
-        call_frame_t *prev = NULL;
-        int           this_call_cnt = 0;
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    int this_call_cnt = 0;
 
-        local = frame->local;
-        prev  = cookie;
+    local = frame->local;
+    prev = cookie;
 
-        if (!local) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "!local, should not happen");
-                goto out;
-        }
+    FRAME_SU_UNDO(frame, dht_local_t);
+    if (!local) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INVALID_VALUE,
+               "!local, should not happen");
+        goto out;
+    }
 
-        this_call_cnt = dht_frame_return (frame);
+    this_call_cnt = dht_frame_return(frame);
 
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: unlink on %s failed (%s)",
-                        local->loc.path, prev->this->name, strerror (op_errno));
-        }
+    if (op_ret == -1) {
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_UNLINK_FAILED,
+               "%s: Rename: unlink on %s failed ", local->loc.path, prev->name);
+    }
 
-        WIPE (&local->preoldparent);
-        WIPE (&local->postoldparent);
-        WIPE (&local->preparent);
-        WIPE (&local->postparent);
+    WIPE(&local->preoldparent);
+    WIPE(&local->postoldparent);
+    WIPE(&local->preparent);
+    WIPE(&local->postparent);
 
-        if (is_last_call (this_call_cnt)) {
-                dht_rename_done (frame, this);
-        }
+    if (is_last_call(this_call_cnt)) {
+        dht_rename_done(frame, this);
+    }
 
 out:
-        return 0;
+    return 0;
 }
 
-
 int
-dht_rename_cleanup (call_frame_t *frame)
+dht_rename_cleanup(call_frame_t *frame)
 {
-        dht_local_t *local = NULL;
-        xlator_t    *this = NULL;
-        xlator_t    *src_hashed = NULL;
-        xlator_t    *src_cached = NULL;
-        xlator_t    *dst_hashed = NULL;
-        xlator_t    *dst_cached = NULL;
-        int          call_cnt = 0;
-        dict_t      *xattr      = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *this = NULL;
+    xlator_t *src_hashed = NULL;
+    xlator_t *src_cached = NULL;
+    xlator_t *dst_hashed = NULL;
+    xlator_t *dst_cached = NULL;
+    int call_cnt = 0;
+    dict_t *xattr = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
 
-        local = frame->local;
-        this  = frame->this;
+    local = frame->local;
+    this = frame->this;
 
-        src_hashed = local->src_hashed;
-        src_cached = local->src_cached;
-        dst_hashed = local->dst_hashed;
-        dst_cached = local->dst_cached;
+    src_hashed = local->src_hashed;
+    src_cached = local->src_cached;
+    dst_hashed = local->dst_hashed;
+    dst_cached = local->dst_cached;
 
-        if (src_cached == dst_cached)
-                goto nolinks;
+    if (src_cached == dst_cached)
+        goto nolinks;
 
-        if (dst_hashed != src_hashed && dst_hashed != src_cached)
-                call_cnt++;
+    if (local->linked && (dst_hashed != src_hashed) &&
+        (dst_hashed != src_cached)) {
+        call_cnt++;
+    }
 
-        if (src_cached != dst_hashed)
-                call_cnt++;
+    if (local->added_link && (src_cached != dst_hashed)) {
+        call_cnt++;
+    }
 
-        local->call_cnt = call_cnt;
+    local->call_cnt = call_cnt;
 
-        if (!call_cnt)
-                goto nolinks;
+    if (!call_cnt)
+        goto nolinks;
 
-        DHT_MARK_FOP_INTERNAL (xattr);
+    DHT_MARK_FOP_INTERNAL(xattr);
 
-        if (dst_hashed != src_hashed && dst_hashed != src_cached) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "unlinking linkfile %s @ %s => %s",
-                        local->loc.path, dst_hashed->name, src_cached->name);
-                STACK_WIND (frame, dht_rename_unlink_cbk,
-                            dst_hashed, dst_hashed->fops->unlink,
-                            &local->loc, 0, xattr);
-        }
+    gf_uuid_unparse(local->loc.inode->gfid, gfid);
 
-        if (src_cached != dst_hashed) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "unlinking link %s => %s (%s)", local->loc.path,
-                        local->loc2.path, src_cached->name);
-                STACK_WIND (frame, dht_rename_unlink_cbk,
-                            src_cached, src_cached->fops->unlink,
-                            &local->loc2, 0, xattr);
-        }
+    if (local->linked && (dst_hashed != src_hashed) &&
+        (dst_hashed != src_cached)) {
+        dict_t *xattr_new = NULL;
 
-        if (xattr)
-                dict_unref (xattr);
+        gf_msg_trace(this->name, 0,
+                     "unlinking linkfile %s @ %s => %s, (gfid = %s)",
+                     local->loc.path, dst_hashed->name, src_cached->name, gfid);
 
-        return 0;
+        xattr_new = dict_copy_with_ref(xattr, NULL);
 
-nolinks:
-        WIPE (&local->preoldparent);
-        WIPE (&local->postoldparent);
-        WIPE (&local->preparent);
-        WIPE (&local->postparent);
+        DHT_MARKER_DONT_ACCOUNT(xattr_new);
 
-        DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
-        DHT_STACK_UNWIND (rename, frame, local->op_ret, local->op_errno,
-                          &local->stbuf, &local->preoldparent,
-                          &local->postoldparent, &local->preparent,
-                          &local->postparent, NULL);
+        FRAME_SU_DO(frame, dht_local_t);
+        STACK_WIND_COOKIE(frame, dht_rename_unlink_cbk, dst_hashed, dst_hashed,
+                          dst_hashed->fops->unlink, &local->loc, 0, xattr_new);
 
-        return 0;
-}
+        dict_unref(xattr_new);
+        xattr_new = NULL;
+    }
 
+    if (local->added_link && (src_cached != dst_hashed)) {
+        dict_t *xattr_new = NULL;
 
-int
-dht_rename_links_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                             int32_t op_ret, int32_t op_errno,
-                             inode_t *inode, struct iatt *stbuf,
-                             struct iatt *preparent, struct iatt *postparent,
-                             dict_t *xdata)
-{
-        call_frame_t *prev = NULL;
-        dht_local_t  *local = NULL;
-
-        prev = cookie;
-        local = frame->local;
-
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "link/file %s on %s failed (%s)",
-                        local->loc.path, prev->this->name, strerror (op_errno));
-        }
+        gf_msg_trace(this->name, 0, "unlinking link %s => %s (%s), (gfid = %s)",
+                     local->loc.path, local->loc2.path, src_cached->name, gfid);
 
-        if (local->linked == _gf_true) {
-                local->linked = _gf_false;
-                dht_linkfile_attr_heal (frame, this);
+        xattr_new = dict_copy_with_ref(xattr, NULL);
+
+        if (gf_uuid_compare(local->loc.pargfid, local->loc2.pargfid) == 0) {
+            DHT_MARKER_DONT_ACCOUNT(xattr_new);
         }
-        DHT_STACK_DESTROY (frame);
+        /* *
+         * The link to file is created using root permission.
+         * Hence deletion should happen using root. Otherwise
+         * it will fail.
+         */
+        FRAME_SU_DO(frame, dht_local_t);
+        STACK_WIND_COOKIE(frame, dht_rename_unlink_cbk, src_cached, src_cached,
+                          src_cached->fops->unlink, &local->loc2, 0, xattr_new);
 
-        return 0;
-}
+        dict_unref(xattr_new);
+        xattr_new = NULL;
+    }
+
+    if (xattr)
+        dict_unref(xattr);
+
+    return 0;
+
+nolinks:
+    WIPE(&local->preoldparent);
+    WIPE(&local->postoldparent);
+    WIPE(&local->preparent);
+    WIPE(&local->postparent);
 
+    dht_rename_unlock(frame, this);
+    return 0;
+}
 
 int
-dht_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                int32_t op_ret, int32_t op_errno, struct iatt *stbuf,
-                struct iatt *preoldparent, struct iatt *postoldparent,
-                struct iatt *prenewparent, struct iatt *postnewparent,
-                dict_t *xdata)
-{
-        dht_local_t  *local = NULL;
-        call_frame_t *prev = NULL;
-        xlator_t     *src_hashed = NULL;
-        xlator_t     *src_cached = NULL;
-        xlator_t     *dst_hashed = NULL;
-        xlator_t     *dst_cached = NULL;
-        xlator_t     *rename_subvol = NULL;
-        call_frame_t *link_frame = NULL;
-        dht_local_t *link_local = NULL;
-        dict_t       *xattr     = NULL;
-
-        local = frame->local;
-        prev = cookie;
-
-        src_hashed = local->src_hashed;
-        src_cached = local->src_cached;
-        dst_hashed = local->dst_hashed;
-        dst_cached = local->dst_cached;
-
-        if (local->linked == _gf_true)
-                FRAME_SU_UNDO (frame, dht_local_t);
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: rename on %s failed (%s)", local->loc.path,
-                        prev->this->name, strerror (op_errno));
-                local->op_ret   = op_ret;
-                local->op_errno = op_errno;
-                goto cleanup;
-        }
+dht_rename_unlink(call_frame_t *frame, xlator_t *this)
+{
+    dht_local_t *local = NULL;
+    xlator_t *src_hashed = NULL;
+    xlator_t *src_cached = NULL;
+    xlator_t *dst_hashed = NULL;
+    xlator_t *dst_cached = NULL;
+    xlator_t *rename_subvol = NULL;
+    dict_t *xattr = NULL;
 
-        if ((src_cached == dst_cached) && (dst_hashed != dst_cached)) {
-                link_frame = copy_frame (frame);
-                if (!link_frame) {
-                        goto err;
-                }
+    local = frame->local;
 
-                /* fop value sent as maxvalue because it is not used
-                   anywhere in this case */
-                link_local = dht_local_init (link_frame, &local->loc2, NULL,
-                                             GF_FOP_MAXVALUE);
-                if (!link_local) {
-                        goto err;
-                }
+    src_hashed = local->src_hashed;
+    src_cached = local->src_cached;
+    dst_hashed = local->dst_hashed;
+    dst_cached = local->dst_cached;
 
-                if (link_local->loc.inode)
-                        inode_unref (link_local->loc.inode);
-                link_local->loc.inode = inode_ref (local->loc.inode);
-                uuid_copy (link_local->gfid, local->loc.inode->gfid);
+    local->call_cnt = 0;
 
-                dht_linkfile_create (link_frame, dht_rename_links_create_cbk,
-                                     this, src_cached, dst_hashed,
-                                     &link_local->loc);
-        }
+    /* NOTE: rename_subvol is the same subvolume from which dht_rename_cbk
+     * is called. since rename has already happened on rename_subvol,
+     * unlink shouldn't be sent for oldpath (either linkfile or cached-file)
+     * on rename_subvol. */
+    if (src_cached == dst_cached)
+        rename_subvol = src_cached;
+    else
+        rename_subvol = dst_hashed;
 
-err:
-        /* Merge attrs only from src_cached. In case there of src_cached !=
-         * dst_hashed, this ignores linkfile attrs. */
-        if (prev->this == src_cached) {
-                dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
-                dht_iatt_merge (this, &local->preoldparent, preoldparent,
-                                prev->this);
-                dht_iatt_merge (this, &local->postoldparent, postoldparent,
-                                prev->this);
-                dht_iatt_merge (this, &local->preparent, prenewparent,
-                                prev->this);
-                dht_iatt_merge (this, &local->postparent, postnewparent,
-                                prev->this);
+    /* TODO: delete files in background */
+
+    if (src_cached != dst_hashed && src_cached != dst_cached)
+        local->call_cnt++;
+
+    if (src_hashed != rename_subvol && src_hashed != src_cached)
+        local->call_cnt++;
+
+    if (dst_cached && dst_cached != dst_hashed && dst_cached != src_cached)
+        local->call_cnt++;
+
+    if (local->call_cnt == 0)
+        goto unwind;
+
+    DHT_MARK_FOP_INTERNAL(xattr);
+
+    if (src_cached != dst_hashed && src_cached != dst_cached) {
+        dict_t *xattr_new = NULL;
+
+        xattr_new = dict_copy_with_ref(xattr, NULL);
+
+        gf_msg_trace(this->name, 0, "deleting old src datafile %s @ %s",
+                     local->loc.path, src_cached->name);
+
+        if (gf_uuid_compare(local->loc.pargfid, local->loc2.pargfid) == 0) {
+            DHT_MARKER_DONT_ACCOUNT(xattr_new);
         }
 
+        DHT_CHANGELOG_TRACK_AS_RENAME(xattr_new, &local->loc, &local->loc2);
+        STACK_WIND_COOKIE(frame, dht_rename_unlink_cbk, src_cached, src_cached,
+                          src_cached->fops->unlink, &local->loc, 0, xattr_new);
 
-        /* NOTE: rename_subvol is the same subvolume from which dht_rename_cbk
-         *       is called. since rename has already happened on rename_subvol,
-         *       unlink should not be sent for oldpath (either linkfile or cached-file)
-         *       on rename_subvol. */
-        if (src_cached == dst_cached)
-                rename_subvol = src_cached;
-        else
-                rename_subvol = dst_hashed;
+        dict_unref(xattr_new);
+        xattr_new = NULL;
+    }
 
-        /* TODO: delete files in background */
+    if (src_hashed != rename_subvol && src_hashed != src_cached) {
+        dict_t *xattr_new = NULL;
 
-        if (src_cached != dst_hashed && src_cached != dst_cached)
-                local->call_cnt++;
+        xattr_new = dict_copy_with_ref(xattr, NULL);
 
-        if (src_hashed != rename_subvol && src_hashed != src_cached)
-                local->call_cnt++;
+        gf_msg_trace(this->name, 0, "deleting old src linkfile %s @ %s",
+                     local->loc.path, src_hashed->name);
 
-        if (dst_cached && dst_cached != dst_hashed && dst_cached != src_cached)
-                local->call_cnt++;
+        DHT_MARKER_DONT_ACCOUNT(xattr_new);
 
-        if (local->call_cnt == 0)
-                goto unwind;
+        STACK_WIND_COOKIE(frame, dht_rename_unlink_cbk, src_hashed, src_hashed,
+                          src_hashed->fops->unlink, &local->loc, 0, xattr_new);
 
-        DHT_MARK_FOP_INTERNAL (xattr);
+        dict_unref(xattr_new);
+        xattr_new = NULL;
+    }
 
-        if (src_cached != dst_hashed && src_cached != dst_cached) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "deleting old src datafile %s @ %s",
-                        local->loc.path, src_cached->name);
+    if (dst_cached && (dst_cached != dst_hashed) &&
+        (dst_cached != src_cached)) {
+        gf_msg_trace(this->name, 0, "deleting old dst datafile %s @ %s",
+                     local->loc2.path, dst_cached->name);
 
-                STACK_WIND (frame, dht_rename_unlink_cbk,
-                            src_cached, src_cached->fops->unlink,
-                            &local->loc, 0, xattr);
-        }
+        STACK_WIND_COOKIE(frame, dht_rename_unlink_cbk, dst_cached, dst_cached,
+                          dst_cached->fops->unlink, &local->loc2, 0, xattr);
+    }
+    if (xattr)
+        dict_unref(xattr);
+    return 0;
 
-        if (src_hashed != rename_subvol && src_hashed != src_cached) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "deleting old src linkfile %s @ %s",
-                        local->loc.path, src_hashed->name);
+unwind:
+    WIPE(&local->preoldparent);
+    WIPE(&local->postoldparent);
+    WIPE(&local->preparent);
+    WIPE(&local->postparent);
 
-                STACK_WIND (frame, dht_rename_unlink_cbk,
-                            src_hashed, src_hashed->fops->unlink,
-                            &local->loc, 0, xattr);
-        }
+    dht_rename_done(frame, this);
 
-        if (dst_cached
-            && (dst_cached != dst_hashed)
-            && (dst_cached != src_cached)) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "deleting old dst datafile %s @ %s",
-                        local->loc2.path, dst_cached->name);
+    return 0;
+}
 
-                STACK_WIND (frame, dht_rename_unlink_cbk,
-                            dst_cached, dst_cached->fops->unlink,
-                            &local->loc2, 0, xattr);
+int
+dht_rename_links_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int32_t op_ret, int32_t op_errno, inode_t *inode,
+                            struct iatt *stbuf, struct iatt *preparent,
+                            struct iatt *postparent, dict_t *xdata)
+{
+    xlator_t *prev = NULL;
+    dht_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
+
+    prev = cookie;
+    local = frame->local;
+    main_frame = local->main_frame;
+
+    /* TODO: Handle this case in lookup-optimize */
+    if (op_ret == -1) {
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_CREATE_LINK_FAILED,
+               "link/file %s on %s failed", local->loc.path, prev->name);
+    }
+
+    if (local->linked == _gf_true) {
+        local->linked = _gf_false;
+        dht_linkfile_attr_heal(frame, this);
+    }
+
+    dht_rename_unlink(main_frame, this);
+    DHT_STACK_DESTROY(frame);
+    return 0;
+}
+
+int
+dht_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, struct iatt *stbuf,
+               struct iatt *preoldparent, struct iatt *postoldparent,
+               struct iatt *prenewparent, struct iatt *postnewparent,
+               dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    xlator_t *src_cached = NULL;
+    xlator_t *dst_hashed = NULL;
+    xlator_t *dst_cached = NULL;
+    call_frame_t *link_frame = NULL;
+    dht_local_t *link_local = NULL;
+
+    local = frame->local;
+    prev = cookie;
+
+    src_cached = local->src_cached;
+    dst_hashed = local->dst_hashed;
+    dst_cached = local->dst_cached;
+
+    if (local->linked == _gf_true)
+        FRAME_SU_UNDO(frame, dht_local_t);
+
+    /* It is a critical failure iff we fail to rename the cached file
+     * if the rename of the linkto failed, it is not a critical failure,
+     * and we do not want to lose the created hard link for the new
+     * name as that could have been read by other clients.
+     *
+     * NOTE: If another client is attempting the same oldname -> newname
+     * rename, and finds both file names as existing, and are hard links
+     * to each other, then FUSE would send in an unlink for oldname. In
+     * this time duration if we treat the linkto as a critical error and
+     * unlink the newname we created, we would have effectively lost the
+     * file to rename operations.
+     *
+     * Repercussions of treating this as a non-critical error is that
+     * we could leave behind a stale linkto file and/or not create the new
+     * linkto file, the second case would be rectified by a subsequent
+     * lookup, the first case by a rebalance, like for all stale linkto
+     * files */
+
+    if (op_ret == -1) {
+        /* Critical failure: unable to rename the cached file */
+        if (prev == src_cached) {
+            gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_RENAME_FAILED,
+                   "%s: Rename on %s failed, (gfid = %s) ", local->loc.path,
+                   prev->name,
+                   local->loc.inode ? uuid_utoa(local->loc.inode->gfid) : "");
+            local->op_ret = op_ret;
+            local->op_errno = op_errno;
+            goto cleanup;
+        } else {
+            /* Non-critical failure, unable to rename the linkto
+             * file
+             */
+            gf_msg(this->name, GF_LOG_INFO, op_errno, DHT_MSG_RENAME_FAILED,
+                   "%s: Rename (linkto file) on %s failed, "
+                   "(gfid = %s) ",
+                   local->loc.path, prev->name,
+                   local->loc.inode ? uuid_utoa(local->loc.inode->gfid) : "");
+        }
+    }
+    if (xdata) {
+        if (!local->xattr)
+            local->xattr = dict_ref(xdata);
+        else
+            local->xattr = dict_copy_with_ref(xdata, local->xattr);
+    }
+
+    /* Merge attrs only from src_cached. In case there of src_cached !=
+     * dst_hashed, this ignores linkfile attrs. */
+    if (prev == src_cached) {
+        dht_iatt_merge(this, &local->stbuf, stbuf);
+        dht_iatt_merge(this, &local->preoldparent, preoldparent);
+        dht_iatt_merge(this, &local->postoldparent, postoldparent);
+        dht_iatt_merge(this, &local->preparent, prenewparent);
+        dht_iatt_merge(this, &local->postparent, postnewparent);
+    }
+
+    /* Create the linkto file for the dst file */
+    if ((src_cached == dst_cached) && (dst_hashed != dst_cached)) {
+        link_frame = copy_frame(frame);
+        if (!link_frame) {
+            goto unlink;
         }
-        if (xattr)
-                dict_unref (xattr);
-        return 0;
 
-unwind:
-        WIPE (&local->preoldparent);
-        WIPE (&local->postoldparent);
-        WIPE (&local->preparent);
-        WIPE (&local->postparent);
-        if (xattr)
-                dict_unref (xattr);
+        /* fop value sent as maxvalue because it is not used
+         * anywhere in this case */
+        link_local = dht_local_init(link_frame, &local->loc2, NULL,
+                                    GF_FOP_MAXVALUE);
+        if (!link_local) {
+            goto unlink;
+        }
 
-        dht_rename_done (frame, this);
+        if (link_local->loc.inode)
+            inode_unref(link_local->loc.inode);
+        link_local->loc.inode = inode_ref(local->loc.inode);
+        link_local->main_frame = frame;
+        link_local->stbuf = local->stbuf;
+        gf_uuid_copy(link_local->gfid, local->loc.inode->gfid);
 
+        dht_linkfile_create(link_frame, dht_rename_links_create_cbk, this,
+                            src_cached, dst_hashed, &link_local->loc);
         return 0;
+    }
+
+unlink:
+
+    if (link_frame) {
+        DHT_STACK_DESTROY(link_frame);
+    }
+    dht_rename_unlink(frame, this);
+    return 0;
 
 cleanup:
-        if (xattr)
-                dict_unref (xattr);
-        dht_rename_cleanup (frame);
+    dht_rename_cleanup(frame);
 
-        return 0;
+    return 0;
 }
 
+int
+dht_do_rename(call_frame_t *frame)
+{
+    dht_local_t *local = NULL;
+    xlator_t *dst_hashed = NULL;
+    xlator_t *src_cached = NULL;
+    xlator_t *dst_cached = NULL;
+    xlator_t *this = NULL;
+    xlator_t *rename_subvol = NULL;
+
+    local = frame->local;
+    this = frame->this;
+
+    dst_hashed = local->dst_hashed;
+    dst_cached = local->dst_cached;
+    src_cached = local->src_cached;
+
+    if (src_cached == dst_cached)
+        rename_subvol = src_cached;
+    else
+        rename_subvol = dst_hashed;
+
+    if ((src_cached != dst_hashed) && (rename_subvol == dst_hashed)) {
+        DHT_MARKER_DONT_ACCOUNT(local->xattr_req);
+    }
+
+    if (rename_subvol == src_cached) {
+        DHT_CHANGELOG_TRACK_AS_RENAME(local->xattr_req, &local->loc,
+                                      &local->loc2);
+    }
+
+    gf_msg_trace(this->name, 0, "renaming %s => %s (%s)", local->loc.path,
+                 local->loc2.path, rename_subvol->name);
+
+    if (local->linked == _gf_true)
+        FRAME_SU_DO(frame, dht_local_t);
+    STACK_WIND_COOKIE(frame, dht_rename_cbk, rename_subvol, rename_subvol,
+                      rename_subvol->fops->rename, &local->loc, &local->loc2,
+                      local->xattr_req);
+    return 0;
+}
 
 int
-dht_do_rename (call_frame_t *frame)
+dht_rename_link_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, inode_t *inode,
+                    struct iatt *stbuf, struct iatt *preparent,
+                    struct iatt *postparent, dict_t *xdata)
 {
-        dht_local_t *local = NULL;
-        xlator_t    *dst_hashed = NULL;
-        xlator_t    *src_cached = NULL;
-        xlator_t    *dst_cached = NULL;
-        xlator_t    *this = NULL;
-        xlator_t    *rename_subvol = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
 
+    local = frame->local;
+    prev = cookie;
 
-        local = frame->local;
-        this  = frame->this;
+    if (op_ret == -1) {
+        gf_msg_debug(this->name, 0, "link/file on %s failed (%s)", prev->name,
+                     strerror(op_errno));
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+        local->added_link = _gf_false;
+    } else
+        dht_iatt_merge(this, &local->stbuf, stbuf);
 
-        dst_hashed = local->dst_hashed;
-        dst_cached = local->dst_cached;
-        src_cached = local->src_cached;
+    if (local->op_ret == -1)
+        goto cleanup;
 
-        if (src_cached == dst_cached)
-                rename_subvol = src_cached;
-        else
-                rename_subvol = dst_hashed;
+    dht_do_rename(frame);
 
-        gf_log (this->name, GF_LOG_TRACE,
-                "renaming %s => %s (%s)",
-                local->loc.path, local->loc2.path, rename_subvol->name);
+    return 0;
 
-        if (local->linked == _gf_true)
-                FRAME_SU_DO (frame, dht_local_t);
-        STACK_WIND (frame, dht_rename_cbk,
-                    rename_subvol, rename_subvol->fops->rename,
-                    &local->loc, &local->loc2, NULL);
+cleanup:
+    dht_rename_cleanup(frame);
 
-        return 0;
+    return 0;
 }
 
-
 int
-dht_rename_links_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno,
-                      inode_t *inode, struct iatt *stbuf,
-                      struct iatt *preparent, struct iatt *postparent,
-                      dict_t *xdata)
-{
-        dht_local_t  *local = NULL;
-        call_frame_t *prev = NULL;
-        int           this_call_cnt  = 0;
-
-
-        local = frame->local;
-        prev = cookie;
-
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "link/file on %s failed (%s)",
-                        prev->this->name, strerror (op_errno));
-                local->op_ret   = -1;
-                if (op_errno != ENOENT)
-                        local->op_errno = op_errno;
-        } else if (local->src_cached == prev->this) {
-                /* merge of attr returned only from linkfile creation */
-                dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
-        }
+dht_rename_linkto_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, inode_t *inode,
+                      struct iatt *stbuf, struct iatt *preparent,
+                      struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    xlator_t *src_cached = NULL;
+    dict_t *xattr = NULL;
 
-        this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt)) {
-                if (local->op_ret == -1)
-                        goto cleanup;
+    local = frame->local;
+    DHT_MARK_FOP_INTERNAL(xattr);
+    prev = cookie;
+    src_cached = local->src_cached;
 
-                dht_do_rename (frame);
-        }
+    if (op_ret == -1) {
+        gf_msg_debug(this->name, 0, "link/file on %s failed (%s)", prev->name,
+                     strerror(op_errno));
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+    }
 
-        return 0;
+    /* If linkto creation failed move to failure cleanup code,
+     * instead of continuing with creating the link file */
+    if (local->op_ret != 0) {
+        goto cleanup;
+    }
+
+    gf_msg_trace(this->name, 0, "link %s => %s (%s)", local->loc.path,
+                 local->loc2.path, src_cached->name);
+    if (gf_uuid_compare(local->loc.pargfid, local->loc2.pargfid) == 0) {
+        DHT_MARKER_DONT_ACCOUNT(xattr);
+    }
+
+    local->added_link = _gf_true;
+
+    STACK_WIND_COOKIE(frame, dht_rename_link_cbk, src_cached, src_cached,
+                      src_cached->fops->link, &local->loc, &local->loc2, xattr);
+
+    if (xattr)
+        dict_unref(xattr);
+
+    return 0;
 
 cleanup:
-        dht_rename_cleanup (frame);
+    dht_rename_cleanup(frame);
 
-        return 0;
-}
+    if (xattr)
+        dict_unref(xattr);
 
+    return 0;
+}
 
 int
-dht_rename_unlink_links_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                             int32_t op_ret, int32_t op_errno,
-                             struct iatt *preparent, struct iatt *postparent,
-                             dict_t *xdata)
+dht_rename_unlink_links_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int32_t op_ret, int32_t op_errno,
+                            struct iatt *preparent, struct iatt *postparent,
+                            dict_t *xdata)
 {
-	dht_local_t  *local = NULL;
-	call_frame_t *prev = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
 
+    local = frame->local;
+    prev = cookie;
 
-	local = frame->local;
-	prev = cookie;
+    if ((op_ret == -1) && (op_errno != ENOENT)) {
+        gf_msg_debug(this->name, 0, "unlink of %s on %s failed (%s)",
+                     local->loc2.path, prev->name, strerror(op_errno));
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+    }
 
-	if ((op_ret == -1) && (op_errno != ENOENT)) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"unlink of %s on %s failed (%s)",
-			local->loc2.path, prev->this->name,
-                        strerror (op_errno));
-		local->op_ret   = -1;
-		local->op_errno = op_errno;
-	}
+    if (local->op_ret == -1)
+        goto cleanup;
 
-        if (local->op_ret == -1)
-                goto cleanup;
+    dht_do_rename(frame);
 
-        dht_do_rename (frame);
-
-	return 0;
+    return 0;
 
 cleanup:
-        dht_rename_cleanup (frame);
+    dht_rename_cleanup(frame);
 
-	return 0;
+    return 0;
 }
 
-
 int
-dht_rename_create_links (call_frame_t *frame)
+dht_rename_create_links(call_frame_t *frame)
 {
-        dht_local_t *local = NULL;
-        xlator_t    *this = NULL;
-        xlator_t    *src_hashed = NULL;
-        xlator_t    *src_cached = NULL;
-        xlator_t    *dst_hashed = NULL;
-        xlator_t    *dst_cached = NULL;
-        int          call_cnt = 0;
-        dict_t      *xattr = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *this = NULL;
+    xlator_t *src_hashed = NULL;
+    xlator_t *src_cached = NULL;
+    xlator_t *dst_hashed = NULL;
+    xlator_t *dst_cached = NULL;
+    int call_cnt = 0;
+    dict_t *xattr = NULL;
 
+    local = frame->local;
+    this = frame->this;
 
-        local = frame->local;
-        this  = frame->this;
+    src_hashed = local->src_hashed;
+    src_cached = local->src_cached;
+    dst_hashed = local->dst_hashed;
+    dst_cached = local->dst_cached;
 
-        src_hashed = local->src_hashed;
-        src_cached = local->src_cached;
-        dst_hashed = local->dst_hashed;
-        dst_cached = local->dst_cached;
+    DHT_MARK_FOP_INTERNAL(xattr);
 
-        DHT_MARK_FOP_INTERNAL (xattr);
+    if (src_cached == dst_cached) {
+        dict_t *xattr_new = NULL;
 
-        if (src_cached == dst_cached) {
-                if (dst_hashed == dst_cached)
-                        goto nolinks;
+        if (dst_hashed == dst_cached)
+            goto nolinks;
 
-		gf_log (this->name, GF_LOG_TRACE,
-			"unlinking dst linkfile %s @ %s",
-			local->loc2.path, dst_hashed->name);
+        xattr_new = dict_copy_with_ref(xattr, NULL);
 
-		STACK_WIND (frame, dht_rename_unlink_links_cbk,
-			    dst_hashed, dst_hashed->fops->unlink,
-			    &local->loc2, 0, xattr);
-                return 0;
-        }
+        gf_msg_trace(this->name, 0, "unlinking dst linkfile %s @ %s",
+                     local->loc2.path, dst_hashed->name);
 
-        if (dst_hashed != src_hashed && dst_hashed != src_cached)
-                call_cnt++;
+        DHT_MARKER_DONT_ACCOUNT(xattr_new);
 
-        if (src_cached != dst_hashed)
-                call_cnt++;
+        STACK_WIND_COOKIE(frame, dht_rename_unlink_links_cbk, dst_hashed,
+                          dst_hashed, dst_hashed->fops->unlink, &local->loc2, 0,
+                          xattr_new);
 
-        local->call_cnt = call_cnt;
+        dict_unref(xattr_new);
+        if (xattr)
+            dict_unref(xattr);
+
+        return 0;
+    }
+
+    if (src_cached != dst_hashed) {
+        /* needed to create the link file */
+        call_cnt++;
+        if (dst_hashed != src_hashed)
+            /* needed to create the linkto file */
+            call_cnt++;
+    }
+
+    /* We should not have any failures post the link creation, as this
+     * introduces the newname into the namespace. Clients could have cached
+     * the existence of the newname and may start taking actions based on
+     * the same. Hence create the linkto first, and then attempt the link.
+     *
+     * NOTE: If another client is attempting the same oldname -> newname
+     * rename, and finds both file names as existing, and are hard links
+     * to each other, then FUSE would send in an unlink for oldname. In
+     * this time duration if we treat the linkto as a critical error and
+     * unlink the newname we created, we would have effectively lost the
+     * file to rename operations. */
+    if (dst_hashed != src_hashed && src_cached != dst_hashed) {
+        gf_msg_trace(this->name, 0, "linkfile %s @ %s => %s", local->loc.path,
+                     dst_hashed->name, src_cached->name);
+
+        memcpy(local->gfid, local->loc.inode->gfid, 16);
+        dht_linkfile_create(frame, dht_rename_linkto_cbk, this, src_cached,
+                            dst_hashed, &local->loc);
+    } else if (src_cached != dst_hashed) {
+        dict_t *xattr_new = NULL;
+
+        xattr_new = dict_copy_with_ref(xattr, NULL);
+
+        gf_msg_trace(this->name, 0, "link %s => %s (%s)", local->loc.path,
+                     local->loc2.path, src_cached->name);
+        if (gf_uuid_compare(local->loc.pargfid, local->loc2.pargfid) == 0) {
+            DHT_MARKER_DONT_ACCOUNT(xattr_new);
+        }
 
-        if (dst_hashed != src_hashed && dst_hashed != src_cached) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "linkfile %s @ %s => %s",
-                        local->loc.path, dst_hashed->name, src_cached->name);
-                memcpy (local->gfid, local->loc.inode->gfid, 16);
-		dht_linkfile_create (frame, dht_rename_links_cbk, this,
-				     src_cached, dst_hashed, &local->loc);
-	}
+        local->added_link = _gf_true;
 
-	if (src_cached != dst_hashed) {
-		gf_log (this->name, GF_LOG_TRACE,
-			"link %s => %s (%s)", local->loc.path,
-			local->loc2.path, src_cached->name);
-		STACK_WIND (frame, dht_rename_links_cbk,
-			    src_cached, src_cached->fops->link,
-			    &local->loc, &local->loc2, xattr);
-	}
+        STACK_WIND_COOKIE(frame, dht_rename_link_cbk, src_cached, src_cached,
+                          src_cached->fops->link, &local->loc, &local->loc2,
+                          xattr_new);
+
+        dict_unref(xattr_new);
+    }
 
 nolinks:
-        if (!call_cnt) {
-                /* skip to next step */
-                dht_do_rename (frame);
+    if (!call_cnt) {
+        /* skip to next step */
+        dht_do_rename(frame);
+    }
+    if (xattr)
+        dict_unref(xattr);
+
+    return 0;
+}
+
+int
+dht_rename_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, inode_t *inode,
+                      struct iatt *stbuf, dict_t *xattr,
+                      struct iatt *postparent)
+{
+    dht_local_t *local = NULL;
+    int call_cnt = 0;
+    dht_conf_t *conf = NULL;
+    char gfid_local[GF_UUID_BUF_SIZE] = {0};
+    char gfid_server[GF_UUID_BUF_SIZE] = {0};
+    int child_index = -1;
+    gf_boolean_t is_src = _gf_false;
+    loc_t *loc = NULL;
+
+    child_index = (long)cookie;
+
+    local = frame->local;
+    conf = this->private;
+
+    is_src = (child_index == 0);
+    if (is_src)
+        loc = &local->loc;
+    else
+        loc = &local->loc2;
+
+    if (op_ret >= 0) {
+        if (is_src)
+            local->src_cached = dht_subvol_get_cached(this, local->loc.inode);
+        else {
+            if (loc->inode)
+                gf_uuid_unparse(loc->inode->gfid, gfid_local);
+
+            gf_msg_debug(this->name, 0,
+                         "dst_cached before lookup: %s, "
+                         "(path:%s)(gfid:%s),",
+                         local->loc2.path,
+                         local->dst_cached ? local->dst_cached->name : NULL,
+                         local->dst_cached ? gfid_local : NULL);
+
+            local->dst_cached = dht_subvol_get_cached(this,
+                                                      local->loc2_copy.inode);
+
+            gf_uuid_unparse(stbuf->ia_gfid, gfid_local);
+
+            gf_msg_debug(this->name, GF_LOG_WARNING,
+                         "dst_cached after lookup: %s, "
+                         "(path:%s)(gfid:%s)",
+                         local->loc2.path,
+                         local->dst_cached ? local->dst_cached->name : NULL,
+                         local->dst_cached ? gfid_local : NULL);
+
+            if ((local->loc2.inode == NULL) ||
+                gf_uuid_compare(stbuf->ia_gfid, local->loc2.inode->gfid)) {
+                if (local->loc2.inode != NULL) {
+                    inode_unlink(local->loc2.inode, local->loc2.parent,
+                                 local->loc2.name);
+                    inode_unref(local->loc2.inode);
+                }
+
+                local->loc2.inode = inode_link(local->loc2_copy.inode,
+                                               local->loc2_copy.parent,
+                                               local->loc2_copy.name, stbuf);
+                gf_uuid_copy(local->loc2.gfid, stbuf->ia_gfid);
+            }
+        }
+    }
+
+    if (op_ret < 0) {
+        if (is_src) {
+            /* The meaning of is_linkfile is overloaded here. For locking
+             * to work properly both rebalance and rename should acquire
+             * lock on datafile. The reason for sending this lookup is to
+             * find out whether we've acquired a lock on data file.
+             * Between the lookup before rename and this rename, the
+             * file could be migrated by a rebalance process and now this
+             * file this might be a linkto file. We verify that by sending
+             * this lookup. However, if this lookup fails we cannot really
+             * say whether we've acquired lock on a datafile or linkto file.
+             * So, we act conservatively and _assume_
+             * that this is a linkfile and fail the rename operation.
+             */
+            local->is_linkfile = _gf_true;
+            local->op_errno = op_errno;
+        } else {
+            if (local->dst_cached)
+                gf_msg_debug(this->name, op_errno,
+                             "file %s (gfid:%s) was present "
+                             "(hashed-subvol=%s, "
+                             "cached-subvol=%s) before rename,"
+                             " but lookup failed",
+                             local->loc2.path,
+                             uuid_utoa(local->loc2.inode->gfid),
+                             local->dst_hashed->name, local->dst_cached->name);
+            if (dht_inode_missing(op_errno))
+                local->dst_cached = NULL;
+        }
+    } else if (is_src && xattr &&
+               check_is_linkfile(inode, stbuf, xattr, conf->link_xattr_name)) {
+        local->is_linkfile = _gf_true;
+        /* Found linkto file instead of data file, passdown ENOENT
+         * based on the above comment */
+        local->op_errno = ENOENT;
+    }
+
+    if (!local->is_linkfile && (op_ret >= 0) &&
+        gf_uuid_compare(loc->gfid, stbuf->ia_gfid)) {
+        gf_uuid_unparse(loc->gfid, gfid_local);
+        gf_uuid_unparse(stbuf->ia_gfid, gfid_server);
+
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_GFID_MISMATCH,
+               "path:%s, received a different gfid, local_gfid= %s"
+               " server_gfid: %s",
+               local->loc.path, gfid_local, gfid_server);
+
+        /* Will passdown ENOENT anyway since the file we sent on
+         * rename is replaced with a different file */
+        local->op_errno = ENOENT;
+        /* Since local->is_linkfile is used here to detect failure,
+         * marking this to true */
+        local->is_linkfile = _gf_true;
+    }
+
+    call_cnt = dht_frame_return(frame);
+    if (is_last_call(call_cnt)) {
+        if (local->is_linkfile) {
+            local->op_ret = -1;
+            goto fail;
         }
-        if (xattr)
-                dict_unref (xattr);
 
-        return 0;
-}
+        dht_rename_create_links(frame);
+    }
 
+    return 0;
+fail:
+    dht_rename_unlock(frame, this);
+    return 0;
+}
 
 int
-dht_rename (call_frame_t *frame, xlator_t *this,
-            loc_t *oldloc, loc_t *newloc, dict_t *xdata)
-{
-        xlator_t    *src_cached = NULL;
-        xlator_t    *src_hashed = NULL;
-        xlator_t    *dst_cached = NULL;
-        xlator_t    *dst_hashed = NULL;
-        int          op_errno = -1;
-        int          ret = -1;
-        dht_local_t *local = NULL;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (oldloc, err);
-        VALIDATE_OR_GOTO (newloc, err);
-
-        src_hashed = dht_subvol_get_hashed (this, oldloc);
-        if (!src_hashed) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "no subvolume in layout for path=%s",
-                        oldloc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
+dht_rename_file_lock1_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    char src_gfid[GF_UUID_BUF_SIZE] = {0};
+    char dst_gfid[GF_UUID_BUF_SIZE] = {0};
+    int ret = 0;
+    loc_t *loc = NULL;
+    xlator_t *subvol = NULL;
+
+    local = frame->local;
+
+    if (op_ret < 0) {
+        uuid_utoa_r(local->loc.inode->gfid, src_gfid);
+
+        if (local->loc2.inode)
+            uuid_utoa_r(local->loc2.inode->gfid, dst_gfid);
+
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_INODE_LK_ERROR,
+               "protecting namespace of %s failed"
+               "rename (%s:%s:%s %s:%s:%s)",
+               local->current == &local->lock[0] ? local->loc.path
+                                                 : local->loc2.path,
+               local->loc.path, src_gfid, local->src_hashed->name,
+               local->loc2.path, dst_gfid,
+               local->dst_hashed ? local->dst_hashed->name : NULL);
+
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+        goto err;
+    }
+
+    if (local->current == &local->lock[0]) {
+        loc = &local->loc2;
+        subvol = local->dst_hashed;
+        local->current = &local->lock[1];
+    } else {
+        loc = &local->loc;
+        subvol = local->src_hashed;
+        local->current = &local->lock[0];
+    }
+
+    ret = dht_protect_namespace(frame, loc, subvol, &local->current->ns,
+                                dht_rename_lock_cbk);
+    if (ret < 0) {
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    return 0;
+err:
+    /* No harm in calling an extra unlock */
+    dht_rename_unlock(frame, this);
+    return 0;
+}
 
-        src_cached = dht_subvol_get_cached (this, oldloc->inode);
-        if (!src_cached) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "no cached subvolume for path=%s", oldloc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
+int32_t
+dht_rename_file_protect_namespace(call_frame_t *frame, void *cookie,
+                                  xlator_t *this, int32_t op_ret,
+                                  int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    char src_gfid[GF_UUID_BUF_SIZE] = {0};
+    char dst_gfid[GF_UUID_BUF_SIZE] = {0};
+    int ret = 0;
+    loc_t *loc = NULL;
+    xlator_t *subvol = NULL;
+
+    local = frame->local;
+
+    if (op_ret < 0) {
+        uuid_utoa_r(local->loc.inode->gfid, src_gfid);
+
+        if (local->loc2.inode)
+            uuid_utoa_r(local->loc2.inode->gfid, dst_gfid);
+
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_INODE_LK_ERROR,
+               "acquiring inodelk failed "
+               "rename (%s:%s:%s %s:%s:%s)",
+               local->loc.path, src_gfid, local->src_cached->name,
+               local->loc2.path, dst_gfid,
+               local->dst_cached ? local->dst_cached->name : NULL);
+
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+
+        goto err;
+    }
+
+    /* Locks on src and dst needs to ordered which otherwise might cause
+     * deadlocks when rename (src, dst) and rename (dst, src) is done from
+     * two different clients
+     */
+    ret = dht_order_rename_lock(frame, &loc, &subvol);
+    if (ret) {
+        local->op_errno = ENOMEM;
+        goto err;
+    }
+
+    ret = dht_protect_namespace(frame, loc, subvol, &local->current->ns,
+                                dht_rename_file_lock1_cbk);
+    if (ret < 0) {
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    return 0;
 
-        dst_hashed = dht_subvol_get_hashed (this, newloc);
-        if (!dst_hashed) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "no subvolume in layout for path=%s",
-                        newloc->path);
-                op_errno = EINVAL;
-                goto err;
-        }
+err:
+    /* Its fine to call unlock even when no locks are acquired, as we check
+     * for lock->locked before winding a unlock call.
+     */
+    dht_rename_unlock(frame, this);
 
-        if (newloc->inode)
-                dst_cached = dht_subvol_get_cached (this, newloc->inode);
+    return 0;
+}
 
-        local = dht_local_init (frame, oldloc, NULL, GF_FOP_RENAME);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        /* cached_subvol will be set from dht_local_init, reset it to NULL,
-           as the logic of handling rename is different  */
-        local->cached_subvol = NULL;
-
-        ret = loc_copy (&local->loc2, newloc);
-        if (ret == -1) {
-                op_errno = ENOMEM;
-                goto err;
+int32_t
+dht_rename_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    char src_gfid[GF_UUID_BUF_SIZE] = {0};
+    char dst_gfid[GF_UUID_BUF_SIZE] = {0};
+    dict_t *xattr_req = NULL;
+    dht_conf_t *conf = NULL;
+    int i = 0;
+    xlator_t *subvol = NULL;
+    dht_lock_t *lock = NULL;
+
+    local = frame->local;
+    conf = this->private;
+
+    if (op_ret < 0) {
+        uuid_utoa_r(local->loc.inode->gfid, src_gfid);
+
+        if (local->loc2.inode)
+            uuid_utoa_r(local->loc2.inode->gfid, dst_gfid);
+
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_INODE_LK_ERROR,
+               "protecting namespace of %s failed. "
+               "rename (%s:%s:%s %s:%s:%s)",
+               local->current == &local->lock[0] ? local->loc.path
+                                                 : local->loc2.path,
+               local->loc.path, src_gfid, local->src_hashed->name,
+               local->loc2.path, dst_gfid,
+               local->dst_hashed ? local->dst_hashed->name : NULL);
+
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+
+        goto done;
+    }
+
+    xattr_req = dict_new();
+    if (xattr_req == NULL) {
+        local->op_ret = -1;
+        local->op_errno = ENOMEM;
+        goto done;
+    }
+
+    op_ret = dict_set_uint32(xattr_req, conf->link_xattr_name, 256);
+    if (op_ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = -op_ret;
+        goto done;
+    }
+
+    /* dst_cached might've changed. This normally happens for two reasons:
+     * 1. rebalance migrated dst
+     * 2. Another parallel rename was done overwriting dst
+     *
+     * Doing a lookup on local->loc2 when dst exists, but is associated
+     * with a different gfid will result in an ESTALE error. So, do a fresh
+     * lookup with a new inode on dst-path and handle change of dst-cached
+     * in the cbk. Also, to identify dst-cached changes we do a lookup on
+     * "this" rather than the subvol.
+     */
+    loc_copy(&local->loc2_copy, &local->loc2);
+    inode_unref(local->loc2_copy.inode);
+    local->loc2_copy.inode = inode_new(local->loc.inode->table);
+
+    /* Why not use local->lock.locks[?].loc for lookup post lock phase
+     * ---------------------------------------------------------------
+     * "layout.parent_layout.locks[?].loc" does not have the name and pargfid
+     * populated.
+     * Reason: If we had populated the name and pargfid, server might
+     * resolve to a successful lookup even if there is a file with same name
+     * with a different gfid(unlink & create) as server does name based
+     * resolution on first priority. And this can result in operating on a
+     * different inode entirely.
+     *
+     * Now consider a scenario where source file was renamed by some other
+     * client to a new name just before this lock was granted. So if a
+     * lookup would be done on local->lock[0].layout.parent_layout.locks[?].loc,
+     * server will send success even if the entry was renamed (since server will
+     * do a gfid based resolution). So once a lock is granted, make sure the
+     * file exists with the name that the client requested with.
+     * */
+
+    local->call_cnt = 2;
+    for (i = 0; i < 2; i++) {
+        if (i == 0) {
+            lock = local->rename_inodelk_backward_compatible[0];
+            if (gf_uuid_compare(local->loc.gfid, lock->loc.gfid) == 0)
+                subvol = lock->xl;
+            else {
+                lock = local->rename_inodelk_backward_compatible[1];
+                subvol = lock->xl;
+            }
+        } else {
+            subvol = this;
         }
 
-        local->src_hashed = src_hashed;
-        local->src_cached = src_cached;
-        local->dst_hashed = dst_hashed;
-        local->dst_cached = dst_cached;
+        STACK_WIND_COOKIE(frame, dht_rename_lookup_cbk, (void *)(long)i, subvol,
+                          subvol->fops->lookup,
+                          (i == 0) ? &local->loc : &local->loc2_copy,
+                          xattr_req);
+    }
 
-        gf_log (this->name, GF_LOG_TRACE,
-                "renaming %s (hash=%s/cache=%s) => %s (hash=%s/cache=%s)",
-                oldloc->path, src_hashed->name, src_cached->name,
-                newloc->path, dst_hashed->name,
-                dst_cached ? dst_cached->name : "<nul>");
+    dict_unref(xattr_req);
+    return 0;
 
-        if (IA_ISDIR (oldloc->inode->ia_type)) {
-                dht_rename_dir (frame, this);
-        } else {
-                local->op_ret = 0;
-                dht_rename_create_links (frame);
+done:
+    /* Its fine to call unlock even when no locks are acquired, as we check
+     * for lock->locked before winding a unlock call.
+     */
+    dht_rename_unlock(frame, this);
+
+    if (xattr_req)
+        dict_unref(xattr_req);
+
+    return 0;
+}
+
+int
+dht_rename_lock(call_frame_t *frame)
+{
+    dht_local_t *local = NULL;
+    int count = 1, ret = -1;
+    dht_lock_t **lk_array = NULL;
+
+    local = frame->local;
+
+    if (local->dst_cached)
+        count++;
+
+    lk_array = GF_CALLOC(count, sizeof(*lk_array), gf_common_mt_pointer);
+    if (lk_array == NULL)
+        goto err;
+
+    lk_array[0] = dht_lock_new(frame->this, local->src_cached, &local->loc,
+                               F_WRLCK, DHT_FILE_MIGRATE_DOMAIN, NULL,
+                               FAIL_ON_ANY_ERROR);
+    if (lk_array[0] == NULL)
+        goto err;
+
+    if (local->dst_cached) {
+        /* dst might be removed by the time inodelk reaches bricks,
+         * which can result in ESTALE errors. POSIX imposes no
+         * restriction for dst to be present for renames to be
+         * successful. So, we'll ignore ESTALE errors. As far as
+         * synchronization on dst goes, we'll achieve the same by
+         * holding entrylk on parent directory of dst in the namespace
+         * of basename(dst). Also, there might not be quorum in cluster
+         * xlators like EC/disperse on errno, in which case they return
+         * EIO. For eg., in a disperse (4 + 2), 3 might return success
+         * and three might return ESTALE. Disperse, having no Quorum
+         * unwinds inodelk with EIO. So, ignore EIO too.
+         */
+        lk_array[1] = dht_lock_new(frame->this, local->dst_cached, &local->loc2,
+                                   F_WRLCK, DHT_FILE_MIGRATE_DOMAIN, NULL,
+                                   IGNORE_ENOENT_ESTALE_EIO);
+        if (lk_array[1] == NULL)
+            goto err;
+    }
+
+    local->rename_inodelk_backward_compatible = lk_array;
+    local->rename_inodelk_bc_count = count;
+
+    /* retaining inodelks for the sake of backward compatibility. Please
+     * make sure to remove this inodelk once all of 3.10, 3.12 and 3.13
+     * reach EOL. Better way of getting synchronization would be to acquire
+     * entrylks on src and dst parent directories in the namespace of
+     * basenames of src and dst
+     */
+    ret = dht_blocking_inodelk(frame, lk_array, count,
+                               dht_rename_file_protect_namespace);
+    if (ret < 0) {
+        local->rename_inodelk_backward_compatible = NULL;
+        local->rename_inodelk_bc_count = 0;
+        goto err;
+    }
+
+    return 0;
+err:
+    if (lk_array != NULL) {
+        int tmp_count = 0, i = 0;
+
+        for (i = 0; (i < count) && (lk_array[i]); i++, tmp_count++)
+            ;
+
+        dht_lock_array_free(lk_array, tmp_count);
+        GF_FREE(lk_array);
+    }
+
+    return -1;
+}
+
+int
+dht_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+           dict_t *xdata)
+{
+    xlator_t *src_cached = NULL;
+    xlator_t *src_hashed = NULL;
+    xlator_t *dst_cached = NULL;
+    xlator_t *dst_hashed = NULL;
+    int op_errno = -1;
+    int ret = -1;
+    dht_local_t *local = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    char newgfid[GF_UUID_BUF_SIZE] = {0};
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(oldloc, err);
+    VALIDATE_OR_GOTO(newloc, err);
+
+    gf_uuid_unparse(oldloc->inode->gfid, gfid);
+
+    src_hashed = dht_subvol_get_hashed(this, oldloc);
+    if (!src_hashed) {
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_RENAME_FAILED,
+               "No hashed subvolume in layout for path=%s,"
+               "(gfid = %s)",
+               oldloc->path, gfid);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    src_cached = dht_subvol_get_cached(this, oldloc->inode);
+    if (!src_cached) {
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_RENAME_FAILED,
+               "No cached subvolume for path = %s,"
+               "(gfid = %s)",
+               oldloc->path, gfid);
+
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    dst_hashed = dht_subvol_get_hashed(this, newloc);
+    if (!dst_hashed) {
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_RENAME_FAILED,
+               "No hashed subvolume in layout for path=%s", newloc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (newloc->inode)
+        dst_cached = dht_subvol_get_cached(this, newloc->inode);
+
+    local = dht_local_init(frame, oldloc, NULL, GF_FOP_RENAME);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+    /* cached_subvol will be set from dht_local_init, reset it to NULL,
+       as the logic of handling rename is different  */
+    local->cached_subvol = NULL;
+
+    ret = loc_copy(&local->loc2, newloc);
+    if (ret == -1) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->src_hashed = src_hashed;
+    local->src_cached = src_cached;
+    local->dst_hashed = dst_hashed;
+    local->dst_cached = dst_cached;
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    if (newloc->inode)
+        gf_uuid_unparse(newloc->inode->gfid, newgfid);
+
+    gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_RENAME_INFO,
+           "renaming %s (%s) (hash=%s/cache=%s) => %s (%s) "
+           "(hash=%s/cache=%s) ",
+           oldloc->path, gfid, src_hashed->name, src_cached->name, newloc->path,
+           newloc->inode ? newgfid : NULL, dst_hashed->name,
+           dst_cached ? dst_cached->name : "<nul>");
+
+    if (IA_ISDIR(oldloc->inode->ia_type)) {
+        dht_rename_dir(frame, this);
+    } else {
+        local->op_ret = 0;
+        ret = dht_rename_lock(frame);
+        if (ret < 0) {
+            op_errno = ENOMEM;
+            goto err;
         }
+    }
 
-        return 0;
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (rename, frame, -1, op_errno, NULL, NULL, NULL, NULL,
-                          NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(rename, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL,
+                     NULL);
 
-        return 0;
+    return 0;
+}
+
+int
+dht_pt_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+              dict_t *xdata)
+{
+    gf_boolean_t free_xdata = _gf_false;
+
+    /* Just a pass through */
+    if (!IA_ISDIR(oldloc->inode->ia_type)) {
+        if (!xdata) {
+            free_xdata = _gf_true;
+        }
+        DHT_CHANGELOG_TRACK_AS_RENAME(xdata, oldloc, newloc);
+    }
+    default_rename(frame, this, oldloc, newloc, xdata);
+    if (free_xdata && xdata) {
+        dict_unref(xdata);
+        xdata = NULL;
+    }
+    return 0;
 }
diff --git a/xlators/cluster/dht/src/dht-selfheal.c b/xlators/cluster/dht/src/dht-selfheal.c
index 3fe96b1c716..3e24065227c 100644
--- a/xlators/cluster/dht/src/dht-selfheal.c
+++ b/xlators/cluster/dht/src/dht-selfheal.c
@@ -8,1022 +8,2593 @@
   cases as published by the Free Software Foundation.
 */
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-
-#include "glusterfs.h"
-#include "xlator.h"
-#include "dht-common.h"
-#include "glusterfs-acl.h"
-
-#define DHT_SET_LAYOUT_RANGE(layout,i,srt,chunk,cnt,path)    do {       \
-                layout->list[i].start = srt;                            \
-                layout->list[i].stop  = srt + chunk - 1;                \
-                                                                        \
-                gf_log (this->name, GF_LOG_TRACE,                       \
-                        "gave fix: %u - %u on %s for %s",               \
-                        layout->list[i].start, layout->list[i].stop,    \
-                        layout->list[i].xlator->name, path);            \
-        } while (0)
-
-#define DHT_RESET_LAYOUT_RANGE(layout)    do {                          \
-                int cnt = 0;                                            \
-                for (cnt = 0; cnt < layout->cnt; cnt++ ) {              \
-                        layout->list[cnt].start = 0;                    \
-                        layout->list[cnt].stop  = 0;                    \
-                }                                                       \
-        } while (0)
+#include "dht-lock.h"
+
+#define DHT_SET_LAYOUT_RANGE(layout, i, srt, chunk, path)                      \
+    do {                                                                       \
+        layout->list[i].start = srt;                                           \
+        layout->list[i].stop = srt + chunk - 1;                                \
+        layout->list[i].commit_hash = layout->commit_hash;                     \
+                                                                               \
+        gf_msg_trace(this->name, 0,                                            \
+                     "gave fix: 0x%x - 0x%x, with commit-hash 0x%x"            \
+                     " on %s for %s",                                          \
+                     layout->list[i].start, layout->list[i].stop,              \
+                     layout->list[i].commit_hash,                              \
+                     layout->list[i].xlator->name, path);                      \
+    } while (0)
+
+#define DHT_RESET_LAYOUT_RANGE(layout)                                         \
+    do {                                                                       \
+        int cnt = 0;                                                           \
+        for (cnt = 0; cnt < layout->cnt; cnt++) {                              \
+            layout->list[cnt].start = 0;                                       \
+            layout->list[cnt].stop = 0;                                        \
+        }                                                                      \
+    } while (0)
+
+static int
+dht_selfheal_layout_lock(call_frame_t *frame, dht_layout_t *layout,
+                         gf_boolean_t newdir, dht_selfheal_layout_t healer,
+                         dht_need_heal_t should_heal);
 
 static uint32_t
-dht_overlap_calc (dht_layout_t *old, int o, dht_layout_t *new, int n)
+dht_overlap_calc(dht_layout_t *old, int o, dht_layout_t *new, int n)
 {
-	if (o >= old->cnt || n >= new->cnt)
-		return 0;
+    if (o >= old->cnt || n >= new->cnt)
+        return 0;
 
-	if (old->list[o].err > 0 || new->list[n].err > 0)
-		return 0;
+    if (old->list[o].err > 0 || new->list[n].err > 0)
+        return 0;
 
-        if (old->list[o].start == old->list[o].stop) {
-                return 0;
-        }
+    if (old->list[o].start == old->list[o].stop) {
+        return 0;
+    }
 
-        if (new->list[n].start == new->list[n].stop) {
-                return 0;
-        }
+    if (new->list[n].start == new->list[n].stop) {
+        return 0;
+    }
 
-	if ((old->list[o].start > new->list[n].stop) ||
-	    (old->list[o].stop < new->list[n].start))
-		return 0;
+    if ((old->list[o].start > new->list[n].stop) ||
+        (old->list[o].stop < new->list[n].start))
+        return 0;
 
-	return min (old->list[o].stop, new->list[n].stop) -
-	        max (old->list[o].start, new->list[n].start) + 1;
+    return min(old->list[o].stop, new->list[n].stop) -
+           max(old->list[o].start, new->list[n].start) + 1;
 }
 
+int
+dht_selfheal_unlock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    DHT_STACK_DESTROY(frame);
+    return 0;
+}
 
 int
-dht_selfheal_dir_finish (call_frame_t *frame, xlator_t *this, int ret)
+dht_selfheal_dir_finish(call_frame_t *frame, xlator_t *this, int ret,
+                        int invoke_cbk)
 {
-        dht_local_t  *local = NULL;
+    dht_local_t *local = NULL, *lock_local = NULL;
+    call_frame_t *lock_frame = NULL;
+    int lock_count = 0;
 
-        local = frame->local;
-        local->selfheal.dir_cbk (frame, NULL, frame->this, ret,
-                                 local->op_errno, NULL);
+    local = frame->local;
 
-        return 0;
-}
+    /* Unlock entrylk */
+    dht_unlock_entrylk_wrapper(frame, &local->lock[0].ns.directory_ns);
+
+    /* Unlock inodelk */
+    lock_count = dht_lock_count(local->lock[0].ns.parent_layout.locks,
+                                local->lock[0].ns.parent_layout.lk_count);
+    if (lock_count == 0)
+        goto done;
+
+    lock_frame = copy_frame(frame);
+    if (lock_frame == NULL) {
+        goto done;
+    }
+
+    lock_local = dht_local_init(lock_frame, &local->loc, NULL,
+                                lock_frame->root->op);
+    if (lock_local == NULL) {
+        goto done;
+    }
 
+    lock_local->lock[0].ns.parent_layout.locks = local->lock[0]
+                                                     .ns.parent_layout.locks;
+    lock_local->lock[0]
+        .ns.parent_layout.lk_count = local->lock[0].ns.parent_layout.lk_count;
+
+    local->lock[0].ns.parent_layout.locks = NULL;
+    local->lock[0].ns.parent_layout.lk_count = 0;
+
+    dht_unlock_inodelk(lock_frame, lock_local->lock[0].ns.parent_layout.locks,
+                       lock_local->lock[0].ns.parent_layout.lk_count,
+                       dht_selfheal_unlock_cbk);
+    lock_frame = NULL;
+
+done:
+    if (invoke_cbk)
+        local->selfheal.dir_cbk(frame, NULL, frame->this, ret, local->op_errno,
+                                NULL);
+    if (lock_frame != NULL) {
+        DHT_STACK_DESTROY(lock_frame);
+    }
+
+    return 0;
+}
 
 int
-dht_selfheal_dir_xattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                            int op_ret, int op_errno, dict_t *xdata)
+dht_refresh_layout_done(call_frame_t *frame)
 {
-        dht_local_t  *local = NULL;
-        call_frame_t *prev = NULL;
-        xlator_t     *subvol = NULL;
-        int           i = 0;
-        dht_layout_t *layout = NULL;
-        int           err = 0;
-        int           this_call_cnt = 0;
-
-        local = frame->local;
-        layout = local->selfheal.layout;
-        prev = cookie;
-        subvol = prev->this;
-
-        if (op_ret == 0)
-                err = 0;
-        else
-                err = op_errno;
+    int ret = -1;
+    dht_layout_t *refreshed = NULL, *heal = NULL;
+    dht_local_t *local = NULL;
+    dht_need_heal_t should_heal = NULL;
+    dht_selfheal_layout_t healer = NULL;
 
-        for (i = 0; i < layout->cnt; i++) {
-                if (layout->list[i].xlator == subvol) {
-                        layout->list[i].err = err;
-                        break;
-                }
-        }
+    local = frame->local;
 
-        this_call_cnt = dht_frame_return (frame);
+    refreshed = local->selfheal.refreshed_layout;
+    heal = local->selfheal.layout;
 
-        if (is_last_call (this_call_cnt)) {
-                dht_selfheal_dir_finish (frame, this, 0);
-        }
+    healer = local->selfheal.healer;
+    should_heal = local->selfheal.should_heal;
 
-        return 0;
-}
+    ret = dht_layout_sort(refreshed);
+    if (ret == -1) {
+        gf_smsg(frame->this->name, GF_LOG_WARNING, 0,
+                DHT_MSG_LAYOUT_SORT_FAILED, NULL);
+        goto err;
+    }
+
+    if (should_heal(frame, &heal, &refreshed)) {
+        healer(frame, &local->loc, heal);
+    } else {
+        local->selfheal.layout = NULL;
+        local->selfheal.refreshed_layout = NULL;
+        local->selfheal.layout = refreshed;
+
+        dht_layout_unref(frame->this, heal);
+
+        dht_selfheal_dir_finish(frame, frame->this, 0, 1);
+    }
+
+    return 0;
 
+err:
+    dht_selfheal_dir_finish(frame, frame->this, -1, 1);
+    return 0;
+}
 
 int
-dht_selfheal_dir_xattr_persubvol (call_frame_t *frame, loc_t *loc,
-                                  dht_layout_t *layout, int i,
-                                  xlator_t *req_subvol)
+dht_refresh_layout_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno, inode_t *inode,
+                       struct iatt *stbuf, dict_t *xattr,
+                       struct iatt *postparent)
 {
-        xlator_t          *subvol = NULL;
-        dict_t            *xattr = NULL;
-        int                ret = 0;
-        xlator_t          *this = NULL;
-        int32_t           *disk_layout = NULL;
-        dht_local_t       *local = NULL;
-        dht_conf_t        *conf = NULL;
-
-        local = frame->local;
-        if (req_subvol)
-                subvol = req_subvol;
-        else
-                subvol = layout->list[i].xlator;
-        this = frame->this;
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+    dht_layout_t *layout = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO("dht", this, err);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, err);
+    GF_VALIDATE_OR_GOTO("dht", this->private, err);
+
+    local = frame->local;
+    prev = cookie;
+
+    layout = local->selfheal.refreshed_layout;
+
+    LOCK(&frame->lock);
+    {
+        op_ret = dht_layout_merge(this, layout, prev, op_ret, op_errno, xattr);
+
+        dht_iatt_merge(this, &local->stbuf, stbuf);
+
+        if (op_ret == -1) {
+            gf_uuid_unparse(local->loc.gfid, gfid);
+            local->op_errno = op_errno;
+            gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                    DHT_MSG_FILE_LOOKUP_FAILED, "path=%s", local->loc.path,
+                    "name=%s", prev->name, "gfid=%s", gfid, NULL);
+
+            goto unlock;
+        }
 
-        GF_VALIDATE_OR_GOTO ("", this, err);
-        GF_VALIDATE_OR_GOTO (this->name, layout, err);
-        GF_VALIDATE_OR_GOTO (this->name, local, err);
-        GF_VALIDATE_OR_GOTO (this->name, subvol, err);
-        VALIDATE_OR_GOTO (this->private, err);
+        local->op_ret = 0;
+    }
+unlock:
+    UNLOCK(&frame->lock);
 
-        conf = this->private;
+    this_call_cnt = dht_frame_return(frame);
 
-        xattr = get_new_dict ();
-        if (!xattr) {
-                goto err;
+    if (is_last_call(this_call_cnt)) {
+        if (local->op_ret == 0) {
+            local->refresh_layout_done(frame);
+        } else {
+            goto err;
         }
+    }
 
-        ret = dht_disk_layout_extract (this, layout, i, &disk_layout);
-        if (ret == -1) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: (subvol %s) failed to extract disk layout",
-                        loc->path, subvol->name);
-                goto err;
-        }
+    return 0;
 
-        ret = dict_set_bin (xattr, conf->xattr_name, disk_layout, 4 * 4);
-        if (ret == -1) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: (subvol %s) failed to set xattr dictionary",
-                        loc->path, subvol->name);
-                goto err;
+err:
+    if (local) {
+        local->refresh_layout_unlock(frame, this, -1, 1);
+    }
+    return 0;
+}
+
+int
+dht_refresh_layout(call_frame_t *frame)
+{
+    int call_cnt = 0;
+    int i = 0, ret = -1;
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *this = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+
+    this = frame->this;
+    conf = this->private;
+    local = frame->local;
+
+    call_cnt = conf->subvolume_cnt;
+    local->call_cnt = call_cnt;
+    local->op_ret = -1;
+
+    if (local->selfheal.refreshed_layout) {
+        dht_layout_unref(this, local->selfheal.refreshed_layout);
+        local->selfheal.refreshed_layout = NULL;
+    }
+
+    local->selfheal.refreshed_layout = dht_layout_new(this,
+                                                      conf->subvolume_cnt);
+    if (!local->selfheal.refreshed_layout) {
+        gf_uuid_unparse(local->loc.gfid, gfid);
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_MEM_ALLOC_FAILED,
+                "path=%s", local->loc.path, "gfid=%s", gfid, NULL);
+        goto out;
+    }
+
+    if (local->xattr != NULL) {
+        dict_del(local->xattr, conf->xattr_name);
+    }
+
+    if (local->xattr_req == NULL) {
+        gf_uuid_unparse(local->loc.gfid, gfid);
+        local->xattr_req = dict_new();
+        if (local->xattr_req == NULL) {
+            gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+                    "path=%s", local->loc.path, "gfid=%s", gfid, NULL);
+            goto out;
         }
-        disk_layout = NULL;
+    }
 
-        gf_log (this->name, GF_LOG_TRACE,
-                "setting hash range %u - %u (type %d) on subvolume %s for %s",
-                layout->list[i].start, layout->list[i].stop,
-                layout->type, subvol->name, loc->path);
+    if (dict_get(local->xattr_req, conf->xattr_name) == 0) {
+        ret = dict_set_uint32(local->xattr_req, conf->xattr_name, 4 * 4);
+        if (ret)
+            gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+                    "path=%s", local->loc.path, "key=%s", conf->xattr_name,
+                    NULL);
+    }
 
-        dict_ref (xattr);
+    for (i = 0; i < call_cnt; i++) {
+        STACK_WIND_COOKIE(frame, dht_refresh_layout_cbk, conf->subvolumes[i],
+                          conf->subvolumes[i],
+                          conf->subvolumes[i]->fops->lookup, &local->loc,
+                          local->xattr_req);
+    }
 
-        if (!uuid_is_null (local->gfid))
-                uuid_copy (loc->gfid, local->gfid);
+    return 0;
 
-        STACK_WIND (frame, dht_selfheal_dir_xattr_cbk,
-                    subvol, subvol->fops->setxattr,
-                    loc, xattr, 0, NULL);
+out:
+    if (local) {
+        local->refresh_layout_unlock(frame, this, -1, 1);
+    }
+    return 0;
+}
 
-        dict_unref (xattr);
+int32_t
+dht_selfheal_layout_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                             int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
 
-        return 0;
+    local = frame->local;
+
+    if (!local) {
+        goto err;
+    }
+
+    if (op_ret < 0) {
+        local->op_errno = op_errno;
+        goto err;
+    }
+
+    local->refresh_layout_unlock = dht_selfheal_dir_finish;
+    local->refresh_layout_done = dht_refresh_layout_done;
+
+    dht_refresh_layout(frame);
+    return 0;
 
 err:
-        if (xattr)
-                dict_destroy (xattr);
+    dht_selfheal_dir_finish(frame, this, -1, 1);
+    return 0;
+}
 
-        GF_FREE (disk_layout);
+gf_boolean_t
+dht_should_heal_layout(call_frame_t *frame, dht_layout_t **heal,
+                       dht_layout_t **ondisk)
+{
+    gf_boolean_t fixit = _gf_true;
+    dht_local_t *local = NULL;
+    int heal_missing_dirs = 0;
+
+    local = frame->local;
+
+    if ((heal == NULL) || (*heal == NULL) || (ondisk == NULL) ||
+        (*ondisk == NULL))
+        goto out;
+
+    dht_layout_anomalies(
+        frame->this, &local->loc, *ondisk, &local->selfheal.hole_cnt,
+        &local->selfheal.overlaps_cnt, &local->selfheal.missing_cnt,
+        &local->selfheal.down, &local->selfheal.misc, NULL);
+
+    /* Directories might've been created as part of this self-heal. We've to
+     * sync non-layout xattrs and set range 0-0 on new directories
+     */
+    heal_missing_dirs = local->selfheal.force_mkdir
+                            ? local->selfheal.force_mkdir
+                            : dht_layout_missing_dirs(*heal);
+
+    if ((local->selfheal.hole_cnt == 0) &&
+        (local->selfheal.overlaps_cnt == 0) && heal_missing_dirs) {
+        dht_layout_t *tmp = NULL;
+
+        /* Just added a brick and need to set 0-0 range on this brick.
+         * But ondisk layout is well-formed. So, swap layouts "heal" and
+         * "ondisk". Now "ondisk" layout will be used for healing
+         * xattrs. If there are any non-participating subvols in
+         * "ondisk" layout, dht_selfheal_dir_xattr_persubvol will set
+         * 0-0 and non-layout xattrs. This way we won't end up in
+         * "corrupting" already set and well-formed "ondisk" layout.
+         */
+        tmp = *heal;
+        *heal = *ondisk;
+        *ondisk = tmp;
+
+        /* Current selfheal code, heals non-layout xattrs only after
+         * an add-brick. In fact non-layout xattrs are considered as
+         * secondary citizens which are healed only if layout xattrs
+         * need to be healed. This is wrong, since for eg., quota can be
+         * set when layout is well-formed, but a node is down. Also,
+         * just for healing non-layout xattrs, we don't need locking.
+         * This issue is _NOT FIXED_ by this patch.
+         */
+    }
+
+    fixit = (local->selfheal.hole_cnt || local->selfheal.overlaps_cnt ||
+             heal_missing_dirs);
 
-        dht_selfheal_dir_xattr_cbk (frame, subvol, frame->this,
-                                    -1, ENOMEM, NULL);
-        return 0;
+out:
+    return fixit;
 }
 
 int
-dht_fix_dir_xattr (call_frame_t *frame, loc_t *loc, dht_layout_t *layout)
+dht_layout_span(dht_layout_t *layout)
 {
-        dht_local_t *local = NULL;
-        int          i = 0;
-        int          count = 0;
-        xlator_t    *this = NULL;
-        dht_conf_t  *conf = NULL;
-        dht_layout_t *dummy = NULL;
+    int i = 0, count = 0;
 
-        local = frame->local;
-        this = frame->this;
-        conf = this->private;
+    for (i = 0; i < layout->cnt; i++) {
+        if (layout->list[i].err)
+            continue;
 
-        gf_log (this->name, GF_LOG_DEBUG,
-                "writing the new range for all subvolumes");
+        if (layout->list[i].start != layout->list[i].stop)
+            count++;
+    }
 
-        local->call_cnt = count = conf->subvolume_cnt;
+    return count;
+}
 
-        for (i = 0; i < layout->cnt; i++) {
-                dht_selfheal_dir_xattr_persubvol (frame, loc, layout, i, NULL);
+int
+dht_decommissioned_bricks_in_layout(xlator_t *this, dht_layout_t *layout)
+{
+    dht_conf_t *conf = NULL;
+    int count = 0, i = 0, j = 0;
 
-                if (--count == 0)
-                        goto out;
+    if ((this == NULL) || (layout == NULL))
+        goto out;
+
+    conf = this->private;
+
+    for (i = 0; i < layout->cnt; i++) {
+        for (j = 0; j < conf->subvolume_cnt; j++) {
+            if (conf->decommissioned_bricks[j] &&
+                conf->decommissioned_bricks[j] == layout->list[i].xlator) {
+                count++;
+            }
         }
-        /* if we are here, subvolcount > layout_count. subvols-per-directory
-         * option might be set here. We need to clear out layout from the
-         * non-participating subvolumes, else it will result in overlaps */
-        dummy = dht_layout_new (this, 1);
-        if (!dummy)
-                goto out;
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                if (_gf_false ==
-                    dht_is_subvol_in_layout (layout, conf->subvolumes[i])) {
-                        dht_selfheal_dir_xattr_persubvol (frame, loc, dummy, 0,
-                                                          conf->subvolumes[i]);
-                        if (--count == 0)
-                                break;
-                }
+    }
+
+out:
+    return count;
+}
+
+dht_distribution_type_t
+dht_distribution_type(xlator_t *this, dht_layout_t *layout)
+{
+    dht_distribution_type_t type = GF_DHT_EQUAL_DISTRIBUTION;
+    int i = 0;
+    uint32_t start_range = 0, range = 0, diff = 0;
+
+    if ((this == NULL) || (layout == NULL) || (layout->cnt < 1)) {
+        goto out;
+    }
+
+    for (i = 0; i < layout->cnt; i++) {
+        if (start_range == 0) {
+            start_range = layout->list[i].stop - layout->list[i].start;
+            continue;
         }
 
-        dht_layout_unref (this, dummy);
+        range = layout->list[i].stop - layout->list[i].start;
+        diff = (range >= start_range) ? range - start_range
+                                      : start_range - range;
+
+        if ((range != 0) && (diff > layout->cnt)) {
+            type = GF_DHT_WEIGHTED_DISTRIBUTION;
+            break;
+        }
+    }
+
 out:
-        return 0;
+    return type;
 }
 
-int
-dht_selfheal_dir_xattr (call_frame_t *frame, loc_t *loc, dht_layout_t *layout)
+gf_boolean_t
+dht_should_fix_layout(call_frame_t *frame, dht_layout_t **inmem,
+                      dht_layout_t **ondisk)
 {
-        dht_local_t *local = NULL;
-        int          missing_xattr = 0;
-        int          i = 0;
-        xlator_t    *this = NULL;
-        dht_conf_t   *conf = NULL;
-        dht_layout_t *dummy = NULL;
+    gf_boolean_t fixit = _gf_true;
 
-        local = frame->local;
-        this = frame->this;
-        conf = this->private;
+    dht_local_t *local = NULL;
+    int layout_span = 0;
+    int decommissioned_bricks = 0;
+    dht_conf_t *conf = NULL;
+    dht_distribution_type_t inmem_dist_type = 0;
+    dht_distribution_type_t ondisk_dist_type = 0;
 
-        for (i = 0; i < layout->cnt; i++) {
-                if (layout->list[i].err != -1 || !layout->list[i].stop) {
-                        /* err != -1 would mean xattr present on the directory
-                         * or the directory is non existent.
-                         * !layout->list[i].stop would mean layout absent
-                         */
+    conf = frame->this->private;
 
-                        continue;
-                }
-                missing_xattr++;
-        }
+    local = frame->local;
 
-        gf_log (this->name, GF_LOG_TRACE,
-                "%d subvolumes missing xattr for %s",
-                missing_xattr, loc->path);
+    if ((inmem == NULL) || (*inmem == NULL) || (ondisk == NULL) ||
+        (*ondisk == NULL))
+        goto out;
 
-        if (missing_xattr == 0) {
-                dht_selfheal_dir_finish (frame, this, 0);
-                return 0;
-        }
+    dht_layout_anomalies(frame->this, &local->loc, *ondisk,
+                         &local->selfheal.hole_cnt,
+                         &local->selfheal.overlaps_cnt, NULL,
+                         &local->selfheal.down, &local->selfheal.misc, NULL);
 
-        local->call_cnt = missing_xattr;
+    if (local->selfheal.down || local->selfheal.misc) {
+        fixit = _gf_false;
+        goto out;
+    }
 
-        for (i = 0; i < layout->cnt; i++) {
-                if (layout->list[i].err != -1 || !layout->list[i].stop)
-                        continue;
+    if (local->selfheal.hole_cnt || local->selfheal.overlaps_cnt)
+        goto out;
 
-                dht_selfheal_dir_xattr_persubvol (frame, loc, layout, i, NULL);
+    /* If commit hashes are being updated, let it through */
+    if ((*inmem)->commit_hash != (*ondisk)->commit_hash)
+        goto out;
+
+    layout_span = dht_layout_span(*ondisk);
+
+    decommissioned_bricks = dht_decommissioned_bricks_in_layout(frame->this,
+                                                                *ondisk);
+    inmem_dist_type = dht_distribution_type(frame->this, *inmem);
+    ondisk_dist_type = dht_distribution_type(frame->this, *ondisk);
+
+    if ((decommissioned_bricks == 0) &&
+        (layout_span ==
+         (conf->subvolume_cnt - conf->decommission_subvols_cnt)) &&
+        (inmem_dist_type == ondisk_dist_type))
+        fixit = _gf_false;
 
-                if (--missing_xattr == 0)
-                        break;
-        }
-        dummy = dht_layout_new (this, 1);
-        if (!dummy)
-                goto out;
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                if (_gf_false ==
-                    dht_is_subvol_in_layout (layout, conf->subvolumes[i])) {
-                        dht_selfheal_dir_xattr_persubvol (frame, loc, dummy, 0,
-                                                          conf->subvolumes[i]);
-                }
-        }
-        dht_layout_unref (this, dummy);
 out:
-        return 0;
+
+    return fixit;
 }
 
-int
-dht_selfheal_dir_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                              int op_ret, int op_errno, struct iatt *statpre,
-                              struct iatt *statpost, dict_t *xdata)
+static int
+dht_selfheal_layout_lock(call_frame_t *frame, dht_layout_t *layout,
+                         gf_boolean_t newdir, dht_selfheal_layout_t healer,
+                         dht_need_heal_t should_heal)
 {
-        dht_local_t   *local = NULL;
-        dht_layout_t  *layout = NULL;
-        int            this_call_cnt = 0;
+    dht_local_t *local = NULL;
+    int count = 1, ret = -1, i = 0;
+    dht_lock_t **lk_array = NULL;
+    dht_conf_t *conf = NULL;
+    dht_layout_t *tmp = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
 
-        local  = frame->local;
-        layout = local->selfheal.layout;
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO(frame->this->name, frame->local, err);
 
-        this_call_cnt = dht_frame_return (frame);
+    local = frame->local;
 
-        if (is_last_call (this_call_cnt)) {
-                dht_selfheal_dir_xattr (frame, &local->loc, layout);
-        }
+    conf = frame->this->private;
 
-        return 0;
-}
+    local->selfheal.healer = healer;
+    local->selfheal.should_heal = should_heal;
 
+    tmp = local->selfheal.layout;
+    local->selfheal.layout = dht_layout_ref(frame->this, layout);
+    dht_layout_unref(frame->this, tmp);
 
-int
-dht_selfheal_dir_setattr (call_frame_t *frame, loc_t *loc, struct iatt *stbuf,
-                          int32_t valid, dht_layout_t *layout)
-{
-        int           missing_attr = 0;
-        int           i     = 0;
-        dht_local_t  *local = NULL;
-        xlator_t     *this = NULL;
+    if (!newdir) {
+        count = conf->subvolume_cnt;
 
-        local = frame->local;
-        this = frame->this;
+        lk_array = GF_CALLOC(count, sizeof(*lk_array), gf_common_mt_char);
+        if (lk_array == NULL) {
+            gf_uuid_unparse(local->stbuf.ia_gfid, gfid);
+            gf_smsg("dht", GF_LOG_ERROR, ENOMEM, DHT_MSG_MEM_ALLOC_FAILED,
+                    "lk_array-gfid=%s", gfid, "path=%s", local->loc.path, NULL);
+            goto err;
+        }
 
-        for (i = 0; i < layout->cnt; i++) {
-                if (layout->list[i].err == -1)
-                        missing_attr++;
+        for (i = 0; i < count; i++) {
+            lk_array[i] = dht_lock_new(
+                frame->this, conf->subvolumes[i], &local->loc, F_WRLCK,
+                DHT_LAYOUT_HEAL_DOMAIN, NULL, FAIL_ON_ANY_ERROR);
+            if (lk_array[i] == NULL) {
+                gf_uuid_unparse(local->stbuf.ia_gfid, gfid);
+                gf_smsg(THIS->name, GF_LOG_ERROR, ENOMEM,
+                        DHT_MSG_MEM_ALLOC_FAILED, "lk_array-gfid=%s", gfid,
+                        "path=%s", local->loc.path, NULL);
+                goto err;
+            }
+        }
+    } else {
+        count = 1;
+        lk_array = GF_CALLOC(count, sizeof(*lk_array), gf_common_mt_char);
+        if (lk_array == NULL) {
+            gf_uuid_unparse(local->stbuf.ia_gfid, gfid);
+            gf_smsg(THIS->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_MEM_ALLOC_FAILED,
+                    "lk_array-gfid=%s", gfid, "path=%s", local->loc.path, NULL);
+            goto err;
         }
 
-        if (missing_attr == 0) {
-                dht_selfheal_dir_xattr (frame, loc, layout);
-                return 0;
+        lk_array[0] = dht_lock_new(frame->this, local->hashed_subvol,
+                                   &local->loc, F_WRLCK, DHT_LAYOUT_HEAL_DOMAIN,
+                                   NULL, FAIL_ON_ANY_ERROR);
+        if (lk_array[0] == NULL) {
+            gf_uuid_unparse(local->stbuf.ia_gfid, gfid);
+            gf_smsg(THIS->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_MEM_ALLOC_FAILED,
+                    "lk_array-gfid=%s", gfid, "path=%s", local->loc.path, NULL);
+            goto err;
         }
+    }
 
-        if (!uuid_is_null (local->gfid))
-                uuid_copy (loc->gfid, local->gfid);
+    local->lock[0].layout.my_layout.locks = lk_array;
+    local->lock[0].layout.my_layout.lk_count = count;
 
-        local->call_cnt = missing_attr;
-        for (i = 0; i < layout->cnt; i++) {
-                if (layout->list[i].err == -1) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "setattr for %s on subvol %s",
-                                loc->path, layout->list[i].xlator->name);
-
-                        STACK_WIND (frame, dht_selfheal_dir_setattr_cbk,
-                                    layout->list[i].xlator,
-                                    layout->list[i].xlator->fops->setattr,
-                                    loc, stbuf, valid, NULL);
-                }
+    ret = dht_blocking_inodelk(frame, lk_array, count,
+                               dht_selfheal_layout_lock_cbk);
+    if (ret < 0) {
+        local->lock[0].layout.my_layout.locks = NULL;
+        local->lock[0].layout.my_layout.lk_count = 0;
+        goto err;
+    }
+
+    return 0;
+err:
+    if (lk_array != NULL) {
+        dht_lock_array_free(lk_array, count);
+        GF_FREE(lk_array);
+    }
+
+    return -1;
+}
+
+static int
+dht_selfheal_dir_xattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int op_ret, int op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *subvol = NULL;
+    struct iatt *stbuf = NULL;
+    int i = 0;
+    int ret = 0;
+    dht_layout_t *layout = NULL;
+    int err = 0;
+    int this_call_cnt = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    layout = local->selfheal.layout;
+    subvol = cookie;
+
+    if (op_ret == 0) {
+        err = 0;
+    } else {
+        gf_uuid_unparse(local->loc.gfid, gfid);
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                DHT_MSG_DIR_SELFHEAL_XATTR_FAILED, "name=%s", subvol->name,
+                "path=%s", local->loc.path, "gfid=%s", gfid, NULL);
+        err = op_errno;
+    }
+
+    ret = dict_get_bin(xdata, DHT_IATT_IN_XDATA_KEY, (void **)&stbuf);
+    if (ret < 0) {
+        gf_uuid_unparse(local->loc.gfid, gfid);
+        gf_msg_debug(this->name, 0,
+                     "key = %s not present in dict"
+                     ", path:%s gfid:%s",
+                     DHT_IATT_IN_XDATA_KEY, local->loc.path, gfid);
+    }
+
+    for (i = 0; i < layout->cnt; i++) {
+        if (layout->list[i].xlator == subvol) {
+            layout->list[i].err = err;
+            break;
         }
+    }
 
-        return 0;
+    LOCK(&frame->lock);
+    {
+        dht_iatt_merge(this, &local->stbuf, stbuf);
+    }
+    UNLOCK(&frame->lock);
+
+    this_call_cnt = dht_frame_return(frame);
+
+    if (is_last_call(this_call_cnt)) {
+        dht_selfheal_dir_finish(frame, this, 0, 1);
+    }
+
+    return 0;
 }
 
+/* Code is required to set user xattr to local->xattr
+ */
 int
-dht_selfheal_dir_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                            int op_ret, int op_errno,
-                            inode_t *inode, struct iatt *stbuf,
-                            struct iatt *preparent, struct iatt *postparent,
-                            dict_t *xdata)
+dht_set_user_xattr(dict_t *dict, char *k, data_t *v, void *data)
 {
-        dht_local_t   *local = NULL;
-        dht_layout_t  *layout = NULL;
-        call_frame_t  *prev = NULL;
-        xlator_t      *subvol = NULL;
-        int            i = 0;
-        int            this_call_cnt = 0;
-
-
-        local  = frame->local;
-        layout = local->selfheal.layout;
-        prev   = cookie;
-        subvol = prev->this;
-
-        if ((op_ret == 0) || ((op_ret == -1) && (op_errno == EEXIST))) {
-                for (i = 0; i < layout->cnt; i++) {
-                        if (layout->list[i].xlator == subvol) {
-                                layout->list[i].err = -1;
-                                break;
-                        }
-                }
-        }
+    dict_t *set_xattr = data;
+    int ret = -1;
 
-        if (op_ret) {
-                gf_log (this->name, ((op_errno == EEXIST) ? GF_LOG_DEBUG :
-                                     GF_LOG_WARNING),
-                        "selfhealing directory %s failed: %s",
-                        local->loc.path, strerror (op_errno));
-                goto out;
+    ret = dict_set(set_xattr, k, v);
+    return ret;
+}
+
+static int
+dht_selfheal_dir_xattr_persubvol(call_frame_t *frame, loc_t *loc,
+                                 dht_layout_t *layout, int i,
+                                 xlator_t *req_subvol)
+{
+    xlator_t *subvol = NULL;
+    dict_t *xattr = NULL;
+    dict_t *xdata = NULL;
+    int ret = 0;
+    xlator_t *this = NULL;
+    int32_t *disk_layout = NULL;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    data_t *data = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    if (req_subvol)
+        subvol = req_subvol;
+    else
+        subvol = layout->list[i].xlator;
+    this = frame->this;
+
+    GF_VALIDATE_OR_GOTO("", this, err);
+    GF_VALIDATE_OR_GOTO(this->name, layout, err);
+    GF_VALIDATE_OR_GOTO(this->name, local, err);
+    GF_VALIDATE_OR_GOTO(this->name, subvol, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+
+    xattr = dict_new();
+    if (!xattr) {
+        goto err;
+    }
+
+    xdata = dict_new();
+    if (!xdata)
+        goto err;
+
+    ret = dict_set_str(xdata, GLUSTERFS_INTERNAL_FOP_KEY, "yes");
+    if (ret < 0) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+                "path=%s", loc->path, "key=%s", GLUSTERFS_INTERNAL_FOP_KEY,
+                "gfid=%s", gfid, NULL);
+        goto err;
+    }
+
+    ret = dict_set_int8(xdata, DHT_IATT_IN_XDATA_KEY, 1);
+    if (ret < 0) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+                "path=%s", loc->path, "key=%s", DHT_IATT_IN_XDATA_KEY,
+                "gfid=%s", gfid, NULL);
+        goto err;
+    }
+
+    gf_uuid_unparse(loc->inode->gfid, gfid);
+
+    ret = dht_disk_layout_extract(this, layout, i, &disk_layout);
+    if (ret == -1) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0,
+                DHT_MSG_DIR_SELFHEAL_XATTR_FAILED,
+                "extract-disk-layout-failed, path=%s", loc->path, "subvol=%s",
+                subvol->name, "gfid=%s", gfid, NULL);
+        goto err;
+    }
+
+    ret = dict_set_bin(xattr, conf->xattr_name, disk_layout, 4 * 4);
+    if (ret == -1) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0,
+                DHT_MSG_DIR_SELFHEAL_XATTR_FAILED, "path=%s", loc->path,
+                "subvol=%s", subvol->name,
+                "set-xattr-dictionary-failed"
+                "gfid=%s",
+                gfid, NULL);
+        goto err;
+    }
+    disk_layout = NULL;
+
+    gf_msg_trace(this->name, 0,
+                 "setting hash range 0x%x - 0x%x (type %d) on subvolume %s"
+                 " for %s",
+                 layout->list[i].start, layout->list[i].stop, layout->type,
+                 subvol->name, loc->path);
+
+    if (local->xattr) {
+        data = dict_get(local->xattr, QUOTA_LIMIT_KEY);
+        if (data) {
+            ret = dict_add(xattr, QUOTA_LIMIT_KEY, data);
+            if (ret) {
+                gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                        "path=%s", loc->path, "key=%s", QUOTA_LIMIT_KEY, NULL);
+            }
+        }
+        data = dict_get(local->xattr, QUOTA_LIMIT_OBJECTS_KEY);
+        if (data) {
+            ret = dict_add(xattr, QUOTA_LIMIT_OBJECTS_KEY, data);
+            if (ret) {
+                gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                        "path=%s", loc->path, "key=%s", QUOTA_LIMIT_OBJECTS_KEY,
+                        NULL);
+            }
         }
+    }
+
+    if (!gf_uuid_is_null(local->gfid))
+        gf_uuid_copy(loc->gfid, local->gfid);
+
+    STACK_WIND_COOKIE(frame, dht_selfheal_dir_xattr_cbk, (void *)subvol, subvol,
+                      subvol->fops->setxattr, loc, xattr, 0, xdata);
+
+    dict_unref(xattr);
+    dict_unref(xdata);
 
-        dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
-        dht_iatt_merge (this, &local->preparent, preparent, prev->this);
-        dht_iatt_merge (this, &local->postparent, postparent, prev->this);
+    return 0;
 
+err:
+    if (xattr)
+        dict_unref(xattr);
+    if (xdata)
+        dict_unref(xdata);
+
+    GF_FREE(disk_layout);
+
+    dht_selfheal_dir_xattr_cbk(frame, (void *)subvol, frame->this, -1, ENOMEM,
+                               NULL);
+    return 0;
+}
+
+static int
+dht_fix_dir_xattr(call_frame_t *frame, loc_t *loc, dht_layout_t *layout)
+{
+    dht_local_t *local = NULL;
+    int i = 0;
+    int count = 0;
+    xlator_t *this = NULL;
+    dht_conf_t *conf = NULL;
+    dht_layout_t *dummy = NULL;
+
+    local = frame->local;
+    this = frame->this;
+    conf = this->private;
+
+    gf_msg_debug(this->name, 0, "%s: Writing the new range for all subvolumes",
+                 loc->path);
+
+    local->call_cnt = count = conf->subvolume_cnt;
+
+    if (gf_log_get_loglevel() >= GF_LOG_DEBUG)
+        dht_log_new_layout_for_dir_selfheal(this, loc, layout);
+
+    for (i = 0; i < layout->cnt; i++) {
+        dht_selfheal_dir_xattr_persubvol(frame, loc, layout, i, NULL);
+
+        if (--count == 0)
+            goto out;
+    }
+    /* if we are here, subvolcount > layout_count. subvols-per-directory
+     * option might be set here. We need to clear out layout from the
+     * non-participating subvolumes, else it will result in overlaps */
+    dummy = dht_layout_new(this, 1);
+    if (!dummy)
+        goto out;
+    dummy->commit_hash = layout->commit_hash;
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (_gf_false == dht_is_subvol_in_layout(layout, conf->subvolumes[i])) {
+            dht_selfheal_dir_xattr_persubvol(frame, loc, dummy, 0,
+                                             conf->subvolumes[i]);
+            if (--count == 0)
+                break;
+        }
+    }
+
+    dht_layout_unref(this, dummy);
 out:
-        this_call_cnt = dht_frame_return (frame);
+    return 0;
+}
 
-        if (is_last_call (this_call_cnt)) {
-                dht_selfheal_dir_setattr (frame, &local->loc, &local->stbuf, 0xffffff, layout);
+static int
+dht_selfheal_dir_xattr(call_frame_t *frame, loc_t *loc, dht_layout_t *layout)
+{
+    dht_local_t *local = NULL;
+    int missing_xattr = 0;
+    int i = 0;
+    xlator_t *this = NULL;
+    dht_conf_t *conf = NULL;
+    dht_layout_t *dummy = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {
+        0,
+    };
+
+    local = frame->local;
+    this = frame->this;
+    conf = this->private;
+
+    for (i = 0; i < layout->cnt; i++) {
+        if (layout->list[i].err != -1 || !layout->list[i].stop) {
+            /* err != -1 would mean xattr present on the directory
+             * or the directory is non existent.
+             * !layout->list[i].stop would mean layout absent
+             */
+
+            continue;
+        }
+        missing_xattr++;
+    }
+    /* Also account for subvolumes with no-layout. Used for zero'ing out
+     * the layouts and for setting quota key's if present */
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (_gf_false == dht_is_subvol_in_layout(layout, conf->subvolumes[i])) {
+            missing_xattr++;
         }
+    }
+    gf_msg_trace(this->name, 0, "%d subvolumes missing xattr for %s",
+                 missing_xattr, loc->path);
 
+    if (missing_xattr == 0) {
+        dht_selfheal_dir_finish(frame, this, 0, 1);
         return 0;
+    }
+
+    local->call_cnt = missing_xattr;
+
+    if (gf_log_get_loglevel() >= GF_LOG_DEBUG)
+        dht_log_new_layout_for_dir_selfheal(this, loc, layout);
+
+    for (i = 0; i < layout->cnt; i++) {
+        if (layout->list[i].err != -1 || !layout->list[i].stop)
+            continue;
+
+        dht_selfheal_dir_xattr_persubvol(frame, loc, layout, i, NULL);
+
+        if (--missing_xattr == 0)
+            break;
+    }
+    dummy = dht_layout_new(this, 1);
+    if (!dummy) {
+        gf_uuid_unparse(loc->gfid, gfid);
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_DUMMY_ALLOC_FAILED,
+                "path=%s", loc->path, "gfid=%s", gfid, NULL);
+        goto out;
+    }
+    for (i = 0; i < conf->subvolume_cnt && missing_xattr; i++) {
+        if (_gf_false == dht_is_subvol_in_layout(layout, conf->subvolumes[i])) {
+            dht_selfheal_dir_xattr_persubvol(frame, loc, dummy, 0,
+                                             conf->subvolumes[i]);
+            missing_xattr--;
+        }
+    }
+
+    dht_layout_unref(this, dummy);
+out:
+    return 0;
 }
 
-void
-dht_selfheal_dir_mkdir_setacl (dict_t *xattr, dict_t *dict)
+int
+dht_selfheal_dir_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                             int op_ret, int op_errno, struct iatt *statpre,
+                             struct iatt *statpost, dict_t *xdata)
 {
-        data_t          *acl_default = NULL;
-        data_t          *acl_access = NULL;
-        xlator_t        *this = NULL;
-        int     ret = -1;
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
+    int this_call_cnt = 0, ret = -1;
 
-        GF_ASSERT (xattr);
-        GF_ASSERT (dict);
+    local = frame->local;
+    layout = local->selfheal.layout;
 
-        this = THIS;
-        GF_ASSERT (this);
+    this_call_cnt = dht_frame_return(frame);
 
-        acl_default = dict_get (xattr, POSIX_ACL_DEFAULT_XATTR);
+    if (is_last_call(this_call_cnt)) {
+        if (!local->heal_layout) {
+            gf_msg_trace(this->name, 0, "Skip heal layout for %s gfid = %s ",
+                         local->loc.path, uuid_utoa(local->gfid));
 
-        if (!acl_default) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "ACL_DEFAULT xattr not present");
-                goto cont;
+            dht_selfheal_dir_finish(frame, this, 0, 1);
+            return 0;
         }
-        ret = dict_set (dict, POSIX_ACL_DEFAULT_XATTR, acl_default);
-        if (ret)
-                gf_log (this->name, GF_LOG_WARNING,
-                        "Could not set ACL_DEFAULT xattr");
-cont:
-        acl_access = dict_get (xattr, POSIX_ACL_ACCESS_XATTR);
-        if (!acl_access) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "ACL_ACCESS xattr not present");
-                goto out;
+        ret = dht_selfheal_layout_lock(frame, layout, _gf_false,
+                                       dht_selfheal_dir_xattr,
+                                       dht_should_heal_layout);
+
+        if (ret < 0) {
+            dht_selfheal_dir_finish(frame, this, -1, 1);
         }
-        ret = dict_set (dict, POSIX_ACL_ACCESS_XATTR, acl_access);
-        if (ret)
-                gf_log (this->name, GF_LOG_WARNING,
-                        "Could not set ACL_ACCESS xattr");
+    }
 
-out:
-        return;
+    return 0;
 }
 
 int
-dht_selfheal_dir_mkdir (call_frame_t *frame, loc_t *loc,
-                        dht_layout_t *layout, int force)
+dht_selfheal_dir_setattr(call_frame_t *frame, loc_t *loc, struct iatt *stbuf,
+                         int32_t valid, dht_layout_t *layout)
 {
-        int           missing_dirs = 0;
-        int           i     = 0;
-        int           ret   = -1;
-        dht_local_t  *local = NULL;
-        xlator_t     *this = NULL;
-        dict_t       *dict = NULL;
+    int missing_attr = 0;
+    int i = 0, ret = -1;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    xlator_t *this = NULL;
+    int cnt = 0;
+
+    local = frame->local;
+    this = frame->this;
+    conf = this->private;
+
+    /* We need to heal the attrs if:
+     * 1. Any directories were missing - the newly created dirs will need
+     *    to have the correct attrs set
+     * 2. An existing dir does not have the correct permissions -they may
+     *    have been changed when a brick was down.
+     */
+
+    for (i = 0; i < layout->cnt; i++) {
+        if (layout->list[i].err == -1)
+            missing_attr++;
+    }
+
+    if ((missing_attr == 0) && (local->need_attrheal == 0)) {
+        if (!local->heal_layout) {
+            gf_msg_trace(this->name, 0, "Skip heal layout for %s gfid = %s ",
+                         loc->path, uuid_utoa(loc->gfid));
+            dht_selfheal_dir_finish(frame, this, 0, 1);
+            return 0;
+        }
+        ret = dht_selfheal_layout_lock(frame, layout, _gf_false,
+                                       dht_selfheal_dir_xattr,
+                                       dht_should_heal_layout);
+
+        if (ret < 0) {
+            dht_selfheal_dir_finish(frame, this, -1, 1);
+        }
+
+        return 0;
+    }
+
+    cnt = local->call_cnt = conf->subvolume_cnt;
 
-        local = frame->local;
-        this = frame->this;
+    for (i = 0; i < cnt; i++) {
+        STACK_WIND(frame, dht_selfheal_dir_setattr_cbk, layout->list[i].xlator,
+                   layout->list[i].xlator->fops->setattr, loc, stbuf, valid,
+                   NULL);
+    }
 
+    return 0;
+}
+
+static int
+dht_selfheal_dir_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int op_ret, int op_errno, inode_t *inode,
+                           struct iatt *stbuf, struct iatt *preparent,
+                           struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
+    xlator_t *prev = NULL;
+    xlator_t *subvol = NULL;
+    int i = 0, ret = -1;
+    int this_call_cnt = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    layout = local->selfheal.layout;
+    prev = cookie;
+    subvol = prev;
+
+    if ((op_ret == 0) || ((op_ret == -1) && (op_errno == EEXIST))) {
         for (i = 0; i < layout->cnt; i++) {
-                if (layout->list[i].err == ENOENT || force)
-                        missing_dirs++;
+            if (layout->list[i].xlator == subvol) {
+                layout->list[i].err = -1;
+                break;
+            }
         }
+    }
+
+    if (op_ret) {
+        gf_uuid_unparse(local->loc.gfid, gfid);
+        gf_smsg(this->name,
+                ((op_errno == EEXIST) ? GF_LOG_DEBUG : GF_LOG_WARNING),
+                op_errno, DHT_MSG_DIR_SELFHEAL_FAILED, "path=%s",
+                local->loc.path, "gfid=%s", gfid, NULL);
+        goto out;
+    }
+    dht_iatt_merge(this, &local->preparent, preparent);
+    dht_iatt_merge(this, &local->postparent, postparent);
+    ret = 0;
 
-        if (missing_dirs == 0) {
-                dht_selfheal_dir_setattr (frame, loc, &local->stbuf, 0xffffffff, layout);
-                return 0;
-        }
+out:
+    this_call_cnt = dht_frame_return(frame);
 
-        local->call_cnt = missing_dirs;
-        if (!uuid_is_null (local->gfid)) {
-                dict = dict_new ();
-                if (!dict)
-                        return -1;
-
-                ret = dict_set_static_bin (dict, "gfid-req", local->gfid, 16);
-                if (ret)
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "%s: failed to set gfid in dict", loc->path);
-        } else if (local->params) {
-                /* Send the dictionary from higher layers directly */
-                dict = dict_ref (local->params);
-        }
-        /* Set acls */
-        if (local->xattr && dict)
-                dht_selfheal_dir_mkdir_setacl (local->xattr, dict);
+    if (is_last_call(this_call_cnt)) {
+        dht_selfheal_dir_finish(frame, this, ret, 0);
+        dht_selfheal_dir_setattr(frame, &local->loc, &local->stbuf, 0xffffff,
+                                 layout);
+    }
 
+    return 0;
+}
+
+static int
+dht_selfheal_dir_mkdir_lookup_done(call_frame_t *frame, xlator_t *this)
+{
+    dht_local_t *local = NULL;
+    int i = 0;
+    dict_t *dict = NULL;
+    dht_layout_t *layout = NULL;
+    loc_t *loc = NULL;
+    int cnt = 0;
+    int ret = -1;
+
+    VALIDATE_OR_GOTO(this->private, err);
+
+    local = frame->local;
+    layout = local->layout;
+    loc = &local->loc;
+
+    if (!gf_uuid_is_null(local->gfid)) {
+        dict = dict_new();
         if (!dict)
-                gf_log (this->name, GF_LOG_WARNING,
-                        "dict is NULL, need to make sure gfids are same");
+            return -1;
 
-        for (i = 0; i < layout->cnt; i++) {
-                if (layout->list[i].err == ENOENT || force) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "creating directory %s on subvol %s",
-                                loc->path, layout->list[i].xlator->name);
-
-                        STACK_WIND (frame, dht_selfheal_dir_mkdir_cbk,
-                                    layout->list[i].xlator,
-                                    layout->list[i].xlator->fops->mkdir,
-                                    loc,
-                                    st_mode_from_ia (local->stbuf.ia_prot,
-                                                     local->stbuf.ia_type),
-                                    0, dict);
-                }
+        ret = dict_set_gfuuid(dict, "gfid-req", local->gfid, true);
+        if (ret)
+            gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+                    "path=%s", loc->path, "key=gfid-req", NULL);
+    } else if (local->params) {
+        /* Send the dictionary from higher layers directly */
+
+        dict = dict_ref(local->params);
+    }
+    /* Code to update all extended attributed from local->xattr
+       to dict
+    */
+    dht_dir_set_heal_xattr(this, local, dict, local->xattr, NULL, NULL);
+
+    if (!dict) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_IS_NULL, NULL);
+        dict = dict_new();
+        if (!dict)
+            return -1;
+    }
+    ret = dict_set_flag(dict, GF_INTERNAL_CTX_KEY, GF_DHT_HEAL_DIR);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, "key=%s",
+                GF_INTERNAL_CTX_KEY, "path=%s", loc->path, NULL);
+        /* We can still continue. As heal can still happen
+         * unless quota limits have reached for the dir.
+         */
+    }
+
+    cnt = layout->cnt;
+    for (i = 0; i < cnt; i++) {
+        if (layout->list[i].err == ESTALE || layout->list[i].err == ENOENT ||
+            local->selfheal.force_mkdir) {
+            gf_msg_debug(this->name, 0, "Creating directory %s on subvol %s",
+                         loc->path, layout->list[i].xlator->name);
+
+            STACK_WIND_COOKIE(
+                frame, dht_selfheal_dir_mkdir_cbk, layout->list[i].xlator,
+                layout->list[i].xlator, layout->list[i].xlator->fops->mkdir,
+                loc,
+                st_mode_from_ia(local->stbuf.ia_prot, local->stbuf.ia_type), 0,
+                dict);
         }
+    }
 
-        if (dict)
-                dict_unref (dict);
+    if (dict)
+        dict_unref(dict);
 
-        return 0;
-}
+    return 0;
 
+err:
+    dht_selfheal_dir_finish(frame, this, -1, 1);
+    return 0;
+}
 
-int
-dht_selfheal_layout_alloc_start (xlator_t *this, loc_t *loc,
-                                 dht_layout_t *layout)
+static int
+dht_selfheal_dir_mkdir_lookup_cbk(call_frame_t *frame, void *cookie,
+                                  xlator_t *this, int op_ret, int op_errno,
+                                  inode_t *inode, struct iatt *stbuf,
+                                  dict_t *xattr, struct iatt *postparent)
 {
-        int           start = 0;
-        uint32_t      hashval = 0;
-        int           ret = 0;
+    dht_local_t *local = NULL;
+    int i = 0;
+    int this_call_cnt = 0;
+    int missing_dirs = 0;
+    dht_layout_t *layout = NULL;
+    xlator_t *prev = 0;
+    loc_t *loc = NULL;
+    char gfid_local[GF_UUID_BUF_SIZE] = {0};
+    int index = -1;
+
+    VALIDATE_OR_GOTO(this->private, err);
+
+    local = frame->local;
+    layout = local->layout;
+    loc = &local->loc;
+    prev = cookie;
+
+    if (!gf_uuid_is_null(local->gfid))
+        gf_uuid_unparse(local->gfid, gfid_local);
+
+    LOCK(&frame->lock);
+    {
+        index = dht_layout_index_for_subvol(layout, prev);
+        if ((op_ret < 0) && (op_errno == ENOENT || op_errno == ESTALE)) {
+            local->selfheal.hole_cnt = !local->selfheal.hole_cnt
+                                           ? 1
+                                           : local->selfheal.hole_cnt + 1;
+            /* the status might have changed. Update the layout with the
+             * new status
+             */
+            if (index >= 0) {
+                layout->list[index].err = op_errno;
+            }
+        }
 
-        ret = dht_hash_compute (this, layout->type, loc->path, &hashval);
-        if (ret == 0) {
-                start = (hashval % layout->cnt);
+        if (!op_ret) {
+            dht_iatt_merge(this, &local->stbuf, stbuf);
+            if (prev == local->mds_subvol) {
+                dict_unref(local->xattr);
+                local->xattr = dict_ref(xattr);
+            }
+            /* the status might have changed. Update the layout with the
+             * new status
+             */
+            if (index >= 0) {
+                layout->list[index].err = -1;
+            }
         }
+    }
+    UNLOCK(&frame->lock);
+
+    this_call_cnt = dht_frame_return(frame);
+
+    if (is_last_call(this_call_cnt)) {
+        if (local->selfheal.hole_cnt == layout->cnt) {
+            gf_msg_debug(this->name, op_errno,
+                         "Lookup failed, an rmdir could have "
+                         "deleted this entry %s",
+                         loc->name);
+            local->op_errno = op_errno;
+            goto err;
+        } else {
+            for (i = 0; i < layout->cnt; i++) {
+                if (layout->list[i].err == ENOENT ||
+                    layout->list[i].err == ESTALE ||
+                    local->selfheal.force_mkdir)
+                    missing_dirs++;
+            }
+
+            if (missing_dirs == 0) {
+                dht_selfheal_dir_finish(frame, this, 0, 0);
+                dht_selfheal_dir_setattr(frame, loc, &local->stbuf, 0xffffffff,
+                                         layout);
+                return 0;
+            }
 
-        return start;
+            local->call_cnt = missing_dirs;
+            dht_selfheal_dir_mkdir_lookup_done(frame, this);
+        }
+    }
+
+    return 0;
+
+err:
+    dht_selfheal_dir_finish(frame, this, -1, 1);
+    return 0;
 }
 
-static inline int
-dht_get_layout_count (xlator_t *this, dht_layout_t *layout, int new_layout)
+static int
+dht_selfheal_dir_mkdir_lock_cbk(call_frame_t *frame, void *cookie,
+                                xlator_t *this, int32_t op_ret,
+                                int32_t op_errno, dict_t *xdata)
 {
-        int i = 0;
-        int j = 0;
-        int err = 0;
-        int count = 0;
-        dht_conf_t *conf = NULL;
-
-        /* Gets in use only for replace-brick, remove-brick */
-        conf = this->private;
-        for (i = 0; i < layout->cnt; i++) {
-                for (j = 0; j < conf->subvolume_cnt; j++) {
-                        if (conf->decommissioned_bricks[j] &&
-                            conf->decommissioned_bricks[j] == layout->list[i].xlator) {
-                                layout->list[i].err = EINVAL;
-                                break;
-                        }
-                }
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int i = 0;
+    int ret = -1;
+    xlator_t *mds_subvol = NULL;
+
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+    local = frame->local;
+    mds_subvol = local->mds_subvol;
+
+    local->call_cnt = conf->subvolume_cnt;
+
+    if (op_ret < 0) {
+        if (op_errno == EINVAL) {
+            local->call_cnt = 1;
+            dht_selfheal_dir_mkdir_lookup_done(frame, this);
+            return 0;
         }
 
-        for (i = 0; i < layout->cnt; i++) {
-                err = layout->list[i].err;
-                if (err == -1 || err == 0 || err == ENOENT) {
-			/* Setting list[i].err = -1 is an indication for
-			   dht_selfheal_layout_new_directory() to assign
-			   a range. We set it to -1 based on any one of
-			   the three criteria:
-
-			   - err == -1 already, which means directory
-			     existed but layout was not set on it.
-
-			   - err == 0, which means directory exists and
-			     has an old layout piece which will be
-			     overwritten now.
-
-			   - err == ENOENT, which means directory does
-			     not exist (possibly racing with mkdir or
-			     finishing half done mkdir). The missing
-			     directory will be attempted to be recreated.
-
-			     It is important to note that it is safe
-			     to race with mkdir() as self-heal and
-			     mkdir are idempotent operations. Both will
-			     strive to set the directory and layouts to
-			     the same final state.
-			*/
-                        count++;
-			if (!err)
-				layout->list[i].err = -1;
-                }
+        gf_smsg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_ENTRYLK_ERROR,
+                "path=%s", local->loc.path, NULL);
+
+        local->op_errno = op_errno;
+        goto err;
+    }
+
+    /* After getting locks, perform lookup again to ensure that the
+       directory was not deleted by a racing rmdir
+    */
+    if (!local->xattr_req)
+        local->xattr_req = dict_new();
+
+    ret = dict_set_int32(local->xattr_req, "list-xattr", 1);
+    if (ret)
+        gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, "path=%s",
+                local->loc.path, NULL);
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (mds_subvol && conf->subvolumes[i] == mds_subvol) {
+            STACK_WIND_COOKIE(frame, dht_selfheal_dir_mkdir_lookup_cbk,
+                              conf->subvolumes[i], conf->subvolumes[i],
+                              conf->subvolumes[i]->fops->lookup, &local->loc,
+                              local->xattr_req);
+        } else {
+            STACK_WIND_COOKIE(frame, dht_selfheal_dir_mkdir_lookup_cbk,
+                              conf->subvolumes[i], conf->subvolumes[i],
+                              conf->subvolumes[i]->fops->lookup, &local->loc,
+                              NULL);
         }
+    }
+
+    return 0;
+
+err:
+    dht_selfheal_dir_finish(frame, this, -1, 1);
+    return 0;
+}
 
-        /* no subvolume has enough space, but can't stop directory creation */
-        if (!count || !new_layout) {
-                for (i = 0; i < layout->cnt; i++) {
-                        err = layout->list[i].err;
-                        if (err == ENOSPC) {
-                                layout->list[i].err = -1;
-                                count++;
-                        }
+static int
+dht_selfheal_dir_mkdir(call_frame_t *frame, loc_t *loc, dht_layout_t *layout,
+                       int force)
+{
+    int missing_dirs = 0;
+    int i = 0;
+    int op_errno = 0;
+    int ret = -1;
+    dht_local_t *local = NULL;
+    xlator_t *this = NULL;
+    dht_conf_t *conf = NULL;
+
+    local = frame->local;
+    this = frame->this;
+    conf = this->private;
+
+    local->selfheal.force_mkdir = force;
+    local->selfheal.hole_cnt = 0;
+
+    for (i = 0; i < layout->cnt; i++) {
+        if (layout->list[i].err == ENOENT || force)
+            missing_dirs++;
+    }
+
+    if (missing_dirs == 0) {
+        /* We don't need to create any directories. Proceed to heal the
+         * attrs and xattrs
+         */
+        if (!__is_root_gfid(local->stbuf.ia_gfid)) {
+            if (local->need_xattr_heal) {
+                local->need_xattr_heal = 0;
+                ret = dht_dir_xattr_heal(this, local, &op_errno);
+                if (ret) {
+                    gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                            DHT_MSG_DIR_XATTR_HEAL_FAILED, "path=%s",
+                            local->loc.path, "gfid=%s", local->gfid, NULL);
                 }
+            } else {
+                if (!gf_uuid_is_null(local->gfid))
+                    gf_uuid_copy(loc->gfid, local->gfid);
+
+                ret = dht_common_mark_mdsxattr(frame, NULL, 0);
+                if (!ret)
+                    return 0;
+
+                gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_SET_XATTR_FAILED,
+                        "path=%s", local->loc.path, "gfid=%s", local->gfid,
+                        NULL);
+            }
+        }
+        dht_selfheal_dir_setattr(frame, loc, &local->stbuf, 0xffffffff, layout);
+        return 0;
+    }
+
+    /* MDS xattr is populated only while DHT is having more than one
+     subvol.In case of graph switch while adding more dht subvols need to
+     consider hash subvol as a MDS to avoid MDS check failure at the time
+     of running fop on directory
+    */
+    if (!dict_get(local->xattr, conf->mds_xattr_key) &&
+        (conf->subvolume_cnt > 1)) {
+        if (local->hashed_subvol == NULL) {
+            local->hashed_subvol = dht_subvol_get_hashed(this, loc);
+            if (local->hashed_subvol == NULL) {
+                local->op_errno = EINVAL;
+                gf_smsg(this->name, GF_LOG_WARNING, local->op_errno,
+                        DHT_MSG_HASHED_SUBVOL_GET_FAILED, "gfid=%s",
+                        loc->pargfid, "name=%s", loc->name, "path=%s",
+                        loc->path, NULL);
+                goto err;
+            }
+        }
+        ret = dht_inode_ctx_mdsvol_set(local->inode, this,
+                                       local->hashed_subvol);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SET_INODE_CTX_FAILED,
+                   "Failed to set hashed subvol for %s on inode vol is %s",
+                   local->loc.path,
+                   local->hashed_subvol ? local->hashed_subvol->name : "NULL");
+            goto err;
+        }
+    }
+
+    if (local->hashed_subvol == NULL) {
+        local->hashed_subvol = dht_subvol_get_hashed(this, loc);
+        if (local->hashed_subvol == NULL) {
+            local->op_errno = EINVAL;
+            gf_smsg(this->name, GF_LOG_WARNING, local->op_errno,
+                    DHT_MSG_HASHED_SUBVOL_GET_FAILED, "gfid=%s", loc->pargfid,
+                    "name=%s", loc->name, "path=%s", loc->path, NULL);
+            goto err;
         }
+    }
 
-        /* if layout->spread_cnt is set, check if it is <= available
-         * subvolumes (down brick and decommissioned bricks are considered
-         * un-availbale). Else return count (available up bricks) */
-        count = ((layout->spread_cnt &&
-                 (layout->spread_cnt <= count)) ?
-                 layout->spread_cnt : ((count) ? count : 1));
+    local->current = &local->lock[0];
+    ret = dht_protect_namespace(frame, loc, local->hashed_subvol,
+                                &local->current->ns,
+                                dht_selfheal_dir_mkdir_lock_cbk);
 
-        return count;
+    if (ret < 0)
+        goto err;
+
+    return 0;
+err:
+    return -1;
+}
+
+static int
+dht_selfheal_layout_alloc_start(xlator_t *this, loc_t *loc,
+                                dht_layout_t *layout)
+{
+    int start = 0;
+    uint32_t hashval = 0;
+    int ret = 0;
+    const char *str = NULL;
+    dht_conf_t *conf = NULL;
+    char buf[UUID_CANONICAL_FORM_LEN + 1] = {
+        0,
+    };
+
+    conf = this->private;
+
+    if (conf->randomize_by_gfid) {
+        str = uuid_utoa_r(loc->gfid, buf);
+    } else {
+        str = loc->path;
+    }
+
+    ret = dht_hash_compute(this, layout->type, str, &hashval);
+    if (ret == 0) {
+        start = (hashval % layout->cnt);
+    }
+
+    return start;
 }
 
+static int
+dht_get_layout_count(xlator_t *this, dht_layout_t *layout, int new_layout)
+{
+    int i = 0;
+    int j = 0;
+    int err = 0;
+    int count = 0;
+    dht_conf_t *conf = NULL;
+
+    /* Gets in use only for replace-brick, remove-brick */
+    conf = this->private;
+    for (i = 0; i < layout->cnt; i++) {
+        for (j = 0; j < conf->subvolume_cnt; j++) {
+            if (conf->decommissioned_bricks[j] &&
+                conf->decommissioned_bricks[j] == layout->list[i].xlator) {
+                layout->list[i].err = EINVAL;
+                break;
+            }
+        }
+    }
+
+    for (i = 0; i < layout->cnt; i++) {
+        err = layout->list[i].err;
+        if (err == -1 || err == 0 || err == ENOENT) {
+            /* Take this with a pinch of salt. The behaviour seems
+             * to be slightly different when this function is
+             * invoked from mkdir codepath. For eg., err == 0 in
+             * mkdir codepath means directory created but xattr
+             * is not set yet.
+             */
+
+            /* Setting list[i].err = -1 is an indication for
+               dht_selfheal_layout_new_directory() to assign
+               a range. We set it to -1 based on any one of
+               the three criteria:
+
+               - err == -1 already, which means directory
+                 existed but layout was not set on it.
+
+               - err == 0, which means directory exists and
+                 has an old layout piece which will be
+                 overwritten now.
+
+               - err == ENOENT, which means directory does
+                 not exist (possibly racing with mkdir or
+                 finishing half done mkdir). The missing
+                 directory will be attempted to be recreated.
+            */
+            count++;
+            if (!err)
+                layout->list[i].err = -1;
+        }
+    }
+
+    /* no subvolume has enough space, but can't stop directory creation */
+    if (!count || !new_layout) {
+        for (i = 0; i < layout->cnt; i++) {
+            err = layout->list[i].err;
+            if (err == ENOSPC) {
+                layout->list[i].err = -1;
+                count++;
+            }
+        }
+    }
 
-void dht_selfheal_layout_new_directory (call_frame_t *frame, loc_t *loc,
-					dht_layout_t *new_layout);
+    /* if layout->spread_cnt is set, check if it is <= available
+     * subvolumes (down brick and decommissioned bricks are considered
+     * un-available). Else return count (available up bricks) */
+    count = ((layout->spread_cnt && (layout->spread_cnt <= count))
+                 ? layout->spread_cnt
+                 : ((count) ? count : 1));
 
-void dht_layout_entry_swap (dht_layout_t *layout, int i, int j);
-void dht_layout_range_swap (dht_layout_t *layout, int i, int j);
+    return count;
+}
+
+void
+dht_selfheal_layout_new_directory(call_frame_t *frame, loc_t *loc,
+                                  dht_layout_t *new_layout);
+
+void
+dht_layout_range_swap(dht_layout_t *layout, int i, int j);
 
 /*
  * It's a bit icky using local variables in a macro, but it makes the rest
  * of the code a lot clearer.
  */
-#define OV_ENTRY(x,y)      table[x*new->cnt+y]
+#define OV_ENTRY(x, y) table[x * new->cnt + y]
 
-void
-dht_selfheal_layout_maximize_overlap (call_frame_t *frame, loc_t *loc,
-				      dht_layout_t *new, dht_layout_t *old)
+static void
+dht_selfheal_layout_maximize_overlap(call_frame_t *frame, loc_t *loc,
+                                     dht_layout_t *new, dht_layout_t *old)
 {
-	int           i            = 0;
-	int           j            = 0;
-	uint32_t      curr_overlap = 0;
-	uint32_t      max_overlap  = 0;
-	int           max_overlap_idx = -1;
-	uint32_t      overlap      = 0;
-        uint32_t     *table = NULL;
-
-	dht_layout_sort_volname (old);
-	/* Now both old_layout->list[] and new_layout->list[]
-	   are match the same xlators/subvolumes. i.e,
-	   old_layout->[i] and new_layout->[i] are referring
-	   to the same subvolumes
-	*/
-
-        /* Build a table of overlaps between new[i] and old[j]. */
-        table = alloca(sizeof(overlap)*old->cnt*new->cnt);
-        if (!table) {
-                return;
+    int i = 0;
+    int j = 0;
+    uint32_t curr_overlap = 0;
+    uint32_t max_overlap = 0;
+    int max_overlap_idx = -1;
+    uint32_t overlap = 0;
+    uint32_t *table = NULL;
+
+    dht_layout_sort_volname(old);
+    /* Now both old_layout->list[] and new_layout->list[]
+       are match the same xlators/subvolumes. i.e,
+       old_layout->[i] and new_layout->[i] are referring
+       to the same subvolumes
+    */
+
+    /* Build a table of overlaps between new[i] and old[j]. */
+    table = alloca(sizeof(overlap) * old->cnt * new->cnt);
+    if (!table) {
+        return;
+    }
+    memset(table, 0, sizeof(overlap) * old->cnt * new->cnt);
+    for (i = 0; i < new->cnt; ++i) {
+        for (j = 0; j < old->cnt; ++j) {
+            OV_ENTRY(i, j) = dht_overlap_calc(old, j, new, i);
         }
-        memset(table,0,sizeof(overlap)*old->cnt*new->cnt);
-        for (i = 0; i < new->cnt; ++i) {
-                for (j = 0; j < old->cnt; ++j) {
-                        OV_ENTRY(i,j) = dht_overlap_calc(old,j,new,i);
-                }
+    }
+
+    for (i = 0; i < new->cnt; i++) {
+        if (new->list[i].err > 0) {
+            /* Subvol might be marked for decommission
+               with EINVAL, or some other serious error
+               marked with positive errno.
+            */
+            continue;
         }
 
-	for (i = 0; i < new->cnt; i++) {
-		if (new->list[i].err > 0) {
-			/* Subvol might be marked for decommission
-			   with EINVAL, or some other serious error
-			   marked with positive errno.
-			*/
-			continue;
-                }
-
-                max_overlap = 0;
-                max_overlap_idx = i;
-                for (j = (i + 1); j < new->cnt; ++j) {
-                        if (new->list[j].err > 0) {
-			        /* Subvol might be marked for decommission
-			        with EINVAL, or some other serious error
-			        marked with positive errno.
-			        */
-			        continue;
-                        }
-                        /* Calculate the overlap now. */
-                        curr_overlap = OV_ENTRY(i,i) + OV_ENTRY(j,j);
-                        /* Calculate the overlap after the proposed swap. */
-                        overlap = OV_ENTRY(i,j) + OV_ENTRY(j,i);
-                        /* Are we better than status quo? */
-                        if (overlap > curr_overlap) {
-                                overlap -= curr_overlap;
-                                /* Are we better than the previous choice? */
-                                if (overlap > max_overlap) {
-                                        max_overlap = overlap;
-                                        max_overlap_idx = j;
-                                }
-                        }
+        max_overlap = 0;
+        max_overlap_idx = i;
+        for (j = (i + 1); j < new->cnt; ++j) {
+            if (new->list[j].err > 0) {
+                /* Subvol might be marked for decommission
+                with EINVAL, or some other serious error
+                marked with positive errno.
+                */
+                continue;
+            }
+            /* Calculate the overlap now. */
+            curr_overlap = OV_ENTRY(i, i) + OV_ENTRY(j, j);
+            /* Calculate the overlap after the proposed swap. */
+            overlap = OV_ENTRY(i, j) + OV_ENTRY(j, i);
+            /* Are we better than status quo? */
+            if (overlap > curr_overlap) {
+                overlap -= curr_overlap;
+                /* Are we better than the previous choice? */
+                if (overlap > max_overlap) {
+                    max_overlap = overlap;
+                    max_overlap_idx = j;
                 }
+            }
+        }
 
-		if (max_overlap_idx != i) {
- 			dht_layout_range_swap (new, i, max_overlap_idx);
-                        /* Need to swap the table values too. */
-                        for (j = 0; j < old->cnt; ++j) {
-                                overlap = OV_ENTRY(i,j);
-                                OV_ENTRY(i,j) = OV_ENTRY(max_overlap_idx,j);
-                                OV_ENTRY(max_overlap_idx,j) = overlap;
-                        }
-                }
-	}
+        if (max_overlap_idx != i) {
+            dht_layout_range_swap(new, i, max_overlap_idx);
+            /* Need to swap the table values too. */
+            for (j = 0; j < old->cnt; ++j) {
+                overlap = OV_ENTRY(i, j);
+                OV_ENTRY(i, j) = OV_ENTRY(max_overlap_idx, j);
+                OV_ENTRY(max_overlap_idx, j) = overlap;
+            }
+        }
+    }
 }
 
-
-dht_layout_t *
-dht_fix_layout_of_directory (call_frame_t *frame, loc_t *loc,
-                             dht_layout_t *layout)
+static dht_layout_t *
+dht_fix_layout_of_directory(call_frame_t *frame, loc_t *loc,
+                            dht_layout_t *layout)
 {
-        int           i            = 0;
-        xlator_t     *this         = NULL;
-        dht_layout_t *new_layout   = NULL;
-        dht_conf_t   *priv         = NULL;
-        dht_local_t  *local        = NULL;
-        uint32_t      subvol_down  = 0;
-        int           ret          = 0;
-
-        this  = frame->this;
-        priv  = this->private;
-        local = frame->local;
-
-        if (layout->type == DHT_HASH_TYPE_DM_USER) {
-                gf_log (THIS->name, GF_LOG_DEBUG, "leaving %s alone",
-                        loc->path);
-                goto done;
+    int i = 0;
+    xlator_t *this = NULL;
+    dht_layout_t *new_layout = NULL;
+    dht_conf_t *priv = NULL;
+    dht_local_t *local = NULL;
+    uint32_t subvol_down = 0;
+    gf_boolean_t maximize_overlap = _gf_true;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    this = frame->this;
+    priv = this->private;
+    local = frame->local;
+
+    if (layout->type == DHT_HASH_TYPE_DM_USER) {
+        gf_msg_debug(THIS->name, 0, "leaving %s alone", loc->path);
+        goto done;
+    }
+
+    new_layout = dht_layout_new(this, priv->subvolume_cnt);
+    if (!new_layout) {
+        gf_uuid_unparse(loc->gfid, gfid);
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_MEM_ALLOC_FAILED,
+                "new_layout, path=%s", loc->path, "gfid=%s", gfid, NULL);
+        goto done;
+    }
+
+    /* If a subvolume is down, do not re-write the layout. */
+    dht_layout_anomalies(this, loc, layout, NULL, NULL, NULL, &subvol_down,
+                         NULL, NULL);
+
+    if (subvol_down) {
+        gf_uuid_unparse(loc->gfid, gfid);
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LAYOUT_FIX_FAILED,
+                "subvol-down=%u", subvol_down, "Skipping-fix-layout", "path=%s",
+                loc->path, "gfid=%s", gfid, NULL);
+        GF_FREE(new_layout);
+        return NULL;
+    }
+
+    for (i = 0; i < new_layout->cnt; i++) {
+        if (layout->list[i].err != ENOSPC)
+            new_layout->list[i].err = layout->list[i].err;
+        else
+            new_layout->list[i].err = -1;
+
+        new_layout->list[i].xlator = layout->list[i].xlator;
+    }
+
+    new_layout->commit_hash = layout->commit_hash;
+
+    if (priv->du_stats) {
+        for (i = 0; i < priv->subvolume_cnt; ++i) {
+            gf_smsg(this->name, GF_LOG_DEBUG, 0, DHT_MSG_SUBVOL_INFO,
+                    "index=%d", i, "name=%s", priv->subvolumes[i]->name,
+                    "chunks=%u", priv->du_stats[i].chunks, "path=%s", loc->path,
+                    NULL);
+
+            /* Maximize overlap if the bricks are all the same
+             *  size.
+             * This is probably not going to be very common on
+             * live setups but will benefit our regression tests
+             */
+            if (i && (priv->du_stats[i].chunks != priv->du_stats[0].chunks)) {
+                maximize_overlap = _gf_false;
+            }
         }
+    } else {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_NO_DISK_USAGE_STATUS,
+                NULL);
+    }
+
+    /* First give it a layout as though it is a new directory. This
+       ensures rotation to kick in */
+    dht_layout_sort_volname(new_layout);
+    dht_selfheal_layout_new_directory(frame, loc, new_layout);
+
+    /* Maximize overlap if weighted-rebalance is disabled */
+    if (!priv->do_weighting)
+        maximize_overlap = _gf_true;
+
+    /* Now selectively re-assign ranges only when it helps */
+    if (maximize_overlap) {
+        dht_selfheal_layout_maximize_overlap(frame, loc, new_layout, layout);
+    }
+done:
+    if (new_layout) {
+        /* Make sure the extra 'ref' for existing layout is removed */
+        dht_layout_unref(this, local->layout);
 
-        new_layout = dht_layout_new (this, priv->subvolume_cnt);
-        if (!new_layout)
-                goto done;
+        local->layout = new_layout;
+    }
 
-        /* If a subvolume is down, do not re-write the layout. */
-        ret = dht_layout_anomalies (this, loc, layout, NULL, NULL, NULL,
-                                    &subvol_down, NULL, NULL);
+    return local->layout;
+}
 
-        if (subvol_down || (ret == -1)) {
-                gf_log (this->name, GF_LOG_WARNING, "%u subvolume(s) are down"
-                        ". Skipping fix layout.", subvol_down);
-                GF_FREE (new_layout);
-                return NULL;
-        }
+/*
+ * Having to call this 2x for each entry in the layout is pretty horrible, but
+ * that's what all of this layout-sorting nonsense gets us.
+ */
+static uint32_t
+dht_get_chunks_from_xl(xlator_t *parent, xlator_t *child)
+{
+    dht_conf_t *priv = parent->private;
+    xlator_list_t *trav;
+    uint32_t index = 0;
 
-        for (i = 0; i < new_layout->cnt; i++) {
-		if (layout->list[i].err != ENOSPC)
-			new_layout->list[i].err = layout->list[i].err;
-		else
-			new_layout->list[i].err = -1;
+    if (!priv->du_stats) {
+        return 0;
+    }
 
-		new_layout->list[i].xlator = layout->list[i].xlator;
+    for (trav = parent->children; trav; trav = trav->next) {
+        if (trav->xlator == child) {
+            return priv->du_stats[index].chunks;
         }
+        ++index;
+    }
 
-	/* First give it a layout as though it is a new directory. This
-	   ensures rotation to kick in */
-        dht_layout_sort_volname (new_layout);
-	dht_selfheal_layout_new_directory (frame, loc, new_layout);
+    return 0;
+}
 
-	/* Now selectively re-assign ranges only when it helps */
-	dht_selfheal_layout_maximize_overlap (frame, loc, new_layout, layout);
+void
+dht_selfheal_layout_new_directory(call_frame_t *frame, loc_t *loc,
+                                  dht_layout_t *layout)
+{
+    xlator_t *this = NULL;
+    double chunk = 0;
+    int i = 0;
+    uint32_t start = 0;
+    int bricks_to_use = 0;
+    int err = 0;
+    int start_subvol = 0;
+    uint32_t curr_size;
+    uint32_t range_size;
+    uint64_t total_size = 0;
+    int real_i;
+    dht_conf_t *priv;
+    gf_boolean_t weight_by_size;
+    int bricks_used = 0;
+
+    this = frame->this;
+    priv = this->private;
+    weight_by_size = priv->do_weighting;
+
+    bricks_to_use = dht_get_layout_count(this, layout, 1);
+    GF_ASSERT(bricks_to_use > 0);
+
+    bricks_used = 0;
+    for (i = 0; i < layout->cnt; ++i) {
+        err = layout->list[i].err;
+        if ((err != -1) && (err != ENOENT)) {
+            continue;
+        }
+        curr_size = dht_get_chunks_from_xl(this, layout->list[i].xlator);
+        if (!curr_size) {
+            weight_by_size = _gf_false;
+            break;
+        }
+        total_size += curr_size;
+        if (++bricks_used >= bricks_to_use) {
+            break;
+        }
+    }
+
+    if (weight_by_size && total_size) {
+        /* We know total_size is not zero. */
+        chunk = ((double)0xffffffff) / ((double)total_size);
+        gf_msg_debug(this->name, 0,
+                     "chunk size = 0xffffffff / %" PRIu64 " = %f", total_size,
+                     chunk);
+    } else {
+        weight_by_size = _gf_false;
+        chunk = ((unsigned long)0xffffffff) / bricks_to_use;
+    }
+
+    start_subvol = dht_selfheal_layout_alloc_start(this, loc, layout);
+
+    /* clear out the range, as we are re-computing here */
+    DHT_RESET_LAYOUT_RANGE(layout);
+
+    /*
+     * OK, what's this "real_i" stuff about?  This used to be two loops -
+     * from start_subvol to layout->cnt-1, then from 0 to start_subvol-1.
+     * That way is practically an open invitation to bugs when only one
+     * of the loops is updated.  Using real_i and modulo operators to make
+     * it one loop avoids this problem.  Remember, folks: it's everyone's
+     * responsibility to help stamp out copy/paste abuse.
+     */
+    bricks_used = 0;
+    for (real_i = 0; real_i < layout->cnt; real_i++) {
+        i = (real_i + start_subvol) % layout->cnt;
+        err = layout->list[i].err;
+        if ((err != -1) && (err != ENOENT)) {
+            continue;
+        }
+        if (weight_by_size) {
+            curr_size = dht_get_chunks_from_xl(this, layout->list[i].xlator);
+            if (!curr_size) {
+                continue;
+            }
+        } else {
+            curr_size = 1;
+        }
+        range_size = chunk * curr_size;
+        gf_msg_debug(this->name, 0, "assigning range size 0x%x to %s",
+                     range_size, layout->list[i].xlator->name);
+        DHT_SET_LAYOUT_RANGE(layout, i, start, range_size, loc->path);
+        if (++bricks_used >= bricks_to_use) {
+            layout->list[i].stop = 0xffffffff;
+            goto done;
+        }
+        start += range_size;
+    }
 
 done:
-        if (new_layout) {
-                /* Now that the new layout has all the proper layout, change the
-                   inode context */
-                dht_layout_set (this, loc->inode, new_layout);
-
-                /* Make sure the extra 'ref' for existing layout is removed */
-                dht_layout_unref (this, local->layout);
+    return;
+}
 
-                local->layout = new_layout;
+static int
+dht_selfheal_dir_getafix(call_frame_t *frame, loc_t *loc, dht_layout_t *layout)
+{
+    dht_local_t *local = NULL;
+    uint32_t holes = 0;
+    int ret = -1;
+    int i = -1;
+    uint32_t overlaps = 0;
+
+    local = frame->local;
+
+    holes = local->selfheal.hole_cnt;
+    overlaps = local->selfheal.overlaps_cnt;
+
+    if (holes || overlaps) {
+        /* If the layout has anomalies which would change the hash
+         * ranges, then we need to reset the commit_hash for this
+         * directory, as the layout would change and things may not
+         * be in place as expected */
+        layout->commit_hash = DHT_LAYOUT_HASH_INVALID;
+        dht_selfheal_layout_new_directory(frame, loc, layout);
+        ret = 0;
+    }
+
+    for (i = 0; i < layout->cnt; i++) {
+        /* directory not present */
+        if (layout->list[i].err == ENOENT) {
+            ret = 0;
+            break;
         }
+    }
 
-        return local->layout;
-}
+    /* TODO: give a fix to these non-virgins */
 
+    return ret;
+}
 
-void
-dht_selfheal_layout_new_directory (call_frame_t *frame, loc_t *loc,
-                                   dht_layout_t *layout)
+int
+dht_selfheal_new_directory(call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk,
+                           dht_layout_t *layout)
 {
-        xlator_t    *this = NULL;
-        uint32_t     chunk = 0;
-        int          i = 0;
-        uint32_t     start = 0;
-        int          cnt = 0;
-        int          err = 0;
-        int          start_subvol = 0;
-
-        this = frame->this;
-
-        cnt = dht_get_layout_count (this, layout, 1);
-
-        chunk = ((unsigned long) 0xffffffff) / ((cnt) ? cnt : 1);
-
-        start_subvol = dht_selfheal_layout_alloc_start (this, loc, layout);
-
-        /* clear out the range, as we are re-computing here */
-        DHT_RESET_LAYOUT_RANGE (layout);
-        for (i = start_subvol; i < layout->cnt; i++) {
-                err = layout->list[i].err;
-                if (err == -1 || err == ENOENT) {
-                        DHT_SET_LAYOUT_RANGE(layout, i, start, chunk,
-                                             cnt, loc->path);
-                        if (--cnt == 0) {
-                                layout->list[i].stop = 0xffffffff;
-                                goto done;
-                        }
-                        start += chunk;
-                }
-        }
+    dht_local_t *local = NULL;
+    int ret = 0;
+    inode_t *linked_inode = NULL, *inode = NULL;
+    loc_t *loc = NULL;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    int32_t op_errno = EIO;
 
-        for (i = 0; i < start_subvol; i++) {
-                err = layout->list[i].err;
-                if (err == -1 || err == ENOENT) {
-                        DHT_SET_LAYOUT_RANGE(layout, i, start, chunk,
-                                             cnt, loc->path);
-                        if (--cnt == 0) {
-                                layout->list[i].stop = 0xffffffff;
-                                goto done;
-                        }
-                        start += chunk;
-                }
-        }
+    local = frame->local;
 
-done:
-        return;
+    loc = &local->loc;
+
+    gf_uuid_unparse(local->stbuf.ia_gfid, gfid);
+    gf_uuid_unparse(loc->parent->gfid, pgfid);
+
+    linked_inode = inode_link(loc->inode, loc->parent, loc->name,
+                              &local->stbuf);
+    if (!linked_inode) {
+        gf_smsg(frame->this->name, GF_LOG_WARNING, 0, DHT_MSG_LINK_INODE_FAILED,
+                "pgfid=%s", pgfid, "name=%s", loc->name, "gfid=%s", gfid, NULL);
+        ret = -1;
+        goto out;
+    }
+
+    inode = loc->inode;
+    loc->inode = linked_inode;
+    inode_unref(inode);
+
+    local->selfheal.dir_cbk = dir_cbk;
+    local->selfheal.layout = dht_layout_ref(frame->this, layout);
+
+    dht_layout_sort_volname(layout);
+    dht_selfheal_layout_new_directory(frame, &local->loc, layout);
+
+    op_errno = ENOMEM;
+    ret = dht_selfheal_layout_lock(frame, layout, _gf_true,
+                                   dht_selfheal_dir_xattr,
+                                   dht_should_heal_layout);
+
+out:
+    if (ret < 0) {
+        dir_cbk(frame, NULL, frame->this, -1, op_errno, NULL);
+    }
+
+    return 0;
 }
 
 int
-dht_selfheal_dir_getafix (call_frame_t *frame, loc_t *loc,
-                          dht_layout_t *layout)
+dht_fix_directory_layout(call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk,
+                         dht_layout_t *layout)
 {
-        dht_local_t *local = NULL;
-        uint32_t     holes = 0;
-        int          ret = -1;
-        int          i = -1;
-        uint32_t     overlaps = 0;
-
-        local = frame->local;
+    dht_local_t *local = NULL;
+    dht_layout_t *tmp_layout = NULL;
+    int ret = 0;
 
-        holes = local->selfheal.hole_cnt;
-        overlaps = local->selfheal.overlaps_cnt;
+    local = frame->local;
 
-        if (holes || overlaps) {
-                dht_selfheal_layout_new_directory (frame, loc, layout);
-                ret = 0;
-        }
+    local->selfheal.dir_cbk = dir_cbk;
+    local->selfheal.layout = dht_layout_ref(frame->this, layout);
 
-        for (i = 0; i < layout->cnt; i++) {
-                /* directory not present */
-                if (layout->list[i].err == ENOENT) {
-                        ret = 0;
-                        break;
-                }
-        }
+    /* No layout sorting required here */
+    tmp_layout = dht_fix_layout_of_directory(frame, &local->loc, layout);
+    if (!tmp_layout) {
+        return -1;
+    }
 
-        /* TODO: give a fix to these non-virgins */
+    ret = dht_selfheal_layout_lock(frame, tmp_layout, _gf_false,
+                                   dht_fix_dir_xattr, dht_should_fix_layout);
 
-        return ret;
+    return ret;
 }
 
 int
-dht_selfheal_new_directory (call_frame_t *frame,
-                            dht_selfheal_dir_cbk_t dir_cbk,
-                            dht_layout_t *layout)
+dht_selfheal_directory(call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk,
+                       loc_t *loc, dht_layout_t *layout)
 {
-        dht_local_t *local = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *this = NULL;
+    uint32_t down = 0;
+    uint32_t misc = 0;
+    int ret = 0;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    inode_t *linked_inode = NULL, *inode = NULL;
+
+    local = frame->local;
+    this = frame->this;
+
+    local->selfheal.dir_cbk = dir_cbk;
+    local->selfheal.layout = dht_layout_ref(this, layout);
+
+    if (local->need_attrheal) {
+        if (__is_root_gfid(local->stbuf.ia_gfid)) {
+            local->stbuf.ia_gid = local->prebuf.ia_gid;
+            local->stbuf.ia_uid = local->prebuf.ia_uid;
+
+            local->stbuf.ia_ctime = local->prebuf.ia_ctime;
+            local->stbuf.ia_ctime_nsec = local->prebuf.ia_ctime_nsec;
+            local->stbuf.ia_prot = local->prebuf.ia_prot;
+
+        } else if (!IA_ISINVAL(local->mds_stbuf.ia_type)) {
+            local->stbuf = local->mds_stbuf;
+        }
+    }
+
+    if (!__is_root_gfid(local->stbuf.ia_gfid)) {
+        gf_uuid_unparse(local->stbuf.ia_gfid, gfid);
+        gf_uuid_unparse(loc->parent->gfid, pgfid);
+
+        linked_inode = inode_link(loc->inode, loc->parent, loc->name,
+                                  &local->stbuf);
+        if (!linked_inode) {
+            gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LINK_INODE_FAILED,
+                    "pgfid=%s", pgfid, "name=%s", loc->name, "gfid=%s", gfid,
+                    NULL);
+            ret = 0;
+            goto sorry_no_fix;
+        }
 
-        local = frame->local;
+        inode = loc->inode;
+        loc->inode = linked_inode;
+        inode_unref(inode);
+    }
+
+    if (local->need_xattr_heal && (local->mds_xattr)) {
+        dht_dir_set_heal_xattr(this, local, local->xattr, local->mds_xattr,
+                               NULL, NULL);
+        dict_unref(local->mds_xattr);
+        local->mds_xattr = NULL;
+    }
+
+    dht_layout_anomalies(this, loc, layout, &local->selfheal.hole_cnt,
+                         &local->selfheal.overlaps_cnt,
+                         &local->selfheal.missing_cnt, &local->selfheal.down,
+                         &local->selfheal.misc, NULL);
+
+    down = local->selfheal.down;
+    misc = local->selfheal.misc;
+
+    if (down) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_SELFHEAL_FAILED,
+                "path=%s", loc->path, "subvol-down=%d", down, "Not-fixing",
+                "gfid=%s", gfid, NULL);
+        ret = 0;
+        goto sorry_no_fix;
+    }
+
+    if (misc) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_SELFHEAL_FAILED,
+                "path=%s", loc->path, "misc=%d", misc, "unrecoverable-errors",
+                "gfid=%s", gfid, NULL);
+
+        ret = 0;
+        goto sorry_no_fix;
+    }
+
+    dht_layout_sort_volname(layout);
+    local->heal_layout = _gf_true;
+
+    /* Ignore return value as it can be inferred from result of
+     * dht_layout_anomalies
+     */
+    dht_selfheal_dir_getafix(frame, loc, layout);
+
+    if (!(local->selfheal.hole_cnt || local->selfheal.overlaps_cnt ||
+          local->selfheal.missing_cnt)) {
+        local->heal_layout = _gf_false;
+    }
+
+    ret = dht_selfheal_dir_mkdir(frame, loc, layout, 0);
+    if (ret < 0) {
+        ret = 0;
+        goto sorry_no_fix;
+    }
+
+    return 0;
 
-        local->selfheal.dir_cbk = dir_cbk;
-        local->selfheal.layout = dht_layout_ref (frame->this, layout);
+sorry_no_fix:
+    /* TODO: need to put appropriate local->op_errno */
+    dht_selfheal_dir_finish(frame, this, ret, 1);
 
-        dht_layout_sort_volname (layout);
-        dht_selfheal_layout_new_directory (frame, &local->loc, layout);
-        dht_selfheal_dir_xattr (frame, &local->loc, layout);
-        return 0;
+    return 0;
 }
 
 int
-dht_fix_directory_layout (call_frame_t *frame,
-                          dht_selfheal_dir_cbk_t dir_cbk,
-                          dht_layout_t *layout)
+dht_selfheal_restore(call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk,
+                     loc_t *loc, dht_layout_t *layout)
 {
-        dht_local_t  *local = NULL;
-        dht_layout_t *tmp_layout = NULL;
+    int ret = 0;
+    dht_local_t *local = NULL;
 
-        local = frame->local;
+    local = frame->local;
 
-        local->selfheal.dir_cbk = dir_cbk;
-        local->selfheal.layout = dht_layout_ref (frame->this, layout);
+    local->selfheal.dir_cbk = dir_cbk;
+    local->selfheal.layout = dht_layout_ref(frame->this, layout);
 
-        /* No layout sorting required here */
-        tmp_layout = dht_fix_layout_of_directory (frame, &local->loc, layout);
-        if (!tmp_layout) {
-                return -1;
+    ret = dht_selfheal_dir_mkdir(frame, loc, layout, 1);
+
+    return ret;
+}
+
+int
+dht_dir_heal_xattrs(void *data)
+{
+    call_frame_t *frame = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *subvol = NULL;
+    xlator_t *mds_subvol = NULL;
+    xlator_t *this = NULL;
+    dht_conf_t *conf = NULL;
+    dict_t *user_xattr = NULL;
+    dict_t *internal_xattr = NULL;
+    dict_t *mds_xattr = NULL;
+    dict_t *xdata = NULL;
+    int call_cnt = 0;
+    int ret = -1;
+    int uret = 0;
+    int uflag = 0;
+    int i = 0;
+    int xattr_hashed = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    int32_t allzero[1] = {0};
+
+    GF_VALIDATE_OR_GOTO("dht", data, out);
+
+    frame = data;
+    local = frame->local;
+    this = frame->this;
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, local, out);
+    mds_subvol = local->mds_subvol;
+    conf = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, conf, out);
+    gf_uuid_unparse(local->loc.gfid, gfid);
+
+    if (!mds_subvol) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_NO_MDS_SUBVOL, "path=%s",
+                local->loc.path, "gfid=%s", gfid, NULL);
+        goto out;
+    }
+
+    if ((local->loc.inode && gf_uuid_is_null(local->loc.inode->gfid)) ||
+        gf_uuid_is_null(local->loc.gfid)) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_GFID_NOT_PRESENT,
+                "skip-heal path=%s", local->loc.path, "gfid=%s", gfid, NULL);
+        goto out;
+    }
+
+    internal_xattr = dict_new();
+    if (!internal_xattr) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_CREATE_FAILED,
+                "dictionary", NULL);
+        goto out;
+    }
+    xdata = dict_new();
+    if (!xdata) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_CREATE_FAILED,
+                "dictionary", NULL);
+        goto out;
+    }
+
+    call_cnt = conf->subvolume_cnt;
+
+    user_xattr = dict_new();
+    if (!user_xattr) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_CREATE_FAILED,
+                "dictionary", NULL);
+        goto out;
+    }
+
+    ret = syncop_listxattr(local->mds_subvol, &local->loc, &mds_xattr, NULL,
+                           NULL);
+    if (ret < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_LIST_XATTRS_FAILED,
+                "path=%s", local->loc.path, "name=%s", local->mds_subvol->name,
+                NULL);
+    }
+
+    if (!mds_xattr)
+        goto out;
+
+    dht_dir_set_heal_xattr(this, local, user_xattr, mds_xattr, &uret, &uflag);
+
+    /* To set quota related xattr need to set GLUSTERFS_INTERNAL_FOP_KEY
+     * key value to 1
+     */
+    if (dict_get(user_xattr, QUOTA_LIMIT_KEY) ||
+        dict_get(user_xattr, QUOTA_LIMIT_OBJECTS_KEY)) {
+        ret = dict_set_int32(xdata, GLUSTERFS_INTERNAL_FOP_KEY, 1);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                    "key=%s", GLUSTERFS_INTERNAL_FOP_KEY, "path=%s",
+                    local->loc.path, NULL);
+            goto out;
         }
-        dht_fix_dir_xattr (frame, &local->loc, tmp_layout);
+    }
+    if (uret <= 0 && !uflag)
+        goto out;
+
+    for (i = 0; i < call_cnt; i++) {
+        subvol = conf->subvolumes[i];
+        if (subvol == mds_subvol)
+            continue;
+        if (uret || uflag) {
+            /* Custom xattr heal is required - let posix handle it */
+            ret = dict_set_int8(xdata, "sync_backend_xattrs", _gf_true);
+            if (ret) {
+                gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+                        "path=%s", local->loc.path, "key=%s",
+                        "sync_backend_xattrs", NULL);
+                goto out;
+            }
+
+            ret = syncop_setxattr(subvol, &local->loc, user_xattr, 0, xdata,
+                                  NULL);
+            if (ret) {
+                xattr_hashed = 1;
+                gf_smsg(this->name, GF_LOG_ERROR, -ret,
+                        DHT_MSG_DIR_XATTR_HEAL_FAILED,
+                        "set-user-xattr-failed path=%s", local->loc.path,
+                        "subvol=%s", subvol->name, "gfid=%s", gfid, NULL);
+            } else {
+                dict_del(xdata, "sync_backend_xattrs");
+            }
+        }
+    }
+    /* After heal all custom xattr reset internal MDS xattr to 0 */
+    if (!xattr_hashed) {
+        ret = dht_dict_set_array(internal_xattr, conf->mds_xattr_key, allzero,
+                                 1);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_WARNING, ENOMEM, DHT_MSG_DICT_SET_FAILED,
+                    "key=%s", conf->mds_xattr_key, "path=%s", local->loc.path,
+                    NULL);
+            goto out;
+        }
+        ret = syncop_setxattr(mds_subvol, &local->loc, internal_xattr, 0, NULL,
+                              NULL);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, -ret,
+                    DHT_MSG_DIR_XATTR_HEAL_FAILED, "path=%s", local->loc.path,
+                    "subvol=%s", mds_subvol->name, "gfid=%s", gfid, NULL);
+        }
+    }
 
-        return 0;
+out:
+    if (user_xattr)
+        dict_unref(user_xattr);
+    if (mds_xattr)
+        dict_unref(mds_xattr);
+    if (internal_xattr)
+        dict_unref(internal_xattr);
+    if (xdata)
+        dict_unref(xdata);
+    return 0;
 }
 
+int
+dht_dir_heal_xattrs_done(int ret, call_frame_t *sync_frame, void *data)
+{
+    DHT_STACK_DESTROY(sync_frame);
+    return 0;
+}
 
 int
-dht_selfheal_directory (call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk,
-                        loc_t *loc, dht_layout_t *layout)
+dht_dir_attr_heal(void *data)
 {
-        dht_local_t *local    = NULL;
-        uint32_t     down     = 0;
-        uint32_t     misc     = 0;
-        int          ret      = 0;
-        xlator_t    *this     = NULL;
-
-        local = frame->local;
-        this = frame->this;
-
-        dht_layout_anomalies (this, loc, layout,
-                              &local->selfheal.hole_cnt,
-                              &local->selfheal.overlaps_cnt,
-                              NULL, &local->selfheal.down,
-                              &local->selfheal.misc, NULL);
-
-        down     = local->selfheal.down;
-        misc     = local->selfheal.misc;
-
-        local->selfheal.dir_cbk = dir_cbk;
-        local->selfheal.layout = dht_layout_ref (this, layout);
-
-        if (down) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%d subvolumes down -- not fixing", down);
-                ret = 0;
-                goto sorry_no_fix;
+    call_frame_t *frame = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *subvol = NULL;
+    xlator_t *mds_subvol = NULL;
+    xlator_t *this = NULL;
+    dht_conf_t *conf = NULL;
+    int call_cnt = 0;
+    int ret = -1;
+    int i = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    GF_VALIDATE_OR_GOTO("dht", data, out);
+
+    frame = data;
+    local = frame->local;
+    this = frame->this;
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", local, out);
+    conf = this->private;
+    GF_VALIDATE_OR_GOTO("dht", conf, out);
+
+    mds_subvol = local->mds_subvol;
+    call_cnt = conf->subvolume_cnt;
+
+    if (!__is_root_gfid(local->stbuf.ia_gfid) && (!mds_subvol)) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_NO_MDS_SUBVOL, "path=%s",
+                local->loc.path, "gfid=%s", gfid, NULL);
+        goto out;
+    }
+
+    if (!__is_root_gfid(local->stbuf.ia_gfid)) {
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            if (conf->subvolumes[i] == mds_subvol) {
+                if (!conf->subvolume_status[i]) {
+                    gf_smsg(this->name, GF_LOG_ERROR, 0,
+                            DHT_MSG_MDS_DOWN_UNABLE_TO_SET, "path=%s",
+                            local->loc.path, "gfid=%s", gfid, NULL);
+                    goto out;
+                }
+            }
         }
-
-        if (misc) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%d subvolumes have unrecoverable errors", misc);
-                ret = 0;
-                goto sorry_no_fix;
+    }
+
+    for (i = 0; i < call_cnt; i++) {
+        subvol = conf->subvolumes[i];
+        if (!subvol || subvol == mds_subvol)
+            continue;
+        if (__is_root_gfid(local->stbuf.ia_gfid)) {
+            ret = syncop_setattr(
+                subvol, &local->loc, &local->stbuf,
+                (GF_SET_ATTR_UID | GF_SET_ATTR_GID | GF_SET_ATTR_MODE), NULL,
+                NULL, NULL, NULL);
+        } else {
+            ret = syncop_setattr(
+                subvol, &local->loc, &local->mds_stbuf,
+                (GF_SET_ATTR_UID | GF_SET_ATTR_GID | GF_SET_ATTR_MODE), NULL,
+                NULL, NULL, NULL);
         }
 
-        dht_layout_sort_volname (layout);
-        ret = dht_selfheal_dir_getafix (frame, loc, layout);
+        if (ret) {
+            gf_uuid_unparse(local->loc.gfid, gfid);
 
-        if (ret == -1) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "not able to form layout for the directory");
-                goto sorry_no_fix;
+            gf_smsg(this->name, GF_LOG_ERROR, -ret,
+                    DHT_MSG_DIR_ATTR_HEAL_FAILED, "path=%s", local->loc.path,
+                    "subvol=%s", subvol->name, "gfid=%s", gfid, NULL);
         }
+    }
+out:
+    return 0;
+}
 
-        dht_selfheal_dir_mkdir (frame, loc, layout, 0);
+int
+dht_dir_attr_heal_done(int ret, call_frame_t *sync_frame, void *data)
+{
+    DHT_STACK_DESTROY(sync_frame);
+    return 0;
+}
 
-        return 0;
+/* EXIT: dht_update_commit_hash_for_layout */
+static int
+dht_update_commit_hash_for_layout_done(call_frame_t *frame, void *cookie,
+                                       xlator_t *this, int32_t op_ret,
+                                       int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
 
-sorry_no_fix:
-        /* TODO: need to put appropriate local->op_errno */
-        dht_selfheal_dir_finish (frame, this, ret);
+    local = frame->local;
 
-        return 0;
+    /* preserve oldest error */
+    if (op_ret && !local->op_ret) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+    }
+
+    DHT_STACK_UNWIND(setxattr, frame, local->op_ret, local->op_errno, NULL);
+
+    return 0;
 }
 
+static int
+dht_update_commit_hash_for_layout_unlock(call_frame_t *frame, xlator_t *this)
+{
+    dht_local_t *local = NULL;
+    int ret = 0;
+
+    local = frame->local;
+
+    ret = dht_unlock_inodelk(frame, local->lock[0].layout.my_layout.locks,
+                             local->lock[0].layout.my_layout.lk_count,
+                             dht_update_commit_hash_for_layout_done);
+    if (ret < 0) {
+        /* preserve oldest error, just ... */
+        if (!local->op_ret) {
+            local->op_errno = errno;
+            local->op_ret = -1;
+        }
 
-int
-dht_selfheal_restore (call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk,
-                      loc_t *loc, dht_layout_t *layout)
+        gf_smsg(this->name, GF_LOG_WARNING, errno, DHT_MSG_WIND_UNLOCK_FAILED,
+                "path=%s", local->loc.path, NULL);
+
+        dht_update_commit_hash_for_layout_done(frame, NULL, this, 0, 0, NULL);
+    }
+
+    return 0;
+}
+
+static int
+dht_update_commit_hash_for_layout_cbk(call_frame_t *frame, void *cookie,
+                                      xlator_t *this, int op_ret, int op_errno,
+                                      dict_t *xdata)
 {
-        int          ret = 0;
-        dht_local_t *local    = NULL;
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+
+    local = frame->local;
 
-        local = frame->local;
+    LOCK(&frame->lock);
+    /* store first failure, just because */
+    if (op_ret && !local->op_ret) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+    }
+    UNLOCK(&frame->lock);
 
-        local->selfheal.dir_cbk = dir_cbk;
-        local->selfheal.layout = dht_layout_ref (frame->this, layout);
+    this_call_cnt = dht_frame_return(frame);
 
-        ret = dht_selfheal_dir_mkdir (frame, loc, layout, 1);
+    if (is_last_call(this_call_cnt)) {
+        dht_update_commit_hash_for_layout_unlock(frame, this);
+    }
 
-        return ret;
+    return 0;
 }
 
-int
-dht_dir_attr_heal (void *data)
+static int
+dht_update_commit_hash_for_layout_resume(call_frame_t *frame, void *cookie,
+                                         xlator_t *this, int32_t op_ret,
+                                         int32_t op_errno, dict_t *xdata)
 {
-        call_frame_t    *frame = NULL;
-        dht_local_t     *local = NULL;
-        xlator_t        *subvol = NULL;
-        xlator_t        *this  = NULL;
-        dht_conf_t      *conf  = NULL;
-        int              call_cnt = 0;
-        int              ret   = -1;
-        int              i     = 0;
-
-        GF_VALIDATE_OR_GOTO ("dht", data, out);
-
-        frame = data;
-        local = frame->local;
-        this = frame->this;
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-        GF_VALIDATE_OR_GOTO ("dht", local, out);
-        conf = this->private;
-        GF_VALIDATE_OR_GOTO ("dht", conf, out);
-
-        call_cnt = conf->subvolume_cnt;
-
-        for (i = 0; i < call_cnt; i++) {
-                subvol = conf->subvolumes[i];
-                if (!subvol || (subvol == dht_first_up_subvol (this)))
-                        continue;
-                ret = syncop_setattr (subvol, &local->loc, &local->stbuf,
-                                      (GF_SET_ATTR_UID | GF_SET_ATTR_GID),
-                                      NULL, NULL);
-                if (ret)
-                        gf_log ("dht", GF_LOG_ERROR, "Failed to set uid/gid on"
-                                " %s on %s subvol (%s)", local->loc.path,
-                                subvol->name, strerror (errno));
+    dht_local_t *local = NULL;
+    int count = 1, ret = -1, i = 0, j = 0;
+    dht_conf_t *conf = NULL;
+    dht_layout_t *layout = NULL;
+    int32_t *disk_layout = NULL;
+    dict_t **xattr = NULL;
+
+    local = frame->local;
+    conf = frame->this->private;
+    count = conf->local_subvols_cnt;
+    layout = local->layout;
+
+    if (op_ret < 0) {
+        goto err_done;
+    }
+
+    /* We precreate the xattr list as we cannot change call count post the
+     * first wind as we may never continue from there. So we finish prep
+     * work before winding the setxattrs */
+    xattr = GF_CALLOC(count, sizeof(*xattr), gf_common_mt_char);
+    if (!xattr) {
+        local->op_errno = errno;
+
+        gf_smsg(this->name, GF_LOG_WARNING, errno, DHT_MSG_COMMIT_HASH_FAILED,
+                "allocation-failed path=%s", local->loc.path, NULL);
+
+        goto err;
+    }
+
+    for (i = 0; i < count; i++) {
+        /* find the layout index for the subvolume */
+        ret = dht_layout_index_for_subvol(layout, conf->local_subvols[i]);
+        if (ret < 0) {
+            local->op_errno = ENOENT;
+
+            gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_COMMIT_HASH_FAILED,
+                    "path=%s", local->loc.path, "subvol=%s",
+                    conf->local_subvols[i]->name, "find-disk-layout-failed",
+                    NULL);
+
+            goto err;
         }
-out:
-        return 0;
+        j = ret;
+
+        /* update the commit hash for the layout */
+        layout->list[j].commit_hash = layout->commit_hash;
+
+        /* extract the current layout */
+        ret = dht_disk_layout_extract(this, layout, j, &disk_layout);
+        if (ret == -1) {
+            local->op_errno = errno;
+
+            gf_smsg(this->name, GF_LOG_WARNING, errno,
+                    DHT_MSG_COMMIT_HASH_FAILED, "path=%s", local->loc.path,
+                    "subvol=%s", conf->local_subvols[i]->name,
+                    "extract-disk-layout-failed", NULL);
+
+            goto err;
+        }
+
+        xattr[i] = dict_new();
+        if (!xattr[i]) {
+            local->op_errno = errno;
+
+            gf_smsg(this->name, GF_LOG_WARNING, errno,
+                    DHT_MSG_COMMIT_HASH_FAILED, "path=%s Allocation-failed",
+                    local->loc.path, NULL);
+
+            goto err;
+        }
+
+        ret = dict_set_bin(xattr[i], conf->xattr_name, disk_layout, 4 * 4);
+        if (ret != 0) {
+            local->op_errno = ENOMEM;
+
+            gf_smsg(this->name, GF_LOG_WARNING, 0,
+                    DHT_MSG_DIR_SELFHEAL_XATTR_FAILED, "path=%s",
+                    local->loc.path, "subvol=%s", conf->local_subvols[i]->name,
+                    "set-xattr-failed", NULL);
+
+            goto err;
+        }
+        disk_layout = NULL;
+
+        gf_msg_trace(this->name, 0,
+                     "setting commit hash %u on subvolume %s"
+                     " for %s",
+                     layout->list[j].commit_hash, conf->local_subvols[i]->name,
+                     local->loc.path);
+    }
+
+    /* wind the setting of the commit hash across the local subvols */
+    local->call_cnt = count;
+    local->op_ret = 0;
+    local->op_errno = 0;
+    for (i = 0; i < count; i++) {
+        STACK_WIND(frame, dht_update_commit_hash_for_layout_cbk,
+                   conf->local_subvols[i],
+                   conf->local_subvols[i]->fops->setxattr, &local->loc,
+                   xattr[i], 0, NULL);
+    }
+    for (i = 0; i < count; i++)
+        dict_unref(xattr[i]);
+    GF_FREE(xattr);
+
+    return 0;
+err:
+    if (xattr) {
+        for (i = 0; i < count; i++) {
+            if (xattr[i])
+                dict_unref(xattr[i]);
+        }
+
+        GF_FREE(xattr);
+    }
+
+    GF_FREE(disk_layout);
+
+    local->op_ret = -1;
+
+    dht_update_commit_hash_for_layout_unlock(frame, this);
+
+    return 0;
+err_done:
+    local->op_ret = -1;
+
+    dht_update_commit_hash_for_layout_done(frame, NULL, this, 0, 0, NULL);
+
+    return 0;
 }
 
+/* ENTER: dht_update_commit_hash_for_layout (see EXIT above)
+ * This function is invoked from rebalance only.
+ * As a result, the check here is simple enough to see if defrag is present
+ * in the conf, as other data would be populated appropriately if so.
+ * If ever this was to be used in other code paths, checks would need to
+ * change.
+ *
+ * Functional details:
+ *  - Lock the inodes on the subvols that we want the commit hash updated
+ *  - Update each layout with the inode layout, modified to take in the new
+ *    commit hash.
+ *  - Unlock and return.
+ */
 int
-dht_dir_attr_heal_done (int ret, call_frame_t *sync_frame, void *data)
+dht_update_commit_hash_for_layout(call_frame_t *frame)
 {
-        DHT_STACK_DESTROY (sync_frame);
-        return 0;
+    dht_local_t *local = NULL;
+    int count = 1, ret = -1, i = 0;
+    dht_lock_t **lk_array = NULL;
+    dht_conf_t *conf = NULL;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO(frame->this->name, frame->local, err);
+
+    local = frame->local;
+    conf = frame->this->private;
+
+    if (!conf->defrag)
+        goto err;
+
+    count = conf->local_subvols_cnt;
+    lk_array = GF_CALLOC(count, sizeof(*lk_array), gf_common_mt_char);
+    if (lk_array == NULL)
+        goto err;
+
+    for (i = 0; i < count; i++) {
+        lk_array[i] = dht_lock_new(frame->this, conf->local_subvols[i],
+                                   &local->loc, F_WRLCK, DHT_LAYOUT_HEAL_DOMAIN,
+                                   NULL, FAIL_ON_ANY_ERROR);
+        if (lk_array[i] == NULL)
+            goto err;
+    }
+
+    local->lock[0].layout.my_layout.locks = lk_array;
+    local->lock[0].layout.my_layout.lk_count = count;
+
+    ret = dht_blocking_inodelk(frame, lk_array, count,
+                               dht_update_commit_hash_for_layout_resume);
+    if (ret < 0) {
+        local->lock[0].layout.my_layout.locks = NULL;
+        local->lock[0].layout.my_layout.lk_count = 0;
+        goto err;
+    }
+
+    return 0;
+err:
+    if (lk_array != NULL) {
+        dht_lock_array_free(lk_array, count);
+        GF_FREE(lk_array);
+    }
+
+    return -1;
 }
diff --git a/xlators/cluster/dht/src/dht-shared.c b/xlators/cluster/dht/src/dht-shared.c
index 70aac7710fe..bb72b0ffbb5 100644
--- a/xlators/cluster/dht/src/dht-shared.c
+++ b/xlators/cluster/dht/src/dht-shared.c
@@ -8,16 +8,14 @@
   cases as published by the Free Software Foundation.
 */
 
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
 /* TODO: add NS locking */
-
-#include "statedump.h"
+#include <glusterfs/statedump.h>
 #include "dht-common.h"
+#include "dht-messages.h"
+
+#ifndef MAX
+#define MAX(a, b) (((a) > (b)) ? (a) : (b))
+#endif
 
 /* TODO:
    - use volumename in xattr instead of "dht"
@@ -25,734 +23,1082 @@
    - handle all cases in self heal layout reconstruction
    - complete linkfile selfheal
 */
-struct volume_options options[];
 
-void
-dht_layout_dump (dht_layout_t  *layout, const char *prefix)
+static void
+dht_layout_dump(dht_layout_t *layout, const char *prefix)
 {
-
-        char    key[GF_DUMP_MAX_BUF_LEN];
-        int     i = 0;
-
-        if (!layout)
-                goto out;
-        if (!prefix)
-                goto out;
-
-        gf_proc_dump_build_key(key, prefix, "cnt");
-        gf_proc_dump_write(key, "%d", layout->cnt);
-        gf_proc_dump_build_key(key, prefix, "preset");
-        gf_proc_dump_write(key, "%d", layout->preset);
-        gf_proc_dump_build_key(key, prefix, "gen");
-        gf_proc_dump_write(key, "%d", layout->gen);
-        if (layout->type != IA_INVAL) {
-                gf_proc_dump_build_key(key, prefix, "inode type");
-                gf_proc_dump_write(key, "%d", layout->type);
-        }
-
-        if  (!IA_ISDIR (layout->type))
-                goto out;
-
-        for (i = 0; i < layout->cnt; i++) {
-                gf_proc_dump_build_key(key, prefix,"list[%d].err", i);
-                gf_proc_dump_write(key, "%d", layout->list[i].err);
-                gf_proc_dump_build_key(key, prefix,"list[%d].start", i);
-                gf_proc_dump_write(key, "%u", layout->list[i].start);
-                gf_proc_dump_build_key(key, prefix,"list[%d].stop", i);
-                gf_proc_dump_write(key, "%u", layout->list[i].stop);
-                if (layout->list[i].xlator) {
-                        gf_proc_dump_build_key(key, prefix,
-                                               "list[%d].xlator.type", i);
-                        gf_proc_dump_write(key, "%s",
-                                           layout->list[i].xlator->type);
-                        gf_proc_dump_build_key(key, prefix,
-                                               "list[%d].xlator.name", i);
-                        gf_proc_dump_write(key, "%s",
-                                           layout->list[i].xlator->name);
-                }
+    char key[GF_DUMP_MAX_BUF_LEN];
+    int i = 0;
+
+    if (!layout)
+        goto out;
+
+    gf_proc_dump_build_key(key, prefix, "cnt");
+    gf_proc_dump_write(key, "%d", layout->cnt);
+    gf_proc_dump_build_key(key, prefix, "preset");
+    gf_proc_dump_write(key, "%d", layout->preset);
+    gf_proc_dump_build_key(key, prefix, "gen");
+    gf_proc_dump_write(key, "%d", layout->gen);
+    if (layout->type != IA_INVAL) {
+        gf_proc_dump_build_key(key, prefix, "inode type");
+        gf_proc_dump_write(key, "%d", layout->type);
+    }
+
+    if (!IA_ISDIR(layout->type))
+        goto out;
+
+    for (i = 0; i < layout->cnt; i++) {
+        gf_proc_dump_build_key(key, prefix, "list[%d].err", i);
+        gf_proc_dump_write(key, "%d", layout->list[i].err);
+        gf_proc_dump_build_key(key, prefix, "list[%d].start", i);
+        gf_proc_dump_write(key, "0x%x", layout->list[i].start);
+        gf_proc_dump_build_key(key, prefix, "list[%d].stop", i);
+        gf_proc_dump_write(key, "0x%x", layout->list[i].stop);
+        if (layout->list[i].xlator) {
+            gf_proc_dump_build_key(key, prefix, "list[%d].xlator.type", i);
+            gf_proc_dump_write(key, "%s", layout->list[i].xlator->type);
+            gf_proc_dump_build_key(key, prefix, "list[%d].xlator.name", i);
+            gf_proc_dump_write(key, "%s", layout->list[i].xlator->name);
         }
+    }
 
 out:
-        return;
+    return;
 }
 
-
 int32_t
-dht_priv_dump (xlator_t *this)
+dht_priv_dump(xlator_t *this)
 {
-        char            key_prefix[GF_DUMP_MAX_BUF_LEN];
-        char            key[GF_DUMP_MAX_BUF_LEN];
-        int             i = 0;
-        dht_conf_t      *conf = NULL;
-        int             ret = -1;
+    char key_prefix[GF_DUMP_MAX_BUF_LEN];
+    char key[GF_DUMP_MAX_BUF_LEN];
+    int i = 0;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
 
-        if (!this)
-                goto out;
+    if (!this)
+        goto out;
 
-        conf = this->private;
-        if (!conf)
-                goto out;
+    conf = this->private;
+    if (!conf)
+        goto out;
 
-        ret = TRY_LOCK(&conf->subvolume_lock);
-        if (ret != 0) {
-                return ret;
+    ret = TRY_LOCK(&conf->subvolume_lock);
+    if (ret != 0) {
+        return ret;
+    }
+
+    gf_proc_dump_add_section("xlator.cluster.dht.%s.priv", this->name);
+    gf_proc_dump_build_key(key_prefix, "xlator.cluster.dht", "%s.priv",
+                           this->name);
+    gf_proc_dump_write("subvol_cnt", "%d", conf->subvolume_cnt);
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        snprintf(key, sizeof(key), "subvolumes[%d]", i);
+        gf_proc_dump_write(key, "%s.%s", conf->subvolumes[i]->type,
+                           conf->subvolumes[i]->name);
+        if (conf->file_layouts && conf->file_layouts[i]) {
+            snprintf(key, sizeof(key), "file_layouts[%d]", i);
+            dht_layout_dump(conf->file_layouts[i], key);
         }
-
-        gf_proc_dump_add_section("xlator.cluster.dht.%s.priv", this->name);
-        gf_proc_dump_build_key(key_prefix,"xlator.cluster.dht","%s.priv",
-                               this->name);
-        gf_proc_dump_write("subvol_cnt","%d", conf->subvolume_cnt);
+        if (conf->dir_layouts && conf->dir_layouts[i]) {
+            snprintf(key, sizeof(key), "dir_layouts[%d]", i);
+            dht_layout_dump(conf->dir_layouts[i], key);
+        }
+        if (conf->subvolume_status) {
+            snprintf(key, sizeof(key), "subvolume_status[%d]", i);
+            gf_proc_dump_write(key, "%d", (int)conf->subvolume_status[i]);
+        }
+    }
+
+    gf_proc_dump_write("search_unhashed", "%d", conf->search_unhashed);
+    gf_proc_dump_write("gen", "%d", conf->gen);
+    gf_proc_dump_write("min_free_disk", "%lf", conf->min_free_disk);
+    gf_proc_dump_write("min_free_inodes", "%lf", conf->min_free_inodes);
+    gf_proc_dump_write("disk_unit", "%c", conf->disk_unit);
+    gf_proc_dump_write("refresh_interval", "%d", conf->refresh_interval);
+    gf_proc_dump_write("unhashed_sticky_bit", "%d", conf->unhashed_sticky_bit);
+    gf_proc_dump_write("use-readdirp", "%d", conf->use_readdirp);
+
+    if (conf->du_stats && conf->subvolume_status) {
         for (i = 0; i < conf->subvolume_cnt; i++) {
-                sprintf (key, "subvolumes[%d]", i);
-                gf_proc_dump_write(key, "%s.%s", conf->subvolumes[i]->type,
-                                   conf->subvolumes[i]->name);
-                if (conf->file_layouts && conf->file_layouts[i]){
-                        sprintf (key, "file_layouts[%d]", i);
-                        dht_layout_dump(conf->file_layouts[i], key);
-                }
-                if (conf->dir_layouts && conf->dir_layouts[i]) {
-                        sprintf (key, "dir_layouts[%d]", i);
-                        dht_layout_dump(conf->dir_layouts[i], key);
-                }
-                if (conf->subvolume_status) {
-
-                        sprintf (key, "subvolume_status[%d]", i);
-                        gf_proc_dump_write(key, "%d",
-                                           (int)conf->subvolume_status[i]);
-                }
+            if (!conf->subvolume_status[i])
+                continue;
 
-        }
+            snprintf(key, sizeof(key), "subvolumes[%d]", i);
+            gf_proc_dump_write(key, "%s", conf->subvolumes[i]->name);
+
+            snprintf(key, sizeof(key), "du_stats[%d].avail_percent", i);
+            gf_proc_dump_write(key, "%lf", conf->du_stats[i].avail_percent);
+
+            snprintf(key, sizeof(key), "du_stats[%d].avail_space", i);
+            gf_proc_dump_write(key, "%" PRIu64, conf->du_stats[i].avail_space);
+
+            snprintf(key, sizeof(key), "du_stats[%d].avail_inodes", i);
+            gf_proc_dump_write(key, "%lf", conf->du_stats[i].avail_inodes);
 
-        gf_proc_dump_write("search_unhashed", "%d", conf->search_unhashed);
-        gf_proc_dump_write("gen", "%d", conf->gen);
-        gf_proc_dump_write("min_free_disk", "%lf", conf->min_free_disk);
-	gf_proc_dump_write("min_free_inodes", "%lf", conf->min_free_inodes);
-        gf_proc_dump_write("disk_unit", "%c", conf->disk_unit);
-        gf_proc_dump_write("refresh_interval", "%d", conf->refresh_interval);
-        gf_proc_dump_write("unhashed_sticky_bit", "%d", conf->unhashed_sticky_bit);
-        if (conf ->du_stats) {
-                gf_proc_dump_write("du_stats.avail_percent", "%lf",
-                                   conf->du_stats->avail_percent);
-                gf_proc_dump_write("du_stats.avail_space", "%lu",
-                                   conf->du_stats->avail_space);
-		gf_proc_dump_write("du_stats.avail_inodes", "%lf",
-                                   conf->du_stats->avail_inodes);
-                gf_proc_dump_write("du_stats.log", "%lu", conf->du_stats->log);
+            snprintf(key, sizeof(key), "du_stats[%d].log", i);
+            gf_proc_dump_write(key, "%" PRIu32, conf->du_stats[i].log);
         }
+    }
 
-        if (conf->last_stat_fetch.tv_sec)
-                gf_proc_dump_write("last_stat_fetch", "%s",
-                                    ctime(&conf->last_stat_fetch.tv_sec));
+    if (conf->last_stat_fetch)
+        gf_proc_dump_write("last_stat_fetch", "%s",
+                           ctime(&conf->last_stat_fetch));
 
-        UNLOCK(&conf->subvolume_lock);
+    UNLOCK(&conf->subvolume_lock);
 
 out:
-        return ret;
+    return ret;
 }
 
 int32_t
-dht_inodectx_dump (xlator_t *this, inode_t *inode)
+dht_inodectx_dump(xlator_t *this, inode_t *inode)
 {
-        int             ret = -1;
-        dht_layout_t    *layout = NULL;
+    int ret = -1;
+    dht_layout_t *layout = NULL;
 
-        if (!this)
-                goto out;
-        if (!inode)
-                goto out;
+    if (!this)
+        goto out;
+    if (!inode)
+        goto out;
 
-        ret = dht_inode_ctx_layout_get (inode, this, &layout);
+    ret = dht_inode_ctx_layout_get(inode, this, &layout);
 
-        if ((ret != 0) || !layout)
-                return ret;
+    if ((ret != 0) || !layout)
+        return ret;
 
-        gf_proc_dump_add_section("xlator.cluster.dht.%s.inode", this->name);
-        dht_layout_dump(layout, "layout");
+    gf_proc_dump_add_section("xlator.cluster.dht.%s.inode", this->name);
+    dht_layout_dump(layout, "layout");
 
 out:
-        return ret;
+    return ret;
 }
 
 void
-dht_fini (xlator_t *this)
+dht_fini(xlator_t *this)
 {
-        int         i = 0;
-        dht_conf_t *conf = NULL;
+    int i = 0;
+    dht_conf_t *conf = NULL;
+
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+
+    conf = this->private;
+    this->private = NULL;
+    if (conf) {
+        if (conf->file_layouts) {
+            for (i = 0; i < conf->subvolume_cnt; i++) {
+                GF_FREE(conf->file_layouts[i]);
+            }
+            GF_FREE(conf->file_layouts);
+        }
 
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        dict_unref(conf->leaf_to_subvol);
 
-        conf = this->private;
-        this->private = NULL;
-        if (conf) {
-                if (conf->file_layouts) {
-                        for (i = 0; i < conf->subvolume_cnt; i++) {
-                                GF_FREE (conf->file_layouts[i]);
-                        }
-                        GF_FREE (conf->file_layouts);
-                }
+        /* allocated in dht_init_subvolumes() */
+        GF_FREE(conf->subvolumes);
+        GF_FREE(conf->subvolume_status);
+        GF_FREE(conf->last_event);
+        GF_FREE(conf->subvol_up_time);
+        GF_FREE(conf->du_stats);
+        GF_FREE(conf->decommissioned_bricks);
 
-                GF_FREE (conf->subvolumes);
+        /* allocated in dht_init() */
+        GF_FREE(conf->mds_xattr_key);
+        GF_FREE(conf->link_xattr_name);
+        GF_FREE(conf->commithash_xattr_name);
+        GF_FREE(conf->wild_xattr_name);
 
-                GF_FREE (conf->subvolume_status);
+        /* allocated in dht_init_regex() */
+        if (conf->rsync_regex_valid)
+            regfree(&conf->rsync_regex);
+        if (conf->extra_regex_valid)
+            regfree(&conf->extra_regex);
 
-                GF_FREE (conf);
-        }
+        synclock_destroy(&conf->link_lock);
+
+        if (conf->lock_pool)
+            mem_pool_destroy(conf->lock_pool);
+
+        GF_FREE(conf);
+    }
 out:
-        return;
+    return;
 }
 
 int32_t
-mem_acct_init (xlator_t *this)
+mem_acct_init(xlator_t *this)
 {
-        int     ret = -1;
+    int ret = -1;
 
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
 
-        ret = xlator_mem_acct_init (this, gf_dht_mt_end + 1);
+    ret = xlator_mem_acct_init(this, gf_dht_mt_end + 1);
 
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
-                        "failed");
-                return ret;
-        }
-out:
+    if (ret != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_NO_MEMORY,
+               "Memory accounting init failed");
         return ret;
+    }
+out:
+    return ret;
 }
 
-
-int
-dht_parse_decommissioned_bricks (xlator_t *this, dht_conf_t *conf,
-                                 const char *bricks)
+static int
+dht_parse_decommissioned_bricks(xlator_t *this, dht_conf_t *conf,
+                                const char *bricks)
 {
-        int         i  = 0;
-        int         ret  = -1;
-        char       *tmpstr = NULL;
-        char       *dup_brick = NULL;
-        char       *node = NULL;
-
-        if (!conf || !bricks)
-                goto out;
-
-        dup_brick = gf_strdup (bricks);
-        node = strtok_r (dup_brick, ",", &tmpstr);
-        while (node) {
-                for (i = 0; i < conf->subvolume_cnt; i++) {
-                        if (!strcmp (conf->subvolumes[i]->name, node)) {
-                                conf->decommissioned_bricks[i] =
-                                        conf->subvolumes[i];
-                                        conf->decommission_subvols_cnt++;
-                                gf_log (this->name, GF_LOG_INFO,
-                                        "decommissioning subvolume %s",
-                                        conf->subvolumes[i]->name);
-                                break;
-                        }
-                }
-                if (i == conf->subvolume_cnt) {
-                        /* Wrong node given. */
-                        goto out;
-                }
-                node = strtok_r (NULL, ",", &tmpstr);
+    int i = 0;
+    int ret = -1;
+    char *tmpstr = NULL;
+    char *dup_brick = NULL;
+    char *node = NULL;
+
+    if (!conf || !bricks)
+        goto out;
+
+    dup_brick = gf_strdup(bricks);
+    if (dup_brick == NULL) {
+        goto out;
+    }
+
+    node = strtok_r(dup_brick, ",", &tmpstr);
+    while (node) {
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            if (!strcmp(conf->subvolumes[i]->name, node)) {
+                conf->decommissioned_bricks[i] = conf->subvolumes[i];
+                conf->decommission_subvols_cnt++;
+                gf_msg(this->name, GF_LOG_INFO, 0,
+                       DHT_MSG_SUBVOL_DECOMMISSION_INFO,
+                       "decommissioning subvolume %s",
+                       conf->subvolumes[i]->name);
+                break;
+            }
+        }
+        if (i == conf->subvolume_cnt) {
+            /* Wrong node given. */
+            goto out;
         }
+        node = strtok_r(NULL, ",", &tmpstr);
+    }
 
-        ret = 0;
-        conf->decommission_in_progress = 1;
+    ret = 0;
+    conf->decommission_in_progress = 1;
 out:
-        GF_FREE (dup_brick);
+    GF_FREE(dup_brick);
 
-        return ret;
+    return ret;
 }
 
-
-int
-dht_decommissioned_remove (xlator_t *this, dht_conf_t *conf)
+static void
+dht_decommissioned_remove(xlator_t *this, dht_conf_t *conf)
 {
-        int         i  = 0;
-        int         ret  = -1;
-
-        if (!conf)
-                goto out;
+    int i = 0;
 
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                if (conf->decommissioned_bricks[i]) {
-                        conf->decommissioned_bricks[i] = NULL;
-                        conf->decommission_subvols_cnt--;
-                }
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (conf->decommissioned_bricks[i]) {
+            conf->decommissioned_bricks[i] = NULL;
+            conf->decommission_subvols_cnt--;
         }
-
-        ret = 0;
-out:
-
-        return ret;
+    }
 }
-void
-dht_init_regex (xlator_t *this, dict_t *odict, char *name,
-                regex_t *re, gf_boolean_t *re_valid)
+
+static void
+dht_init_regex(xlator_t *this, dict_t *odict, char *name, regex_t *re,
+               gf_boolean_t *re_valid, dht_conf_t *conf)
 {
-        char    *temp_str;
+    char *temp_str = NULL;
 
-        if (dict_get_str (odict, name, &temp_str) != 0) {
-                if (strcmp(name,"rsync-hash-regex")) {
-                        return;
-                }
-                temp_str = "^\\.(.+)\\.[^.]+$";
+    if (dict_get_str(odict, name, &temp_str) != 0) {
+        if (strcmp(name, "rsync-hash-regex")) {
+            return;
         }
+        temp_str = "^\\.(.+)\\.[^.]+$";
+    }
 
+    LOCK(&conf->lock);
+    {
         if (*re_valid) {
-                regfree(re);
-                *re_valid = _gf_false;
+            regfree(re);
+            *re_valid = _gf_false;
         }
 
-        if (!strcmp(temp_str,"none")) {
-                return;
+        if (!strcmp(temp_str, "none")) {
+            goto unlock;
         }
 
-        if (regcomp(re,temp_str,REG_EXTENDED) == 0) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "using regex %s = %s", name, temp_str);
-                *re_valid = _gf_true;
-        }
-        else {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "compiling regex %s failed", temp_str);
+        if (regcomp(re, temp_str, REG_EXTENDED) == 0) {
+            gf_msg_debug(this->name, 0, "using regex %s = %s", name, temp_str);
+            *re_valid = _gf_true;
+        } else {
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_REGEX_INFO,
+                   "compiling regex %s failed", temp_str);
         }
+    }
+unlock:
+    UNLOCK(&conf->lock);
 }
 
 int
-dht_reconfigure (xlator_t *this, dict_t *options)
+dht_set_subvol_range(xlator_t *this)
 {
-        dht_conf_t      *conf = NULL;
-        char            *temp_str = NULL;
-        gf_boolean_t     search_unhashed;
-        int              ret = -1;
-
-        GF_VALIDATE_OR_GOTO ("dht", this, out);
-        GF_VALIDATE_OR_GOTO ("dht", options, out);
-
-        conf = this->private;
-        if (!conf)
-                return 0;
-
-        if (dict_get_str (options, "lookup-unhashed", &temp_str) == 0) {
-                /* If option is not "auto", other options _should_ be boolean*/
-                if (strcasecmp (temp_str, "auto")) {
-                        if (!gf_string2boolean (temp_str, &search_unhashed)) {
-                                gf_log(this->name, GF_LOG_DEBUG, "Reconfigure:"
-                                       " lookup-unhashed reconfigured (%s)",
-                                       temp_str);
-                                conf->search_unhashed = search_unhashed;
-                        } else {
-                                gf_log(this->name, GF_LOG_ERROR, "Reconfigure:"
-                                       " lookup-unhashed should be boolean,"
-                                       " not (%s), defaulting to (%d)",
-                                       temp_str, conf->search_unhashed);
-                                //return -1;
-                                ret = -1;
-                                goto out;
-                        }
-                } else {
-                        gf_log(this->name, GF_LOG_DEBUG, "Reconfigure:"
-                               " lookup-unhashed reconfigured auto ");
-                        conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_AUTO;
-                }
-        }
+    int ret = -1;
+    dht_conf_t *conf = NULL;
 
-	GF_OPTION_RECONF ("min-free-disk", conf->min_free_disk, options,
-                          percent_or_size, out);
-        /* option can be any one of percent or bytes */
-        conf->disk_unit = 0;
-        if (conf->min_free_disk < 100.0)
-                conf->disk_unit = 'p';
+    conf = this->private;
 
-	GF_OPTION_RECONF ("min-free-inodes", conf->min_free_inodes, options,
-                          percent, out);
+    if (!conf)
+        goto out;
 
-        GF_OPTION_RECONF ("directory-layout-spread", conf->dir_spread_cnt,
-                          options, uint32, out);
+    conf->leaf_to_subvol = dict_new();
+    if (!conf->leaf_to_subvol)
+        goto out;
 
-        GF_OPTION_RECONF ("readdir-optimize", conf->readdir_optimize, options,
-                          bool, out);
-        if (conf->defrag) {
-                GF_OPTION_RECONF ("rebalance-stats", conf->defrag->stats,
-                                  options, bool, out);
+    ret = glusterfs_reachable_leaves(this, conf->leaf_to_subvol);
+
+out:
+    return ret;
+}
+
+static int
+dht_configure_throttle(xlator_t *this, dht_conf_t *conf, char *temp_str)
+{
+    int rebal_thread_count = 0;
+    int ret = 0;
+
+    pthread_mutex_lock(&conf->defrag->dfq_mutex);
+    {
+        if (!strcasecmp(temp_str, "lazy")) {
+            conf->defrag->recon_thread_count = 1;
+        } else if (!strcasecmp(temp_str, "normal")) {
+            conf->defrag->recon_thread_count = 2;
+        } else if (!strcasecmp(temp_str, "aggressive")) {
+            conf->defrag->recon_thread_count = MAX(MAX_REBAL_THREADS - 4, 4);
+        } else if ((gf_string2int(temp_str, &rebal_thread_count) == 0)) {
+            if ((rebal_thread_count > 0) &&
+                (rebal_thread_count <= MAX_REBAL_THREADS)) {
+                conf->defrag->recon_thread_count = rebal_thread_count;
+                pthread_mutex_unlock(&conf->defrag->dfq_mutex);
+                gf_msg(this->name, GF_LOG_INFO, 0, 0,
+                       "rebal thread count configured to %d",
+                       rebal_thread_count);
+                goto out;
+            } else {
+                pthread_mutex_unlock(&conf->defrag->dfq_mutex);
+                gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INVALID_OPTION,
+                       "Invalid option: Reconfigure: "
+                       "rebal-throttle should be "
+                       "within range of 0 and maximum number of"
+                       " cores available");
+                ret = -1;
+                goto out;
+            }
+        } else {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INVALID_OPTION,
+                   "Invalid option: Reconfigure: "
+                   "rebal-throttle should be {lazy|normal|aggressive}"
+                   " or a number up to the number of cores available,"
+                   " not (%s), defaulting to (%d)",
+                   temp_str, conf->dthrottle);
+            ret = -1;
         }
+    }
+    pthread_mutex_unlock(&conf->defrag->dfq_mutex);
+
+out:
+    return ret;
+}
+
+int
+dht_reconfigure(xlator_t *this, dict_t *options)
+{
+    dht_conf_t *conf = NULL;
+    char *temp_str = NULL;
+    gf_boolean_t search_unhashed;
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", options, out);
+
+    conf = this->private;
+    if (!conf)
+        return 0;
 
-        if (dict_get_str (options, "decommissioned-bricks", &temp_str) == 0) {
-                ret = dht_parse_decommissioned_bricks (this, conf, temp_str);
-                if (ret == -1)
-                        goto out;
+    if (dict_get_str(options, "lookup-unhashed", &temp_str) == 0) {
+        /* If option is not "auto", other options _should_ be boolean*/
+        if (strcasecmp(temp_str, "auto")) {
+            if (!gf_string2boolean(temp_str, &search_unhashed)) {
+                gf_msg_debug(this->name, 0,
+                             "Reconfigure: "
+                             "lookup-unhashed reconfigured(%s)",
+                             temp_str);
+                conf->search_unhashed = search_unhashed;
+            } else {
+                gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INVALID_OPTION,
+                       "Invalid option: Reconfigure: "
+                       "lookup-unhashed should be boolean,"
+                       " not (%s), defaulting to (%d)",
+                       temp_str, conf->search_unhashed);
+                ret = -1;
+                goto out;
+            }
         } else {
-                ret = dht_decommissioned_remove (this, conf);
-                if (ret == -1)
-                        goto out;
+            gf_msg_debug(this->name, 0,
+                         "Reconfigure:"
+                         " lookup-unhashed reconfigured auto ");
+            conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_AUTO;
         }
+    }
+
+    GF_OPTION_RECONF("lookup-optimize", conf->lookup_optimize, options, bool,
+                     out);
+
+    GF_OPTION_RECONF("min-free-disk", conf->min_free_disk, options,
+                     percent_or_size, out);
+    /* option can be any one of percent or bytes */
+    conf->disk_unit = 0;
+    if (conf->min_free_disk < 100.0)
+        conf->disk_unit = 'p';
+
+    GF_OPTION_RECONF("min-free-inodes", conf->min_free_inodes, options, percent,
+                     out);
 
-        dht_init_regex (this, options, "rsync-hash-regex",
-                        &conf->rsync_regex, &conf->rsync_regex_valid);
-        dht_init_regex (this, options, "extra-hash-regex",
-                        &conf->extra_regex, &conf->extra_regex_valid);
+    GF_OPTION_RECONF("directory-layout-spread", conf->dir_spread_cnt, options,
+                     uint32, out);
 
-        ret = 0;
+    GF_OPTION_RECONF("readdir-optimize", conf->readdir_optimize, options, bool,
+                     out);
+    GF_OPTION_RECONF("randomize-hash-range-by-gfid", conf->randomize_by_gfid,
+                     options, bool, out);
+
+    GF_OPTION_RECONF("lock-migration", conf->lock_migration_enabled, options,
+                     bool, out);
+
+    GF_OPTION_RECONF("force-migration", conf->force_migration, options, bool,
+                     out);
+
+    if (conf->defrag) {
+        if (dict_get_str(options, "rebal-throttle", &temp_str) == 0) {
+            ret = dht_configure_throttle(this, conf, temp_str);
+            if (ret == -1)
+                goto out;
+        }
+    }
+
+    if (conf->defrag) {
+        conf->defrag->lock_migration_enabled = conf->lock_migration_enabled;
+    }
+
+    if (conf->defrag) {
+        GF_OPTION_RECONF("rebalance-stats", conf->defrag->stats, options, bool,
+                         out);
+    }
+
+    if (dict_get_str(options, "decommissioned-bricks", &temp_str) == 0) {
+        ret = dht_parse_decommissioned_bricks(this, conf, temp_str);
+        if (ret == -1)
+            goto out;
+    } else {
+        dht_decommissioned_remove(this, conf);
+    }
+
+    dht_init_regex(this, options, "rsync-hash-regex", &conf->rsync_regex,
+                   &conf->rsync_regex_valid, conf);
+    dht_init_regex(this, options, "extra-hash-regex", &conf->extra_regex,
+                   &conf->extra_regex_valid, conf);
+
+    GF_OPTION_RECONF("weighted-rebalance", conf->do_weighting, options, bool,
+                     out);
+
+    GF_OPTION_RECONF("use-readdirp", conf->use_readdirp, options, bool, out);
+    ret = 0;
 out:
-        return ret;
+    return ret;
 }
 
 static int
-gf_defrag_pattern_list_fill (xlator_t *this, gf_defrag_info_t *defrag, char *data)
+gf_defrag_pattern_list_fill(xlator_t *this, gf_defrag_info_t *defrag,
+                            char *data)
 {
-        int                    ret = -1;
-        char                  *tmp_str = NULL;
-        char                  *tmp_str1 = NULL;
-        char                  *dup_str = NULL;
-        char                  *num = NULL;
-        char                  *pattern_str = NULL;
-        char                  *pattern = NULL;
-        gf_defrag_pattern_list_t *temp_list = NULL;
-        gf_defrag_pattern_list_t *pattern_list = NULL;
-
-        if (!this || !defrag || !data)
-                goto out;
-
-        /* Get the pattern for pattern list. "pattern:<optional-size>"
-         * eg: *avi, *pdf:10MB, *:1TB
-         */
-        pattern_str = strtok_r (data, ",", &tmp_str);
-        while (pattern_str) {
-                dup_str = gf_strdup (pattern_str);
-                pattern_list = GF_CALLOC (1, sizeof (gf_defrag_pattern_list_t),
-                                        1);
-                if (!pattern_list) {
-                        goto out;
-                }
-                pattern = strtok_r (dup_str, ":", &tmp_str1);
-                num = strtok_r (NULL, ":", &tmp_str1);
-                if (!pattern)
-                        goto out;
-                if (!num) {
-                        if (gf_string2bytesize(pattern, &pattern_list->size)
-                             == 0) {
-                                pattern = "*";
-                        }
-                } else if (gf_string2bytesize (num, &pattern_list->size) != 0) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "invalid number format \"%s\"", num);
-                        goto out;
-                }
-                memcpy (pattern_list->path_pattern, pattern, strlen (dup_str));
-
-                if (!defrag->defrag_pattern)
-                        temp_list = NULL;
-                else
-                        temp_list = defrag->defrag_pattern;
-
-                pattern_list->next = temp_list;
-
-                defrag->defrag_pattern = pattern_list;
-                pattern_list = NULL;
-
-                GF_FREE (dup_str);
-                dup_str = NULL;
-
-                pattern_str = strtok_r (NULL, ",", &tmp_str);
+    int ret = -1;
+    char *tmp_str = NULL;
+    char *tmp_str1 = NULL;
+    char *dup_str = NULL;
+    char *num = NULL;
+    char *pattern_str = NULL;
+    char *pattern = NULL;
+    gf_defrag_pattern_list_t *temp_list = NULL;
+    gf_defrag_pattern_list_t *pattern_list = NULL;
+
+    if (!this || !defrag || !data)
+        goto out;
+
+    /* Get the pattern for pattern list. "pattern:<optional-size>"
+     * eg: *avi, *pdf:10MB, *:1TB
+     */
+    pattern_str = strtok_r(data, ",", &tmp_str);
+    while (pattern_str) {
+        dup_str = gf_strdup(pattern_str);
+        if (!dup_str)
+            goto out;
+        pattern_list = GF_CALLOC(1, sizeof(gf_defrag_pattern_list_t), 1);
+        if (!pattern_list) {
+            goto out;
+        }
+        pattern = strtok_r(dup_str, ":", &tmp_str1);
+        num = strtok_r(NULL, ":", &tmp_str1);
+        if (!pattern)
+            goto out;
+        if (!num) {
+            if (gf_string2bytesize_uint64(pattern, &pattern_list->size) == 0) {
+                pattern = "*";
+            }
+        } else if (gf_string2bytesize_uint64(num, &pattern_list->size) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INVALID_OPTION,
+                   "Invalid option. Defrag pattern:"
+                   " Invalid number format \"%s\"",
+                   num);
+            goto out;
         }
+        memcpy(pattern_list->path_pattern, pattern, strlen(dup_str));
+
+        if (!defrag->defrag_pattern)
+            temp_list = NULL;
+        else
+            temp_list = defrag->defrag_pattern;
+
+        pattern_list->next = temp_list;
+
+        defrag->defrag_pattern = pattern_list;
+        pattern_list = NULL;
+
+        GF_FREE(dup_str);
+        dup_str = NULL;
 
-        ret = 0;
+        pattern_str = strtok_r(NULL, ",", &tmp_str);
+    }
+
+    ret = 0;
 out:
-        if (ret)
-                GF_FREE (pattern_list);
-        GF_FREE (dup_str);
+    if (ret)
+        GF_FREE(pattern_list);
+    GF_FREE(dup_str);
 
-        return ret;
+    return ret;
 }
 
-int
-dht_init (xlator_t *this)
+static int
+dht_init_methods(xlator_t *this)
 {
-        dht_conf_t                      *conf           = NULL;
-        char                            *temp_str       = NULL;
-        int                              ret            = -1;
-        int                              i              = 0;
-        gf_defrag_info_t                *defrag         = NULL;
-        int                              cmd            = 0;
-        char                            *node_uuid      = NULL;
+    int ret = -1;
+    dht_conf_t *conf = NULL;
+    dht_methods_t *methods = NULL;
 
+    GF_VALIDATE_OR_GOTO("dht", this, err);
 
-        GF_VALIDATE_OR_GOTO ("dht", this, err);
+    conf = this->private;
+    methods = &(conf->methods);
 
-        if (!this->children) {
-                gf_log (this->name, GF_LOG_CRITICAL,
-                        "Distribute needs more than one subvolume");
-                return -1;
-        }
+    methods->migration_get_dst_subvol = dht_migration_get_dst_subvol;
+    methods->migration_other = NULL;
+    methods->layout_search = dht_layout_search;
 
-        if (!this->parents) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "dangling volume. check volfile");
-        }
+    ret = 0;
+err:
+    return ret;
+}
 
-        conf = GF_CALLOC (1, sizeof (*conf), gf_dht_mt_dht_conf_t);
-        if (!conf) {
-                goto err;
-        }
+int
+dht_init(xlator_t *this)
+{
+    dht_conf_t *conf = NULL;
+    char *temp_str = NULL;
+    int ret = -1;
+    int i = 0;
+    gf_defrag_info_t *defrag = NULL;
+    int cmd = 0;
+    char *node_uuid = NULL;
+    uint32_t commit_hash = 0;
+
+    GF_VALIDATE_OR_GOTO("dht", this, err);
+
+    if (!this->children) {
+        gf_msg(this->name, GF_LOG_CRITICAL, 0, DHT_MSG_INVALID_CONFIGURATION,
+               "Distribute needs more than one subvolume");
+        return -1;
+    }
 
-        ret = dict_get_int32 (this->options, "rebalance-cmd", &cmd);
+    if (!this->parents) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_INVALID_CONFIGURATION,
+               "dangling volume. check volfile");
+    }
 
-        if (cmd) {
-                defrag = GF_CALLOC (1, sizeof (gf_defrag_info_t),
-                                    gf_defrag_info_mt);
+    conf = GF_CALLOC(1, sizeof(*conf), gf_dht_mt_dht_conf_t);
+    if (!conf) {
+        goto err;
+    }
 
-                GF_VALIDATE_OR_GOTO (this->name, defrag, err);
+    LOCK_INIT(&conf->subvolume_lock);
+    LOCK_INIT(&conf->layout_lock);
+    LOCK_INIT(&conf->lock);
+    synclock_init(&conf->link_lock, SYNC_LOCK_DEFAULT);
 
-                LOCK_INIT (&defrag->lock);
+    /* We get the commit-hash to set only for rebalance process */
+    if (dict_get_uint32(this->options, "commit-hash", &commit_hash) == 0) {
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_COMMIT_HASH_INFO,
+               "%s using commit hash %u", __func__, commit_hash);
+        conf->vol_commit_hash = commit_hash;
+        conf->vch_forced = _gf_true;
+    }
 
-                defrag->is_exiting = 0;
+    ret = dict_get_int32(this->options, "rebalance-cmd", &cmd);
 
-                conf->defrag = defrag;
+    if (cmd) {
+        defrag = GF_CALLOC(1, sizeof(gf_defrag_info_t), gf_defrag_info_mt);
 
-                ret = dict_get_str (this->options, "node-uuid", &node_uuid);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_ERROR, "node-uuid not "
-                                "specified");
-                        goto err;
-                }
+        GF_VALIDATE_OR_GOTO(this->name, defrag, err);
 
-                if (uuid_parse (node_uuid, defrag->node_uuid)) {
-                        gf_log (this->name, GF_LOG_ERROR, "Cannot parse "
-                                "glusterd node uuid");
-                        goto err;
-                }
+        LOCK_INIT(&defrag->lock);
 
-                defrag->cmd = cmd;
+        defrag->is_exiting = 0;
 
-                defrag->stats = _gf_false;
+        conf->defrag = defrag;
+        defrag->this = this;
+
+        ret = dict_get_str(this->options, "node-uuid", &node_uuid);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INVALID_CONFIGURATION,
+                   "Invalid volume configuration: "
+                   "node-uuid not specified");
+            goto err;
         }
 
-        conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_ON;
-        if (dict_get_str (this->options, "lookup-unhashed", &temp_str) == 0) {
-                /* If option is not "auto", other options _should_ be boolean */
-                if (strcasecmp (temp_str, "auto"))
-                        gf_string2boolean (temp_str, &conf->search_unhashed);
-                else
-                        conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_AUTO;
+        if (gf_uuid_parse(node_uuid, defrag->node_uuid)) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INVALID_OPTION,
+                   "Invalid option:"
+                   " Cannot parse glusterd node uuid");
+            goto err;
         }
 
-        GF_OPTION_INIT ("unhashed-sticky-bit", conf->unhashed_sticky_bit, bool,
-                        err);
+        defrag->cmd = cmd;
 
-        GF_OPTION_INIT ("use-readdirp", conf->use_readdirp, bool, err);
+        defrag->stats = _gf_false;
 
-	GF_OPTION_INIT ("min-free-disk", conf->min_free_disk, percent_or_size,
-			err);
+        defrag->queue = NULL;
 
-        GF_OPTION_INIT ("min-free-inodes", conf->min_free_inodes, percent,
-                        err);
+        defrag->crawl_done = 0;
 
-        conf->dir_spread_cnt = conf->subvolume_cnt;
-        GF_OPTION_INIT ("directory-layout-spread", conf->dir_spread_cnt,
-                        uint32, err);
+        defrag->global_error = 0;
 
-        GF_OPTION_INIT ("assert-no-child-down", conf->assert_no_child_down,
-                        bool, err);
+        defrag->q_entry_count = 0;
 
-        GF_OPTION_INIT ("readdir-optimize", conf->readdir_optimize, bool, err);
+        defrag->wakeup_crawler = 0;
 
-        if (defrag) {
-                GF_OPTION_INIT ("rebalance-stats", defrag->stats, bool, err);
-                if (dict_get_str (this->options, "rebalance-filter", &temp_str)
-                    == 0) {
-                        if (gf_defrag_pattern_list_fill (this, defrag, temp_str)
-                            == -1) {
-                                gf_log (this->name, GF_LOG_ERROR, "Cannot parse"
-                                        " rebalance-filter (%s)", temp_str);
-                                goto err;
-                        }
-                }
-        }
+        pthread_mutex_init(&defrag->dfq_mutex, 0);
+        pthread_cond_init(&defrag->parallel_migration_cond, 0);
+        pthread_cond_init(&defrag->rebalance_crawler_alarm, 0);
+        pthread_cond_init(&defrag->df_wakeup_thread, 0);
+
+        pthread_mutex_init(&defrag->fc_mutex, 0);
+        pthread_cond_init(&defrag->fc_wakeup_cond, 0);
+
+        defrag->global_error = 0;
+    }
 
-        /* option can be any one of percent or bytes */
-        conf->disk_unit = 0;
-        if (conf->min_free_disk < 100)
-                conf->disk_unit = 'p';
+    conf->use_fallocate = 1;
 
-        ret = dht_init_subvolumes (this, conf);
-        if (ret == -1) {
+    conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_ON;
+    if (dict_get_str(this->options, "lookup-unhashed", &temp_str) == 0) {
+        /* If option is not "auto", other options _should_ be boolean */
+        if (strcasecmp(temp_str, "auto")) {
+            gf_boolean_t search_unhashed_bool;
+            ret = gf_string2boolean(temp_str, &search_unhashed_bool);
+            if (ret == -1) {
                 goto err;
+            }
+            conf->search_unhashed = search_unhashed_bool
+                                        ? GF_DHT_LOOKUP_UNHASHED_ON
+                                        : GF_DHT_LOOKUP_UNHASHED_OFF;
+        } else {
+            conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_AUTO;
         }
+    }
 
-        if (dict_get_str (this->options, "decommissioned-bricks", &temp_str) == 0) {
-                ret = dht_parse_decommissioned_bricks (this, conf, temp_str);
-                if (ret == -1)
-                        goto err;
-        }
+    GF_OPTION_INIT("lookup-optimize", conf->lookup_optimize, bool, err);
 
-        dht_init_regex (this, this->options, "rsync-hash-regex",
-                        &conf->rsync_regex, &conf->rsync_regex_valid);
-        dht_init_regex (this, this->options, "extra-hash-regex",
-                        &conf->extra_regex, &conf->extra_regex_valid);
+    GF_OPTION_INIT("unhashed-sticky-bit", conf->unhashed_sticky_bit, bool, err);
 
-        ret = dht_layouts_init (this, conf);
-        if (ret == -1) {
-                goto err;
-        }
+    GF_OPTION_INIT("use-readdirp", conf->use_readdirp, bool, err);
+
+    GF_OPTION_INIT("min-free-disk", conf->min_free_disk, percent_or_size, err);
+
+    GF_OPTION_INIT("min-free-inodes", conf->min_free_inodes, percent, err);
+
+    conf->dir_spread_cnt = conf->subvolume_cnt;
+    GF_OPTION_INIT("directory-layout-spread", conf->dir_spread_cnt, uint32,
+                   err);
+
+    GF_OPTION_INIT("assert-no-child-down", conf->assert_no_child_down, bool,
+                   err);
+
+    GF_OPTION_INIT("readdir-optimize", conf->readdir_optimize, bool, err);
 
-        LOCK_INIT (&conf->subvolume_lock);
-        LOCK_INIT (&conf->layout_lock);
+    GF_OPTION_INIT("lock-migration", conf->lock_migration_enabled, bool, err);
 
-        conf->gen = 1;
+    GF_OPTION_INIT("force-migration", conf->force_migration, bool, err);
+
+    if (defrag) {
+        defrag->lock_migration_enabled = conf->lock_migration_enabled;
+
+        GF_OPTION_INIT("rebalance-stats", defrag->stats, bool, err);
+        if (dict_get_str(this->options, "rebalance-filter", &temp_str) == 0) {
+            if (gf_defrag_pattern_list_fill(this, defrag, temp_str) == -1) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INVALID_OPTION,
+                       "Invalid option:"
+                       " Cannot parse rebalance-filter (%s)",
+                       temp_str);
 
-        this->local_pool = mem_pool_new (dht_local_t, 512);
-        if (!this->local_pool) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "failed to create local_t's memory pool");
                 goto err;
+            }
         }
-
-        GF_OPTION_INIT ("xattr-name", conf->xattr_name, str, err);
-        gf_asprintf (&conf->link_xattr_name, "%s.linkto", conf->xattr_name);
-        gf_asprintf (&conf->wild_xattr_name, "%s*", conf->xattr_name);
-        if (!conf->link_xattr_name || !conf->wild_xattr_name) {
+    }
+
+    /* option can be any one of percent or bytes */
+    conf->disk_unit = 0;
+    if (conf->min_free_disk < 100)
+        conf->disk_unit = 'p';
+
+    ret = dht_init_subvolumes(this, conf);
+    if (ret == -1) {
+        goto err;
+    }
+
+    if (cmd) {
+        ret = dht_init_local_subvolumes(this, conf);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   DHT_MSG_INIT_LOCAL_SUBVOL_FAILED,
+                   "dht_init_local_subvolumes failed");
+            goto err;
+        }
+    }
+
+    if (dict_get_str(this->options, "decommissioned-bricks", &temp_str) == 0) {
+        ret = dht_parse_decommissioned_bricks(this, conf, temp_str);
+        if (ret == -1)
+            goto err;
+    }
+
+    dht_init_regex(this, this->options, "rsync-hash-regex", &conf->rsync_regex,
+                   &conf->rsync_regex_valid, conf);
+    dht_init_regex(this, this->options, "extra-hash-regex", &conf->extra_regex,
+                   &conf->extra_regex_valid, conf);
+
+    ret = dht_layouts_init(this, conf);
+    if (ret == -1) {
+        goto err;
+    }
+
+    conf->gen = 1;
+
+    this->local_pool = mem_pool_new(dht_local_t, 512);
+    if (!this->local_pool) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+               " DHT initialisation failed. "
+               "failed to create local_t's memory pool");
+        goto err;
+    }
+
+    GF_OPTION_INIT("randomize-hash-range-by-gfid", conf->randomize_by_gfid,
+                   bool, err);
+
+    if (defrag) {
+        GF_OPTION_INIT("rebal-throttle", temp_str, str, err);
+        if (temp_str) {
+            ret = dht_configure_throttle(this, conf, temp_str);
+            if (ret == -1)
                 goto err;
         }
+    }
 
-        this->private = conf;
+    GF_OPTION_INIT("xattr-name", conf->xattr_name, str, err);
+    gf_asprintf(&conf->mds_xattr_key, "%s." DHT_MDS_STR, conf->xattr_name);
+    gf_asprintf(&conf->link_xattr_name, "%s." DHT_LINKFILE_STR,
+                conf->xattr_name);
+    gf_asprintf(&conf->commithash_xattr_name, "%s." DHT_COMMITHASH_STR,
+                conf->xattr_name);
+    gf_asprintf(&conf->wild_xattr_name, "%s*", conf->xattr_name);
+    if (!conf->link_xattr_name || !conf->wild_xattr_name) {
+        goto err;
+    }
 
-        return 0;
+    GF_OPTION_INIT("weighted-rebalance", conf->do_weighting, bool, err);
+
+    conf->lock_pool = mem_pool_new(dht_lock_t, 512);
+    if (!conf->lock_pool) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INIT_FAILED,
+               "failed to create lock mem_pool, failing "
+               "initialization");
+        goto err;
+    }
+
+    this->private = conf;
+
+    if (dht_set_subvol_range(this))
+        goto err;
+
+    if (dht_init_methods(this))
+        goto err;
+
+    return 0;
 
 err:
-        if (conf) {
-                if (conf->file_layouts) {
-                        for (i = 0; i < conf->subvolume_cnt; i++) {
-                                GF_FREE (conf->file_layouts[i]);
-                        }
-                        GF_FREE (conf->file_layouts);
-                }
+    if (conf) {
+        if (conf->file_layouts) {
+            for (i = 0; i < conf->subvolume_cnt; i++) {
+                GF_FREE(conf->file_layouts[i]);
+            }
+            GF_FREE(conf->file_layouts);
+        }
 
-                GF_FREE (conf->subvolumes);
+        GF_FREE(conf->subvolumes);
 
-                GF_FREE (conf->subvolume_status);
+        GF_FREE(conf->subvolume_status);
 
-                GF_FREE (conf->du_stats);
+        GF_FREE(conf->du_stats);
 
-                GF_FREE (conf->defrag);
+        GF_FREE(conf->defrag);
 
-                GF_FREE (conf->xattr_name);
-                GF_FREE (conf->link_xattr_name);
-                GF_FREE (conf->wild_xattr_name);
+        GF_FREE(conf->xattr_name);
+        GF_FREE(conf->link_xattr_name);
+        GF_FREE(conf->wild_xattr_name);
+        GF_FREE(conf->mds_xattr_key);
 
-                GF_FREE (conf);
-        }
+        if (conf->lock_pool)
+            mem_pool_destroy(conf->lock_pool);
 
-        return -1;
-}
+        GF_FREE(conf);
+    }
 
+    return -1;
+}
 
-struct volume_options options[] = {
-        { .key  = {"lookup-unhashed"},
-          .value = {"auto", "yes", "no", "enable", "disable", "1", "0",
-                    "on", "off"},
-          .type = GF_OPTION_TYPE_STR,
-          .default_value = "on",
-          .description = "This option if set to ON, does a lookup through "
-          "all the sub-volumes, in case a lookup didn't return any result "
-          "from the hash subvolume. If set to OFF, it does not do a lookup "
-          "on the remaining subvolumes."
-        },
-        { .key  = {"min-free-disk"},
-          .type = GF_OPTION_TYPE_PERCENT_OR_SIZET,
-          .default_value = "10%",
-          .description = "Percentage/Size of disk space, after which the "
-          "process starts balancing out the cluster, and logs will appear "
-          "in log files",
-        },
-	{ .key  = {"min-free-inodes"},
-          .type = GF_OPTION_TYPE_PERCENT,
-          .default_value = "5%",
-          .description = "after system has only N% of inodes, warnings "
-          "starts to appear in log files",
-        },
-        { .key = {"unhashed-sticky-bit"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "off",
-        },
-        { .key = {"use-readdirp"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "on",
-          .description = "This option if set to ON, forces the use of "
-          "readdirp, and hence also displays the stats of the files."
-        },
-        { .key = {"assert-no-child-down"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "off",
-          .description = "This option if set to ON, in the event of "
-          "CHILD_DOWN, will call exit."
-        },
-        { .key  = {"directory-layout-spread"},
-          .type = GF_OPTION_TYPE_INT,
-          .min  = 1,
-          .validate = GF_OPT_VALIDATE_MIN,
-          .description = "Specifies the directory layout spread."
-        },
-        { .key  = {"decommissioned-bricks"},
-          .type = GF_OPTION_TYPE_ANY,
-          .description = "This option if set to ON, decommissions "
-          "the brick, so that no new data is allowed to be created "
-          "on that brick."
-        },
-        { .key  = {"rebalance-cmd"},
-          .type = GF_OPTION_TYPE_INT,
-        },
-        { .key = {"node-uuid"},
-          .type = GF_OPTION_TYPE_STR,
-        },
-        { .key = {"rebalance-stats"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "off",
-          .description = "This option if set to ON displays and logs the "
-          " time taken for migration of each file, during the rebalance "
-          "process. If set to OFF, the rebalance logs will only display the "
-          "time spent in each directory."
-        },
-        { .key = {"readdir-optimize"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "off",
-          .description = "This option if set to ON enables the optimization "
-          "that allows DHT to requests non-first subvolumes to filter out "
-          "directory entries."
-        },
-        { .key = {"rsync-hash-regex"},
-          .type = GF_OPTION_TYPE_STR,
-          /* Setting a default here doesn't work.  See dht_init_regex. */
-          .description = "Regular expression for stripping temporary-file "
-          "suffix and prefix used by rsync, to prevent relocation when the "
-          "file is renamed."
-        },
-        { .key = {"extra-hash-regex"},
-          .type = GF_OPTION_TYPE_STR,
-          /* Setting a default here doesn't work.  See dht_init_regex. */
-          .description = "Regular expression for stripping temporary-file "
-          "suffix and prefix used by an application, to prevent relocation when "
-          "the file is renamed."
-        },
-        { .key = {"rebalance-filter"},
-          .type = GF_OPTION_TYPE_STR,
-        },
-
-        { .key = {"xattr-name"},
-          .type = GF_OPTION_TYPE_STR,
-          .default_value = "trusted.glusterfs.dht",
-          .description = "Base for extended attributes used by this "
-          "translator instance, to avoid conflicts with others above or "
-          "below it."
-        },
-
-        /* NUFA option */
-        { .key  = {"local-volume-name"},
-          .type = GF_OPTION_TYPE_XLATOR
-        },
-
-        /* switch option */
-        { .key  = {"pattern.switch.case"},
-          .type = GF_OPTION_TYPE_ANY
-        },
-
-        { .key  = {NULL} },
+struct volume_options dht_options[] = {
+    {
+        .key = {"lookup-unhashed"},
+        .value = {"auto", "yes", "no", "enable", "disable", "1", "0", "on",
+                  "off"},
+        .type = GF_OPTION_TYPE_STR,
+        .default_value = "on",
+        .description =
+            "This option if set to ON, does a lookup through "
+            "all the sub-volumes, in case a lookup didn't return any result "
+            "from the hash subvolume. If set to OFF, it does not do a lookup "
+            "on the remaining subvolumes.",
+        .op_version = {1},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE,
+        .level = OPT_STATUS_BASIC,
+    },
+    {.key = {"lookup-optimize"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .description =
+         "This option if set to ON enables the optimization "
+         "of -ve lookups, by not doing a lookup on non-hashed subvolumes for "
+         "files, in case the hashed subvolume does not return any result. "
+         "This option disregards the lookup-unhashed setting, when enabled.",
+     .op_version = {GD_OP_VERSION_3_7_2},
+     .level = OPT_STATUS_ADVANCED,
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC},
+    {.key = {"min-free-disk"},
+     .type = GF_OPTION_TYPE_PERCENT_OR_SIZET,
+     .default_value = "10%",
+     .description =
+         "Percentage/Size of disk space, after which the "
+         "process starts balancing out the cluster, and logs will appear "
+         "in log files",
+     .op_version = {1},
+     .level = OPT_STATUS_BASIC,
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC},
+    {.key = {"min-free-inodes"},
+     .type = GF_OPTION_TYPE_PERCENT,
+     .default_value = "5%",
+     .description = "after system has only N% of inodes, warnings "
+                    "starts to appear in log files",
+     .op_version = {1},
+     .level = OPT_STATUS_BASIC,
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC},
+    {
+        .key = {"unhashed-sticky-bit"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "off",
+    },
+    {.key = {"use-readdirp"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .description = "This option if set to ON, forces the use of "
+                    "readdirp, and hence also displays the stats of the files.",
+     .level = OPT_STATUS_ADVANCED,
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC},
+    {.key = {"assert-no-child-down"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description = "This option if set to ON, in the event of "
+                    "CHILD_DOWN, will call exit."},
+    {
+        .key = {"directory-layout-spread"},
+        .type = GF_OPTION_TYPE_INT,
+        .min = 1,
+        .validate = GF_OPT_VALIDATE_MIN,
+        .description = "Specifies the directory layout spread. Takes number "
+                       "of subvolumes as default value.",
+
+        .op_version = {2},
+    },
+    {
+        .key = {"decommissioned-bricks"},
+        .type = GF_OPTION_TYPE_ANY,
+        .description =
+            "This option if set to ON, decommissions "
+            "the brick, so that no new data is allowed to be created "
+            "on that brick.",
+        .level = OPT_STATUS_ADVANCED,
+    },
+    {
+        .key = {"rebalance-cmd"},
+        .type = GF_OPTION_TYPE_INT,
+    },
+    {
+        .key = {"commit-hash"},
+        .type = GF_OPTION_TYPE_INT,
+    },
+    {
+        .key = {"node-uuid"},
+        .type = GF_OPTION_TYPE_STR,
+    },
+    {
+        .key = {"rebalance-stats"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "off",
+        .description =
+            "This option if set to ON displays and logs the "
+            " time taken for migration of each file, during the rebalance "
+            "process. If set to OFF, the rebalance logs will only display the "
+            "time spent in each directory.",
+        .op_version = {2},
+        .level = OPT_STATUS_BASIC,
+    },
+    {.key = {"readdir-optimize"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description =
+         "This option if set to ON enables the optimization "
+         "that allows DHT to requests non-first subvolumes to filter out "
+         "directory entries.",
+     .op_version = {1},
+     .level = OPT_STATUS_ADVANCED,
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC},
+    {.key = {"rsync-hash-regex"},
+     .type = GF_OPTION_TYPE_STR,
+     /* Setting a default here doesn't work.  See dht_init_regex. */
+     .description =
+         "Regular expression for stripping temporary-file "
+         "suffix and prefix used by rsync, to prevent relocation when the "
+         "file is renamed.",
+     .op_version = {3},
+     .level = OPT_STATUS_BASIC,
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC},
+    {.key = {"extra-hash-regex"},
+     .type = GF_OPTION_TYPE_STR,
+     /* Setting a default here doesn't work.  See dht_init_regex. */
+     .description =
+         "Regular expression for stripping temporary-file "
+         "suffix and prefix used by an application, to prevent relocation when "
+         "the file is renamed.",
+     .op_version = {3},
+     .level = OPT_STATUS_BASIC,
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC},
+    {
+        .key = {"rebalance-filter"},
+        .type = GF_OPTION_TYPE_STR,
+    },
+
+    {
+        .key = {"xattr-name"},
+        .type = GF_OPTION_TYPE_STR,
+        .default_value = "trusted.glusterfs.dht",
+        .description =
+            "Base for extended attributes used by this "
+            "translator instance, to avoid conflicts with others above or "
+            "below it.",
+        .op_version = {3},
+    },
+
+    {.key = {"weighted-rebalance"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .description =
+         "When enabled, files will be allocated to bricks "
+         "with a probability proportional to their size.  Otherwise, all "
+         "bricks will have the same probability (legacy behavior).",
+     .op_version = {GD_OP_VERSION_3_6_0},
+     .level = OPT_STATUS_BASIC,
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC},
+
+    /* NUFA option */
+    {.key = {"local-volume-name"}, .type = GF_OPTION_TYPE_XLATOR},
+
+    /* switch option */
+    {.key = {"pattern.switch.case"}, .type = GF_OPTION_TYPE_ANY},
+
+    {
+        .key = {"randomize-hash-range-by-gfid"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "off",
+        .description =
+            "Use gfid of directory to determine the subvolume "
+            "from which hash ranges are allocated starting with 0. "
+            "Note that we still use a directory/file's name to determine the "
+            "subvolume to which it hashes",
+        .op_version = {GD_OP_VERSION_3_6_0},
+    },
+
+    {.key = {"rebal-throttle"},
+     .type = GF_OPTION_TYPE_STR,
+     .default_value = "normal",
+     .description = " Sets the maximum number of parallel file migrations "
+                    "allowed on a node during the rebalance operation. The"
+                    " default value is normal and allows a max of "
+                    "[($(processing units) - 4) / 2), 2]  files to be "
+                    "migrated at a time. Lazy will allow only one file to "
+                    "be migrated at a time and aggressive will allow "
+                    "max of [($(processing units) - 4) / 2), 4]",
+     .op_version = {GD_OP_VERSION_3_7_0},
+     .level = OPT_STATUS_BASIC,
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC
+
+    },
+
+    {.key = {"lock-migration"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description = " If enabled this feature will migrate the posix locks"
+                    " associated with a file during rebalance",
+     .op_version = {GD_OP_VERSION_3_8_0},
+     .level = OPT_STATUS_ADVANCED,
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC},
+
+    {.key = {"force-migration"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description = "If disabled, rebalance will not migrate files that "
+                    "are being written to by an application",
+     .op_version = {GD_OP_VERSION_4_0_0},
+     .level = OPT_STATUS_ADVANCED,
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC},
+
+    {.key = {NULL}},
 };
+
+#define NUM_DHT_OPTIONS (sizeof(dht_options) / sizeof(dht_options[0]))
+
+extern struct volume_options options[NUM_DHT_OPTIONS]
+    __attribute__((alias("dht_options")));
diff --git a/xlators/cluster/dht/src/dht.c b/xlators/cluster/dht/src/dht.c
index 0349b63a91f..53de8292704 100644
--- a/xlators/cluster/dht/src/dht.c
+++ b/xlators/cluster/dht/src/dht.c
@@ -8,81 +8,116 @@
   cases as published by the Free Software Foundation.
 */
 
+#include "dht-common.h"
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+struct xlator_fops dht_pt_fops = {
+    /* we need to keep mkdir to make sure we
+       have layout on new directory */
+    .mkdir = dht_pt_mkdir,
+    .getxattr = dht_pt_getxattr,
+    .fgetxattr = dht_pt_fgetxattr,
 
-#include "statedump.h"
-#include "dht-common.h"
+    /* required to trace fop properly in changelog */
+    .rename = dht_pt_rename,
 
-class_methods_t class_methods = {
-        .init           = dht_init,
-        .fini           = dht_fini,
-        .reconfigure    = dht_reconfigure,
-        .notify         = dht_notify
+    /* FIXME: commenting the '.lookup()' below made some of
+       the failing tests to pass. I would remove the below
+       line, but keeping it here as a reminder for people
+       to check for issues if they find concerns with DHT
+       pass-through logic  */
+    /*
+      .lookup = dht_lookup,
+      .readdir = dht_readdir,
+      .readdirp = dht_readdirp,
+    */
+    /* Keeping above as commented, mainly to support the
+       usecase of a gluster volume getting to 1x(anytype),
+       due to remove-brick (shrinking) exercise. In that case,
+       we would need above fops to be available, so we can
+       handle the case of dangling linkto files (if any) */
 };
 
 struct xlator_fops fops = {
-        .lookup      = dht_lookup,
-        .mknod       = dht_mknod,
-        .create      = dht_create,
+    .ipc = dht_ipc,
+    .lookup = dht_lookup,
+    .mknod = dht_mknod,
+    .create = dht_create,
 
-        .open        = dht_open,
-        .statfs      = dht_statfs,
-        .opendir     = dht_opendir,
-        .readdir     = dht_readdir,
-        .readdirp    = dht_readdirp,
-        .fsyncdir    = dht_fsyncdir,
-        .symlink     = dht_symlink,
-        .unlink      = dht_unlink,
-        .link        = dht_link,
-        .mkdir       = dht_mkdir,
-        .rmdir       = dht_rmdir,
-        .rename      = dht_rename,
-        .entrylk     = dht_entrylk,
-        .fentrylk    = dht_fentrylk,
+    .open = dht_open,
+    .statfs = dht_statfs,
+    .opendir = dht_opendir,
+    .readdir = dht_readdir,
+    .readdirp = dht_readdirp,
+    .fsyncdir = dht_fsyncdir,
+    .symlink = dht_symlink,
+    .unlink = dht_unlink,
+    .link = dht_link,
+    .mkdir = dht_mkdir,
+    .rmdir = dht_rmdir,
+    .rename = dht_rename,
+    .entrylk = dht_entrylk,
+    .fentrylk = dht_fentrylk,
 
-        /* Inode read operations */
-        .stat        = dht_stat,
-        .fstat       = dht_fstat,
-        .access      = dht_access,
-        .readlink    = dht_readlink,
-        .getxattr    = dht_getxattr,
-        .fgetxattr    = dht_fgetxattr,
-        .readv       = dht_readv,
-        .flush       = dht_flush,
-        .fsync       = dht_fsync,
-        .inodelk     = dht_inodelk,
-        .finodelk    = dht_finodelk,
-        .lk          = dht_lk,
+    /* Inode read operations */
+    .stat = dht_stat,
+    .fstat = dht_fstat,
+    .access = dht_access,
+    .readlink = dht_readlink,
+    .getxattr = dht_getxattr,
+    .fgetxattr = dht_fgetxattr,
+    .readv = dht_readv,
+    .flush = dht_flush,
+    .fsync = dht_fsync,
+    .inodelk = dht_inodelk,
+    .finodelk = dht_finodelk,
+    .lk = dht_lk,
+    .lease = dht_lease,
 
-        /* Inode write operations */
-        .fremovexattr = dht_fremovexattr,
-        .removexattr = dht_removexattr,
-        .setxattr    = dht_setxattr,
-        .fsetxattr   = dht_fsetxattr,
-        .truncate    = dht_truncate,
-        .ftruncate   = dht_ftruncate,
-        .writev      = dht_writev,
-        .xattrop     = dht_xattrop,
-        .fxattrop    = dht_fxattrop,
-        .setattr     = dht_setattr,
-        .fsetattr    = dht_fsetattr,
-	.fallocate   = dht_fallocate,
-	.discard     = dht_discard,
+    /* Inode write operations */
+    .fremovexattr = dht_fremovexattr,
+    .removexattr = dht_removexattr,
+    .setxattr = dht_setxattr,
+    .fsetxattr = dht_fsetxattr,
+    .truncate = dht_truncate,
+    .ftruncate = dht_ftruncate,
+    .writev = dht_writev,
+    .xattrop = dht_xattrop,
+    .fxattrop = dht_fxattrop,
+    .setattr = dht_setattr,
+    .fsetattr = dht_fsetattr,
+    .fallocate = dht_fallocate,
+    .discard = dht_discard,
+    .zerofill = dht_zerofill,
 };
 
 struct xlator_dumpops dumpops = {
-        .priv = dht_priv_dump,
-        .inodectx = dht_inodectx_dump,
+    .priv = dht_priv_dump,
+    .inodectx = dht_inodectx_dump,
 };
 
-
 struct xlator_cbks cbks = {
-//      .release    = dht_release,
-//      .releasedir = dht_releasedir,
-        .forget     = dht_forget
+    .release = dht_release,
+    //      .releasedir = dht_releasedir,
+    .forget = dht_forget,
+};
+
+extern int32_t
+mem_acct_init(xlator_t *this);
+
+extern struct volume_options dht_options[];
+
+xlator_api_t xlator_api = {
+    .init = dht_init,
+    .fini = dht_fini,
+    .notify = dht_notify,
+    .reconfigure = dht_reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .dumpops = &dumpops,
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = dht_options,
+    .identifier = "distribute",
+    .pass_through_fops = &dht_pt_fops,
+    .category = GF_MAINTAINED,
 };
-;
diff --git a/xlators/cluster/dht/src/nufa.c b/xlators/cluster/dht/src/nufa.c
index e934acdf00a..3648a564840 100644
--- a/xlators/cluster/dht/src/nufa.c
+++ b/xlators/cluster/dht/src/nufa.c
@@ -8,669 +8,650 @@
   cases as published by the Free Software Foundation.
 */
 
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
 #include "dht-common.h"
 
 /* TODO: all 'TODO's in dht.c holds good */
 
-extern struct volume_options options[];
+extern struct volume_options dht_options[];
 
 int
-nufa_local_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int op_ret, int op_errno,
-                       inode_t *inode, struct iatt *stbuf, dict_t *xattr,
-                       struct iatt *postparent)
+nufa_local_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, inode_t *inode,
+                      struct iatt *stbuf, dict_t *xattr,
+                      struct iatt *postparent)
 {
-        xlator_t     *subvol      = NULL;
-        char          is_linkfile = 0;
-        char          is_dir      = 0;
-        dht_conf_t   *conf        = NULL;
-        dht_local_t  *local       = NULL;
-        loc_t        *loc         = NULL;
-        int           i           = 0;
-        call_frame_t *prev        = NULL;
-        int           call_cnt    = 0;
-        int           ret         = 0;
-
-        conf  = this->private;
-
-        prev  = cookie;
-        local = frame->local;
-        loc   = &local->loc;
-
-        if (ENTRY_MISSING (op_ret, op_errno)) {
-                if (conf->search_unhashed) {
-                        local->op_errno = ENOENT;
-                        dht_lookup_everywhere (frame, this, loc);
-                        return 0;
-                }
+    xlator_t *subvol = NULL;
+    char is_linkfile = 0;
+    char is_dir = 0;
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    loc_t *loc = NULL;
+    int i = 0;
+    xlator_t *prev = NULL;
+    int call_cnt = 0;
+    int ret = 0;
+
+    conf = this->private;
+
+    prev = cookie;
+    local = frame->local;
+    loc = &local->loc;
+
+    if (ENTRY_MISSING(op_ret, op_errno)) {
+        if (conf->search_unhashed) {
+            local->op_errno = ENOENT;
+            dht_lookup_everywhere(frame, this, loc);
+            return 0;
         }
-
-        if (op_ret == -1)
-                goto out;
-
-        is_linkfile = check_is_linkfile (inode, stbuf, xattr,
-                                         conf->link_xattr_name);
-        is_dir      = check_is_dir (inode, stbuf, xattr);
-
-        if (!is_dir && !is_linkfile) {
-                /* non-directory and not a linkfile */
-                ret = dht_layout_preset (this, prev->this, inode);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "could not set pre-set layout for subvol %s",
-                                prev->this->name);
-                        op_ret   = -1;
-                        op_errno = EINVAL;
-                        goto err;
-                }
-
-                goto out;
+    }
+
+    if (op_ret == -1)
+        goto out;
+
+    is_linkfile = check_is_linkfile(inode, stbuf, xattr, conf->link_xattr_name);
+    is_dir = check_is_dir(inode, stbuf, xattr);
+
+    if (!is_dir && !is_linkfile) {
+        /* non-directory and not a linkfile */
+        ret = dht_layout_preset(this, prev, inode);
+        if (ret < 0) {
+            gf_msg_debug(this->name, 0,
+                         "could not set pre-set layout for subvol"
+                         " %s",
+                         prev->name);
+            op_ret = -1;
+            op_errno = EINVAL;
+            goto err;
         }
 
-        if (is_dir) {
-                call_cnt        = conf->subvolume_cnt;
-                local->call_cnt = call_cnt;
-
-                local->inode = inode_ref (inode);
-                local->xattr = dict_ref (xattr);
-
-                local->op_ret = 0;
-                local->op_errno = 0;
-
-                local->layout = dht_layout_new (this, conf->subvolume_cnt);
-                if (!local->layout) {
-                        op_ret   = -1;
-                        op_errno = ENOMEM;
-                        goto err;
-                }
-
-                for (i = 0; i < call_cnt; i++) {
-                        STACK_WIND (frame, dht_lookup_dir_cbk,
-                                    conf->subvolumes[i],
-                                    conf->subvolumes[i]->fops->lookup,
-                                    &local->loc, local->xattr_req);
-                }
-        }
+        goto out;
+    }
+
+    if (is_dir) {
+        call_cnt = conf->subvolume_cnt;
+        local->call_cnt = call_cnt;
 
-        if (is_linkfile) {
-                subvol = dht_linkfile_subvol (this, inode, stbuf, xattr);
+        local->inode = inode_ref(inode);
+        local->xattr = dict_ref(xattr);
 
-                if (!subvol) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "linkfile not having link subvolume. path=%s",
-                                loc->path);
-                        dht_lookup_everywhere (frame, this, loc);
-                        return 0;
-                }
+        local->op_ret = 0;
+        local->op_errno = 0;
 
-                STACK_WIND (frame, dht_lookup_linkfile_cbk,
-                            subvol, subvol->fops->lookup,
-                            &local->loc, local->xattr_req);
+        local->layout = dht_layout_new(this, conf->subvolume_cnt);
+        if (!local->layout) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            goto err;
         }
 
-        return 0;
+        for (i = 0; i < call_cnt; i++) {
+            STACK_WIND_COOKIE(frame, dht_lookup_dir_cbk, conf->subvolumes[i],
+                              conf->subvolumes[i],
+                              conf->subvolumes[i]->fops->lookup, &local->loc,
+                              local->xattr_req);
+        }
+    }
 
-out:
-        if (!local->hashed_subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no subvolume in layout for path=%s",
-                        local->loc.path);
-                local->op_errno = ENOENT;
-                dht_lookup_everywhere (frame, this, loc);
-                return 0;
+    if (is_linkfile) {
+        subvol = dht_linkfile_subvol(this, inode, stbuf, xattr);
+
+        if (!subvol) {
+            gf_msg_debug(this->name, 0,
+                         "linkfile has no link subvolume. path=%s", loc->path);
+            dht_lookup_everywhere(frame, this, loc);
+            return 0;
         }
 
-        STACK_WIND (frame, dht_lookup_cbk,
-                    local->hashed_subvol, local->hashed_subvol->fops->lookup,
-                    &local->loc, local->xattr_req);
+        STACK_WIND_COOKIE(frame, dht_lookup_linkfile_cbk, subvol, subvol,
+                          subvol->fops->lookup, &local->loc, local->xattr_req);
+    }
+
+    return 0;
 
+out:
+    if (!local->hashed_subvol) {
+        gf_msg_debug(this->name, 0, "no subvolume in layout for path=%s",
+                     local->loc.path);
+        local->op_errno = ENOENT;
+        dht_lookup_everywhere(frame, this, loc);
         return 0;
+    }
+
+    STACK_WIND_COOKIE(frame, dht_lookup_cbk, local->hashed_subvol,
+                      local->hashed_subvol, local->hashed_subvol->fops->lookup,
+                      &local->loc, local->xattr_req);
+
+    return 0;
 
 err:
-        DHT_STACK_UNWIND (lookup, frame, op_ret, op_errno,
-                          inode, stbuf, xattr, postparent);
-        return 0;
+    DHT_STACK_UNWIND(lookup, frame, op_ret, op_errno, inode, stbuf, xattr,
+                     postparent);
+    return 0;
 }
 
 int
-nufa_lookup (call_frame_t *frame, xlator_t *this,
-             loc_t *loc, dict_t *xattr_req)
+nufa_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req)
 {
-        xlator_t     *hashed_subvol = NULL;
-        xlator_t     *subvol = NULL;
-        dht_local_t  *local  = NULL;
-        dht_conf_t   *conf = NULL;
-        int           ret    = -1;
-        int           op_errno = -1;
-        dht_layout_t *layout = NULL;
-        int           i = 0;
-        int           call_cnt = 0;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-        conf = this->private;
-
-        local = dht_local_init (frame, loc, NULL, GF_FOP_LOOKUP);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
+    xlator_t *hashed_subvol = NULL;
+    xlator_t *subvol = NULL;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
+    int op_errno = -1;
+    dht_layout_t *layout = NULL;
+    int i = 0;
+    int call_cnt = 0;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+    VALIDATE_OR_GOTO(loc->path, err);
+
+    conf = this->private;
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_LOOKUP);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    if (xattr_req) {
+        local->xattr_req = dict_ref(xattr_req);
+    } else {
+        local->xattr_req = dict_new();
+    }
+
+    hashed_subvol = dht_subvol_get_hashed(this, &local->loc);
+
+    local->hashed_subvol = hashed_subvol;
+
+    if (is_revalidate(loc)) {
+        layout = local->layout;
+        if (!layout) {
+            gf_msg_debug(this->name, 0,
+                         "revalidate lookup without cache. "
+                         "path=%s",
+                         loc->path);
+            op_errno = EINVAL;
+            goto err;
         }
 
-        if (xattr_req) {
-                local->xattr_req = dict_ref (xattr_req);
-        } else {
-                local->xattr_req = dict_new ();
+        if (layout->gen && (layout->gen < conf->gen)) {
+            gf_msg_debug(this->name, 0, "incomplete layout failure for path=%s",
+                         loc->path);
+            dht_layout_unref(this, local->layout);
+            goto do_fresh_lookup;
         }
 
-        hashed_subvol = dht_subvol_get_hashed (this, &local->loc);
-
-        local->hashed_subvol = hashed_subvol;
-
-        if (is_revalidate (loc)) {
-                layout = local->layout;
-                if (!layout) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "revalidate without cache. path=%s",
-                                loc->path);
-                        op_errno = EINVAL;
-                        goto err;
-                }
-
-                if (layout->gen && (layout->gen < conf->gen)) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "incomplete layout failure for path=%s",
-                                loc->path);
-                        dht_layout_unref (this, local->layout);
-                        goto do_fresh_lookup;
-                }
-
-                local->inode = inode_ref (loc->inode);
-
-                local->call_cnt = layout->cnt;
-                call_cnt = local->call_cnt;
-
-                /* NOTE: we don't require 'trusted.glusterfs.dht.linkto' attribute,
-                 *       revalidates directly go to the cached-subvolume.
-                 */
-                ret = dict_set_uint32 (local->xattr_req,
-                                       conf->xattr_name, 4 * 4);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Failed to set dict value.");
-                        op_errno = -1;
-                        goto err;
-                }
-
-                for (i = 0; i < layout->cnt; i++) {
-                        subvol = layout->list[i].xlator;
-
-                        STACK_WIND (frame, dht_revalidate_cbk,
-                                    subvol, subvol->fops->lookup,
-                                    loc, local->xattr_req);
-
-                        if (!--call_cnt)
-                                break;
-                }
-        } else {
-        do_fresh_lookup:
-                ret = dict_set_uint32 (local->xattr_req,
-                                       conf->xattr_name, 4 * 4);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Failed to set dict value.");
-                        op_errno = -1;
-                        goto err;
-                }
-
-                ret = dict_set_uint32 (local->xattr_req,
-                                       conf->link_xattr_name, 256);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Failed to set dict value.");
-                        op_errno = -1;
-                        goto err;
-                }
-
-                /* Send it to only local volume */
-                STACK_WIND (frame, nufa_local_lookup_cbk,
-                            (xlator_t *)conf->private,
-                            ((xlator_t *)conf->private)->fops->lookup,
-                            loc, local->xattr_req);
-        }
+        local->inode = inode_ref(loc->inode);
 
-        return 0;
+        local->call_cnt = layout->cnt;
+        call_cnt = local->call_cnt;
 
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (lookup, frame, -1, op_errno, NULL, NULL, NULL,
-                          NULL);
-        return 0;
-}
+        /* NOTE: we don't require 'trusted.glusterfs.dht.linkto' attribute,
+         *       revalidates directly go to the cached-subvolume.
+         */
+        ret = dict_set_uint32(local->xattr_req, conf->xattr_name, 4 * 4);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                   "Failed to set dict value.");
+            op_errno = -1;
+            goto err;
+        }
 
-int
-nufa_create_linkfile_create_cbk (call_frame_t *frame, void *cookie,
-                                 xlator_t *this, int op_ret, int op_errno,
-                                 inode_t *inode, struct iatt *stbuf,
-                                 struct iatt *preparent,
-                                 struct iatt *postparent, dict_t *xdata)
-{
-        dht_local_t  *local = NULL;
+        for (i = 0; i < layout->cnt; i++) {
+            subvol = layout->list[i].xlator;
+
+            STACK_WIND_COOKIE(frame, dht_revalidate_cbk, subvol, subvol,
+                              subvol->fops->lookup, loc, local->xattr_req);
 
-        local = frame->local;
+            if (!--call_cnt)
+                break;
+        }
+    } else {
+    do_fresh_lookup:
+        ret = dict_set_uint32(local->xattr_req, conf->xattr_name, 4 * 4);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                   "Failed to set dict value.");
+            op_errno = -1;
+            goto err;
+        }
 
-        if (op_ret == -1)
-                goto err;
+        ret = dict_set_uint32(local->xattr_req, conf->link_xattr_name, 256);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                   "Failed to set dict value.");
+            op_errno = -1;
+            goto err;
+        }
 
-        STACK_WIND (frame, dht_create_cbk,
-                    local->cached_subvol, local->cached_subvol->fops->create,
-                    &local->loc, local->flags, local->mode, local->umask,
-                    local->fd, local->params);
+        /* Send it to only local volume */
+        STACK_WIND_COOKIE(
+            frame, nufa_local_lookup_cbk, ((xlator_t *)conf->private),
+            ((xlator_t *)conf->private),
+            ((xlator_t *)conf->private)->fops->lookup, loc, local->xattr_req);
+    }
 
-        return 0;
+    return 0;
 
 err:
-        DHT_STACK_UNWIND (create, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL, NULL, NULL);
-        return 0;
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+    return 0;
 }
 
 int
-nufa_create (call_frame_t *frame, xlator_t *this,
-             loc_t *loc, int32_t flags, mode_t mode,
-             mode_t umask, fd_t *fd, dict_t *params)
+nufa_create_linkfile_create_cbk(call_frame_t *frame, void *cookie,
+                                xlator_t *this, int op_ret, int op_errno,
+                                inode_t *inode, struct iatt *stbuf,
+                                struct iatt *preparent, struct iatt *postparent,
+                                dict_t *xdata)
 {
-        dht_local_t *local = NULL;
-        dht_conf_t  *conf  = NULL;
-        xlator_t    *subvol = NULL;
-        xlator_t    *avail_subvol = NULL;
-        int          op_errno = -1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
+    dht_local_t *local = NULL;
 
-        conf  = this->private;
+    local = frame->local;
 
-        dht_get_du_info (frame, this, loc);
+    if (op_ret == -1)
+        goto err;
 
-        local = dht_local_init (frame, loc, fd, GF_FOP_CREATE);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
+    STACK_WIND_COOKIE(frame, dht_create_cbk, local->cached_subvol,
+                      local->cached_subvol, local->cached_subvol->fops->create,
+                      &local->loc, local->flags, local->mode, local->umask,
+                      local->fd, local->params);
 
-        subvol = dht_subvol_get_hashed (this, loc);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no subvolume in layout for path=%s",
-                        loc->path);
-                op_errno = ENOENT;
-                goto err;
-        }
+    return 0;
 
-        avail_subvol = conf->private;
-        if (dht_is_subvol_filled (this, (xlator_t *)conf->private)) {
-                avail_subvol =
-                        dht_free_disk_available_subvol (this,
-                                                        (xlator_t *)conf->private,
-                                                        local);
-        }
+err:
+    DHT_STACK_UNWIND(create, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL,
+                     NULL);
+    return 0;
+}
 
-        if (subvol != avail_subvol) {
-                /* create a link file instead of actual file */
-                local->params = dict_ref (params);
-                local->mode = mode;
-                local->flags = flags;
-                local->umask = umask;
-                local->cached_subvol = avail_subvol;
-                dht_linkfile_create (frame, nufa_create_linkfile_create_cbk,
-                                     this, avail_subvol, subvol, loc);
-                return 0;
-        }
+int
+nufa_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+            mode_t mode, mode_t umask, fd_t *fd, dict_t *params)
+{
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    xlator_t *subvol = NULL;
+    xlator_t *avail_subvol = NULL;
+    int op_errno = -1;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+
+    conf = this->private;
+
+    dht_get_du_info(frame, this, loc);
+
+    local = dht_local_init(frame, loc, fd, GF_FOP_CREATE);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = dht_subvol_get_hashed(this, loc);
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no subvolume in layout for path=%s",
+                     loc->path);
+        op_errno = ENOENT;
+        goto err;
+    }
+
+    avail_subvol = conf->private;
+    if (dht_is_subvol_filled(this, (xlator_t *)conf->private)) {
+        avail_subvol = dht_free_disk_available_subvol(
+            this, (xlator_t *)conf->private, local);
+    }
+
+    if (subvol != avail_subvol) {
+        /* create a link file instead of actual file */
+        local->params = dict_ref(params);
+        local->mode = mode;
+        local->flags = flags;
+        local->umask = umask;
+        local->cached_subvol = avail_subvol;
+        dht_linkfile_create(frame, nufa_create_linkfile_create_cbk, this,
+                            avail_subvol, subvol, loc);
+        return 0;
+    }
 
-        gf_log (this->name, GF_LOG_TRACE,
-                "creating %s on %s", loc->path, subvol->name);
+    gf_msg_trace(this->name, 0, "creating %s on %s", loc->path, subvol->name);
 
-        STACK_WIND (frame, dht_create_cbk,
-                    subvol, subvol->fops->create,
-                    loc, flags, mode, umask, fd, params);
+    STACK_WIND_COOKIE(frame, dht_create_cbk, subvol, subvol,
+                      subvol->fops->create, loc, flags, mode, umask, fd,
+                      params);
 
-        return 0;
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (create, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(create, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL,
+                     NULL);
 
-        return 0;
+    return 0;
 }
 
 int
-nufa_mknod_linkfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int op_ret, int op_errno, inode_t *inode,
-                         struct iatt *stbuf, struct iatt *preparent,
-                         struct iatt *postparent, dict_t *xdata)
+nufa_mknod_linkfile_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, inode_t *inode,
+                        struct iatt *stbuf, struct iatt *preparent,
+                        struct iatt *postparent, dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
+    dht_local_t *local = NULL;
 
-        local = frame->local;
-        if (!local || !local->cached_subvol) {
-                op_errno = EINVAL;
-                op_ret = -1;
-                goto err;
-        }
+    local = frame->local;
+    if (!local || !local->cached_subvol) {
+        op_errno = EINVAL;
+        op_ret = -1;
+        goto err;
+    }
 
-        if (op_ret >= 0) {
-                STACK_WIND_COOKIE (frame, dht_newfile_cbk,
-                            (void *)local->cached_subvol, local->cached_subvol,
-                            local->cached_subvol->fops->mknod,
-                            &local->loc, local->mode, local->rdev,
-                            local->umask, local->params);
+    if (op_ret >= 0) {
+        STACK_WIND_COOKIE(
+            frame, dht_newfile_cbk, (void *)local->cached_subvol,
+            local->cached_subvol, local->cached_subvol->fops->mknod,
+            &local->loc, local->mode, local->rdev, local->umask, local->params);
 
-                return 0;
-        }
+        return 0;
+    }
 err:
-        WIPE (postparent);
-        WIPE (preparent);
+    WIPE(postparent);
+    WIPE(preparent);
 
-        DHT_STACK_UNWIND (link, frame, op_ret, op_errno,
-                          inode, stbuf, preparent, postparent, xdata);
-        return 0;
+    DHT_STACK_UNWIND(link, frame, op_ret, op_errno, inode, stbuf, preparent,
+                     postparent, xdata);
+    return 0;
 }
 
-
 int
-nufa_mknod (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, mode_t mode, dev_t rdev, mode_t umask, dict_t *params)
+nufa_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+           dev_t rdev, mode_t umask, dict_t *params)
 {
-        dht_local_t *local = NULL;
-        dht_conf_t  *conf  = NULL;
-        xlator_t    *subvol = NULL;
-        xlator_t    *avail_subvol = NULL;
-        int          op_errno = -1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-
-        conf  = this->private;
-
-        dht_get_du_info (frame, this, loc);
-
-        local = dht_local_init (frame, loc, NULL, GF_FOP_MKNOD);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-        subvol = dht_subvol_get_hashed (this, loc);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no subvolume in layout for path=%s",
-                        loc->path);
-                op_errno = ENOENT;
-                goto err;
-        }
-
-        /* Consider the disksize in consideration */
-        avail_subvol = conf->private;
-        if (dht_is_subvol_filled (this, (xlator_t *)conf->private)) {
-                avail_subvol =
-                        dht_free_disk_available_subvol (this,
-                                                        (xlator_t *)conf->private,
-                                                        local);
-        }
-
-        if (avail_subvol != subvol) {
-                /* Create linkfile first */
-
-                local->params = dict_ref (params);
-                local->mode = mode;
-                local->umask = umask;
-                local->rdev = rdev;
-                local->cached_subvol = avail_subvol;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    xlator_t *subvol = NULL;
+    xlator_t *avail_subvol = NULL;
+    int op_errno = -1;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+
+    conf = this->private;
+
+    dht_get_du_info(frame, this, loc);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_MKNOD);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = dht_subvol_get_hashed(this, loc);
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no subvolume in layout for path=%s",
+                     loc->path);
+        op_errno = ENOENT;
+        goto err;
+    }
+
+    /* Consider the disksize in consideration */
+    avail_subvol = conf->private;
+    if (dht_is_subvol_filled(this, (xlator_t *)conf->private)) {
+        avail_subvol = dht_free_disk_available_subvol(
+            this, (xlator_t *)conf->private, local);
+    }
+
+    if (avail_subvol != subvol) {
+        /* Create linkfile first */
+
+        local->params = dict_ref(params);
+        local->mode = mode;
+        local->umask = umask;
+        local->rdev = rdev;
+        local->cached_subvol = avail_subvol;
+
+        dht_linkfile_create(frame, nufa_mknod_linkfile_cbk, this, avail_subvol,
+                            subvol, loc);
+        return 0;
+    }
 
-                dht_linkfile_create (frame, nufa_mknod_linkfile_cbk, this,
-                                     avail_subvol, subvol, loc);
-                return 0;
-        }
+    gf_msg_trace(this->name, 0, "creating %s on %s", loc->path, subvol->name);
 
-        gf_log (this->name, GF_LOG_TRACE,
-                "creating %s on %s", loc->path, subvol->name);
+    STACK_WIND_COOKIE(frame, dht_newfile_cbk, (void *)subvol, subvol,
+                      subvol->fops->mknod, loc, mode, rdev, umask, params);
 
-        STACK_WIND_COOKIE (frame, dht_newfile_cbk, (void *)subvol, subvol,
-                           subvol->fops->mknod, loc, mode, rdev, umask,
-                           params);
-
-        return 0;
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (mknod, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
-
 gf_boolean_t
-same_first_part (char *str1, char term1, char *str2, char term2)
+same_first_part(char *str1, char term1, char *str2, char term2)
 {
-        gf_boolean_t    ended1;
-        gf_boolean_t    ended2;
-
-        for (;;) {
-                ended1 = ((*str1 == '\0') || (*str1 == term1));
-                ended2 = ((*str2 == '\0') || (*str2 == term2));
-                if (ended1 && ended2) {
-                        return _gf_true;
-                }
-                if (ended1 || ended2 || (*str1 != *str2)) {
-                        return _gf_false;
-                }
-                ++str1;
-                ++str2;
+    gf_boolean_t ended1;
+    gf_boolean_t ended2;
+
+    for (;;) {
+        ended1 = ((*str1 == '\0') || (*str1 == term1));
+        ended2 = ((*str2 == '\0') || (*str2 == term2));
+        if (ended1 && ended2) {
+            return _gf_true;
+        }
+        if (ended1 || ended2 || (*str1 != *str2)) {
+            return _gf_false;
         }
+        ++str1;
+        ++str2;
+    }
 }
 
 typedef struct nufa_args {
-        xlator_t    *this;
-        char        *volname;
-        gf_boolean_t addr_match;
+    xlator_t *this;
+    char *volname;
+    gf_boolean_t addr_match;
 } nufa_args_t;
 
 static void
-nufa_find_local_brick (xlator_t *xl, void *data)
+nufa_find_local_brick(xlator_t *xl, void *data)
 {
-        nufa_args_t     *args = data;
-        xlator_t        *this = args->this;
-        char            *local_volname = args->volname;
-        gf_boolean_t    addr_match = args->addr_match;
-        char            *brick_host = NULL;
-        dht_conf_t      *conf = this->private;
-        int             ret = -1;
-
-        /*This means a local subvol was already found. We pick the first brick
-         * that is local*/
-        if (conf->private)
-                return;
-
-        if (strcmp (xl->name, local_volname) == 0) {
-                conf->private = xl;
-                gf_log (this->name, GF_LOG_INFO, "Using specified subvol %s",
-                        local_volname);
-                return;
-        }
-
-        if (!addr_match)
-                return;
-
-        ret = dict_get_str (xl->options, "remote-host", &brick_host);
-        if ((ret == 0) &&
-            (gf_is_same_address (local_volname, brick_host) ||
-             gf_is_local_addr (brick_host))) {
-                conf->private = xl;
-                gf_log (this->name, GF_LOG_INFO, "Using the first local "
-                        "subvol %s", xl->name);
-                return;
-        }
-
+    nufa_args_t *args = data;
+    xlator_t *this = args->this;
+    char *local_volname = args->volname;
+    gf_boolean_t addr_match = args->addr_match;
+    char *brick_host = NULL;
+    dht_conf_t *conf = this->private;
+    int ret = -1;
+
+    /*This means a local subvol was already found. We pick the first brick
+     * that is local*/
+    if (conf->private)
+        return;
+
+    if (strcmp(xl->name, local_volname) == 0) {
+        conf->private = xl;
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_INFO,
+               "Using specified subvol %s", local_volname);
+        return;
+    }
+
+    if (!addr_match)
+        return;
+
+    ret = dict_get_str(xl->options, "remote-host", &brick_host);
+    if ((ret == 0) && (gf_is_same_address(local_volname, brick_host) ||
+                       gf_is_local_addr(brick_host))) {
+        conf->private = xl;
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_INFO,
+               "Using the first local "
+               "subvol %s",
+               xl->name);
+        return;
+    }
 }
 
 static void
-nufa_to_dht (xlator_t *this)
+nufa_to_dht(xlator_t *this)
 {
-        GF_ASSERT (this);
-        GF_ASSERT (this->fops);
+    GF_ASSERT(this);
+    GF_ASSERT(this->fops);
 
-        this->fops->lookup = dht_lookup;
-        this->fops->create = dht_create;
-        this->fops->mknod  = dht_mknod;
+    this->fops->lookup = dht_lookup;
+    this->fops->create = dht_create;
+    this->fops->mknod = dht_mknod;
 }
 
 int
-nufa_find_local_subvol (xlator_t *this,
-                        void (*fn) (xlator_t *each, void* data), void *data)
+nufa_find_local_subvol(xlator_t *this, void (*fn)(xlator_t *each, void *data),
+                       void *data)
 {
-        int             ret = -1;
-        dht_conf_t      *conf = this->private;
-        xlator_list_t   *trav = NULL;
-        xlator_t        *parent = NULL;
-        xlator_t        *candidate = NULL;
-
-        xlator_foreach_depth_first (this, fn, data);
-        if (!conf->private) {
-                gf_log (this->name, GF_LOG_ERROR, "Couldn't find a local "
-                        "brick");
-                return -1;
+    int ret = -1;
+    dht_conf_t *conf = this->private;
+    xlator_list_t *trav = NULL;
+    xlator_t *parent = NULL;
+    xlator_t *candidate = NULL;
+
+    xlator_foreach_depth_first(this, fn, data);
+    if (!conf->private) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_BRICK_ERROR,
+               "Couldn't find a local "
+               "brick");
+        return -1;
+    }
+
+    candidate = conf->private;
+    trav = candidate->parents;
+    while (trav) {
+        parent = trav->xlator;
+        if (strcmp(parent->type, "cluster/nufa") == 0) {
+            gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_INFO,
+                   "Found local subvol, "
+                   "%s",
+                   candidate->name);
+            ret = 0;
+            conf->private = candidate;
+            break;
         }
 
-        candidate = conf->private;
-        trav = candidate->parents;
-        while (trav) {
-
-                parent = trav->xlator;
-                if (strcmp (parent->type, "cluster/nufa") == 0) {
-                        gf_log (this->name, GF_LOG_INFO, "Found local subvol, "
-                                "%s", candidate->name);
-                        ret = 0;
-                        conf->private = candidate;
-                        break;
-                }
-
-                candidate = parent;
-                trav = parent->parents;
-        }
+        candidate = parent;
+        trav = parent->parents;
+    }
 
-        return ret;
+    return ret;
 }
 
 int
-nufa_init (xlator_t *this)
+nufa_init(xlator_t *this)
 {
-        data_t        *data = NULL;
-        char          *local_volname = NULL;
-        int            ret = -1;
-        char           my_hostname[256];
-        gf_boolean_t   addr_match = _gf_false;
-        nufa_args_t    args = {0, };
-
-        ret = dht_init(this);
-        if (ret) {
-                return ret;
-        }
-
-        if ((data = dict_get (this->options, "local-volume-name"))) {
-                local_volname = data->data;
-
-        } else {
-                addr_match = _gf_true;
-                local_volname = "localhost";
-                ret = gethostname (my_hostname, 256);
-                if (ret == 0)
-                        local_volname = my_hostname;
-
-                else
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "could not find hostname (%s)",
-                                strerror (errno));
-
-        }
-
-        args.this = this;
-        args.volname = local_volname;
-        args.addr_match = addr_match;
-        ret = nufa_find_local_subvol (this, nufa_find_local_brick, &args);
-        if (ret) {
-                gf_log (this->name, GF_LOG_INFO,
-                        "Unable to find local subvolume, switching "
-                        "to dht mode");
-                nufa_to_dht (this);
-        }
-        return 0;
+    data_t *data = NULL;
+    char *local_volname = NULL;
+    int ret = -1;
+    char my_hostname[256];
+    gf_boolean_t addr_match = _gf_false;
+    nufa_args_t args = {
+        0,
+    };
+
+    ret = dht_init(this);
+    if (ret) {
+        return ret;
+    }
+
+    if ((data = dict_get(this->options, "local-volume-name"))) {
+        local_volname = data->data;
+
+    } else {
+        addr_match = _gf_true;
+        local_volname = "localhost";
+        ret = gethostname(my_hostname, 256);
+        if (ret == 0)
+            local_volname = my_hostname;
+
+        else
+            gf_msg(this->name, GF_LOG_WARNING, errno,
+                   DHT_MSG_GET_HOSTNAME_FAILED, "could not find hostname");
+    }
+
+    args.this = this;
+    args.volname = local_volname;
+    args.addr_match = addr_match;
+    ret = nufa_find_local_subvol(this, nufa_find_local_brick, &args);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_INFO,
+               "Unable to find local subvolume, switching "
+               "to dht mode");
+        nufa_to_dht(this);
+    }
+    return 0;
 }
 
-
-class_methods_t class_methods = {
-        .init           = nufa_init,
-        .fini           = dht_fini,
-        .reconfigure    = dht_reconfigure,
-        .notify         = dht_notify
+dht_methods_t dht_methods = {
+    .migration_get_dst_subvol = dht_migration_get_dst_subvol,
+    .layout_search = dht_layout_search,
 };
 
-
 struct xlator_fops fops = {
-        .lookup      = nufa_lookup,
-        .create      = nufa_create,
-        .mknod       = nufa_mknod,
-
-        .stat        = dht_stat,
-        .fstat       = dht_fstat,
-        .truncate    = dht_truncate,
-        .ftruncate   = dht_ftruncate,
-        .access      = dht_access,
-        .readlink    = dht_readlink,
-        .setxattr    = dht_setxattr,
-        .getxattr    = dht_getxattr,
-        .removexattr = dht_removexattr,
-        .open        = dht_open,
-        .readv       = dht_readv,
-        .writev      = dht_writev,
-        .flush       = dht_flush,
-        .fsync       = dht_fsync,
-        .statfs      = dht_statfs,
-        .lk          = dht_lk,
-        .opendir     = dht_opendir,
-        .readdir     = dht_readdir,
-        .readdirp    = dht_readdirp,
-        .fsyncdir    = dht_fsyncdir,
-        .symlink     = dht_symlink,
-        .unlink      = dht_unlink,
-        .link        = dht_link,
-        .mkdir       = dht_mkdir,
-        .rmdir       = dht_rmdir,
-        .rename      = dht_rename,
-        .inodelk     = dht_inodelk,
-        .finodelk    = dht_finodelk,
-        .entrylk     = dht_entrylk,
-        .fentrylk    = dht_fentrylk,
-        .xattrop     = dht_xattrop,
-        .fxattrop    = dht_fxattrop,
-        .setattr     = dht_setattr,
+    .lookup = nufa_lookup,
+    .create = nufa_create,
+    .mknod = nufa_mknod,
+
+    .stat = dht_stat,
+    .fstat = dht_fstat,
+    .truncate = dht_truncate,
+    .ftruncate = dht_ftruncate,
+    .access = dht_access,
+    .readlink = dht_readlink,
+    .setxattr = dht_setxattr,
+    .getxattr = dht_getxattr,
+    .removexattr = dht_removexattr,
+    .open = dht_open,
+    .readv = dht_readv,
+    .writev = dht_writev,
+    .flush = dht_flush,
+    .fsync = dht_fsync,
+    .statfs = dht_statfs,
+    .lk = dht_lk,
+    .opendir = dht_opendir,
+    .readdir = dht_readdir,
+    .readdirp = dht_readdirp,
+    .fsyncdir = dht_fsyncdir,
+    .symlink = dht_symlink,
+    .unlink = dht_unlink,
+    .link = dht_link,
+    .mkdir = dht_mkdir,
+    .rmdir = dht_rmdir,
+    .rename = dht_rename,
+    .inodelk = dht_inodelk,
+    .finodelk = dht_finodelk,
+    .entrylk = dht_entrylk,
+    .fentrylk = dht_fentrylk,
+    .xattrop = dht_xattrop,
+    .fxattrop = dht_fxattrop,
+    .setattr = dht_setattr,
 };
 
-
-struct xlator_cbks cbks = {
-        .forget     = dht_forget
+struct xlator_cbks cbks = {.forget = dht_forget};
+extern int32_t
+mem_acct_init(xlator_t *this);
+
+xlator_api_t xlator_api = {
+    .init = nufa_init,
+    .fini = dht_fini,
+    .notify = dht_notify,
+    .reconfigure = dht_reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = dht_options,
+    .identifier = "nufa",
+    .category = GF_TECH_PREVIEW,
 };
diff --git a/xlators/cluster/dht/src/switch.c b/xlators/cluster/dht/src/switch.c
index d3ea90ba8cf..207d109a025 100644
--- a/xlators/cluster/dht/src/switch.c
+++ b/xlators/cluster/dht/src/switch.c
@@ -8,12 +8,6 @@
   cases as published by the Free Software Foundation.
 */
 
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
 #include "dht-common.h"
 #include "dht-mem-types.h"
 
@@ -22,883 +16,876 @@
 #include <fnmatch.h>
 #include <string.h>
 
-extern struct volume_options options[];
+extern struct volume_options dht_options[];
 
 struct switch_sched_array {
-        xlator_t *xl;
-        int32_t   eligible;
-        int32_t   considered;
+    xlator_t *xl;
+    int32_t eligible;
+    int32_t considered;
 };
 
 /* Select one of this struct based on the path's pattern match */
 struct switch_struct {
-        struct switch_struct      *next;
-        struct switch_sched_array *array;
-        int32_t                    node_index; /* Index of the node in
-                                                  this pattern. */
-        int32_t                    num_child;  /* Total num of child nodes
-                                                  with this pattern. */
-        char                       path_pattern[256];
+    struct switch_struct *next;
+    struct switch_sched_array *array;
+    int32_t node_index; /* Index of the node in
+                           this pattern. */
+    int32_t num_child;  /* Total num of child nodes
+                           with this pattern. */
+    char path_pattern[256];
 };
 
 /* TODO: all 'TODO's in dht.c holds good */
 /* This function should return child node as '*:subvolumes' is inserterd */
 
 static int32_t
-gf_switch_valid_child (xlator_t *this, const char *child)
+gf_switch_valid_child(xlator_t *this, const char *child)
 {
-        xlator_list_t *children = NULL;
-        int32_t        ret = 0;
-
-        children = this->children;
-        while (children) {
-                if (!strcmp (child, children->xlator->name)) {
-                        ret = 1;
-                        break;
-                }
-                children = children->next;
+    xlator_list_t *children = NULL;
+    int32_t ret = 0;
+
+    children = this->children;
+    while (children) {
+        if (!strcmp(child, children->xlator->name)) {
+            ret = 1;
+            break;
         }
+        children = children->next;
+    }
 
-        return ret;
+    return ret;
 }
 
 static xlator_t *
-get_switch_matching_subvol (const char *path, dht_conf_t *conf,
-                            xlator_t *hashed_subvol)
+get_switch_matching_subvol(const char *path, dht_conf_t *conf,
+                           xlator_t *hashed_subvol)
 {
-        struct switch_struct *cond      = NULL;
-        struct switch_struct *trav      = NULL;
-        char                 *pathname  = NULL;
-        int                   idx       = 0;
-        xlator_t             *subvol    = NULL;
-
-        cond = conf->private;
-        subvol = hashed_subvol;
-        if (!cond)
-                goto out;
-
-        pathname = gf_strdup (path);
-        if (!pathname)
-                goto out;
-
-        trav = cond;
-        while (trav) {
-                if (fnmatch (trav->path_pattern,
-                             pathname, FNM_NOESCAPE) == 0) {
-                        for (idx = 0; idx < trav->num_child; idx++) {
-                                if (trav->array[idx].xl == hashed_subvol)
-                                        goto out;
-                        }
-                        idx = trav->node_index++;
-                        trav->node_index %= trav->num_child;
-                        subvol = trav->array[idx].xl;
-                        goto out;
-                }
-                trav = trav->next;
+    struct switch_struct *cond = NULL;
+    struct switch_struct *trav = NULL;
+    char *pathname = NULL;
+    int idx = 0;
+    xlator_t *subvol = NULL;
+
+    cond = conf->private;
+    subvol = hashed_subvol;
+    if (!cond)
+        goto out;
+
+    pathname = gf_strdup(path);
+    if (!pathname)
+        goto out;
+
+    trav = cond;
+    while (trav) {
+        if (fnmatch(trav->path_pattern, pathname, FNM_NOESCAPE) == 0) {
+            for (idx = 0; idx < trav->num_child; idx++) {
+                if (trav->array[idx].xl == hashed_subvol)
+                    goto out;
+            }
+            idx = trav->node_index++;
+            trav->node_index %= trav->num_child;
+            subvol = trav->array[idx].xl;
+            goto out;
         }
+        trav = trav->next;
+    }
 out:
-        GF_FREE (pathname);
+    GF_FREE(pathname);
 
-        return subvol;
+    return subvol;
 }
 
-
 int
-switch_local_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int op_ret, int op_errno,
-                         inode_t *inode, struct iatt *stbuf, dict_t *xattr,
-                         struct iatt *postparent)
+switch_local_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, inode_t *inode,
+                        struct iatt *stbuf, dict_t *xattr,
+                        struct iatt *postparent)
 {
-        xlator_t     *subvol      = NULL;
-        char          is_linkfile = 0;
-        char          is_dir      = 0;
-        dht_conf_t   *conf        = NULL;
-        dht_local_t  *local       = NULL;
-        loc_t        *loc         = NULL;
-        int           i           = 0;
-        call_frame_t *prev        = NULL;
-        int           call_cnt    = 0;
-        int           ret         = 0;
-
-        conf  = this->private;
-
-        prev  = cookie;
-        local = frame->local;
-        loc   = &local->loc;
-
-        if (ENTRY_MISSING (op_ret, op_errno)) {
-                if (conf->search_unhashed) {
-                        local->op_errno = ENOENT;
-                        dht_lookup_everywhere (frame, this, loc);
-                        return 0;
-                }
+    xlator_t *subvol = NULL;
+    char is_linkfile = 0;
+    char is_dir = 0;
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    loc_t *loc = NULL;
+    int i = 0;
+    xlator_t *prev = NULL;
+    int call_cnt = 0;
+    int ret = 0;
+
+    conf = this->private;
+
+    prev = cookie;
+    local = frame->local;
+    loc = &local->loc;
+
+    if (ENTRY_MISSING(op_ret, op_errno)) {
+        if (conf->search_unhashed) {
+            local->op_errno = ENOENT;
+            dht_lookup_everywhere(frame, this, loc);
+            return 0;
         }
-
-        if (op_ret == -1)
-                goto out;
-
-        is_linkfile = check_is_linkfile (inode, stbuf, xattr,
-                                         conf->link_xattr_name);
-        is_dir      = check_is_dir (inode, stbuf, xattr);
-
-        if (!is_dir && !is_linkfile) {
-                /* non-directory and not a linkfile */
-
-                ret = dht_layout_preset (this, prev->this, inode);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "could not set pre-set layout for subvol %s",
-                                prev->this->name);
-                        op_ret   = -1;
-                        op_errno = EINVAL;
-                        goto err;
-                }
-
-                goto out;
+    }
+
+    if (op_ret == -1)
+        goto out;
+
+    is_linkfile = check_is_linkfile(inode, stbuf, xattr, conf->link_xattr_name);
+    is_dir = check_is_dir(inode, stbuf, xattr);
+
+    if (!is_dir && !is_linkfile) {
+        /* non-directory and not a linkfile */
+
+        ret = dht_layout_preset(this, prev, inode);
+        if (ret < 0) {
+            gf_msg_debug(this->name, 0,
+                         "could not set pre-set layout "
+                         "for subvol %s",
+                         prev->name);
+            op_ret = -1;
+            op_errno = EINVAL;
+            goto err;
         }
 
-        if (is_dir) {
-                call_cnt        = conf->subvolume_cnt;
-                local->call_cnt = call_cnt;
+        goto out;
+    }
 
-                local->inode = inode_ref (inode);
-                local->xattr = dict_ref (xattr);
+    if (is_dir) {
+        call_cnt = conf->subvolume_cnt;
+        local->call_cnt = call_cnt;
 
-                local->op_ret = 0;
-                local->op_errno = 0;
+        local->inode = inode_ref(inode);
+        local->xattr = dict_ref(xattr);
 
-                local->layout = dht_layout_new (this, conf->subvolume_cnt);
-                if (!local->layout) {
-                        op_ret   = -1;
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "memory allocation failed :(");
-                        goto err;
-                }
+        local->op_ret = 0;
+        local->op_errno = 0;
 
-                for (i = 0; i < call_cnt; i++) {
-                        STACK_WIND (frame, dht_lookup_dir_cbk,
-                                    conf->subvolumes[i],
-                                    conf->subvolumes[i]->fops->lookup,
-                                    &local->loc, local->xattr_req);
-                }
+        local->layout = dht_layout_new(this, conf->subvolume_cnt);
+        if (!local->layout) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            gf_msg_debug(this->name, 0, "memory allocation failed :(");
+            goto err;
         }
 
-        if (is_linkfile) {
-                subvol = dht_linkfile_subvol (this, inode, stbuf, xattr);
+        for (i = 0; i < call_cnt; i++) {
+            STACK_WIND_COOKIE(frame, dht_lookup_dir_cbk, conf->subvolumes[i],
+                              conf->subvolumes[i],
+                              conf->subvolumes[i]->fops->lookup, &local->loc,
+                              local->xattr_req);
+        }
+    }
 
-                if (!subvol) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "linkfile not having link subvolume. path=%s",
-                                loc->path);
-                        dht_lookup_everywhere (frame, this, loc);
-                        return 0;
-                }
+    if (is_linkfile) {
+        subvol = dht_linkfile_subvol(this, inode, stbuf, xattr);
 
-                STACK_WIND (frame, dht_lookup_linkfile_cbk,
-                            subvol, subvol->fops->lookup,
-                            &local->loc, local->xattr_req);
+        if (!subvol) {
+            gf_msg_debug(this->name, 0,
+                         "linkfile has no link subvolume.path=%s", loc->path);
+            dht_lookup_everywhere(frame, this, loc);
+            return 0;
         }
 
-        return 0;
+        STACK_WIND_COOKIE(frame, dht_lookup_linkfile_cbk, subvol, subvol,
+                          subvol->fops->lookup, &local->loc, local->xattr_req);
+    }
+
+    return 0;
 
 out:
-        if (!local->hashed_subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no subvolume in layout for path=%s",
-                        local->loc.path);
-                local->op_errno = ENOENT;
-                dht_lookup_everywhere (frame, this, loc);
-                return 0;
-        }
+    if (!local->hashed_subvol) {
+        gf_msg_debug(this->name, 0, "no subvolume in layout for path=%s",
+                     local->loc.path);
+        local->op_errno = ENOENT;
+        dht_lookup_everywhere(frame, this, loc);
+        return 0;
+    }
 
-        STACK_WIND (frame, dht_lookup_cbk,
-                    local->hashed_subvol, local->hashed_subvol->fops->lookup,
-                    &local->loc, local->xattr_req);
+    STACK_WIND_COOKIE(frame, dht_lookup_cbk, local->hashed_subvol,
+                      local->hashed_subvol, local->hashed_subvol->fops->lookup,
+                      &local->loc, local->xattr_req);
 
-        return 0;
+    return 0;
 
 err:
-        DHT_STACK_UNWIND (lookup, frame, op_ret, op_errno,
-                          inode, stbuf, xattr, NULL);
-        return 0;
+    DHT_STACK_UNWIND(lookup, frame, op_ret, op_errno, inode, stbuf, xattr,
+                     NULL);
+    return 0;
 }
 
 int
-switch_lookup (call_frame_t *frame, xlator_t *this,
-               loc_t *loc, dict_t *xattr_req)
+switch_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc,
+              dict_t *xattr_req)
 {
-        xlator_t     *hashed_subvol = NULL;
-        xlator_t     *cached_subvol = NULL;
-        xlator_t     *subvol = NULL;
-        dht_local_t  *local  = NULL;
-        dht_conf_t   *conf = NULL;
-        int           ret    = -1;
-        int           op_errno = -1;
-        dht_layout_t *layout = NULL;
-        int           i = 0;
-        int           call_cnt = 0;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-        conf = this->private;
-
-        local = dht_local_init (frame, loc, NULL, GF_FOP_LOOKUP);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
+    xlator_t *hashed_subvol = NULL;
+    xlator_t *cached_subvol = NULL;
+    xlator_t *subvol = NULL;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
+    int op_errno = -1;
+    dht_layout_t *layout = NULL;
+    int i = 0;
+    int call_cnt = 0;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+    VALIDATE_OR_GOTO(loc->path, err);
+
+    conf = this->private;
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_LOOKUP);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    if (xattr_req) {
+        local->xattr_req = dict_ref(xattr_req);
+    } else {
+        local->xattr_req = dict_new();
+    }
+
+    hashed_subvol = dht_subvol_get_hashed(this, &local->loc);
+    cached_subvol = local->cached_subvol;
+
+    local->hashed_subvol = hashed_subvol;
+
+    if (is_revalidate(loc)) {
+        layout = local->layout;
+        if (!layout) {
+            gf_msg_debug(this->name, 0,
+                         "revalidate lookup without cache. path=%s", loc->path);
+            op_errno = EINVAL;
+            goto err;
         }
 
-        if (xattr_req) {
-                local->xattr_req = dict_ref (xattr_req);
-        } else {
-                local->xattr_req = dict_new ();
+        if (layout->gen && (layout->gen < conf->gen)) {
+            gf_msg_debug(this->name, 0, "incomplete layout failure for path=%s",
+                         loc->path);
+            dht_layout_unref(this, local->layout);
+            goto do_fresh_lookup;
         }
 
-        hashed_subvol = dht_subvol_get_hashed (this, &local->loc);
-        cached_subvol = local->cached_subvol;
+        local->inode = inode_ref(loc->inode);
 
-        local->hashed_subvol = hashed_subvol;
-
-        if (is_revalidate (loc)) {
-                layout = local->layout;
-                if (!layout) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "revalidate without cache. path=%s",
-                                loc->path);
-                        op_errno = EINVAL;
-                        goto err;
-                }
+        local->call_cnt = layout->cnt;
+        call_cnt = local->call_cnt;
 
-                if (layout->gen && (layout->gen < conf->gen)) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "incomplete layout failure for path=%s",
-                                loc->path);
-                        dht_layout_unref (this, local->layout);
-                        goto do_fresh_lookup;
-                }
-
-                local->inode    = inode_ref (loc->inode);
-
-                local->call_cnt = layout->cnt;
-                call_cnt = local->call_cnt;
+        /* NOTE: we don't require 'trusted.glusterfs.dht.linkto'
+         * attribute, revalidates directly go to the cached-subvolume.
+         */
+        ret = dict_set_uint32(local->xattr_req, conf->xattr_name, 4 * 4);
+        if (ret < 0)
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+                   "failed to set dict value for %s", conf->xattr_name);
 
-                /* NOTE: we don't require 'trusted.glusterfs.dht.linkto'
-                 * attribute, revalidates directly go to the cached-subvolume.
-                 */
-                ret = dict_set_uint32 (local->xattr_req,
-                                       conf->xattr_name, 4 * 4);
-                if (ret < 0)
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "failed to set dict value for %s",
-                                conf->xattr_name);
+        for (i = 0; i < layout->cnt; i++) {
+            subvol = layout->list[i].xlator;
 
-                for (i = 0; i < layout->cnt; i++) {
-                        subvol = layout->list[i].xlator;
+            STACK_WIND_COOKIE(frame, dht_revalidate_cbk, subvol, subvol,
+                              subvol->fops->lookup, loc, local->xattr_req);
 
-                        STACK_WIND (frame, dht_revalidate_cbk,
-                                    subvol, subvol->fops->lookup,
-                                    loc, local->xattr_req);
+            if (!--call_cnt)
+                break;
+        }
+    } else {
+    do_fresh_lookup:
+        ret = dict_set_uint32(local->xattr_req, conf->xattr_name, 4 * 4);
+        if (ret < 0)
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+                   "failed to set dict value for %s", conf->xattr_name);
+
+        ret = dict_set_uint32(local->xattr_req, conf->link_xattr_name, 256);
+        if (ret < 0)
+            gf_msg(this->name, GF_LOG_WARNING, EINVAL, DHT_MSG_DICT_SET_FAILED,
+                   "failed to set dict value for %s", conf->link_xattr_name);
+
+        if (!hashed_subvol) {
+            gf_msg_debug(this->name, 0,
+                         "no subvolume in layout for path=%s, "
+                         "checking on all the subvols to see if "
+                         "it is a directory",
+                         loc->path);
+            call_cnt = conf->subvolume_cnt;
+            local->call_cnt = call_cnt;
+
+            local->layout = dht_layout_new(this, conf->subvolume_cnt);
+            if (!local->layout) {
+                op_errno = ENOMEM;
+                goto err;
+            }
+
+            for (i = 0; i < call_cnt; i++) {
+                STACK_WIND_COOKIE(frame, dht_lookup_dir_cbk,
+                                  conf->subvolumes[i], conf->subvolumes[i],
+                                  conf->subvolumes[i]->fops->lookup,
+                                  &local->loc, local->xattr_req);
+            }
+            return 0;
+        }
 
-                        if (!--call_cnt)
-                                break;
-                }
+        /*  */
+        cached_subvol = get_switch_matching_subvol(loc->path, conf,
+                                                   hashed_subvol);
+        if (cached_subvol == hashed_subvol) {
+            STACK_WIND_COOKIE(frame, dht_lookup_cbk, hashed_subvol,
+                              hashed_subvol, hashed_subvol->fops->lookup, loc,
+                              local->xattr_req);
         } else {
-        do_fresh_lookup:
-                ret = dict_set_uint32 (local->xattr_req,
-                                       conf->xattr_name, 4 * 4);
-                if (ret < 0)
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "failed to set dict value for %s",
-                                conf->xattr_name);
-
-                ret = dict_set_uint32 (local->xattr_req,
-                                       conf->link_xattr_name, 256);
-                if (ret < 0)
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "failed to set dict value for %s",
-                                conf->link_xattr_name);
-
-                if (!hashed_subvol) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "no subvolume in layout for path=%s, "
-                                "checking on all the subvols to see if "
-                                "it is a directory", loc->path);
-                        call_cnt        = conf->subvolume_cnt;
-                        local->call_cnt = call_cnt;
-
-                        local->layout = dht_layout_new (this,
-                                                        conf->subvolume_cnt);
-                        if (!local->layout) {
-                                op_errno = ENOMEM;
-                                goto err;
-                        }
-
-                        for (i = 0; i < call_cnt; i++) {
-                                STACK_WIND (frame, dht_lookup_dir_cbk,
-                                            conf->subvolumes[i],
-                                            conf->subvolumes[i]->fops->lookup,
-                                            &local->loc, local->xattr_req);
-                        }
-                        return 0;
-                }
-
-                /*  */
-                cached_subvol = get_switch_matching_subvol (loc->path, conf,
-                                                            hashed_subvol);
-                if (cached_subvol == hashed_subvol) {
-                        STACK_WIND (frame, dht_lookup_cbk,
-                                    hashed_subvol,
-                                    hashed_subvol->fops->lookup,
-                                    loc, local->xattr_req);
-                } else {
-                        STACK_WIND (frame, switch_local_lookup_cbk,
-                                    cached_subvol,
-                                    cached_subvol->fops->lookup,
-                                    loc, local->xattr_req);
-                }
+            STACK_WIND_COOKIE(frame, switch_local_lookup_cbk, cached_subvol,
+                              cached_subvol, cached_subvol->fops->lookup, loc,
+                              local->xattr_req);
         }
+    }
 
-        return 0;
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (lookup, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL);
-        return 0;
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+    return 0;
 }
 
 int
-switch_create_linkfile_create_cbk (call_frame_t *frame, void *cookie,
-                                   xlator_t *this, int op_ret, int op_errno,
-                                   inode_t *inode, struct iatt *stbuf,
-                                   struct iatt *preparent,
-                                   struct iatt *postparent, dict_t *xdata)
+switch_create_linkfile_create_cbk(call_frame_t *frame, void *cookie,
+                                  xlator_t *this, int op_ret, int op_errno,
+                                  inode_t *inode, struct iatt *stbuf,
+                                  struct iatt *preparent,
+                                  struct iatt *postparent, dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
+    dht_local_t *local = NULL;
 
-        local = frame->local;
+    local = frame->local;
 
-        if (op_ret == -1)
-                goto err;
+    if (op_ret == -1)
+        goto err;
 
-        STACK_WIND (frame, dht_create_cbk,
-                    local->cached_subvol, local->cached_subvol->fops->create,
-                    &local->loc, local->flags, local->mode, local->umask,
-                    local->fd, local->params);
+    STACK_WIND_COOKIE(frame, dht_create_cbk, local->cached_subvol,
+                      local->cached_subvol, local->cached_subvol->fops->create,
+                      &local->loc, local->flags, local->mode, local->umask,
+                      local->fd, local->params);
 
-        return 0;
+    return 0;
 
 err:
-        DHT_STACK_UNWIND (create, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL, NULL, NULL);
-        return 0;
+    DHT_STACK_UNWIND(create, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL,
+                     NULL);
+    return 0;
 }
 
 int
-switch_create (call_frame_t *frame, xlator_t *this,
-               loc_t *loc, int32_t flags, mode_t mode,
-               mode_t umask, fd_t *fd, dict_t *params)
+switch_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+              mode_t mode, mode_t umask, fd_t *fd, dict_t *params)
 {
-        dht_local_t *local = NULL;
-        dht_conf_t  *conf  = NULL;
-        xlator_t    *subvol = NULL;
-        xlator_t    *avail_subvol = NULL;
-        int          op_errno = -1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-
-        conf  = this->private;
-
-        dht_get_du_info (frame, this, loc);
-
-        local = dht_local_init (frame, loc, fd, GF_FOP_CREATE);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-        subvol = dht_subvol_get_hashed (this, loc);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no subvolume in layout for path=%s",
-                        loc->path);
-                op_errno = ENOENT;
-                goto err;
-        }
-
-        avail_subvol = get_switch_matching_subvol (loc->path, conf, subvol);
-        if (dht_is_subvol_filled (this, avail_subvol)) {
-                avail_subvol =
-                        dht_free_disk_available_subvol (this, avail_subvol,
-                                                        local);
-        }
-
-        if (subvol != avail_subvol) {
-                /* create a link file instead of actual file */
-                local->mode = mode;
-                local->flags = flags;
-                local->umask = umask;
-                local->cached_subvol = avail_subvol;
-                dht_linkfile_create (frame, switch_create_linkfile_create_cbk,
-                                     this, avail_subvol, subvol, loc);
-                return 0;
-        }
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    xlator_t *subvol = NULL;
+    xlator_t *avail_subvol = NULL;
+    int op_errno = -1;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+
+    conf = this->private;
+
+    dht_get_du_info(frame, this, loc);
+
+    local = dht_local_init(frame, loc, fd, GF_FOP_CREATE);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = dht_subvol_get_hashed(this, loc);
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no subvolume in layout for path=%s",
+                     loc->path);
+        op_errno = ENOENT;
+        goto err;
+    }
+
+    avail_subvol = get_switch_matching_subvol(loc->path, conf, subvol);
+    if (dht_is_subvol_filled(this, avail_subvol)) {
+        avail_subvol = dht_free_disk_available_subvol(this, avail_subvol,
+                                                      local);
+    }
+
+    if (subvol != avail_subvol) {
+        /* create a link file instead of actual file */
+        local->mode = mode;
+        local->flags = flags;
+        local->umask = umask;
+        local->cached_subvol = avail_subvol;
+        dht_linkfile_create(frame, switch_create_linkfile_create_cbk, this,
+                            avail_subvol, subvol, loc);
+        return 0;
+    }
 
-        gf_log (this->name, GF_LOG_TRACE,
-                "creating %s on %s", loc->path, subvol->name);
+    gf_msg_trace(this->name, 0, "creating %s on %s", loc->path, subvol->name);
 
-        STACK_WIND (frame, dht_create_cbk,
-                    subvol, subvol->fops->create,
-                    loc, flags, mode, umask, fd, params);
+    STACK_WIND_COOKIE(frame, dht_create_cbk, subvol, subvol,
+                      subvol->fops->create, loc, flags, mode, umask, fd,
+                      params);
 
-        return 0;
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (create, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(create, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL,
+                     NULL);
 
-        return 0;
+    return 0;
 }
 
 int
-switch_mknod_linkfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                           int op_ret, int op_errno, inode_t *inode,
-                           struct iatt *stbuf, struct iatt *preparent,
-                           struct iatt *postparent, dict_t *xdata)
+switch_mknod_linkfile_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int op_ret, int op_errno, inode_t *inode,
+                          struct iatt *stbuf, struct iatt *preparent,
+                          struct iatt *postparent, dict_t *xdata)
 {
-        dht_local_t  *local = NULL;
+    dht_local_t *local = NULL;
 
-        local = frame->local;
-        if (!local || !local->cached_subvol) {
-                op_errno = EINVAL;
-                op_ret = -1;
-                goto err;
-        }
+    local = frame->local;
+    if (!local || !local->cached_subvol) {
+        op_errno = EINVAL;
+        op_ret = -1;
+        goto err;
+    }
 
-        if (op_ret >= 0) {
-                STACK_WIND_COOKIE (frame, dht_newfile_cbk,
-                            (void *)local->cached_subvol, local->cached_subvol,
-                            local->cached_subvol->fops->mknod,
-                            &local->loc, local->mode, local->rdev,
-                            local->umask, local->params);
+    if (op_ret >= 0) {
+        STACK_WIND_COOKIE(
+            frame, dht_newfile_cbk, (void *)local->cached_subvol,
+            local->cached_subvol, local->cached_subvol->fops->mknod,
+            &local->loc, local->mode, local->rdev, local->umask, local->params);
 
-                return 0;
-        }
-err:
-        DHT_STACK_UNWIND (link, frame, op_ret, op_errno,
-                          inode, stbuf, preparent, postparent, xdata);
         return 0;
+    }
+err:
+    DHT_STACK_UNWIND(link, frame, op_ret, op_errno, inode, stbuf, preparent,
+                     postparent, xdata);
+    return 0;
 }
 
-
 int
-switch_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
-              dev_t rdev, mode_t umask, dict_t *params)
+switch_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+             dev_t rdev, mode_t umask, dict_t *params)
 {
-        dht_local_t *local = NULL;
-        dht_conf_t  *conf  = NULL;
-        xlator_t    *subvol = NULL;
-        xlator_t    *avail_subvol = NULL;
-        int          op_errno = -1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-
-        conf  = this->private;
-
-        dht_get_du_info (frame, this, loc);
-
-        local = dht_local_init (frame, loc, NULL, GF_FOP_MKNOD);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-        subvol = dht_subvol_get_hashed (this, loc);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no subvolume in layout for path=%s",
-                        loc->path);
-                op_errno = ENOENT;
-                goto err;
-        }
-
-        /* Consider the disksize in consideration */
-        avail_subvol = get_switch_matching_subvol (loc->path, conf, subvol);
-        if (dht_is_subvol_filled (this, avail_subvol)) {
-                avail_subvol =
-                        dht_free_disk_available_subvol (this, avail_subvol,
-                                                        local);
-        }
-
-        if (avail_subvol != subvol) {
-                /* Create linkfile first */
-
-                local->params = dict_ref (params);
-                local->mode = mode;
-                local->umask = umask;
-                local->rdev = rdev;
-                local->cached_subvol = avail_subvol;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    xlator_t *subvol = NULL;
+    xlator_t *avail_subvol = NULL;
+    int op_errno = -1;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+
+    conf = this->private;
+
+    dht_get_du_info(frame, this, loc);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_MKNOD);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = dht_subvol_get_hashed(this, loc);
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no subvolume in layout for path=%s",
+                     loc->path);
+        op_errno = ENOENT;
+        goto err;
+    }
+
+    /* Consider the disksize in consideration */
+    avail_subvol = get_switch_matching_subvol(loc->path, conf, subvol);
+    if (dht_is_subvol_filled(this, avail_subvol)) {
+        avail_subvol = dht_free_disk_available_subvol(this, avail_subvol,
+                                                      local);
+    }
+
+    if (avail_subvol != subvol) {
+        /* Create linkfile first */
+
+        local->params = dict_ref(params);
+        local->mode = mode;
+        local->umask = umask;
+        local->rdev = rdev;
+        local->cached_subvol = avail_subvol;
+
+        dht_linkfile_create(frame, switch_mknod_linkfile_cbk, this,
+                            avail_subvol, subvol, loc);
+        return 0;
+    }
 
-                dht_linkfile_create (frame, switch_mknod_linkfile_cbk,
-                                     this, avail_subvol, subvol, loc);
-                return 0;
-        }
+    gf_msg_trace(this->name, 0, "creating %s on %s", loc->path, subvol->name);
 
-        gf_log (this->name, GF_LOG_TRACE,
-                "creating %s on %s", loc->path, subvol->name);
+    STACK_WIND_COOKIE(frame, dht_newfile_cbk, (void *)subvol, subvol,
+                      subvol->fops->mknod, loc, mode, rdev, umask, params);
 
-        STACK_WIND_COOKIE (frame, dht_newfile_cbk, (void *)subvol, subvol,
-                           subvol->fops->mknod, loc, mode, rdev, umask,
-                           params);
-
-        return 0;
+    return 0;
 
 err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (mknod, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
-
 void
-switch_fini (xlator_t *this)
+switch_fini(xlator_t *this)
 {
-        dht_conf_t           *conf = NULL;
-        struct switch_struct *trav = NULL;
-        struct switch_struct *prev = NULL;
-
-        conf = this->private;
-
-        if (conf) {
-                trav = (struct switch_struct *)conf->private;
-                conf->private = NULL;
-                while (trav) {
-                        GF_FREE (trav->array);
-                        prev = trav;
-                        trav = trav->next;
-                        GF_FREE (prev);
-                }
+    dht_conf_t *conf = NULL;
+    struct switch_struct *trav = NULL;
+    struct switch_struct *prev = NULL;
+
+    conf = this->private;
+
+    if (conf) {
+        trav = (struct switch_struct *)conf->private;
+        conf->private = NULL;
+        while (trav) {
+            GF_FREE(trav->array);
+            prev = trav;
+            trav = trav->next;
+            GF_FREE(prev);
         }
+    }
 
-        dht_fini(this);
+    dht_fini(this);
 }
 
 int
-set_switch_pattern (xlator_t *this, dht_conf_t *conf,
-                    const char *pattern_str)
+set_switch_pattern(xlator_t *this, dht_conf_t *conf, const char *pattern_str)
 {
-        int                         flag = 0;
-        int                         idx = 0;
-        int                         index = 0;
-        int                         child_count = 0;
-        char                       *tmp = NULL;
-        char                       *tmp1 = NULL;
-        char                       *child = NULL;
-        char                       *tmp_str = NULL;
-        char                       *tmp_str1 = NULL;
-        char                       *dup_str = NULL;
-        char                       *dup_childs = NULL;
-        char                       *switch_str = NULL;
-        char                       *pattern = NULL;
-        char                       *childs = NULL;
-        char                       *option_string = NULL;
-        struct switch_struct        *switch_buf = NULL;
-        struct switch_struct        *switch_opt = NULL;
-        struct switch_struct        *trav = NULL;
-        struct switch_sched_array  *switch_buf_array = NULL;
-        xlator_list_t              *trav_xl = NULL;
-
-        trav_xl = this->children;
-        while (trav_xl) {
-                index++;
-                trav_xl = trav_xl->next;
+    int flag = 0;
+    int idx = 0;
+    int index = 0;
+    int child_count = 0;
+    char *tmp = NULL;
+    char *tmp1 = NULL;
+    char *child = NULL;
+    char *tmp_str = NULL;
+    char *tmp_str1 = NULL;
+    char *dup_str = NULL;
+    char *dup_childs = NULL;
+    char *switch_str = NULL;
+    char *pattern = NULL;
+    char *childs = NULL;
+    char *option_string = NULL;
+    size_t pattern_length;
+    struct switch_struct *switch_buf = NULL;
+    struct switch_struct *switch_opt = NULL;
+    struct switch_struct *trav = NULL;
+    struct switch_sched_array *switch_buf_array = NULL;
+    xlator_list_t *trav_xl = NULL;
+
+    trav_xl = this->children;
+    while (trav_xl) {
+        index++;
+        trav_xl = trav_xl->next;
+    }
+    child_count = index;
+    switch_buf_array = GF_CALLOC((index + 1), sizeof(struct switch_sched_array),
+                                 gf_switch_mt_switch_sched_array);
+    if (!switch_buf_array)
+        goto err;
+
+    trav_xl = this->children;
+    index = 0;
+
+    while (trav_xl) {
+        switch_buf_array[index].xl = trav_xl->xlator;
+        switch_buf_array[index].eligible = 1;
+        trav_xl = trav_xl->next;
+        index++;
+    }
+
+    /*  *jpg:child1,child2;*mpg:child3;*:child4,child5,child6 */
+
+    /* Get the pattern for considering switch case.
+       "option block-size *avi:10MB" etc */
+    option_string = gf_strdup(pattern_str);
+    if (option_string == NULL) {
+        goto err;
+    }
+    switch_str = strtok_r(option_string, ";", &tmp_str);
+    while (switch_str) {
+        dup_str = gf_strdup(switch_str);
+        if (dup_str == NULL) {
+            goto err;
         }
-        child_count = index;
-        switch_buf_array = GF_CALLOC ((index + 1),
-                                      sizeof (struct switch_sched_array),
-                                      gf_switch_mt_switch_sched_array);
-        if (!switch_buf_array)
-                goto err;
-
-        trav_xl = this->children;
-        index = 0;
-
-        while (trav_xl) {
-                switch_buf_array[index].xl = trav_xl->xlator;
-                switch_buf_array[index].eligible = 1;
-                trav_xl = trav_xl->next;
-                index++;
+        switch_opt = GF_CALLOC(1, sizeof(struct switch_struct),
+                               gf_switch_mt_switch_struct);
+        if (!switch_opt) {
+            GF_FREE(dup_str);
+            goto err;
         }
 
-        /*  *jpg:child1,child2;*mpg:child3;*:child4,child5,child6 */
-
-        /* Get the pattern for considering switch case.
-           "option block-size *avi:10MB" etc */
-        option_string = gf_strdup (pattern_str);
-        switch_str = strtok_r (option_string, ";", &tmp_str);
-        while (switch_str) {
-                dup_str = gf_strdup (switch_str);
-                switch_opt = GF_CALLOC (1, sizeof (struct switch_struct),
-                                        gf_switch_mt_switch_struct);
-                if (!switch_opt) {
-                        GF_FREE (dup_str);
-                        goto err;
-                }
+        pattern = strtok_r(dup_str, ":", &tmp_str1);
+        childs = strtok_r(NULL, ":", &tmp_str1);
+        if (strncmp(pattern, "*", 2) == 0) {
+            gf_msg("switch", GF_LOG_INFO, 0, DHT_MSG_SWITCH_PATTERN_INFO,
+                   "'*' pattern will be taken by default "
+                   "for all the unconfigured child nodes,"
+                   " hence neglecting current option");
+            switch_str = strtok_r(NULL, ";", &tmp_str);
+            GF_FREE(switch_opt);
+            switch_opt = NULL;
+            GF_FREE(dup_str);
+            continue;
+        }
+        GF_FREE(dup_str);
+
+        pattern_length = strlen(pattern);
+        if (pattern_length >= (sizeof(switch_opt->path_pattern))) {
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   DHT_MSG_SET_SWITCH_PATTERN_ERROR, "Pattern (%s) too long",
+                   pattern);
+            goto err;
+        }
+        memcpy(switch_opt->path_pattern, pattern, pattern_length);
+        switch_opt->path_pattern[pattern_length] = '\0';
 
-                pattern = strtok_r (dup_str, ":", &tmp_str1);
-                childs = strtok_r (NULL, ":", &tmp_str1);
-                if (strncmp (pattern, "*", 2) == 0) {
-                        gf_log ("switch", GF_LOG_INFO,
-                                "'*' pattern will be taken by default "
-                                "for all the unconfigured child nodes,"
-                                " hence neglecting current option");
-                        switch_str = strtok_r (NULL, ";", &tmp_str);
-                        GF_FREE (switch_opt);
-                        GF_FREE (dup_str);
-                        continue;
-                }
-                memcpy (switch_opt->path_pattern, pattern, strlen (pattern));
-                if (childs) {
-                        dup_childs = gf_strdup (childs);
-                        child = strtok_r (dup_childs, ",", &tmp);
-                        while (child) {
-                                if (gf_switch_valid_child (this, child)) {
-                                        idx++;
-                                        child = strtok_r (NULL, ",", &tmp);
-                                } else {
-                                        gf_log (this->name, GF_LOG_ERROR,
-                                                "%s is not a subvolume of %s. "
-                                                "pattern can only be scheduled "
-                                                "only to a subvolume of %s",
-                                                child, this->name, this->name);
-                                        goto err;
-                                }
-                        }
-                        GF_FREE (dup_childs);
-                        child = strtok_r (childs, ",", &tmp1);
-                        switch_opt->num_child = idx;
-                        switch_opt->array = GF_CALLOC (1, (idx *
-                                                           sizeof (struct switch_sched_array)),
-                                                       gf_switch_mt_switch_sched_array);
-                        if (!switch_opt->array)
-                                goto err;
-                        idx = 0;
-                        while (child) {
-                                for (index = 0; index < child_count; index++) {
-                                        if (strcmp (switch_buf_array[index].xl->name,
-                                                    child) == 0) {
-                                                gf_log ("switch", GF_LOG_DEBUG,
-                                                        "'%s' pattern will be "
-                                                        "scheduled to \"%s\"",
-                                                        switch_opt->path_pattern, child);
-                                                /*
-                                                  if (switch_buf_array[index-1].considered) {
-                                                  gf_log ("switch", GF_LOG_DEBUG,
-                                                  "ambiguity found, exiting");
-                                                  return -1;
-                                                  }
-                                                */
-                                                switch_opt->array[idx].xl = switch_buf_array[index].xl;
-                                                switch_buf_array[index].considered = 1;
-                                                idx++;
-                                                break;
-                                        }
-                                }
-                                child = strtok_r (NULL, ",", &tmp1);
-                        }
+        if (childs) {
+            dup_childs = gf_strdup(childs);
+            if (dup_childs == NULL) {
+                goto err;
+            }
+            child = strtok_r(dup_childs, ",", &tmp);
+            while (child) {
+                if (gf_switch_valid_child(this, child)) {
+                    idx++;
+                    child = strtok_r(NULL, ",", &tmp);
                 } else {
-                        /* error */
-                        gf_log ("switch", GF_LOG_ERROR,
-                                "Check \"scheduler.switch.case\" "
-                                "option in unify volume. Exiting");
-                        goto err;
+                    gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SUBVOL_ERROR,
+                           "%s is not a subvolume of %s. "
+                           "pattern can only be scheduled "
+                           "only to a subvolume of %s",
+                           child, this->name, this->name);
+                    GF_FREE(dup_childs);
+                    goto err;
                 }
-                GF_FREE (dup_str);
-
-                /* Link it to the main structure */
-                if (switch_buf) {
-                        /* there are already few entries */
-                        trav = switch_buf;
-                        while (trav->next)
-                                trav = trav->next;
-                        trav->next = switch_opt;
-                } else {
-                        /* First entry */
-                        switch_buf = switch_opt;
+            }
+            GF_FREE(dup_childs);
+            child = strtok_r(childs, ",", &tmp1);
+            switch_opt->num_child = idx;
+            switch_opt->array = GF_CALLOC(
+                1, (idx * sizeof(struct switch_sched_array)),
+                gf_switch_mt_switch_sched_array);
+            if (!switch_opt->array)
+                goto err;
+            idx = 0;
+            while (child) {
+                for (index = 0; index < child_count; index++) {
+                    if (strcmp(switch_buf_array[index].xl->name, child) == 0) {
+                        gf_msg_debug("switch", 0,
+                                     "'%s' pattern will be "
+                                     "scheduled to \"%s\"",
+                                     switch_opt->path_pattern, child);
+                        /*
+                          if (switch_buf_array[index-1].considered) {
+                          gf_msg_debug ("switch", 0,
+                          "ambiguity found, exiting");
+                          return -1;
+                          }
+                        */
+                        switch_opt->array[idx].xl = switch_buf_array[index].xl;
+                        switch_buf_array[index].considered = 1;
+                        idx++;
+                        break;
+                    }
                 }
-                switch_opt = NULL;
-                switch_str = strtok_r (NULL, ";", &tmp_str);
+                child = strtok_r(NULL, ",", &tmp1);
+            }
+        } else {
+            /* error */
+            gf_msg("switch", GF_LOG_ERROR, 0, DHT_MSG_SET_SWITCH_PATTERN_ERROR,
+                   "Check \"scheduler.switch.case\" "
+                   "option in unify volume. Exiting");
+            goto err;
         }
 
-        /* Now, all the pattern based considerations done, so for all the
-         * remaining pattern, '*' to all the remaining child nodes
-         */
-        {
-                for (index=0; index < child_count; index++) {
-                        /* check for considered flag */
-                        if (switch_buf_array[index].considered)
-                                continue;
-                        flag++;
-                }
-                if (!flag) {
-                        gf_log ("switch", GF_LOG_ERROR,
-                                "No nodes left for pattern '*'. Exiting");
-                        goto err;
-                }
-                switch_opt = GF_CALLOC (1, sizeof (struct switch_struct),
-                                        gf_switch_mt_switch_struct);
-                if (!switch_opt)
-                        goto err;
-
-                /* Add the '*' pattern to the array */
-                memcpy (switch_opt->path_pattern, "*", 2);
-                switch_opt->num_child = flag;
-                switch_opt->array =
-                        GF_CALLOC (1,
-                                   flag * sizeof (struct switch_sched_array),
-                                   gf_switch_mt_switch_sched_array);
-                if (!switch_opt->array)
-                        goto err;
-                flag = 0;
-                for (index=0; index < child_count; index++) {
-                        /* check for considered flag */
-                        if (switch_buf_array[index].considered)
-                                continue;
-                        gf_log ("switch", GF_LOG_DEBUG,
-                                "'%s' pattern will be scheduled to \"%s\"",
-                                switch_opt->path_pattern,
-                                switch_buf_array[index].xl->name);
-                        switch_opt->array[flag].xl =
-                                switch_buf_array[index].xl;
-                        switch_buf_array[index].considered = 1;
-                        flag++;
-                }
-                if (switch_buf) {
-                        /* there are already few entries */
-                        trav = switch_buf;
-                        while (trav->next)
-                                trav = trav->next;
-                        trav->next = switch_opt;
-                } else {
-                        /* First entry */
-                        switch_buf = switch_opt;
-                }
-                switch_opt = NULL;
+        /* Link it to the main structure */
+        if (switch_buf) {
+            /* there are already few entries */
+            trav = switch_buf;
+            while (trav->next)
+                trav = trav->next;
+            trav->next = switch_opt;
+        } else {
+            /* First entry */
+            switch_buf = switch_opt;
+        }
+        switch_opt = NULL;
+        switch_str = strtok_r(NULL, ";", &tmp_str);
+    }
+
+    /* Now, all the pattern based considerations done, so for all the
+     * remaining pattern, '*' to all the remaining child nodes
+     */
+    {
+        for (index = 0; index < child_count; index++) {
+            /* check for considered flag */
+            if (switch_buf_array[index].considered)
+                continue;
+            flag++;
+        }
+        if (!flag) {
+            gf_msg("switch", GF_LOG_ERROR, 0, DHT_MSG_SET_SWITCH_PATTERN_ERROR,
+                   "No nodes left for pattern '*'. Exiting");
+            goto err;
+        }
+        switch_opt = GF_CALLOC(1, sizeof(struct switch_struct),
+                               gf_switch_mt_switch_struct);
+        if (!switch_opt)
+            goto err;
+
+        /* Add the '*' pattern to the array */
+        memcpy(switch_opt->path_pattern, "*", 2);
+        switch_opt->num_child = flag;
+        switch_opt->array = GF_CALLOC(1,
+                                      flag * sizeof(struct switch_sched_array),
+                                      gf_switch_mt_switch_sched_array);
+        if (!switch_opt->array)
+            goto err;
+        flag = 0;
+        for (index = 0; index < child_count; index++) {
+            /* check for considered flag */
+            if (switch_buf_array[index].considered)
+                continue;
+            gf_msg_debug("switch", 0,
+                         "'%s'"
+                         " pattern will be scheduled to \"%s\"",
+                         switch_opt->path_pattern,
+                         switch_buf_array[index].xl->name);
+
+            switch_opt->array[flag].xl = switch_buf_array[index].xl;
+            switch_buf_array[index].considered = 1;
+            flag++;
+        }
+        if (switch_buf) {
+            /* there are already few entries */
+            trav = switch_buf;
+            while (trav->next)
+                trav = trav->next;
+            trav->next = switch_opt;
+        } else {
+            /* First entry */
+            switch_buf = switch_opt;
         }
-        /* */
-        conf->private = switch_buf;
+        switch_opt = NULL;
+    }
+    /* */
+    conf->private = switch_buf;
 
-        return 0;
+    GF_FREE(option_string);
+    return 0;
 err:
-        GF_FREE (switch_buf_array);
-        GF_FREE (switch_opt);
+    GF_FREE(switch_buf_array);
+    GF_FREE(switch_opt);
+    GF_FREE(option_string);
 
-        if (switch_buf) {
-                trav = switch_buf;
-                while (trav) {
-                        GF_FREE (trav->array);
-                        switch_opt = trav;
-                        trav = trav->next;
-                        GF_FREE (switch_opt);
-                }
+    if (switch_buf) {
+        trav = switch_buf;
+        while (trav) {
+            GF_FREE(trav->array);
+            switch_opt = trav;
+            trav = trav->next;
+            GF_FREE(switch_opt);
         }
-        return -1;
+    }
+    return -1;
 }
 
-
 int32_t
-switch_init (xlator_t *this)
+switch_init(xlator_t *this)
 {
-        dht_conf_t            *conf = NULL;
-        data_t                *data = NULL;
-        int                    ret = -1;
+    dht_conf_t *conf = NULL;
+    data_t *data = NULL;
+    int ret = -1;
+
+    ret = dht_init(this);
+    if (ret) {
+        return ret;
+    }
+    conf = this->private;
 
-        ret = dht_init(this);
+    data = dict_get(this->options, "pattern.switch.case");
+    if (data) {
+        /* TODO: */
+        ret = set_switch_pattern(this, conf, data->data);
         if (ret) {
-                return ret;
-        }
-        conf = this->private;
-
-        data = dict_get (this->options, "pattern.switch.case");
-        if (data) {
-                /* TODO: */
-                ret = set_switch_pattern (this, conf, data->data);
-                if (ret) {
-                        goto err;
-                }
+            goto err;
         }
+    }
 
-        this->private = conf;
-        return 0;
+    this->private = conf;
+    return 0;
 
 err:
-        dht_fini(this);
-        return -1;
+    dht_fini(this);
+    return -1;
 }
 
-
-class_methods_t class_methods = {
-        .init           = switch_init,
-        .fini           = switch_fini,
-        .reconfigure    = dht_reconfigure,
-        .notify         = dht_notify
-};
-
-
 struct xlator_fops fops = {
-        .lookup      = switch_lookup,
-        .create      = switch_create,
-        .mknod       = switch_mknod,
-
-        .stat        = dht_stat,
-        .fstat       = dht_fstat,
-        .truncate    = dht_truncate,
-        .ftruncate   = dht_ftruncate,
-        .access      = dht_access,
-        .readlink    = dht_readlink,
-        .setxattr    = dht_setxattr,
-        .getxattr    = dht_getxattr,
-        .removexattr = dht_removexattr,
-        .open        = dht_open,
-        .readv       = dht_readv,
-        .writev      = dht_writev,
-        .flush       = dht_flush,
-        .fsync       = dht_fsync,
-        .statfs      = dht_statfs,
-        .lk          = dht_lk,
-        .opendir     = dht_opendir,
-        .readdir     = dht_readdir,
-        .readdirp    = dht_readdirp,
-        .fsyncdir    = dht_fsyncdir,
-        .symlink     = dht_symlink,
-        .unlink      = dht_unlink,
-        .link        = dht_link,
-        .mkdir       = dht_mkdir,
-        .rmdir       = dht_rmdir,
-        .rename      = dht_rename,
-        .inodelk     = dht_inodelk,
-        .finodelk    = dht_finodelk,
-        .entrylk     = dht_entrylk,
-        .fentrylk    = dht_fentrylk,
-        .xattrop     = dht_xattrop,
-        .fxattrop    = dht_fxattrop,
-        .setattr     = dht_setattr,
+    .lookup = switch_lookup,
+    .create = switch_create,
+    .mknod = switch_mknod,
+
+    .stat = dht_stat,
+    .fstat = dht_fstat,
+    .truncate = dht_truncate,
+    .ftruncate = dht_ftruncate,
+    .access = dht_access,
+    .readlink = dht_readlink,
+    .setxattr = dht_setxattr,
+    .getxattr = dht_getxattr,
+    .removexattr = dht_removexattr,
+    .open = dht_open,
+    .readv = dht_readv,
+    .writev = dht_writev,
+    .flush = dht_flush,
+    .fsync = dht_fsync,
+    .statfs = dht_statfs,
+    .lk = dht_lk,
+    .opendir = dht_opendir,
+    .readdir = dht_readdir,
+    .readdirp = dht_readdirp,
+    .fsyncdir = dht_fsyncdir,
+    .symlink = dht_symlink,
+    .unlink = dht_unlink,
+    .link = dht_link,
+    .mkdir = dht_mkdir,
+    .rmdir = dht_rmdir,
+    .rename = dht_rename,
+    .inodelk = dht_inodelk,
+    .finodelk = dht_finodelk,
+    .entrylk = dht_entrylk,
+    .fentrylk = dht_fentrylk,
+    .xattrop = dht_xattrop,
+    .fxattrop = dht_fxattrop,
+    .setattr = dht_setattr,
 };
 
-
-struct xlator_cbks cbks = {
-        .forget     = dht_forget
+struct xlator_cbks cbks = {.forget = dht_forget};
+extern int32_t
+mem_acct_init(xlator_t *this);
+
+xlator_api_t xlator_api = {
+    .init = switch_init,
+    .fini = switch_fini,
+    .notify = dht_notify,
+    .reconfigure = dht_reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = dht_options,
+    .identifier = "switch",
+    .category = GF_TECH_PREVIEW,
 };
diff --git a/xlators/cluster/dht/src/unittest/dht_layout_mock.c b/xlators/cluster/dht/src/unittest/dht_layout_mock.c
new file mode 100644
index 00000000000..771452963d1
--- /dev/null
+++ b/xlators/cluster/dht/src/unittest/dht_layout_mock.c
@@ -0,0 +1,73 @@
+/*
+  Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/xlator.h>
+#include "dht-common.h"
+#include <glusterfs/byte-order.h>
+
+int
+dht_hash_compute(xlator_t *this, int type, const char *name, uint32_t *hash_p)
+{
+    return 0;
+}
+
+int
+dht_inode_ctx_layout_get(inode_t *inode, xlator_t *this, dht_layout_t **layout)
+{
+    return 0;
+}
+
+int
+dht_inode_ctx_layout_set(inode_t *inode, xlator_t *this,
+                         dht_layout_t *layout_int)
+{
+    return 0;
+}
+
+int
+dict_get_ptr(dict_t *this, char *key, void **ptr)
+{
+    return 0;
+}
+
+int
+dict_get_ptr_and_len(dict_t *this, char *key, void **ptr, int *len)
+{
+    return 0;
+}
+
+int
+_gf_log(const char *domain, const char *file, const char *function,
+        int32_t line, gf_loglevel_t level, const char *fmt, ...)
+{
+    return 0;
+}
+
+int
+_gf_log_callingfn(const char *domain, const char *file, const char *function,
+                  int32_t line, gf_loglevel_t level, const char *fmt, ...)
+{
+    return 0;
+}
+
+void
+gf_uuid_unparse(const uuid_t uu, char *out)
+{
+    // could call a will-return function here
+    // to place the correct data in *out
+}
+
+int
+_gf_msg(const char *domain, const char *file, const char *function,
+        int32_t line, gf_loglevel_t level, int errnum, int trace,
+        uint64_t msgid, const char *fmt, ...)
+{
+    return 0;
+}
diff --git a/xlators/cluster/dht/src/unittest/dht_layout_unittest.c b/xlators/cluster/dht/src/unittest/dht_layout_unittest.c
new file mode 100644
index 00000000000..c94a1d0a2e1
--- /dev/null
+++ b/xlators/cluster/dht/src/unittest/dht_layout_unittest.c
@@ -0,0 +1,127 @@
+/*
+  Copyright (c) 2008-2014 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "dht-common.h"
+#include <glusterfs/logging.h>
+#include <glusterfs/xlator.h>
+
+#include <inttypes.h>
+#include <stdarg.h>
+#include <stddef.h>
+#include <setjmp.h>
+#include <cmocka_pbc.h>
+#include <cmocka.h>
+
+/*
+ * Helper functions
+ */
+
+static xlator_t *
+helper_xlator_init(uint32_t num_types)
+{
+    xlator_t *xl;
+    int i, ret;
+
+    REQUIRE(num_types > 0);
+
+    xl = test_calloc(1, sizeof(xlator_t));
+    assert_non_null(xl);
+    xl->mem_acct->num_types = num_types;
+    xl->mem_acct = test_calloc(sizeof(struct mem_acct) +
+                               sizeof(struct mem_acct_rec) + num_types);
+    assert_non_null(xl->mem_acct);
+
+    xl->ctx = test_calloc(1, sizeof(glusterfs_ctx_t));
+    assert_non_null(xl->ctx);
+
+    for (i = 0; i < num_types; i++) {
+        ret = LOCK_INIT(&(xl->mem_acct.rec[i].lock));
+        assert_false(ret);
+    }
+
+    ENSURE(num_types == xl->mem_acct.num_types);
+    ENSURE(NULL != xl);
+
+    return xl;
+}
+
+static int
+helper_xlator_destroy(xlator_t *xl)
+{
+    int i, ret;
+
+    for (i = 0; i < xl->mem_acct.num_types; i++) {
+        ret = LOCK_DESTROY(&(xl->mem_acct.rec[i].lock));
+        assert_int_equal(ret, 0);
+    }
+
+    free(xl->mem_acct.rec);
+    free(xl->ctx);
+    free(xl);
+    return 0;
+}
+
+/*
+ * Unit tests
+ */
+static void
+test_dht_layout_new(void **state)
+{
+    xlator_t *xl;
+    dht_layout_t *layout;
+    dht_conf_t *conf;
+    int cnt;
+
+    expect_assert_failure(dht_layout_new(NULL, 0));
+    expect_assert_failure(dht_layout_new((xlator_t *)0x12345, -1));
+    xl = helper_xlator_init(10);
+
+    // xl->private is NULL
+    assert_null(xl->private);
+    cnt = 100;
+    layout = dht_layout_new(xl, cnt);
+    assert_non_null(layout);
+    assert_int_equal(layout->type, DHT_HASH_TYPE_DM);
+    assert_int_equal(layout->cnt, cnt);
+    assert_int_equal(GF_ATOMIC_GET(layout->ref), 1);
+    assert_int_equal(layout->gen, 0);
+    assert_int_equal(layout->spread_cnt, 0);
+    free(layout);
+
+    // xl->private is not NULL
+    cnt = 110;
+    conf = (dht_conf_t *)test_calloc(1, sizeof(dht_conf_t));
+    assert_non_null(conf);
+    conf->dir_spread_cnt = 12345;
+    conf->gen = -123;
+    xl->private = conf;
+
+    layout = dht_layout_new(xl, cnt);
+    assert_non_null(layout);
+    assert_int_equal(layout->type, DHT_HASH_TYPE_DM);
+    assert_int_equal(layout->cnt, cnt);
+    assert_int_equal(GF_ATOMIC_GET(layout->ref), 1);
+    assert_int_equal(layout->gen, conf->gen);
+    assert_int_equal(layout->spread_cnt, conf->dir_spread_cnt);
+    free(layout);
+
+    free(conf);
+    helper_xlator_destroy(xl);
+}
+
+int
+main(void)
+{
+    const struct CMUnitTest xlator_dht_layout_tests[] = {
+        unit_test(test_dht_layout_new),
+    };
+
+    return cmocka_run_group_tests(xlator_dht_layout_tests, NULL, NULL);
+}
diff --git a/xlators/cluster/ha/Makefile.am b/xlators/cluster/ec/Makefile.am
index d471a3f9243..d471a3f9243 100644
--- a/xlators/cluster/ha/Makefile.am
+++ b/xlators/cluster/ec/Makefile.am
diff --git a/xlators/cluster/ec/src/Makefile.am b/xlators/cluster/ec/src/Makefile.am
new file mode 100644
index 00000000000..406a636bbc2
--- /dev/null
+++ b/xlators/cluster/ec/src/Makefile.am
@@ -0,0 +1,83 @@
+xlator_LTLIBRARIES = ec.la
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/cluster
+
+ec_sources := ec.c
+ec_sources += ec-data.c
+ec_sources += ec-helpers.c
+ec_sources += ec-common.c
+ec_sources += ec-generic.c
+ec_sources += ec-locks.c
+ec_sources += ec-dir-read.c
+ec_sources += ec-dir-write.c
+ec_sources += ec-inode-read.c
+ec_sources += ec-inode-write.c
+ec_sources += ec-combine.c
+ec_sources += ec-method.c
+ec_sources += ec-galois.c
+ec_sources += ec-code.c
+ec_sources += ec-code-c.c
+ec_sources += ec-gf8.c
+ec_sources += ec-heal.c
+ec_sources += ec-heald.c
+
+ec_headers := ec.h
+ec_headers += ec-mem-types.h
+ec_headers += ec-helpers.h
+ec_headers += ec-data.h
+ec_headers += ec-fops.h
+ec_headers += ec-common.h
+ec_headers += ec-combine.h
+ec_headers += ec-method.h
+ec_headers += ec-galois.h
+ec_headers += ec-code.h
+ec_headers += ec-code-c.h
+ec_headers += ec-gf8.h
+ec_headers += ec-heald.h
+ec_headers += ec-messages.h
+ec_headers += ec-types.h
+
+if ENABLE_EC_DYNAMIC_INTEL
+  ec_sources += ec-code-intel.c
+  ec_headers += ec-code-intel.h
+endif
+
+if ENABLE_EC_DYNAMIC_X64
+  ec_sources += ec-code-x64.c
+  ec_headers += ec-code-x64.h
+endif
+
+if ENABLE_EC_DYNAMIC_SSE
+  ec_sources += ec-code-sse.c
+  ec_headers += ec-code-sse.h
+endif
+
+if ENABLE_EC_DYNAMIC_AVX
+  ec_sources += ec-code-avx.c
+  ec_headers += ec-code-avx.h
+endif
+
+ec_ext_sources = $(top_builddir)/xlators/lib/src/libxlator.c
+
+ec_ext_headers = $(top_builddir)/xlators/lib/src/libxlator.h
+
+ec_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+ec_la_SOURCES = $(ec_sources) $(ec_headers) $(ec_ext_sources) $(ec_ext_headers)
+ec_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+AM_CPPFLAGS = $(GF_CPPFLAGS)
+AM_CPPFLAGS += -I$(top_srcdir)/libglusterfs/src
+AM_CPPFLAGS += -I$(top_srcdir)/xlators/lib/src
+AM_CPPFLAGS += -I$(top_srcdir)/rpc/rpc-lib/src
+AM_CPPFLAGS += -I$(top_srcdir)/rpc/xdr/src
+AM_CPPFLAGS += -I$(top_builddir)/rpc/xdr/src
+AM_CPPFLAGS += -DGLUSTERFS_LIBEXECDIR=\"$(GLUSTERFS_LIBEXECDIR)\"
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
+
+install-data-hook:
+	ln -sf ec.so $(DESTDIR)$(xlatordir)/disperse.so
+
+uninstall-local:
+	rm -f $(DESTDIR)$(xlatordir)/disperse.so
diff --git a/xlators/cluster/ec/src/ec-code-avx.c b/xlators/cluster/ec/src/ec-code-avx.c
new file mode 100644
index 00000000000..70afaa00f54
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code-avx.c
@@ -0,0 +1,109 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <errno.h>
+
+#include "ec-code-intel.h"
+
+static void
+ec_code_avx_prolog(ec_code_builder_t *builder)
+{
+    builder->loop = builder->address;
+}
+
+static void
+ec_code_avx_epilog(ec_code_builder_t *builder)
+{
+    ec_code_intel_op_add_i2r(builder, 32, REG_DX);
+    ec_code_intel_op_add_i2r(builder, 32, REG_DI);
+    ec_code_intel_op_test_i2r(builder, builder->width - 1, REG_DX);
+    ec_code_intel_op_jne(builder, builder->loop);
+
+    ec_code_intel_op_ret(builder, 0);
+}
+
+static void
+ec_code_avx_load(ec_code_builder_t *builder, uint32_t dst, uint32_t idx,
+                 uint32_t bit)
+{
+    if (builder->linear) {
+        ec_code_intel_op_mov_m2avx(
+            builder, REG_SI, REG_DX, 1,
+            idx * builder->width * builder->bits + bit * builder->width, dst);
+    } else {
+        if (builder->base != idx) {
+            ec_code_intel_op_mov_m2r(builder, REG_SI, REG_NULL, 0, idx * 8,
+                                     REG_AX);
+            builder->base = idx;
+        }
+        ec_code_intel_op_mov_m2avx(builder, REG_AX, REG_DX, 1,
+                                   bit * builder->width, dst);
+    }
+}
+
+static void
+ec_code_avx_store(ec_code_builder_t *builder, uint32_t src, uint32_t bit)
+{
+    ec_code_intel_op_mov_avx2m(builder, src, REG_DI, REG_NULL, 0,
+                               bit * builder->width);
+}
+
+static void
+ec_code_avx_copy(ec_code_builder_t *builder, uint32_t dst, uint32_t src)
+{
+    ec_code_intel_op_mov_avx2avx(builder, src, dst);
+}
+
+static void
+ec_code_avx_xor2(ec_code_builder_t *builder, uint32_t dst, uint32_t src)
+{
+    ec_code_intel_op_xor_avx2avx(builder, src, dst);
+}
+
+static void
+ec_code_avx_xor3(ec_code_builder_t *builder, uint32_t dst, uint32_t src1,
+                 uint32_t src2)
+{
+    ec_code_intel_op_mov_avx2avx(builder, src1, dst);
+    ec_code_intel_op_xor_avx2avx(builder, src2, dst);
+}
+
+static void
+ec_code_avx_xorm(ec_code_builder_t *builder, uint32_t dst, uint32_t idx,
+                 uint32_t bit)
+{
+    if (builder->linear) {
+        ec_code_intel_op_xor_m2avx(
+            builder, REG_SI, REG_DX, 1,
+            idx * builder->width * builder->bits + bit * builder->width, dst);
+    } else {
+        if (builder->base != idx) {
+            ec_code_intel_op_mov_m2r(builder, REG_SI, REG_NULL, 0, idx * 8,
+                                     REG_AX);
+            builder->base = idx;
+        }
+        ec_code_intel_op_xor_m2avx(builder, REG_AX, REG_DX, 1,
+                                   bit * builder->width, dst);
+    }
+}
+
+static char *ec_code_avx_needed_flags[] = {"avx2", NULL};
+
+ec_code_gen_t ec_code_gen_avx = {.name = "avx",
+                                 .flags = ec_code_avx_needed_flags,
+                                 .width = 32,
+                                 .prolog = ec_code_avx_prolog,
+                                 .epilog = ec_code_avx_epilog,
+                                 .load = ec_code_avx_load,
+                                 .store = ec_code_avx_store,
+                                 .copy = ec_code_avx_copy,
+                                 .xor2 = ec_code_avx_xor2,
+                                 .xor3 = ec_code_avx_xor3,
+                                 .xorm = ec_code_avx_xorm};
diff --git a/xlators/cluster/ec/src/ec-code-avx.h b/xlators/cluster/ec/src/ec-code-avx.h
new file mode 100644
index 00000000000..fdca4ad2c8f
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code-avx.h
@@ -0,0 +1,18 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_CODE_AVX_H__
+#define __EC_CODE_AVX_H__
+
+#include "ec-code.h"
+
+extern ec_code_gen_t ec_code_gen_avx;
+
+#endif /* __EC_CODE_AVX_H__ */
diff --git a/xlators/cluster/ec/src/ec-code-c.c b/xlators/cluster/ec/src/ec-code-c.c
new file mode 100644
index 00000000000..acdc665c2cf
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code-c.c
@@ -0,0 +1,11679 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <inttypes.h>
+#include <string.h>
+
+#include "ec-method.h"
+#include "ec-code-c.h"
+
+#define WIDTH (EC_METHOD_WORD_SIZE / sizeof(uint64_t))
+
+static void
+gf8_muladd_00(void *out, void *in)
+{
+    memcpy(out, in, EC_METHOD_WORD_SIZE * 8);
+}
+
+static void
+gf8_muladd_01(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        out_ptr[0] ^= in_ptr[0];
+        out_ptr[WIDTH] ^= in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] ^= in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] ^= in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] ^= in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] ^= in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] ^= in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] ^= in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_02(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in7;
+        out1 = in0;
+        out7 = in6;
+        out5 = in4;
+        out6 = in5;
+        out3 = in2 ^ in7;
+        out4 = in3 ^ in7;
+        out2 = in1 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_03(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in0 ^ in7;
+        tmp0 = in2 ^ in7;
+        out1 = in0 ^ in1;
+        out7 = in6 ^ in7;
+        out5 = in4 ^ in5;
+        out6 = in5 ^ in6;
+        out4 = in3 ^ in4 ^ in7;
+        out2 = tmp0 ^ in1;
+        out3 = tmp0 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_04(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in6;
+        out1 = in7;
+        out7 = in5;
+        out6 = in4;
+        tmp0 = in6 ^ in7;
+        out2 = in0 ^ in6;
+        out5 = in3 ^ in7;
+        out3 = tmp0 ^ in1;
+        out4 = tmp0 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_05(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in0 ^ in6;
+        out1 = in1 ^ in7;
+        out7 = in5 ^ in7;
+        out6 = in4 ^ in6;
+        out2 = out0 ^ in2;
+        out3 = out1 ^ in3 ^ in6;
+        out5 = out7 ^ in3;
+        out4 = out6 ^ in2 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_06(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in6 ^ in7;
+        tmp0 = in1 ^ in6;
+        out1 = in0 ^ in7;
+        out7 = in5 ^ in6;
+        out6 = in4 ^ in5;
+        out4 = in2 ^ in3 ^ in6;
+        out5 = in3 ^ in4 ^ in7;
+        out3 = tmp0 ^ in2;
+        out2 = tmp0 ^ out1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_07(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in6;
+        tmp1 = in5 ^ in6;
+        tmp2 = in0 ^ in7;
+        tmp3 = tmp0 ^ in3;
+        out6 = tmp1 ^ in4;
+        out7 = tmp1 ^ in7;
+        out0 = tmp2 ^ in6;
+        out1 = tmp2 ^ in1;
+        out3 = tmp3 ^ in1;
+        out4 = tmp3 ^ in4;
+        out5 = out4 ^ out7 ^ in2;
+        out2 = tmp0 ^ out1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_08(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in5;
+        out1 = in6;
+        out7 = in4;
+        out6 = in3 ^ in7;
+        out3 = in0 ^ in5 ^ in6;
+        out5 = in2 ^ in6 ^ in7;
+        out2 = in5 ^ in7;
+        out4 = out2 ^ in1 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_09(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in0 ^ in5;
+        tmp0 = in3 ^ in6;
+        out1 = in1 ^ in6;
+        out7 = in4 ^ in7;
+        out2 = in2 ^ in5 ^ in7;
+        out3 = tmp0 ^ out0;
+        out6 = tmp0 ^ in7;
+        out4 = out1 ^ out7 ^ in5;
+        out5 = out2 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_0A(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in5 ^ in7;
+        out1 = in0 ^ in6;
+        out7 = in4 ^ in6;
+        out2 = in1 ^ in5;
+        out6 = out0 ^ in3;
+        out3 = out0 ^ out1 ^ in2;
+        out5 = out7 ^ in2 ^ in7;
+        out4 = out2 ^ in3 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_0B(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in5;
+        tmp1 = in0 ^ in6;
+        tmp2 = in4 ^ in7;
+        out0 = in0 ^ in5 ^ in7;
+        out2 = tmp0 ^ in1;
+        out1 = tmp1 ^ in1;
+        out6 = tmp1 ^ out0 ^ in3;
+        out7 = tmp2 ^ in6;
+        out4 = tmp2 ^ out6 ^ in1;
+        out3 = out6 ^ in0 ^ in2;
+        out5 = tmp0 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_0C(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in5 ^ in6;
+        out1 = in6 ^ in7;
+        out7 = in4 ^ in5;
+        tmp0 = in1 ^ in5;
+        tmp1 = in0 ^ in7;
+        out5 = in2 ^ in3 ^ in6;
+        out6 = in3 ^ in4 ^ in7;
+        out2 = tmp1 ^ out0;
+        out4 = tmp0 ^ in2;
+        out3 = tmp0 ^ tmp1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_0D(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in4 ^ in5;
+        tmp1 = in5 ^ in6;
+        out1 = in1 ^ in6 ^ in7;
+        out7 = tmp0 ^ in7;
+        out4 = tmp0 ^ in1 ^ in2;
+        out0 = tmp1 ^ in0;
+        tmp2 = tmp1 ^ in3;
+        out6 = tmp2 ^ out7;
+        out2 = out0 ^ in2 ^ in7;
+        out3 = out0 ^ out1 ^ in3;
+        out5 = tmp2 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_0E(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in1;
+        tmp1 = in2 ^ in5;
+        tmp2 = in5 ^ in6;
+        out1 = in0 ^ in6 ^ in7;
+        out3 = tmp0 ^ tmp1;
+        out2 = tmp0 ^ tmp2;
+        tmp3 = tmp1 ^ in3;
+        out7 = tmp2 ^ in4;
+        out0 = tmp2 ^ in7;
+        out4 = tmp3 ^ in1 ^ in7;
+        out5 = tmp3 ^ out7;
+        out6 = out0 ^ out5 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_0F(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in6 ^ in7;
+        tmp1 = tmp0 ^ in1;
+        tmp2 = tmp0 ^ in5;
+        out1 = tmp1 ^ in0;
+        out7 = tmp2 ^ in4;
+        out0 = tmp2 ^ in0;
+        out6 = out7 ^ in3;
+        out5 = out6 ^ in2 ^ in7;
+        tmp3 = tmp1 ^ out0 ^ in2;
+        out4 = tmp1 ^ out5;
+        out2 = tmp3 ^ in6;
+        out3 = tmp3 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_10(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in4;
+        out1 = in5;
+        out7 = in3 ^ in7;
+        tmp0 = in6 ^ in7;
+        out2 = in4 ^ in6;
+        tmp1 = out2 ^ in5;
+        out6 = tmp0 ^ in2;
+        out3 = tmp0 ^ tmp1;
+        out5 = out2 ^ out3 ^ in1;
+        out4 = tmp1 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_11(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out7 = in3;
+        out0 = in0 ^ in4;
+        out1 = in1 ^ in5;
+        out6 = in2 ^ in7;
+        out4 = in0 ^ in5 ^ in6;
+        out5 = in1 ^ in6 ^ in7;
+        out2 = in2 ^ in4 ^ in6;
+        out3 = in3 ^ in4 ^ in5 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_12(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in4 ^ in7;
+        out1 = in0 ^ in5;
+        out3 = in2 ^ in4 ^ in5;
+        tmp0 = out0 ^ in6;
+        out2 = tmp0 ^ in1;
+        tmp1 = tmp0 ^ in3;
+        out6 = tmp0 ^ out3;
+        out5 = out2 ^ in5;
+        out7 = tmp1 ^ in4;
+        out4 = tmp1 ^ out1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_13(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out7 = in3 ^ in6;
+        tmp0 = in0 ^ in5;
+        tmp1 = in4 ^ in7;
+        out6 = in2 ^ in5 ^ in7;
+        out4 = tmp0 ^ out7 ^ in7;
+        out1 = tmp0 ^ in1;
+        out0 = tmp1 ^ in0;
+        out5 = tmp1 ^ in1 ^ in6;
+        out3 = tmp1 ^ out6 ^ in3;
+        out2 = out5 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_14(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in4 ^ in6;
+        out1 = in5 ^ in7;
+        out2 = in0 ^ in4;
+        tmp0 = out0 ^ in5;
+        out7 = out1 ^ in3;
+        tmp1 = out1 ^ in2;
+        out3 = tmp0 ^ in1;
+        out6 = tmp0 ^ tmp1;
+        out4 = tmp1 ^ out2;
+        out5 = out3 ^ in3 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_15(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out7 = in3 ^ in5;
+        tmp0 = in0 ^ in4;
+        out1 = in1 ^ in5 ^ in7;
+        out5 = in1 ^ in3 ^ in6;
+        out0 = tmp0 ^ in6;
+        out2 = tmp0 ^ in2;
+        out3 = out5 ^ in4 ^ in5;
+        out6 = out2 ^ in0 ^ in7;
+        out4 = tmp0 ^ out6 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_16(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in5;
+        tmp1 = in4 ^ in7;
+        tmp2 = in2 ^ in3 ^ in4;
+        out1 = tmp0 ^ in7;
+        out4 = tmp0 ^ tmp2;
+        out0 = tmp1 ^ in6;
+        tmp3 = tmp1 ^ in1;
+        out6 = out0 ^ in2 ^ in5;
+        out2 = tmp3 ^ in0;
+        out3 = out6 ^ in1;
+        out7 = tmp2 ^ out6;
+        out5 = tmp3 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_17(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in5;
+        tmp1 = in3 ^ in6;
+        tmp2 = tmp0 ^ in4;
+        out4 = tmp0 ^ in0 ^ in3;
+        out7 = tmp1 ^ in5;
+        tmp3 = tmp1 ^ in1;
+        out6 = tmp2 ^ in7;
+        out5 = tmp3 ^ in4;
+        out3 = tmp3 ^ out6;
+        out0 = out3 ^ out4 ^ in1;
+        out2 = out3 ^ out7 ^ in0;
+        out1 = tmp2 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_18(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in4 ^ in5;
+        out1 = in5 ^ in6;
+        tmp0 = in4 ^ in7;
+        out5 = in1 ^ in2 ^ in5;
+        out6 = in2 ^ in3 ^ in6;
+        out2 = tmp0 ^ out1;
+        out7 = tmp0 ^ in3;
+        tmp1 = tmp0 ^ in0;
+        out3 = tmp1 ^ in6;
+        out4 = tmp1 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_19(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out5 = in1 ^ in2;
+        out7 = in3 ^ in4;
+        tmp0 = in0 ^ in7;
+        out6 = in2 ^ in3;
+        out1 = in1 ^ in5 ^ in6;
+        out0 = in0 ^ in4 ^ in5;
+        out4 = tmp0 ^ in1;
+        tmp1 = tmp0 ^ in6;
+        out2 = tmp1 ^ out0 ^ in2;
+        out3 = tmp1 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_1A(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in4 ^ in5;
+        tmp1 = in5 ^ in6;
+        tmp2 = tmp0 ^ in1;
+        out0 = tmp0 ^ in7;
+        out1 = tmp1 ^ in0;
+        tmp3 = tmp1 ^ in3;
+        out5 = tmp2 ^ in2;
+        out2 = tmp2 ^ in6;
+        out7 = tmp3 ^ out0;
+        out6 = tmp3 ^ in2;
+        out4 = tmp3 ^ out2 ^ in0;
+        out3 = tmp0 ^ out1 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_1B(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in4;
+        tmp1 = in2 ^ in5;
+        tmp2 = in3 ^ in6;
+        out5 = tmp0 ^ in1;
+        tmp3 = tmp0 ^ in0;
+        out6 = tmp1 ^ in3;
+        out0 = tmp1 ^ tmp3 ^ in7;
+        out7 = tmp2 ^ in4;
+        tmp4 = out5 ^ in6;
+        out3 = tmp2 ^ tmp3;
+        out2 = tmp4 ^ in5;
+        out4 = tmp4 ^ out3;
+        out1 = tmp3 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_1C(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in3;
+        tmp1 = in4 ^ in6;
+        tmp2 = in5 ^ in7;
+        out6 = tmp0 ^ tmp1;
+        out0 = tmp1 ^ in5;
+        out1 = tmp2 ^ in6;
+        tmp3 = tmp2 ^ in1;
+        tmp4 = tmp2 ^ in4;
+        out2 = tmp4 ^ in0;
+        out7 = tmp4 ^ in3;
+        out5 = tmp0 ^ tmp3;
+        out3 = tmp3 ^ out2;
+        out4 = out3 ^ in2 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_1D(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in3;
+        tmp1 = in0 ^ in4;
+        tmp2 = in3 ^ in4;
+        tmp3 = in2 ^ in7;
+        out3 = tmp0 ^ tmp1;
+        out5 = tmp0 ^ tmp3;
+        tmp4 = tmp1 ^ in5;
+        out6 = tmp2 ^ in2;
+        out7 = tmp2 ^ in5;
+        out2 = tmp3 ^ tmp4;
+        out4 = out3 ^ out6 ^ in6;
+        out0 = tmp4 ^ in6;
+        out1 = out2 ^ out4 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_1E(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in4;
+        tmp1 = in2 ^ in7;
+        tmp2 = tmp0 ^ in1;
+        out3 = tmp1 ^ tmp2;
+        out2 = tmp2 ^ in5;
+        out4 = out3 ^ in3 ^ in6;
+        tmp3 = out4 ^ in7;
+        out6 = tmp3 ^ out2 ^ in4;
+        out7 = tmp1 ^ out6;
+        out0 = out7 ^ in3;
+        out1 = tmp0 ^ out0;
+        out5 = tmp3 ^ out1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_1F(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in4 ^ in6;
+        tmp1 = tmp0 ^ in5;
+        out7 = tmp1 ^ in3;
+        out0 = tmp1 ^ in0 ^ in7;
+        out6 = out7 ^ in2 ^ in6;
+        out1 = out0 ^ in1 ^ in4;
+        out4 = out0 ^ out6 ^ in1;
+        out3 = tmp0 ^ out4;
+        out2 = out4 ^ out7 ^ in7;
+        out5 = out3 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_20(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in4;
+        out0 = in3 ^ in7;
+        tmp0 = in3 ^ in4;
+        tmp1 = in6 ^ in7;
+        out2 = out0 ^ in5;
+        out4 = tmp0 ^ in5;
+        out3 = tmp0 ^ tmp1;
+        out7 = tmp1 ^ in2;
+        out6 = tmp1 ^ in1 ^ in5;
+        out5 = out2 ^ out3 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_21(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in1 ^ in4;
+        tmp0 = in4 ^ in6;
+        out4 = in3 ^ in5;
+        out7 = in2 ^ in6;
+        out0 = in0 ^ in3 ^ in7;
+        out6 = in1 ^ in5 ^ in7;
+        out3 = tmp0 ^ in7;
+        out5 = tmp0 ^ in0;
+        out2 = out4 ^ in2 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_22(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in3;
+        out1 = in0 ^ in4;
+        out7 = in2 ^ in7;
+        out4 = in4 ^ in5 ^ in7;
+        out5 = in0 ^ in5 ^ in6;
+        out6 = in1 ^ in6 ^ in7;
+        out3 = in2 ^ in3 ^ in4 ^ in6;
+        out2 = in1 ^ in3 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_23(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out7 = in2;
+        out0 = in0 ^ in3;
+        out4 = in5 ^ in7;
+        out5 = in0 ^ in6;
+        out6 = in1 ^ in7;
+        out3 = in2 ^ in4 ^ in6;
+        out1 = in0 ^ in1 ^ in4;
+        out2 = out4 ^ out6 ^ in2 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_24(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in4 ^ in7;
+        tmp0 = in3 ^ in4;
+        out0 = in3 ^ in6 ^ in7;
+        out3 = tmp0 ^ in1;
+        tmp1 = out0 ^ in5;
+        out6 = tmp1 ^ out3;
+        out2 = tmp1 ^ in0;
+        out7 = tmp1 ^ in2 ^ in3;
+        out5 = out2 ^ in4;
+        out4 = tmp0 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_25(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in1 ^ in4;
+        tmp0 = in2 ^ in5;
+        out1 = out3 ^ in7;
+        out7 = tmp0 ^ in6;
+        out6 = out1 ^ in5;
+        out4 = out7 ^ in3 ^ in7;
+        out2 = out4 ^ in0;
+        out0 = tmp0 ^ out2;
+        out5 = out0 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_26(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in3 ^ in6;
+        tmp0 = in4 ^ in7;
+        out7 = in2 ^ in5 ^ in7;
+        tmp1 = out0 ^ in0 ^ in5;
+        out1 = tmp0 ^ in0;
+        tmp2 = tmp0 ^ in6;
+        out2 = tmp1 ^ in1;
+        out5 = tmp1 ^ in7;
+        out6 = tmp2 ^ in1;
+        out4 = tmp2 ^ out7;
+        out3 = out0 ^ out6 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_27(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out7 = in2 ^ in5;
+        out0 = in0 ^ in3 ^ in6;
+        out6 = in1 ^ in4 ^ in7;
+        out4 = out7 ^ in6;
+        out2 = out0 ^ out7 ^ in1;
+        out5 = out0 ^ in7;
+        out1 = out6 ^ in0;
+        out3 = out6 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_28(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in3;
+        out1 = in4 ^ in6;
+        out0 = in3 ^ in5 ^ in7;
+        tmp0 = out1 ^ in7;
+        tmp1 = out0 ^ in4;
+        out7 = tmp0 ^ in2;
+        tmp2 = tmp0 ^ in1;
+        out3 = tmp1 ^ in0;
+        out6 = tmp1 ^ tmp2;
+        out4 = tmp2 ^ in3;
+        out5 = out3 ^ in2 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_29(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in2 ^ in3;
+        tmp0 = in1 ^ in3;
+        tmp1 = in4 ^ in6;
+        tmp2 = in0 ^ in4 ^ in7;
+        out6 = tmp0 ^ in5;
+        out4 = tmp0 ^ in6 ^ in7;
+        out1 = tmp1 ^ in1;
+        out7 = tmp1 ^ in2;
+        out3 = tmp2 ^ in5;
+        out5 = tmp2 ^ in2;
+        out0 = out3 ^ in3 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_2A(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in3 ^ in5;
+        tmp0 = in1 ^ in3;
+        tmp1 = in0 ^ in4;
+        out7 = in2 ^ in4 ^ in7;
+        out3 = tmp1 ^ out0 ^ in2;
+        out2 = tmp0 ^ in7;
+        out6 = tmp0 ^ in6;
+        out1 = tmp1 ^ in6;
+        out5 = tmp1 ^ out7 ^ in5;
+        out4 = out1 ^ in0 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_2B(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in1 ^ in6;
+        out7 = in2 ^ in4;
+        tmp0 = in0 ^ in5;
+        tmp1 = in2 ^ in7;
+        out6 = in1 ^ in3;
+        out1 = out4 ^ in0 ^ in4;
+        out3 = tmp0 ^ out7;
+        out0 = tmp0 ^ in3;
+        out5 = tmp1 ^ in0;
+        out2 = tmp1 ^ out6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_2C(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in5;
+        tmp1 = in2 ^ in3 ^ in4;
+        tmp2 = tmp0 ^ in6;
+        out4 = tmp1 ^ in1;
+        out5 = tmp1 ^ in0 ^ in5;
+        tmp3 = tmp2 ^ in4;
+        out6 = tmp2 ^ out4;
+        out7 = tmp3 ^ in7;
+        out2 = tmp3 ^ out5;
+        out3 = out6 ^ in0;
+        out0 = tmp1 ^ out7;
+        out1 = tmp0 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_2D(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in3;
+        out4 = tmp0 ^ in1;
+        tmp1 = tmp0 ^ in0;
+        out2 = tmp1 ^ in6;
+        out5 = tmp1 ^ in4;
+        tmp2 = out2 ^ in2;
+        tmp3 = tmp2 ^ in5;
+        out0 = tmp3 ^ in7;
+        out7 = tmp3 ^ out5;
+        out6 = out4 ^ out7 ^ in6;
+        out3 = tmp2 ^ out6;
+        out1 = out0 ^ out6 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_2E(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in4 ^ in7;
+        out0 = in3 ^ in5 ^ in6;
+        tmp1 = tmp0 ^ in0;
+        tmp2 = tmp0 ^ in2;
+        out1 = tmp1 ^ in6;
+        out4 = tmp2 ^ in1;
+        out7 = tmp2 ^ in5;
+        out3 = out0 ^ out4 ^ in0;
+        out2 = out3 ^ out7 ^ in7;
+        out6 = tmp1 ^ out2;
+        out5 = tmp1 ^ out7 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_2F(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in3;
+        tmp1 = in2 ^ in5;
+        out4 = in1 ^ in2 ^ in7;
+        out6 = in1 ^ in3 ^ in4;
+        out5 = tmp0 ^ in2;
+        tmp2 = tmp0 ^ in6;
+        out7 = tmp1 ^ in4;
+        out0 = tmp2 ^ in5;
+        out2 = tmp2 ^ out4;
+        out1 = tmp2 ^ out6 ^ in7;
+        out3 = tmp1 ^ out1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_30(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in4 ^ in5;
+        tmp0 = in3 ^ in6;
+        tmp1 = in4 ^ in7;
+        out6 = in1 ^ in2 ^ in5;
+        out3 = tmp0 ^ in5;
+        out4 = tmp0 ^ in0;
+        out7 = tmp0 ^ in2;
+        out0 = tmp1 ^ in3;
+        out2 = tmp1 ^ out3;
+        out5 = tmp1 ^ in0 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_31(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in5 ^ in6;
+        tmp0 = in4 ^ in5;
+        tmp1 = in0 ^ in3 ^ in4;
+        tmp2 = out3 ^ in2;
+        out1 = tmp0 ^ in1;
+        out0 = tmp1 ^ in7;
+        out4 = tmp1 ^ in6;
+        out6 = tmp2 ^ in1;
+        out2 = tmp2 ^ out0 ^ in0;
+        out5 = out1 ^ in0 ^ in7;
+        out7 = tmp0 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_32(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in3 ^ in4;
+        out7 = in2 ^ in3;
+        tmp0 = in5 ^ in6;
+        tmp1 = in0 ^ in7;
+        out6 = in1 ^ in2;
+        out1 = in0 ^ in4 ^ in5;
+        out2 = tmp0 ^ out0 ^ in1;
+        out3 = tmp0 ^ out7 ^ in7;
+        out4 = tmp1 ^ in6;
+        out5 = tmp1 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_33(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in3;
+        tmp1 = in0 ^ in4;
+        tmp2 = in1 ^ in5;
+        out6 = in1 ^ in2 ^ in6;
+        out7 = tmp0 ^ in7;
+        out0 = tmp1 ^ in3;
+        out1 = tmp1 ^ tmp2;
+        tmp3 = tmp2 ^ in7;
+        tmp4 = tmp2 ^ in4 ^ in6;
+        out5 = tmp3 ^ in0;
+        out3 = tmp3 ^ out6;
+        out4 = tmp4 ^ out5;
+        out2 = tmp0 ^ tmp4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_34(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in4;
+        tmp1 = in4 ^ in5;
+        tmp2 = tmp0 ^ in1;
+        tmp3 = tmp0 ^ in6;
+        out1 = tmp1 ^ in7;
+        tmp4 = tmp1 ^ in2;
+        out5 = tmp2 ^ in0;
+        out3 = tmp2 ^ out1;
+        out0 = tmp3 ^ in7;
+        out7 = tmp3 ^ tmp4;
+        out6 = tmp4 ^ in1;
+        out2 = out3 ^ out5 ^ in3;
+        out4 = tmp4 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_35(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in6;
+        tmp1 = in5 ^ in7;
+        out7 = tmp0 ^ tmp1 ^ in3;
+        out3 = tmp1 ^ in1;
+        out1 = out3 ^ in4;
+        tmp2 = out1 ^ in7;
+        out5 = tmp2 ^ in0 ^ in3;
+        out6 = tmp0 ^ tmp2;
+        out0 = out3 ^ out5 ^ in6;
+        out4 = tmp0 ^ out0;
+        out2 = out4 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_36(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in0 ^ in2;
+        tmp0 = in1 ^ in3;
+        out0 = in3 ^ in4 ^ in6;
+        out6 = in1 ^ in2 ^ in4;
+        out5 = tmp0 ^ in0;
+        tmp1 = out5 ^ in5;
+        out2 = tmp1 ^ in4;
+        out3 = tmp1 ^ out4;
+        out1 = tmp0 ^ out2 ^ in7;
+        out7 = out3 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_37(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in2;
+        tmp1 = in2 ^ in4;
+        tmp2 = tmp0 ^ in6;
+        out3 = tmp0 ^ in5;
+        out4 = tmp1 ^ in0;
+        out6 = tmp2 ^ in4;
+        out1 = out3 ^ out4 ^ in7;
+        tmp3 = out4 ^ in1 ^ in3;
+        out7 = tmp3 ^ out1;
+        out2 = tmp3 ^ in5;
+        out5 = tmp1 ^ out2;
+        out0 = tmp2 ^ tmp3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_38(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in0 ^ in3;
+        tmp0 = in3 ^ in4;
+        tmp1 = in5 ^ in7;
+        tmp2 = out3 ^ in1;
+        out2 = tmp0 ^ in6;
+        out0 = tmp0 ^ tmp1;
+        out4 = tmp1 ^ tmp2;
+        out7 = out2 ^ in2;
+        out1 = out2 ^ in3 ^ in5;
+        out6 = out4 ^ in0 ^ in2;
+        out5 = tmp2 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_39(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in0;
+        tmp0 = in1 ^ in5;
+        tmp1 = tmp0 ^ in4;
+        out1 = tmp1 ^ in6;
+        out5 = out1 ^ in0 ^ in2;
+        tmp2 = tmp0 ^ out5;
+        out2 = tmp2 ^ in0 ^ in3;
+        out7 = out2 ^ in7;
+        out6 = tmp1 ^ out7;
+        out4 = tmp2 ^ out6;
+        out0 = out4 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_3A(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in1;
+        tmp1 = in0 ^ in2;
+        tmp2 = in3 ^ in4;
+        tmp3 = in1 ^ in6;
+        tmp4 = in3 ^ in7;
+        out4 = tmp0 ^ in5;
+        out5 = tmp1 ^ tmp3;
+        out3 = tmp1 ^ tmp4;
+        out0 = tmp2 ^ in5;
+        out7 = tmp2 ^ in2;
+        tmp5 = tmp3 ^ in4;
+        out2 = tmp4 ^ tmp5;
+        out1 = tmp5 ^ out4;
+        out6 = tmp0 ^ out3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_3B(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in6;
+        tmp1 = in2 ^ in7;
+        tmp2 = tmp0 ^ in3;
+        out3 = tmp1 ^ in0;
+        out6 = tmp1 ^ tmp2;
+        out2 = out6 ^ in4;
+        out7 = tmp0 ^ out2;
+        out0 = out3 ^ out7 ^ in5;
+        out5 = out0 ^ out2 ^ in7;
+        out1 = tmp2 ^ out0;
+        out4 = out1 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_3C(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in3;
+        tmp1 = in2 ^ in7;
+        tmp2 = in1 ^ in6 ^ in7;
+        out2 = tmp0 ^ in4;
+        out3 = tmp0 ^ tmp2;
+        out4 = tmp1 ^ out3 ^ in5;
+        out5 = tmp2 ^ out2 ^ in2;
+        out1 = out4 ^ out5 ^ in6;
+        out0 = out1 ^ in3;
+        out7 = tmp1 ^ out0;
+        out6 = tmp2 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_3D(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in2;
+        tmp1 = tmp0 ^ in3;
+        out2 = tmp1 ^ in4;
+        tmp2 = out2 ^ in5;
+        out4 = tmp2 ^ in1 ^ in6;
+        out5 = out4 ^ in7;
+        out6 = out5 ^ in0;
+        out7 = out6 ^ in1;
+        out0 = tmp0 ^ out7;
+        out1 = tmp1 ^ out5;
+        out3 = tmp2 ^ out6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_3E(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in5;
+        tmp1 = tmp0 ^ in4;
+        out0 = tmp1 ^ in6;
+        out7 = tmp1 ^ in2;
+        out6 = out7 ^ in1 ^ in5 ^ in7;
+        out2 = out6 ^ in0 ^ in2;
+        out4 = out0 ^ out6 ^ in0;
+        out5 = tmp0 ^ out4;
+        out3 = out5 ^ in7;
+        out1 = out3 ^ out6 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_3F(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in1;
+        out3 = tmp0 ^ in2 ^ in6;
+        tmp1 = out3 ^ in5 ^ in7;
+        out4 = tmp1 ^ in4;
+        out5 = tmp1 ^ in3;
+        out1 = out4 ^ in2;
+        out7 = out1 ^ out3 ^ in3;
+        out2 = tmp0 ^ out7 ^ in5;
+        tmp2 = out2 ^ in0;
+        out6 = tmp2 ^ in6;
+        out0 = tmp1 ^ tmp2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_40(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in3 ^ in7;
+        tmp0 = in3 ^ in4;
+        tmp1 = in6 ^ in7;
+        out4 = tmp0 ^ in2;
+        out5 = tmp0 ^ in5;
+        out0 = tmp1 ^ in2;
+        out7 = tmp1 ^ in1 ^ in5;
+        out2 = out0 ^ in4;
+        out3 = out2 ^ out5 ^ in7;
+        out6 = out3 ^ out4 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_41(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in2 ^ in3;
+        tmp0 = in5 ^ in6;
+        tmp1 = in6 ^ in7;
+        out5 = in3 ^ in4;
+        out1 = in1 ^ in3 ^ in7;
+        out6 = in0 ^ in4 ^ in5;
+        out3 = tmp0 ^ in2;
+        out7 = tmp0 ^ in1;
+        out2 = tmp1 ^ in4;
+        out0 = tmp1 ^ in0 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_42(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in2 ^ in6;
+        out5 = in3 ^ in5;
+        out1 = in0 ^ in3 ^ in7;
+        out7 = in1 ^ in5 ^ in7;
+        out4 = in2 ^ in4 ^ in7;
+        out6 = in0 ^ in4 ^ in6;
+        out2 = out0 ^ in1 ^ in4;
+        out3 = out5 ^ in6 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_43(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out5 = in3;
+        out7 = in1 ^ in5;
+        out4 = in2 ^ in7;
+        out6 = in0 ^ in4;
+        out0 = in0 ^ in2 ^ in6;
+        out3 = in5 ^ in6 ^ in7;
+        out2 = in1 ^ in4 ^ in6;
+        out1 = in0 ^ in1 ^ in3 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_44(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in3;
+        out0 = in2 ^ in7;
+        tmp0 = in4 ^ in7;
+        out7 = in1 ^ in6 ^ in7;
+        out6 = in0 ^ in5 ^ in6;
+        out4 = tmp0 ^ in3 ^ in6;
+        out3 = out0 ^ in1 ^ in3 ^ in5;
+        out2 = out0 ^ in0 ^ in4;
+        out5 = tmp0 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_45(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in1 ^ in3;
+        out7 = in1 ^ in6;
+        out5 = in4 ^ in7;
+        out6 = in0 ^ in5;
+        out0 = in0 ^ in2 ^ in7;
+        out4 = in3 ^ in6 ^ in7;
+        out2 = out5 ^ in0;
+        out3 = out0 ^ out6 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_46(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in2;
+        out1 = in0 ^ in3;
+        out7 = in1 ^ in7;
+        out4 = in4 ^ in6;
+        out5 = in5 ^ in7;
+        out6 = in0 ^ in6;
+        out3 = in1 ^ in3 ^ in5;
+        out2 = out4 ^ out6 ^ in1 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_47(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in6;
+        out7 = in1;
+        out5 = in7;
+        out6 = in0;
+        tmp0 = in0 ^ in1;
+        out3 = in1 ^ in5;
+        out0 = in0 ^ in2;
+        out1 = tmp0 ^ in3;
+        out2 = tmp0 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_48(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in3;
+        out1 = in3 ^ in6 ^ in7;
+        out3 = tmp0 ^ in0;
+        out0 = tmp0 ^ out1 ^ in5;
+        tmp1 = out0 ^ in4;
+        out2 = tmp1 ^ in7;
+        out5 = tmp1 ^ in3;
+        out4 = out5 ^ in1;
+        out7 = tmp0 ^ out4;
+        out6 = tmp1 ^ out3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_49(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in0 ^ in2;
+        tmp0 = in2 ^ in5;
+        out2 = in4 ^ in5 ^ in6;
+        tmp1 = tmp0 ^ out2 ^ in3;
+        out7 = out2 ^ in1;
+        out5 = tmp1 ^ in7;
+        out4 = out5 ^ out7 ^ in6;
+        out1 = tmp0 ^ out4;
+        out6 = out1 ^ out7 ^ in0;
+        out0 = tmp1 ^ out6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_4A(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in6;
+        tmp1 = in3 ^ in7;
+        out0 = tmp0 ^ in5;
+        out3 = tmp1 ^ in0;
+        out5 = tmp1 ^ out0;
+        out4 = out0 ^ in1 ^ in4;
+        out1 = out3 ^ in6;
+        out2 = out4 ^ in7;
+        out6 = out1 ^ in4;
+        out7 = tmp0 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_4B(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in0 ^ in7;
+        tmp0 = in1 ^ in5;
+        tmp1 = in2 ^ in6;
+        tmp2 = out3 ^ in3;
+        out7 = tmp0 ^ in4;
+        out4 = tmp0 ^ tmp1;
+        tmp3 = tmp1 ^ in0;
+        out6 = tmp2 ^ in4;
+        out5 = tmp2 ^ tmp3;
+        out1 = tmp2 ^ in1 ^ in6;
+        out2 = out7 ^ in6 ^ in7;
+        out0 = tmp3 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_4C(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in3 ^ in6;
+        tmp0 = in2 ^ in5;
+        tmp1 = out1 ^ in5 ^ in7;
+        out0 = tmp0 ^ in7;
+        tmp2 = tmp0 ^ in4;
+        out6 = tmp1 ^ in0;
+        out2 = tmp2 ^ in0;
+        out5 = tmp2 ^ in6;
+        out3 = tmp0 ^ out6 ^ in1;
+        out7 = out0 ^ out5 ^ in1;
+        out4 = tmp1 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_4D(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in5;
+        tmp1 = in1 ^ in6;
+        out4 = in1 ^ in3 ^ in5;
+        tmp2 = tmp0 ^ in7;
+        out2 = tmp0 ^ in4;
+        out1 = tmp1 ^ in3;
+        out7 = tmp1 ^ in4;
+        out0 = tmp2 ^ in2;
+        out6 = tmp2 ^ in3;
+        out5 = out7 ^ in1 ^ in2;
+        out3 = tmp1 ^ out0 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_4E(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in2 ^ in5;
+        out7 = in1 ^ in4 ^ in7;
+        out1 = in0 ^ in3 ^ in6;
+        out5 = out0 ^ in6;
+        out4 = out7 ^ in5;
+        out3 = out1 ^ in1;
+        out6 = out1 ^ in7;
+        out2 = out4 ^ in0 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_4F(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out5 = in2 ^ in6;
+        out7 = in1 ^ in4;
+        out3 = in0 ^ in1 ^ in6;
+        out4 = in1 ^ in5 ^ in7;
+        out0 = in0 ^ in2 ^ in5;
+        out6 = in0 ^ in3 ^ in7;
+        out1 = out3 ^ in3;
+        out2 = out4 ^ in0 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_50(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in2 ^ in7;
+        tmp0 = in3 ^ in5;
+        out0 = out2 ^ in4 ^ in6;
+        out1 = tmp0 ^ in7;
+        tmp1 = tmp0 ^ in6;
+        out3 = out0 ^ in3;
+        out7 = tmp1 ^ in1;
+        tmp2 = tmp1 ^ in0;
+        out5 = out3 ^ in1 ^ in2;
+        out4 = tmp2 ^ in2;
+        out6 = tmp2 ^ out3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_51(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in7;
+        out3 = in2 ^ in4 ^ in6 ^ in7;
+        out0 = out3 ^ in0;
+        out6 = out0 ^ in5;
+        out4 = out6 ^ in3 ^ in7;
+        out1 = out0 ^ out4 ^ in1;
+        out7 = out1 ^ in6;
+        out5 = out7 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_52(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in1 ^ in2;
+        tmp0 = in2 ^ in4;
+        tmp1 = in3 ^ in5;
+        tmp2 = in3 ^ in6;
+        tmp3 = in0 ^ in7;
+        out0 = tmp0 ^ in6;
+        out6 = tmp0 ^ tmp3;
+        out7 = tmp1 ^ in1;
+        out1 = tmp1 ^ tmp3;
+        out3 = tmp2 ^ in4;
+        out5 = tmp2 ^ in1 ^ in7;
+        out4 = tmp2 ^ out1 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_53(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in1;
+        out3 = in4 ^ in6;
+        out0 = out3 ^ in0 ^ in2;
+        out6 = out0 ^ in7;
+        out4 = out6 ^ in5;
+        out7 = out0 ^ out4 ^ in1 ^ in3;
+        out1 = out7 ^ in0;
+        out5 = out7 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_54(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in3 ^ in5;
+        tmp0 = in1 ^ in3;
+        tmp1 = in2 ^ in4;
+        tmp2 = in0 ^ in7;
+        out5 = in1 ^ in4 ^ in6;
+        out4 = tmp2 ^ out1;
+        out7 = tmp0 ^ in6;
+        out3 = tmp0 ^ tmp1;
+        out0 = tmp1 ^ in7;
+        tmp3 = tmp2 ^ in2;
+        out2 = tmp3 ^ in6;
+        out6 = tmp3 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_55(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in3;
+        tmp1 = in1 ^ in4;
+        tmp2 = in6 ^ in7;
+        out7 = tmp0 ^ tmp2;
+        out1 = tmp0 ^ in5;
+        out3 = tmp1 ^ in2;
+        out5 = tmp1 ^ in5 ^ in6;
+        out2 = tmp2 ^ in0;
+        out4 = out5 ^ out7 ^ in0;
+        out6 = out2 ^ in2 ^ in5;
+        out0 = out5 ^ out6 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_56(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in2 ^ in4;
+        tmp0 = in0 ^ in2;
+        out4 = in0 ^ in5;
+        out7 = in1 ^ in3;
+        out5 = in1 ^ in6;
+        out6 = tmp0 ^ in7;
+        out2 = tmp0 ^ out5;
+        out1 = out4 ^ in3;
+        out3 = out7 ^ in4 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_57(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in5;
+        tmp1 = in1 ^ in7;
+        out0 = in0 ^ in2 ^ in4;
+        out5 = in1 ^ in5 ^ in6;
+        out4 = tmp0 ^ in4;
+        out1 = tmp0 ^ in1 ^ in3;
+        out2 = tmp0 ^ out5;
+        out3 = tmp1 ^ in4;
+        out7 = tmp1 ^ in3;
+        out6 = tmp1 ^ out2 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_58(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in2 ^ in5;
+        tmp0 = in2 ^ in3 ^ in4;
+        out5 = tmp0 ^ in1;
+        out6 = tmp0 ^ in0 ^ in5;
+        out3 = out6 ^ in7;
+        tmp1 = out2 ^ out5;
+        out7 = tmp1 ^ in6;
+        out4 = tmp1 ^ out3 ^ in3;
+        out0 = out4 ^ out7 ^ in0;
+        out1 = tmp0 ^ out0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_59(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in5;
+        tmp0 = in0 ^ in5 ^ in7;
+        out3 = tmp0 ^ in2 ^ in4;
+        out0 = out3 ^ in6;
+        tmp1 = out0 ^ in7;
+        out6 = tmp1 ^ in3;
+        out5 = out6 ^ in0 ^ in1 ^ in6;
+        out4 = tmp0 ^ out5;
+        out1 = tmp1 ^ out4;
+        out7 = out1 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_5A(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in2;
+        tmp1 = in2 ^ in5;
+        out5 = tmp0 ^ in3;
+        out4 = tmp0 ^ in0;
+        tmp2 = tmp1 ^ in4;
+        out2 = tmp1 ^ in1 ^ in7;
+        out7 = tmp2 ^ out5;
+        out6 = out4 ^ out7 ^ in5;
+        out0 = tmp2 ^ in6;
+        out1 = out0 ^ out6 ^ in7;
+        out3 = tmp1 ^ out6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_5B(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in3;
+        tmp1 = in0 ^ in4;
+        tmp2 = in1 ^ in5;
+        out5 = tmp0 ^ tmp2;
+        tmp3 = tmp1 ^ in6;
+        out3 = tmp1 ^ in5;
+        out2 = tmp2 ^ in7;
+        tmp4 = out3 ^ in2;
+        out7 = out2 ^ in3 ^ in4;
+        out0 = tmp4 ^ in6;
+        out6 = tmp0 ^ tmp3;
+        out4 = tmp2 ^ tmp4;
+        out1 = tmp3 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_5C(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in6;
+        tmp1 = in0 ^ in2 ^ in5;
+        out1 = tmp0 ^ in5;
+        tmp2 = tmp0 ^ in1;
+        out2 = tmp1 ^ in6;
+        out6 = tmp1 ^ in3;
+        out4 = tmp2 ^ in0;
+        out7 = tmp2 ^ in4;
+        out3 = tmp1 ^ out7;
+        out0 = out3 ^ out4 ^ in7;
+        out5 = out0 ^ in1 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_5D(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in1;
+        tmp1 = in0 ^ in6;
+        out2 = tmp1 ^ in5;
+        tmp2 = out2 ^ in3;
+        out6 = tmp2 ^ in2;
+        out1 = tmp0 ^ tmp2;
+        tmp3 = out1 ^ in4 ^ in5;
+        out4 = tmp3 ^ in0;
+        out7 = tmp3 ^ in7;
+        tmp4 = out4 ^ out6;
+        out5 = tmp4 ^ in7;
+        out0 = tmp0 ^ out5;
+        out3 = tmp1 ^ tmp4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_5E(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in5;
+        tmp1 = in3 ^ in5;
+        tmp2 = in1 ^ in7;
+        out7 = in1 ^ in3 ^ in4;
+        out0 = tmp0 ^ in4;
+        tmp3 = tmp1 ^ in0;
+        out5 = tmp2 ^ in2;
+        out1 = tmp3 ^ in6;
+        out6 = tmp0 ^ tmp3;
+        tmp4 = tmp2 ^ out1;
+        out3 = tmp4 ^ in4;
+        out4 = tmp1 ^ tmp4;
+        out2 = tmp0 ^ out4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_5F(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in5;
+        tmp1 = in0 ^ in6;
+        tmp2 = tmp0 ^ in7;
+        tmp3 = tmp1 ^ in3;
+        out2 = tmp1 ^ tmp2;
+        out5 = tmp2 ^ in2;
+        out6 = tmp3 ^ in2;
+        out3 = out2 ^ in4;
+        out4 = out3 ^ in5;
+        out1 = tmp0 ^ tmp3;
+        out7 = tmp3 ^ out4;
+        out0 = out4 ^ out5 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_60(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in2 ^ in5;
+        tmp0 = in3 ^ in6;
+        out1 = in3 ^ in4 ^ in7;
+        out7 = out4 ^ in1;
+        tmp1 = out4 ^ in4;
+        out0 = tmp0 ^ in2;
+        out5 = tmp0 ^ in0;
+        out2 = tmp0 ^ tmp1;
+        out3 = tmp1 ^ in7;
+        out6 = out3 ^ out7 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_61(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in5;
+        out4 = tmp0 ^ in4;
+        tmp1 = out4 ^ in3;
+        out3 = tmp1 ^ in7;
+        out2 = tmp1 ^ in2 ^ in6;
+        out1 = tmp0 ^ out3 ^ in1;
+        out0 = out2 ^ out4 ^ in0;
+        out7 = tmp1 ^ out1;
+        out6 = out0 ^ out1 ^ in2;
+        out5 = tmp0 ^ out0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_62(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in4 ^ in5;
+        tmp0 = in0 ^ in3 ^ in4;
+        out1 = tmp0 ^ in7;
+        out5 = tmp0 ^ in6;
+        tmp1 = out1 ^ in0;
+        tmp2 = tmp1 ^ out3;
+        out4 = tmp2 ^ in2;
+        tmp3 = tmp2 ^ in1;
+        out0 = out4 ^ in5 ^ in6;
+        out7 = tmp3 ^ out0;
+        out6 = tmp0 ^ tmp3;
+        out2 = tmp1 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_63(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in4;
+        tmp1 = in1 ^ in7;
+        out3 = tmp0 ^ in5;
+        tmp2 = out3 ^ in6;
+        out4 = out3 ^ in2 ^ in7;
+        out5 = tmp2 ^ in0;
+        tmp3 = out5 ^ in3;
+        out0 = tmp3 ^ out4;
+        out2 = tmp1 ^ tmp2;
+        out6 = tmp1 ^ tmp3;
+        tmp4 = tmp0 ^ out2;
+        out1 = tmp4 ^ out5;
+        out7 = tmp4 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_64(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in2 ^ in3;
+        out1 = in3 ^ in4;
+        out7 = in1 ^ in2;
+        tmp0 = in4 ^ in5;
+        tmp1 = in0 ^ in7;
+        out4 = in5 ^ in6 ^ in7;
+        out2 = tmp0 ^ out0 ^ in0;
+        out3 = tmp0 ^ out7 ^ in6;
+        out5 = tmp1 ^ in6;
+        out6 = tmp1 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_65(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in3;
+        tmp1 = in4 ^ in5;
+        tmp2 = in6 ^ in7;
+        out7 = in1 ^ in2 ^ in7;
+        out1 = in1 ^ in3 ^ in4;
+        out0 = tmp0 ^ in2;
+        out2 = tmp0 ^ tmp1;
+        out4 = tmp1 ^ tmp2;
+        tmp3 = tmp2 ^ in0;
+        out3 = out4 ^ out7 ^ in3;
+        out5 = tmp3 ^ in5;
+        out6 = tmp3 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_66(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in2;
+        tmp1 = in2 ^ in3;
+        tmp2 = in0 ^ in4;
+        out7 = tmp0 ^ in6;
+        out0 = tmp1 ^ in7;
+        out1 = tmp2 ^ in3;
+        tmp3 = tmp2 ^ in6;
+        tmp4 = out1 ^ in5;
+        out5 = tmp3 ^ in7;
+        out4 = tmp3 ^ tmp4;
+        out2 = tmp0 ^ tmp4 ^ in7;
+        out6 = tmp1 ^ out2 ^ in4;
+        out3 = tmp3 ^ out6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_67(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in3;
+        tmp1 = tmp0 ^ in1;
+        tmp2 = tmp0 ^ in7;
+        out1 = tmp1 ^ in4;
+        out0 = tmp2 ^ in2;
+        tmp3 = out1 ^ in7;
+        out2 = tmp3 ^ in5;
+        out3 = out2 ^ in0 ^ in6;
+        out7 = tmp1 ^ out0 ^ in6;
+        out5 = tmp1 ^ out3;
+        out4 = tmp2 ^ out5;
+        out6 = tmp3 ^ out4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_68(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in4;
+        tmp1 = in2 ^ in3 ^ in5;
+        tmp2 = tmp0 ^ in1;
+        tmp3 = tmp0 ^ in6;
+        out0 = tmp1 ^ in6;
+        out6 = tmp2 ^ in0;
+        out7 = tmp1 ^ tmp2;
+        out1 = tmp3 ^ in7;
+        out2 = out1 ^ in2;
+        out4 = tmp2 ^ out2;
+        out3 = out4 ^ out6 ^ in3;
+        out5 = tmp3 ^ out3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_69(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in6 ^ in7;
+        out2 = tmp0 ^ in3 ^ in4;
+        out1 = out2 ^ in1;
+        out3 = out2 ^ in0 ^ in2;
+        out4 = out1 ^ in2 ^ in3;
+        out6 = out1 ^ in0 ^ in7;
+        out7 = out4 ^ in5 ^ in6;
+        out5 = out4 ^ out6 ^ in5;
+        out0 = tmp0 ^ out5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_6A(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in6;
+        out3 = in0 ^ in4 ^ in6;
+        tmp1 = tmp0 ^ in3;
+        out4 = tmp1 ^ in1;
+        tmp2 = tmp1 ^ in7;
+        out2 = out4 ^ in4;
+        out0 = tmp2 ^ in5;
+        out5 = tmp2 ^ out3;
+        out7 = out2 ^ in3 ^ in5;
+        out1 = tmp0 ^ out5;
+        out6 = tmp1 ^ out7 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_6B(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in4 ^ in6;
+        out2 = tmp0 ^ in1 ^ in3;
+        out4 = out2 ^ in2;
+        tmp1 = out2 ^ in0;
+        out7 = out4 ^ in3 ^ in5 ^ in7;
+        out1 = tmp1 ^ in7;
+        out3 = tmp1 ^ in1;
+        out6 = tmp1 ^ in5;
+        out0 = tmp1 ^ out7 ^ in6;
+        out5 = tmp0 ^ out0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_6C(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in1;
+        tmp0 = in2 ^ in3;
+        out5 = in0 ^ in2;
+        out1 = in3 ^ in4 ^ in6;
+        tmp1 = out5 ^ in1;
+        out0 = tmp0 ^ in5;
+        out6 = tmp0 ^ tmp1;
+        out3 = tmp1 ^ in4;
+        out7 = out3 ^ in0;
+        out2 = out6 ^ out7 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_6D(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in1 ^ in4;
+        tmp0 = in0 ^ in2;
+        tmp1 = out4 ^ in3;
+        out7 = out4 ^ in2 ^ in7;
+        out5 = tmp0 ^ in5;
+        out3 = tmp0 ^ tmp1;
+        out1 = tmp1 ^ in6;
+        out0 = out5 ^ in3;
+        out2 = out3 ^ out7 ^ in4;
+        out6 = out1 ^ in0 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_6E(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in3;
+        tmp1 = in0 ^ in4;
+        out4 = tmp0 ^ in7;
+        out6 = tmp0 ^ in0 ^ in5;
+        out5 = tmp1 ^ in2;
+        tmp2 = tmp1 ^ in3;
+        out3 = tmp2 ^ out4;
+        out1 = tmp2 ^ in6;
+        out2 = tmp0 ^ out5;
+        out0 = out2 ^ out3 ^ in5;
+        out7 = out1 ^ out2 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_6F(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in7;
+        tmp1 = tmp0 ^ in4;
+        tmp2 = tmp0 ^ in0 ^ in2;
+        out4 = tmp1 ^ in1;
+        out0 = tmp2 ^ in5;
+        out3 = out4 ^ in0;
+        out2 = out3 ^ in7;
+        out1 = out2 ^ in6;
+        out6 = out1 ^ in4 ^ in5;
+        out7 = tmp2 ^ out1;
+        out5 = tmp1 ^ out0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_70(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in2;
+        tmp0 = in2 ^ in4;
+        out2 = in2 ^ in3 ^ in5;
+        tmp1 = tmp0 ^ in6;
+        tmp2 = out2 ^ in7;
+        out0 = tmp1 ^ in3;
+        out4 = tmp1 ^ in0;
+        out7 = tmp2 ^ in1;
+        out6 = out4 ^ in1;
+        out5 = out7 ^ in0 ^ in2;
+        out1 = tmp0 ^ tmp2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_71(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in3 ^ in5;
+        out3 = in2 ^ in3;
+        tmp0 = in0 ^ in2;
+        tmp1 = out2 ^ in1;
+        out4 = tmp0 ^ in6;
+        tmp2 = tmp0 ^ in1;
+        out7 = tmp1 ^ in2;
+        out1 = tmp1 ^ in4 ^ in7;
+        out0 = out4 ^ in3 ^ in4;
+        out6 = tmp2 ^ in4;
+        out5 = tmp2 ^ out3 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_72(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in7;
+        tmp0 = in0 ^ in4;
+        tmp1 = tmp0 ^ in3 ^ in7;
+        out1 = tmp1 ^ in5;
+        out5 = out1 ^ in1;
+        tmp2 = tmp0 ^ out5;
+        out2 = tmp2 ^ in2;
+        out7 = out2 ^ in6;
+        out6 = tmp1 ^ out7;
+        out4 = tmp2 ^ out6;
+        out0 = out4 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_73(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in3 ^ in7;
+        out2 = out3 ^ in1 ^ in5;
+        out1 = out2 ^ in0 ^ in4;
+        out5 = out1 ^ in5;
+        out6 = out1 ^ out3 ^ in2;
+        out0 = out2 ^ out6 ^ in6;
+        out7 = out0 ^ out1 ^ in3;
+        out4 = out0 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_74(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in4;
+        tmp1 = in1 ^ in2 ^ in6;
+        out4 = in0 ^ in4 ^ in7;
+        out5 = in0 ^ in1 ^ in5;
+        out0 = tmp0 ^ in2;
+        out1 = tmp0 ^ in5;
+        out3 = tmp1 ^ in7;
+        out6 = tmp1 ^ in0;
+        out2 = tmp1 ^ out5 ^ in3;
+        out7 = out3 ^ in3 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_75(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in0 ^ in7;
+        tmp0 = in1 ^ in3;
+        out5 = in0 ^ in1;
+        out7 = tmp0 ^ in2;
+        tmp1 = tmp0 ^ in4;
+        out6 = out5 ^ in2;
+        tmp2 = out7 ^ in6;
+        out1 = tmp1 ^ in5;
+        out0 = tmp1 ^ out6;
+        out3 = tmp2 ^ in7;
+        out2 = tmp2 ^ out6 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_76(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in1 ^ in6;
+        tmp0 = in0 ^ in5;
+        tmp1 = in3 ^ in7;
+        tmp2 = tmp0 ^ in4;
+        tmp3 = tmp1 ^ in2;
+        out5 = tmp2 ^ in1;
+        out1 = tmp2 ^ in3;
+        out0 = tmp3 ^ in4;
+        out4 = out1 ^ in5;
+        out7 = tmp3 ^ out3;
+        out2 = tmp0 ^ out7;
+        out6 = tmp1 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_77(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in0 ^ in3;
+        tmp0 = in1 ^ in4;
+        tmp1 = in1 ^ in6;
+        tmp2 = out4 ^ in5;
+        out5 = tmp0 ^ in0;
+        out1 = tmp0 ^ tmp2;
+        out3 = tmp1 ^ in3;
+        out2 = tmp1 ^ tmp2 ^ in7;
+        out7 = out3 ^ in2;
+        tmp3 = out7 ^ in6;
+        out6 = tmp2 ^ tmp3;
+        out0 = tmp3 ^ out5 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_78(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in3;
+        tmp1 = in2 ^ in7;
+        tmp2 = in0 ^ in5 ^ in6;
+        out2 = tmp1 ^ in3;
+        out3 = tmp2 ^ in2;
+        out5 = out3 ^ in1 ^ in3;
+        out0 = tmp0 ^ out3 ^ in4;
+        out1 = tmp1 ^ out0;
+        out4 = out1 ^ out5 ^ in5;
+        out7 = tmp0 ^ out4;
+        out6 = tmp2 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_79(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in3 ^ in7;
+        tmp0 = in3 ^ in4;
+        tmp1 = in1 ^ in5;
+        tmp2 = tmp1 ^ in2;
+        out4 = tmp2 ^ in0 ^ in7;
+        tmp3 = out4 ^ in5;
+        out5 = tmp3 ^ out2 ^ in6;
+        out7 = tmp0 ^ tmp2;
+        out6 = tmp0 ^ tmp3;
+        out3 = tmp1 ^ out5;
+        out0 = out3 ^ in4;
+        out1 = tmp3 ^ out0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_7A(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in2;
+        out2 = tmp0 ^ in3;
+        tmp1 = out2 ^ in4;
+        out4 = tmp1 ^ in0 ^ in5;
+        out5 = out4 ^ in6;
+        out6 = out5 ^ in7;
+        out7 = out6 ^ in0;
+        out0 = out7 ^ in1;
+        out1 = tmp0 ^ out6;
+        out3 = tmp1 ^ out6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_7B(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in1 ^ in3;
+        tmp0 = in0 ^ in5;
+        out4 = tmp0 ^ out2 ^ in2;
+        tmp1 = out4 ^ in4;
+        out6 = tmp1 ^ in7;
+        out5 = tmp1 ^ in5 ^ in6;
+        out0 = out6 ^ in1 ^ in6;
+        tmp2 = out0 ^ in2;
+        out1 = tmp2 ^ in1;
+        out3 = tmp2 ^ in4;
+        out7 = tmp0 ^ out5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_7C(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in5;
+        tmp1 = tmp0 ^ in4;
+        out0 = tmp1 ^ in2;
+        out1 = tmp1 ^ in6;
+        out7 = out0 ^ in1 ^ in5 ^ in7;
+        out5 = out1 ^ out7 ^ in0;
+        out3 = out5 ^ in6;
+        out6 = tmp0 ^ out5;
+        out2 = out6 ^ in1;
+        out4 = out2 ^ out7 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_7D(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in2;
+        tmp1 = tmp0 ^ in3;
+        tmp2 = tmp0 ^ in6;
+        out7 = tmp1 ^ in4;
+        tmp3 = tmp2 ^ in0;
+        out5 = tmp3 ^ in7;
+        out4 = tmp3 ^ in2 ^ in5;
+        out2 = tmp1 ^ out5;
+        out6 = tmp2 ^ out2;
+        out0 = out4 ^ out7 ^ in6;
+        out1 = tmp3 ^ out0;
+        out3 = out6 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_7E(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in4;
+        tmp1 = in0 ^ in5;
+        out1 = tmp0 ^ tmp1 ^ in6;
+        out3 = tmp1 ^ in1;
+        out4 = out1 ^ in1 ^ in7;
+        tmp2 = out4 ^ in3;
+        out5 = tmp2 ^ in2;
+        out6 = tmp0 ^ out5;
+        out7 = tmp1 ^ out4 ^ in2;
+        out2 = out6 ^ in5 ^ in7;
+        out0 = tmp2 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_7F(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in7;
+        tmp1 = tmp0 ^ in3 ^ in5;
+        tmp2 = tmp1 ^ in0;
+        out0 = tmp2 ^ in4;
+        out6 = tmp2 ^ in1;
+        out3 = tmp0 ^ out6;
+        tmp3 = out3 ^ in6;
+        out1 = tmp3 ^ in4;
+        out2 = tmp3 ^ in5;
+        out4 = tmp3 ^ in7;
+        out5 = tmp1 ^ out1;
+        out7 = out0 ^ out4 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_80(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in3;
+        tmp1 = in4 ^ in5;
+        out1 = in2 ^ in6 ^ in7;
+        out5 = tmp0 ^ in4;
+        tmp2 = tmp0 ^ in1;
+        out6 = tmp1 ^ in3;
+        out7 = tmp1 ^ in0 ^ in6;
+        out4 = tmp2 ^ in7;
+        out3 = tmp2 ^ out6;
+        out2 = out3 ^ out5 ^ in6;
+        out0 = out2 ^ in3 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_81(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in4 ^ in6;
+        tmp1 = tmp0 ^ in3;
+        out6 = tmp1 ^ in5;
+        out5 = out6 ^ in2 ^ in6;
+        out3 = out5 ^ in1;
+        out2 = tmp0 ^ out3;
+        out1 = out3 ^ out6 ^ in7;
+        out4 = tmp1 ^ out1;
+        out7 = out2 ^ out4 ^ in0;
+        out0 = out7 ^ in1 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_82(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in1 ^ in2;
+        tmp0 = in6 ^ in7;
+        out5 = in2 ^ in3;
+        out6 = in3 ^ in4;
+        out7 = in0 ^ in4 ^ in5;
+        out0 = in1 ^ in5 ^ in6;
+        out1 = tmp0 ^ in0 ^ in2;
+        out2 = tmp0 ^ in3 ^ in5;
+        out3 = tmp0 ^ out0 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_83(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in1;
+        tmp1 = in2 ^ in5;
+        tmp2 = in3 ^ in6;
+        out4 = in1 ^ in2 ^ in4;
+        out0 = tmp0 ^ in5 ^ in6;
+        out5 = tmp1 ^ in3;
+        tmp3 = tmp1 ^ in7;
+        out6 = tmp2 ^ in4;
+        out2 = tmp2 ^ tmp3;
+        tmp4 = tmp3 ^ out4;
+        out1 = tmp3 ^ out0;
+        out3 = tmp4 ^ in3;
+        out7 = tmp0 ^ tmp4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_84(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in2 ^ in6;
+        out6 = in3 ^ in5;
+        out0 = in1 ^ in5 ^ in7;
+        out7 = in0 ^ in4 ^ in6;
+        out4 = in1 ^ in3 ^ in6;
+        out5 = in2 ^ in4 ^ in7;
+        out2 = out6 ^ in0 ^ in1;
+        out3 = out5 ^ in5 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_85(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in6;
+        tmp1 = in3 ^ in6;
+        tmp2 = tmp0 ^ in4;
+        out1 = tmp0 ^ in2;
+        out6 = tmp1 ^ in5;
+        out4 = tmp2 ^ in3;
+        tmp3 = out1 ^ out6;
+        out2 = tmp3 ^ in0;
+        out3 = tmp2 ^ tmp3 ^ in7;
+        out7 = out2 ^ out3 ^ in1;
+        out5 = tmp1 ^ out3;
+        out0 = tmp2 ^ out7 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_86(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out6 = in3;
+        out7 = in0 ^ in4;
+        out0 = in1 ^ in5;
+        out5 = in2 ^ in7;
+        out3 = in4 ^ in5 ^ in6;
+        out1 = in0 ^ in2 ^ in6;
+        out4 = in1 ^ in6 ^ in7;
+        out2 = in0 ^ in3 ^ in5 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_87(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out6 = in3 ^ in6;
+        tmp0 = in0 ^ in1;
+        out7 = in0 ^ in4 ^ in7;
+        out5 = in2 ^ in5 ^ in7;
+        out3 = out6 ^ in4 ^ in5;
+        out0 = tmp0 ^ in5;
+        tmp1 = tmp0 ^ in6;
+        out2 = out5 ^ in0 ^ in3;
+        out1 = tmp1 ^ in2;
+        out4 = tmp1 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_88(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in2 ^ in7;
+        tmp0 = in5 ^ in6;
+        out0 = in1 ^ in6 ^ in7;
+        out6 = in4 ^ in5 ^ in7;
+        out3 = out0 ^ out1 ^ in0 ^ in4;
+        out7 = tmp0 ^ in0;
+        tmp1 = tmp0 ^ in3;
+        out2 = out0 ^ in3;
+        out4 = tmp1 ^ in2;
+        out5 = tmp1 ^ out6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_89(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in7;
+        tmp1 = in2 ^ in7;
+        tmp2 = tmp0 ^ in6;
+        out1 = tmp1 ^ in1;
+        out7 = tmp2 ^ in5;
+        out0 = tmp2 ^ in1;
+        out2 = out1 ^ in3 ^ in6;
+        out6 = out7 ^ in0 ^ in4;
+        out5 = out6 ^ in3;
+        out3 = tmp0 ^ out2 ^ in4;
+        out4 = tmp1 ^ out5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_8A(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in1 ^ in6;
+        out7 = in0 ^ in5;
+        out2 = in3 ^ in6;
+        out6 = in4 ^ in7;
+        out1 = in0 ^ in2 ^ in7;
+        out3 = out0 ^ out6 ^ in0;
+        out4 = out1 ^ out7 ^ in6;
+        out5 = out2 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_8B(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in1;
+        tmp1 = in3 ^ in6;
+        tmp2 = in5 ^ in7;
+        tmp3 = tmp0 ^ in7;
+        out0 = tmp0 ^ in6;
+        out2 = tmp1 ^ in2;
+        out5 = tmp1 ^ tmp2;
+        out7 = tmp2 ^ in0;
+        tmp4 = tmp3 ^ in4;
+        out1 = tmp3 ^ in2;
+        out6 = tmp4 ^ out0;
+        out4 = out6 ^ in2 ^ in5;
+        out3 = tmp1 ^ tmp4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_8C(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in2;
+        out0 = in1 ^ in7;
+        out7 = in0 ^ in6;
+        out5 = in4 ^ in6;
+        out6 = in5 ^ in7;
+        out2 = out0 ^ in0 ^ in3;
+        out3 = out5 ^ out7 ^ in2 ^ in7;
+        out4 = out6 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_8D(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in1 ^ in2;
+        tmp0 = in6 ^ in7;
+        out0 = in0 ^ in1 ^ in7;
+        out5 = in4 ^ in5 ^ in6;
+        out6 = tmp0 ^ in5;
+        out7 = tmp0 ^ in0;
+        out4 = tmp0 ^ out5 ^ in3;
+        out2 = out0 ^ in2 ^ in3;
+        out3 = out2 ^ in1 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_8E(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in1;
+        out4 = in5;
+        out7 = in0;
+        out5 = in6;
+        out6 = in7;
+        out3 = in0 ^ in4;
+        out1 = in0 ^ in2;
+        out2 = in0 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_8F(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in0 ^ in1;
+        tmp0 = in0 ^ in3;
+        out4 = in4 ^ in5;
+        out7 = in0 ^ in7;
+        out5 = in5 ^ in6;
+        out6 = in6 ^ in7;
+        out1 = out0 ^ in2;
+        out2 = tmp0 ^ in2;
+        out3 = tmp0 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_90(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in2;
+        tmp1 = in2 ^ in6 ^ in7;
+        out3 = tmp0 ^ in7;
+        out1 = tmp1 ^ in5;
+        tmp2 = out1 ^ in4;
+        out6 = tmp2 ^ in3;
+        out5 = out6 ^ in1;
+        out4 = out5 ^ in0;
+        out0 = tmp0 ^ tmp2;
+        out7 = tmp0 ^ out4;
+        out2 = tmp1 ^ out5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_91(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in4;
+        tmp1 = tmp0 ^ in3 ^ in5;
+        out2 = tmp1 ^ in1;
+        out6 = tmp1 ^ in7;
+        tmp2 = out2 ^ in5 ^ in7;
+        out3 = tmp2 ^ in4;
+        out5 = tmp2 ^ in6;
+        out1 = tmp1 ^ out5 ^ in2;
+        tmp3 = out1 ^ in0;
+        out4 = tmp3 ^ in3;
+        out0 = tmp0 ^ tmp3;
+        out7 = tmp2 ^ tmp3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_92(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in1;
+        tmp0 = in4 ^ in5;
+        tmp1 = tmp0 ^ in1;
+        out2 = tmp0 ^ in3 ^ in7;
+        out0 = tmp1 ^ in6;
+        out7 = out2 ^ in0;
+        out4 = out0 ^ in0 ^ in2;
+        out5 = out4 ^ out7 ^ in5;
+        out6 = tmp1 ^ out5;
+        out1 = out6 ^ out7 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_93(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in1 ^ in3;
+        tmp0 = in2 ^ in7;
+        tmp1 = out3 ^ in6;
+        tmp2 = tmp0 ^ in4;
+        out5 = tmp0 ^ tmp1;
+        out6 = tmp2 ^ in3;
+        out2 = out6 ^ in5;
+        out0 = out2 ^ out5 ^ in0;
+        out7 = tmp1 ^ out0;
+        out1 = tmp2 ^ out0;
+        out4 = out1 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_94(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in2 ^ in6;
+        tmp0 = in1 ^ in4 ^ in5;
+        out1 = out3 ^ in5;
+        out5 = tmp0 ^ out3;
+        out0 = tmp0 ^ in7;
+        out4 = tmp0 ^ in0 ^ in3;
+        out6 = out1 ^ in3 ^ in7;
+        out2 = out4 ^ in6;
+        out7 = out0 ^ out2 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_95(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in3;
+        out3 = tmp0 ^ in6;
+        tmp1 = tmp0 ^ in7;
+        tmp2 = out3 ^ in0;
+        out6 = tmp1 ^ in5;
+        tmp3 = tmp2 ^ in4;
+        out7 = tmp3 ^ in2;
+        tmp4 = tmp3 ^ in5;
+        out2 = tmp4 ^ in1;
+        tmp5 = out2 ^ in6;
+        out0 = tmp1 ^ tmp5;
+        out1 = tmp5 ^ out7;
+        out4 = tmp2 ^ out1;
+        out5 = tmp4 ^ out4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_96(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in6 ^ in7;
+        tmp0 = in1 ^ in5;
+        tmp1 = in5 ^ in6;
+        out6 = out3 ^ in2 ^ in3;
+        out0 = tmp0 ^ in4;
+        tmp2 = tmp1 ^ in2;
+        out4 = out0 ^ in0 ^ in7;
+        out1 = tmp2 ^ in0;
+        out5 = tmp2 ^ in1;
+        out7 = tmp0 ^ out4 ^ in3;
+        out2 = tmp1 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_97(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in4;
+        tmp1 = in2 ^ in6;
+        out3 = in3 ^ in6 ^ in7;
+        out7 = tmp0 ^ in3;
+        tmp2 = tmp0 ^ in5;
+        out5 = tmp1 ^ in1;
+        out6 = tmp1 ^ out3;
+        out0 = tmp2 ^ in1;
+        out2 = tmp2 ^ out3 ^ in2;
+        tmp3 = out0 ^ in4;
+        out4 = tmp3 ^ in7;
+        out1 = tmp1 ^ tmp3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_98(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in5 ^ in7;
+        tmp1 = in1 ^ in4 ^ in7;
+        out1 = tmp0 ^ in2;
+        out0 = tmp1 ^ in6;
+        out2 = tmp1 ^ in3;
+        out6 = out0 ^ out1 ^ in1;
+        out5 = tmp0 ^ out2;
+        out3 = tmp1 ^ out6 ^ in0;
+        out7 = out0 ^ out5 ^ in0;
+        out4 = out6 ^ out7 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_99(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in3;
+        out5 = in1 ^ in3 ^ in4;
+        out6 = in2 ^ in4 ^ in5;
+        out4 = tmp0 ^ in2;
+        tmp1 = tmp0 ^ in6;
+        tmp2 = out5 ^ in7;
+        out7 = tmp1 ^ in5;
+        out0 = tmp1 ^ tmp2;
+        out2 = tmp2 ^ in2;
+        out3 = out0 ^ out6 ^ in3;
+        out1 = tmp1 ^ out3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_9A(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in3 ^ in4;
+        tmp0 = in0 ^ in5;
+        tmp1 = in1 ^ in6;
+        out5 = in1 ^ in3 ^ in5;
+        tmp2 = tmp0 ^ in7;
+        out3 = tmp0 ^ tmp1;
+        out0 = tmp1 ^ in4;
+        out7 = tmp2 ^ in3;
+        out1 = tmp2 ^ in2;
+        out6 = out0 ^ in1 ^ in2;
+        out4 = out1 ^ in4 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_9B(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out5 = in1 ^ in3;
+        tmp0 = in3 ^ in5;
+        out6 = in2 ^ in4;
+        out4 = in0 ^ in2 ^ in7;
+        out7 = tmp0 ^ in0;
+        out2 = out6 ^ in3;
+        out1 = out4 ^ in1 ^ in5;
+        out3 = out7 ^ in1 ^ in6;
+        out0 = tmp0 ^ out3 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_9C(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in2 ^ in5;
+        tmp0 = in0 ^ in3 ^ in6;
+        out3 = out1 ^ in0;
+        out6 = out1 ^ in6;
+        out7 = tmp0 ^ in7;
+        out4 = out7 ^ in4;
+        out2 = out4 ^ in1;
+        out0 = tmp0 ^ out2;
+        out5 = out0 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_9D(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out6 = in2 ^ in5;
+        tmp0 = in0 ^ in3;
+        out5 = in1 ^ in4 ^ in7;
+        out1 = out6 ^ in1;
+        out3 = tmp0 ^ out6;
+        out7 = tmp0 ^ in6;
+        out0 = out5 ^ in0;
+        out4 = out7 ^ in7;
+        out2 = out5 ^ out7 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_9E(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in1 ^ in4;
+        tmp0 = in0 ^ in5;
+        out6 = in2 ^ in6;
+        out7 = in0 ^ in3 ^ in7;
+        out4 = in0 ^ in4 ^ in6;
+        out5 = in1 ^ in5 ^ in7;
+        out1 = tmp0 ^ in2;
+        out3 = tmp0 ^ in7;
+        out2 = out4 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_9F(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out6 = in2;
+        out7 = in0 ^ in3;
+        tmp0 = in0 ^ in1;
+        out4 = in0 ^ in6;
+        out5 = in1 ^ in7;
+        out1 = tmp0 ^ in2 ^ in5;
+        out2 = out7 ^ in2 ^ in4 ^ in6;
+        out3 = out7 ^ in5 ^ in7;
+        out0 = tmp0 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_A0(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in6;
+        out2 = tmp0 ^ in7;
+        tmp1 = tmp0 ^ in5;
+        out6 = out2 ^ in3 ^ in4;
+        out0 = tmp1 ^ in3;
+        tmp2 = out0 ^ in2;
+        out3 = tmp2 ^ in7;
+        tmp3 = tmp2 ^ in1;
+        out5 = tmp3 ^ in0;
+        out4 = tmp3 ^ out6;
+        out7 = out5 ^ out6 ^ in1;
+        out1 = tmp1 ^ out4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_A1(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in5;
+        tmp1 = tmp0 ^ in1;
+        tmp2 = tmp0 ^ in4;
+        out4 = tmp1 ^ in7;
+        out7 = tmp2 ^ in0;
+        out6 = tmp2 ^ out4 ^ in3;
+        out3 = out4 ^ in6;
+        out2 = out3 ^ in5;
+        out1 = out2 ^ in4;
+        out5 = out1 ^ out6 ^ in0;
+        out0 = tmp1 ^ out5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_A2(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in6;
+        tmp0 = in1 ^ in3 ^ in5;
+        out3 = tmp0 ^ in6;
+        out4 = tmp0 ^ in2 ^ in4;
+        out0 = out3 ^ in7;
+        out6 = out0 ^ in4;
+        out1 = out0 ^ out4 ^ in0;
+        out7 = out1 ^ in5;
+        out5 = out7 ^ in3 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_A3(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in2 ^ in6;
+        out3 = in1 ^ in5 ^ in6;
+        tmp0 = out2 ^ in0;
+        out4 = out2 ^ out3 ^ in3;
+        tmp1 = tmp0 ^ in4;
+        out0 = tmp0 ^ out4 ^ in7;
+        out5 = tmp1 ^ in3;
+        out7 = tmp1 ^ in5;
+        out1 = tmp1 ^ in1 ^ in7;
+        out6 = tmp1 ^ out0 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_A4(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in3;
+        tmp1 = in2 ^ in4;
+        tmp2 = in2 ^ in5;
+        tmp3 = in0 ^ in7;
+        out0 = tmp0 ^ in5;
+        out6 = tmp0 ^ in6 ^ in7;
+        out1 = tmp1 ^ in6;
+        out7 = tmp1 ^ tmp3;
+        out3 = tmp2 ^ in3;
+        tmp4 = tmp2 ^ out1;
+        out2 = tmp3 ^ in1;
+        out5 = tmp4 ^ out7;
+        out4 = tmp4 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_A5(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in2 ^ in5;
+        tmp0 = in1 ^ in6;
+        tmp1 = in0 ^ in1;
+        tmp2 = in2 ^ in4;
+        out6 = in1 ^ in3 ^ in7;
+        out4 = tmp0 ^ in5;
+        out1 = tmp0 ^ tmp2;
+        out0 = tmp1 ^ in3 ^ in5;
+        out2 = tmp1 ^ in2 ^ in7;
+        out7 = tmp2 ^ in0;
+        out5 = tmp0 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_A6(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in0;
+        out3 = in3 ^ in5 ^ in7;
+        out1 = in0 ^ in2 ^ in4 ^ in6;
+        out0 = out3 ^ in1;
+        out7 = out1 ^ in7;
+        out6 = out0 ^ in6;
+        out5 = out7 ^ in5;
+        out4 = out6 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_A7(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in0 ^ in2;
+        out3 = in5 ^ in7;
+        out7 = out2 ^ in4 ^ in6;
+        out6 = out3 ^ in1 ^ in3;
+        out1 = out7 ^ in1;
+        out5 = out7 ^ in7;
+        out0 = out6 ^ in0;
+        out4 = out6 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_A8(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in4;
+        tmp1 = in1 ^ in6;
+        tmp2 = in0 ^ in2 ^ in7;
+        out1 = tmp0 ^ in7;
+        out4 = tmp0 ^ in6;
+        out0 = tmp1 ^ in3;
+        out2 = tmp1 ^ in5;
+        out6 = tmp1 ^ in4;
+        out7 = tmp2 ^ in5;
+        out3 = tmp2 ^ out0 ^ in6;
+        out5 = out7 ^ in2 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_A9(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in2 ^ in6;
+        out6 = in1 ^ in4;
+        out7 = in0 ^ in2 ^ in5;
+        out5 = in0 ^ in3 ^ in7;
+        out2 = out4 ^ in1 ^ in5;
+        out1 = out6 ^ in2 ^ in7;
+        out0 = out2 ^ out7 ^ in3;
+        out3 = out1 ^ in0 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_AA(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in2;
+        tmp1 = in1 ^ in3;
+        tmp2 = in6 ^ in7;
+        out1 = tmp0 ^ in4 ^ in7;
+        out3 = tmp1 ^ in0;
+        out0 = tmp1 ^ tmp2;
+        out2 = tmp2 ^ in5;
+        out7 = tmp0 ^ out2;
+        out6 = out1 ^ out7 ^ in1;
+        out5 = out0 ^ out6 ^ in0;
+        out4 = out5 ^ out7 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_AB(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in0 ^ in1;
+        tmp0 = in1 ^ in4;
+        tmp1 = in0 ^ in7;
+        out6 = tmp0 ^ in5;
+        out1 = tmp0 ^ tmp1 ^ in2;
+        out5 = tmp1 ^ in3 ^ in4;
+        out0 = tmp0 ^ out5 ^ in6;
+        out4 = out0 ^ out3 ^ in2;
+        out2 = out4 ^ in3 ^ in5;
+        out7 = tmp1 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_AC(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in1 ^ in3;
+        out1 = in2 ^ in4;
+        tmp0 = in0 ^ in2;
+        out4 = in4 ^ in7;
+        out5 = in0 ^ in5;
+        out6 = in1 ^ in6;
+        out7 = tmp0 ^ in7;
+        out3 = tmp0 ^ in3 ^ in6;
+        out2 = out5 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_AD(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in7;
+        out5 = in0;
+        out6 = in1;
+        out7 = in0 ^ in2;
+        out0 = in0 ^ in1 ^ in3;
+        out2 = out7 ^ in1 ^ in5;
+        out1 = in1 ^ in2 ^ in4;
+        out3 = out7 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_AE(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in3 ^ in4;
+        tmp0 = in0 ^ in4;
+        tmp1 = in0 ^ in7;
+        out0 = in1 ^ in3 ^ in7;
+        out1 = tmp0 ^ in2;
+        out5 = tmp0 ^ in5;
+        tmp2 = tmp1 ^ in6;
+        out2 = tmp1 ^ in5;
+        out3 = tmp2 ^ in3;
+        out7 = tmp2 ^ in2;
+        out6 = tmp2 ^ out2 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_AF(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in3;
+        tmp0 = in0 ^ in7;
+        out5 = in0 ^ in4;
+        out6 = in1 ^ in5;
+        out7 = in0 ^ in2 ^ in6;
+        out0 = tmp0 ^ in1 ^ in3;
+        out3 = tmp0 ^ in6;
+        out2 = tmp0 ^ in2 ^ in5;
+        out1 = out5 ^ in1 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_B0(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in4;
+        tmp1 = in3 ^ in6;
+        out2 = tmp0 ^ in7;
+        tmp2 = tmp0 ^ tmp1;
+        out0 = tmp2 ^ in5;
+        out3 = tmp2 ^ in2;
+        out6 = out3 ^ in6;
+        tmp3 = out6 ^ in0 ^ in1;
+        out7 = tmp3 ^ in5;
+        out5 = tmp3 ^ out2;
+        out1 = out0 ^ out5 ^ in0;
+        out4 = tmp1 ^ out5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_B1(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in4;
+        out2 = tmp0 ^ in2 ^ in7;
+        tmp1 = out2 ^ in6;
+        out1 = tmp1 ^ in5;
+        out3 = tmp1 ^ in7;
+        out4 = tmp1 ^ in0;
+        out6 = out3 ^ in3;
+        out0 = out6 ^ in0 ^ in2 ^ in5;
+        out5 = tmp1 ^ out0 ^ in1;
+        out7 = tmp0 ^ out5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_B2(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in4;
+        tmp0 = in4 ^ in7;
+        tmp1 = in1 ^ in3 ^ in6;
+        out3 = tmp0 ^ tmp1;
+        tmp2 = tmp1 ^ in0;
+        out0 = out3 ^ in5;
+        out4 = tmp2 ^ in2;
+        tmp3 = out4 ^ in6;
+        out5 = tmp0 ^ tmp3;
+        out1 = tmp3 ^ out0;
+        tmp4 = out1 ^ in7;
+        out7 = tmp4 ^ in3;
+        out6 = tmp2 ^ tmp4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_B3(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in2 ^ in4;
+        tmp0 = in0 ^ in5;
+        tmp1 = in1 ^ in6;
+        out3 = tmp1 ^ in4 ^ in7;
+        tmp2 = tmp0 ^ out3;
+        out0 = tmp2 ^ in3;
+        out1 = tmp2 ^ in2;
+        out5 = out0 ^ in2 ^ in6;
+        out7 = tmp1 ^ out5;
+        out4 = out7 ^ in1 ^ in5 ^ in7;
+        out6 = tmp0 ^ out4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_B4(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in0 ^ in1;
+        out5 = out4 ^ in2;
+        tmp0 = out4 ^ in4;
+        out6 = out5 ^ in0 ^ in3;
+        out7 = tmp0 ^ out6;
+        out2 = tmp0 ^ in6 ^ in7;
+        out3 = out7 ^ in0 ^ in7;
+        out0 = out5 ^ out7 ^ in5;
+        out1 = out0 ^ out6 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_B5(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in1;
+        tmp1 = in2 ^ in4;
+        out4 = tmp0 ^ in4;
+        out3 = tmp1 ^ in7;
+        tmp2 = out4 ^ in5;
+        out7 = out3 ^ in0 ^ in3;
+        out0 = tmp2 ^ in3;
+        out2 = tmp0 ^ out3 ^ in6;
+        out5 = tmp1 ^ tmp2;
+        out6 = out2 ^ out7 ^ in2;
+        out1 = tmp0 ^ out0 ^ out6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_B6(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in3 ^ in4;
+        tmp0 = in1 ^ in2;
+        tmp1 = in0 ^ in4;
+        tmp2 = in3 ^ in5;
+        tmp3 = out3 ^ in1 ^ in7;
+        out5 = tmp0 ^ tmp1;
+        out6 = tmp0 ^ tmp2;
+        out2 = tmp1 ^ in6;
+        out4 = tmp1 ^ tmp3;
+        out0 = tmp3 ^ in5;
+        out1 = out2 ^ in2 ^ in5;
+        out7 = tmp2 ^ out1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_B7(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in4;
+        tmp0 = in0 ^ in4;
+        out2 = tmp0 ^ in2 ^ in6;
+        tmp1 = out2 ^ in7;
+        out1 = out2 ^ in1 ^ in5;
+        out7 = tmp1 ^ in3;
+        out5 = out1 ^ in6;
+        out6 = tmp0 ^ out1 ^ in3;
+        out0 = tmp1 ^ out6;
+        out4 = out0 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_B8(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in4;
+        tmp1 = in2 ^ in5;
+        out2 = tmp0 ^ in5;
+        out4 = tmp1 ^ in0;
+        tmp2 = tmp1 ^ in7;
+        out6 = tmp2 ^ out2;
+        out7 = out4 ^ in3;
+        out1 = tmp2 ^ in4;
+        out3 = tmp0 ^ out7;
+        out0 = out3 ^ out4 ^ in6;
+        out5 = out0 ^ in0 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_B9(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in2;
+        tmp1 = in4 ^ in5;
+        out4 = tmp0 ^ tmp1;
+        tmp2 = tmp0 ^ in3 ^ in7;
+        out3 = out4 ^ in1;
+        out7 = tmp2 ^ in5;
+        out2 = out3 ^ in0;
+        out1 = out2 ^ in7;
+        out6 = out1 ^ in5 ^ in6;
+        out0 = tmp2 ^ out6;
+        out5 = tmp1 ^ out0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_BA(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in5 ^ in7;
+        out2 = tmp0 ^ in4;
+        tmp1 = out2 ^ in2;
+        out1 = tmp1 ^ in0;
+        out6 = tmp1 ^ in1;
+        out4 = out1 ^ in3 ^ in4;
+        tmp2 = out4 ^ out6;
+        out7 = out4 ^ in6 ^ in7;
+        out5 = tmp2 ^ in6;
+        out3 = tmp0 ^ tmp2;
+        out0 = out6 ^ out7 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_BB(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in2 ^ in4 ^ in5 ^ in7;
+        tmp0 = out2 ^ in1;
+        out4 = out2 ^ in0 ^ in3;
+        out1 = tmp0 ^ in0;
+        out6 = tmp0 ^ in6;
+        out3 = out1 ^ in2;
+        tmp1 = out4 ^ out6 ^ in4;
+        out0 = tmp1 ^ in7;
+        out5 = tmp1 ^ in5;
+        out7 = tmp0 ^ tmp1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_BC(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in2;
+        tmp1 = in2 ^ in4;
+        out0 = in1 ^ in3 ^ in4;
+        out6 = in1 ^ in2 ^ in7;
+        out7 = tmp0 ^ in3;
+        out5 = tmp0 ^ out6 ^ in6;
+        out1 = tmp1 ^ in5;
+        tmp2 = out1 ^ out5 ^ in1;
+        out3 = tmp2 ^ in3;
+        out4 = tmp1 ^ tmp2;
+        out2 = tmp2 ^ out6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_BD(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in3;
+        tmp1 = in1 ^ in4;
+        out0 = tmp0 ^ tmp1;
+        out7 = tmp0 ^ in2 ^ in7;
+        out1 = tmp1 ^ in2 ^ in5;
+        tmp2 = out1 ^ in0;
+        out2 = tmp2 ^ in6;
+        out3 = out2 ^ in1 ^ in7;
+        out4 = out3 ^ in2;
+        out5 = tmp1 ^ out4;
+        out6 = tmp2 ^ out4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_BE(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in3 ^ in6;
+        out4 = tmp0 ^ in5;
+        out7 = tmp0 ^ in2;
+        out3 = out4 ^ in4;
+        out1 = out3 ^ out7 ^ in0;
+        out2 = out3 ^ in3 ^ in7;
+        out0 = out2 ^ out4 ^ in1;
+        out5 = tmp0 ^ out0;
+        out6 = out1 ^ out5 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_BF(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in4;
+        out3 = tmp0 ^ in5 ^ in6;
+        out4 = out3 ^ in3;
+        tmp1 = out3 ^ in7;
+        out2 = tmp1 ^ in2;
+        out5 = tmp1 ^ in1;
+        tmp2 = out2 ^ in5;
+        out7 = tmp2 ^ in3 ^ in4;
+        tmp3 = tmp0 ^ out5;
+        out0 = tmp3 ^ out4;
+        out1 = tmp2 ^ tmp3;
+        out6 = tmp3 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_C0(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out5 = in2 ^ in5;
+        tmp0 = in1 ^ in4;
+        tmp1 = in3 ^ in6;
+        out0 = out5 ^ in1;
+        out4 = tmp0 ^ in7;
+        out3 = tmp0 ^ tmp1;
+        out1 = tmp1 ^ in2;
+        out6 = tmp1 ^ in0;
+        out7 = out4 ^ in0;
+        out2 = out4 ^ out5 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_C1(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out5 = in2;
+        tmp0 = in0 ^ in1;
+        out4 = in1 ^ in7;
+        out6 = in0 ^ in3;
+        out3 = in1 ^ in4 ^ in6;
+        tmp1 = tmp0 ^ in2;
+        out7 = tmp0 ^ in4;
+        out0 = tmp1 ^ in5;
+        out1 = tmp1 ^ out6 ^ in6;
+        out2 = out6 ^ out7 ^ in5 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_C2(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in1 ^ in3 ^ in4;
+        tmp0 = in0 ^ in3 ^ in6;
+        out5 = in2 ^ in4 ^ in5;
+        tmp1 = out4 ^ in7;
+        out1 = tmp0 ^ in2;
+        out6 = tmp0 ^ in5;
+        out2 = out5 ^ in3;
+        out7 = tmp0 ^ tmp1;
+        out3 = tmp1 ^ in2 ^ in6;
+        out0 = tmp1 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_C3(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in1 ^ in3;
+        tmp0 = in0 ^ in2;
+        tmp1 = in3 ^ in5;
+        out5 = in2 ^ in4;
+        tmp2 = tmp0 ^ out4;
+        out2 = tmp1 ^ in4;
+        out6 = tmp1 ^ in0;
+        out0 = tmp1 ^ tmp2 ^ in7;
+        out1 = tmp2 ^ in6;
+        out7 = out1 ^ out5 ^ in3;
+        out3 = tmp0 ^ out7 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_C4(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in7;
+        out3 = tmp0 ^ in4;
+        tmp1 = tmp0 ^ in2;
+        out1 = tmp1 ^ in6;
+        out5 = tmp1 ^ in5;
+        out4 = out1 ^ out3 ^ in1;
+        out0 = out4 ^ in4 ^ in5;
+        out2 = out0 ^ out3 ^ in0;
+        out7 = out1 ^ out2 ^ in7;
+        out6 = tmp1 ^ out0 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_C5(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in4 ^ in7;
+        tmp0 = in3 ^ in7;
+        out4 = in1 ^ in2 ^ in6;
+        out6 = in0 ^ in3 ^ in4;
+        out5 = tmp0 ^ in2;
+        out1 = tmp0 ^ out4;
+        out0 = out4 ^ in0 ^ in5;
+        out2 = out0 ^ out5 ^ in4;
+        out7 = tmp0 ^ out2 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_C6(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in5 ^ in6;
+        tmp1 = in1 ^ in7;
+        tmp2 = tmp0 ^ in0;
+        tmp3 = tmp0 ^ tmp1;
+        tmp4 = tmp2 ^ in4;
+        out0 = tmp3 ^ in2;
+        out6 = tmp4 ^ in3;
+        out2 = out6 ^ in2;
+        out7 = tmp1 ^ tmp4;
+        out3 = tmp2 ^ out2;
+        tmp5 = out3 ^ in5;
+        out5 = tmp5 ^ in7;
+        out4 = tmp3 ^ tmp5;
+        out1 = tmp4 ^ out5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_C7(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in2 ^ in4;
+        tmp0 = in3 ^ in5;
+        tmp1 = out3 ^ in7;
+        out6 = tmp0 ^ in0 ^ in4;
+        out5 = tmp1 ^ in3;
+        out2 = out6 ^ in6;
+        out7 = out2 ^ in1 ^ in3;
+        out0 = tmp1 ^ out7;
+        out1 = tmp0 ^ out0;
+        out4 = out1 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_C8(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in1 ^ in2;
+        out1 = in2 ^ in3;
+        tmp0 = in5 ^ in6;
+        tmp1 = in0 ^ in7;
+        out2 = out1 ^ in1 ^ in4;
+        out4 = tmp0 ^ in4;
+        out5 = tmp0 ^ in7;
+        out6 = tmp1 ^ in6;
+        out7 = tmp1 ^ in1;
+        out3 = out2 ^ in0 ^ in2 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_C9(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in5 ^ in6;
+        out7 = in0 ^ in1;
+        tmp0 = in1 ^ in3;
+        out5 = in6 ^ in7;
+        out6 = in0 ^ in7;
+        out0 = out7 ^ in2;
+        out3 = out7 ^ in4 ^ in5;
+        out1 = tmp0 ^ in2;
+        out2 = tmp0 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_CA(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in7;
+        tmp1 = in2 ^ in7;
+        tmp2 = tmp0 ^ in6;
+        out0 = tmp1 ^ in1;
+        tmp3 = tmp1 ^ in3;
+        out6 = tmp2 ^ in5;
+        out7 = tmp2 ^ in1;
+        out2 = tmp3 ^ in4;
+        out5 = out6 ^ in0 ^ in4;
+        out4 = out5 ^ in3;
+        out1 = tmp0 ^ tmp3;
+        out3 = tmp3 ^ out5 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_CB(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in4 ^ in7;
+        tmp1 = in5 ^ in7;
+        out7 = in0 ^ in1 ^ in6;
+        out5 = tmp0 ^ in6;
+        out2 = tmp0 ^ in3;
+        out6 = tmp1 ^ in0;
+        out4 = tmp1 ^ in3 ^ in6;
+        tmp2 = out5 ^ out7 ^ in2;
+        out1 = tmp2 ^ out2;
+        out0 = tmp2 ^ in4;
+        out3 = tmp2 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_CC(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in5;
+        tmp1 = in1 ^ in6;
+        out1 = in2 ^ in3 ^ in7;
+        out5 = tmp0 ^ in6;
+        out0 = tmp1 ^ in2;
+        tmp2 = out5 ^ in0 ^ in7;
+        out3 = tmp2 ^ in4;
+        out6 = tmp0 ^ out3;
+        out7 = tmp1 ^ tmp2 ^ in3;
+        tmp3 = out1 ^ out6;
+        out4 = tmp2 ^ tmp3;
+        out2 = tmp3 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_CD(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out5 = in3 ^ in6;
+        tmp0 = in0 ^ in1;
+        tmp1 = in2 ^ in7;
+        out6 = in0 ^ in4 ^ in7;
+        out2 = tmp0 ^ out5 ^ in4;
+        out7 = tmp0 ^ in5;
+        out0 = tmp0 ^ in2 ^ in6;
+        out4 = tmp1 ^ in5;
+        out1 = tmp1 ^ in1 ^ in3;
+        out3 = out6 ^ in5 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_CE(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in5;
+        tmp1 = tmp0 ^ in3;
+        out4 = tmp1 ^ in4;
+        tmp2 = out4 ^ in6;
+        out3 = tmp2 ^ in0;
+        out5 = tmp2 ^ in2;
+        out2 = out3 ^ in5 ^ in7;
+        out6 = tmp1 ^ out2;
+        out7 = out2 ^ out4 ^ in1;
+        out1 = tmp2 ^ out6;
+        out0 = tmp0 ^ out7 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_CF(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in6;
+        tmp1 = in0 ^ in1 ^ in5;
+        out4 = in2 ^ in3 ^ in5;
+        out5 = tmp0 ^ in4;
+        out7 = tmp1 ^ in6;
+        out1 = tmp1 ^ out4 ^ in7;
+        tmp2 = out5 ^ in0;
+        out2 = tmp2 ^ in7;
+        out3 = tmp2 ^ out4;
+        out6 = tmp0 ^ out2 ^ in5;
+        out0 = tmp0 ^ out1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_D0(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in3;
+        tmp1 = in1 ^ in4;
+        tmp2 = in2 ^ in5;
+        out7 = tmp0 ^ tmp1;
+        out0 = tmp1 ^ tmp2;
+        tmp3 = tmp2 ^ in3;
+        out1 = tmp3 ^ in6;
+        tmp4 = out1 ^ in1;
+        out2 = tmp4 ^ in7;
+        out3 = out2 ^ in2;
+        out4 = tmp0 ^ out3;
+        out5 = tmp3 ^ out3;
+        out6 = tmp4 ^ out4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_D1(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in5 ^ in6;
+        tmp1 = tmp0 ^ in1;
+        out1 = tmp1 ^ in2;
+        out2 = tmp1 ^ in7;
+        out3 = out2 ^ in3;
+        out5 = out3 ^ in2;
+        tmp2 = out3 ^ in0;
+        out4 = tmp2 ^ in4;
+        out7 = tmp0 ^ out4;
+        out6 = tmp2 ^ out1 ^ in6;
+        out0 = out2 ^ out6 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_D2(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in5 ^ in6;
+        out2 = tmp0 ^ in2 ^ in3;
+        out1 = out2 ^ in0;
+        out3 = out2 ^ in1;
+        out4 = out1 ^ in1 ^ in2;
+        out6 = out1 ^ in6 ^ in7;
+        out7 = out4 ^ in4 ^ in5;
+        out5 = out4 ^ out6 ^ in4;
+        out0 = tmp0 ^ out5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_D3(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in3 ^ in5 ^ in6;
+        tmp0 = out2 ^ in2;
+        tmp1 = tmp0 ^ in1;
+        out1 = tmp1 ^ in0;
+        out3 = tmp1 ^ in3;
+        out4 = out1 ^ in2 ^ in4;
+        tmp2 = out4 ^ in5;
+        out7 = tmp2 ^ in7;
+        out0 = tmp0 ^ out7;
+        tmp3 = out0 ^ in0;
+        out5 = tmp3 ^ in6;
+        out6 = tmp2 ^ tmp3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_D4(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in3 ^ in5;
+        tmp0 = in1 ^ in5;
+        tmp1 = tmp0 ^ in2;
+        out4 = tmp1 ^ in0;
+        tmp2 = tmp1 ^ in6;
+        out2 = out4 ^ in3 ^ in7;
+        out0 = tmp2 ^ in4;
+        out5 = tmp2 ^ out3;
+        out1 = tmp0 ^ out5 ^ in7;
+        out6 = tmp0 ^ out2 ^ in4;
+        out7 = tmp1 ^ out6 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_D5(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in5;
+        tmp0 = in0 ^ in4;
+        tmp1 = tmp0 ^ in1 ^ in5;
+        out4 = tmp1 ^ in2;
+        out0 = out4 ^ in6;
+        tmp2 = tmp0 ^ out0;
+        out5 = tmp2 ^ in3;
+        out1 = out5 ^ in7;
+        out6 = tmp1 ^ out1;
+        out7 = tmp2 ^ out6;
+        out2 = out7 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_D6(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in2 ^ in4 ^ in6;
+        out5 = tmp0 ^ in3;
+        out0 = tmp0 ^ in5 ^ in7;
+        out3 = out0 ^ out5 ^ in2;
+        tmp1 = out3 ^ in0;
+        out1 = tmp1 ^ in6;
+        out2 = tmp1 ^ in7;
+        out4 = tmp1 ^ in1;
+        out6 = tmp1 ^ in4;
+        out7 = tmp0 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_D7(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in3;
+        out3 = in2 ^ in5 ^ in7;
+        out2 = tmp0 ^ in5;
+        tmp1 = tmp0 ^ out3 ^ in1;
+        out1 = tmp1 ^ in6;
+        out4 = tmp1 ^ in4;
+        tmp2 = out1 ^ in4;
+        out6 = tmp2 ^ in1;
+        out7 = tmp2 ^ in2;
+        out0 = tmp2 ^ in3;
+        out5 = tmp2 ^ in0 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_D8(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in0;
+        out5 = in1;
+        tmp0 = in1 ^ in2;
+        out6 = in0 ^ in2;
+        out0 = tmp0 ^ in4;
+        tmp1 = tmp0 ^ in3;
+        out7 = tmp1 ^ out6;
+        out2 = tmp1 ^ in6;
+        out3 = out7 ^ in7;
+        out1 = tmp1 ^ in1 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_D9(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in0 ^ in4;
+        out5 = in1 ^ in5;
+        out2 = in1 ^ in3 ^ in6;
+        out3 = in0 ^ in1 ^ in7;
+        out6 = in0 ^ in2 ^ in6;
+        out0 = out4 ^ in1 ^ in2;
+        out1 = out5 ^ in2 ^ in3;
+        out7 = out3 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_DA(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out5 = in1 ^ in4;
+        tmp0 = in2 ^ in7;
+        tmp1 = in0 ^ in2 ^ in3;
+        out0 = tmp0 ^ out5;
+        out4 = tmp0 ^ tmp1;
+        out2 = tmp0 ^ in3 ^ in6;
+        out1 = tmp1 ^ in5;
+        out3 = tmp1 ^ in1;
+        out6 = out1 ^ in3;
+        out7 = out3 ^ in2 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_DB(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in1;
+        tmp1 = in1 ^ in5;
+        tmp2 = in3 ^ in7;
+        out3 = tmp0 ^ in2;
+        out5 = tmp1 ^ in4;
+        out6 = tmp1 ^ out3 ^ in6;
+        out2 = tmp2 ^ in6;
+        tmp3 = tmp2 ^ in4;
+        tmp4 = out3 ^ in3;
+        out4 = tmp3 ^ in0;
+        out1 = tmp4 ^ in5;
+        out0 = tmp3 ^ tmp4;
+        out7 = tmp0 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_DC(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in2;
+        tmp1 = in0 ^ in3;
+        out6 = tmp0 ^ in4;
+        tmp2 = tmp0 ^ in7;
+        out3 = tmp1 ^ in6;
+        tmp3 = tmp1 ^ in1;
+        out1 = tmp1 ^ tmp2 ^ in5;
+        out4 = tmp2 ^ in6;
+        out2 = tmp3 ^ in2;
+        out7 = tmp3 ^ in5;
+        out5 = tmp2 ^ out2;
+        out0 = out2 ^ out3 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_DD(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in0 ^ in6;
+        out2 = in0 ^ in1 ^ in3;
+        out6 = out3 ^ in2 ^ in4;
+        out7 = out2 ^ in5 ^ in7;
+        out0 = out6 ^ in1;
+        out4 = out6 ^ in7;
+        out5 = out7 ^ in0;
+        out1 = out5 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_DE(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in3 ^ in6;
+        tmp1 = in3 ^ in4 ^ in7;
+        out4 = tmp0 ^ in0;
+        out5 = tmp1 ^ in1;
+        out3 = out4 ^ in7;
+        out2 = out3 ^ in6;
+        out1 = out2 ^ in5;
+        out6 = tmp1 ^ out1;
+        out0 = tmp0 ^ out5;
+        out7 = out0 ^ out1 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_DF(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in0 ^ in3 ^ in7;
+        tmp0 = out2 ^ in1 ^ in5;
+        out1 = tmp0 ^ in2;
+        out7 = tmp0 ^ in6;
+        out5 = tmp0 ^ in0 ^ in4;
+        tmp1 = out1 ^ out5 ^ in6;
+        out4 = tmp1 ^ in3;
+        out6 = tmp1 ^ in5;
+        tmp2 = tmp1 ^ in7;
+        out0 = tmp2 ^ in1;
+        out3 = tmp2 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_E0(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in1 ^ in7;
+        tmp0 = in2 ^ in4;
+        out4 = out3 ^ in3 ^ in5;
+        out2 = tmp0 ^ in1;
+        tmp1 = tmp0 ^ in6;
+        out0 = out4 ^ in2;
+        out6 = out4 ^ in0;
+        out1 = tmp1 ^ in3;
+        out5 = tmp1 ^ in0;
+        out7 = out5 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_E1(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in1 ^ in4;
+        tmp0 = in1 ^ in7;
+        out3 = tmp0 ^ in3;
+        tmp1 = out3 ^ in5;
+        out4 = tmp1 ^ in4;
+        tmp2 = tmp1 ^ in0;
+        out0 = tmp2 ^ in2;
+        out6 = tmp2 ^ in6;
+        tmp3 = out0 ^ out4 ^ in6;
+        out5 = tmp3 ^ in5;
+        out7 = tmp0 ^ tmp3;
+        out1 = tmp2 ^ out5 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_E2(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in1 ^ in2;
+        out4 = in1 ^ in5;
+        out2 = in2 ^ in4 ^ in7;
+        out5 = in0 ^ in2 ^ in6;
+        out0 = out3 ^ in3 ^ in5;
+        out7 = out3 ^ in0 ^ in4;
+        out6 = out2 ^ out7 ^ in3;
+        out1 = out5 ^ in3 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_E3(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in4 ^ in7;
+        tmp0 = in1 ^ in3;
+        out3 = tmp0 ^ in2;
+        tmp1 = out3 ^ in0;
+        out0 = tmp1 ^ in5;
+        tmp2 = tmp1 ^ in4;
+        out1 = tmp2 ^ in6;
+        tmp3 = tmp2 ^ in3;
+        out7 = tmp3 ^ in7;
+        out6 = out1 ^ out2 ^ in2;
+        tmp4 = tmp0 ^ out0;
+        out5 = tmp4 ^ in6;
+        out4 = tmp3 ^ tmp4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_E4(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in6;
+        tmp0 = in0 ^ in4;
+        tmp1 = tmp0 ^ in2 ^ in6;
+        out2 = tmp1 ^ in1;
+        out7 = out2 ^ in5;
+        tmp2 = tmp0 ^ out7;
+        out4 = tmp2 ^ in3;
+        out0 = out4 ^ in7;
+        out6 = tmp1 ^ out0;
+        out5 = tmp2 ^ out6;
+        out1 = out5 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_E5(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in3 ^ in6;
+        tmp0 = in0 ^ in1;
+        tmp1 = in5 ^ in7;
+        out2 = tmp0 ^ in4 ^ in6;
+        tmp2 = tmp1 ^ out2;
+        out6 = tmp2 ^ in3;
+        out7 = tmp2 ^ in2;
+        out0 = out6 ^ in2 ^ in4;
+        out5 = out6 ^ in1 ^ in2;
+        out1 = tmp0 ^ out5 ^ in5;
+        out4 = tmp1 ^ out1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_E6(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in2 ^ in6 ^ in7;
+        out2 = out3 ^ in0 ^ in4;
+        out4 = out3 ^ in1 ^ in5;
+        out1 = out2 ^ in3;
+        out7 = out2 ^ out4 ^ in2;
+        out0 = out4 ^ in3 ^ in7;
+        out5 = out1 ^ in4;
+        out6 = out0 ^ out2 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_E7(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in3;
+        out3 = tmp0 ^ in6 ^ in7;
+        tmp1 = out3 ^ in0;
+        out5 = tmp1 ^ in5;
+        tmp2 = tmp1 ^ in4;
+        tmp3 = out5 ^ in7;
+        out1 = tmp2 ^ in1;
+        out0 = tmp3 ^ in1;
+        out6 = out1 ^ in2;
+        out2 = tmp0 ^ tmp2;
+        tmp4 = tmp3 ^ out6;
+        out4 = tmp4 ^ in6;
+        out7 = tmp4 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_E8(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in3 ^ in6;
+        tmp0 = in4 ^ in7;
+        out1 = in2 ^ in3 ^ in4;
+        out5 = tmp0 ^ in0;
+        tmp1 = tmp0 ^ in1;
+        tmp2 = tmp1 ^ in5;
+        out0 = tmp1 ^ out1;
+        out2 = tmp2 ^ in2;
+        out6 = tmp2 ^ out5;
+        tmp3 = out6 ^ in6;
+        out3 = tmp3 ^ in7;
+        out7 = tmp3 ^ in2 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_E9(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in1;
+        tmp1 = in3 ^ in6;
+        tmp2 = tmp0 ^ in6;
+        out4 = tmp1 ^ in4;
+        out6 = tmp2 ^ in5;
+        out7 = tmp2 ^ in2 ^ in7;
+        out3 = out6 ^ in3 ^ in7;
+        out0 = tmp1 ^ out7;
+        out2 = out3 ^ out4 ^ in0;
+        out5 = tmp0 ^ out2;
+        out1 = out0 ^ out5 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_EA(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in6 ^ in7;
+        out5 = in0 ^ in7;
+        out6 = in0 ^ in1;
+        out0 = in1 ^ in2 ^ in3;
+        out2 = in2 ^ in4 ^ in5;
+        out7 = out6 ^ in2;
+        out1 = out0 ^ out6 ^ in4;
+        out3 = out7 ^ in5 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_EB(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in4 ^ in5;
+        tmp0 = in0 ^ in1;
+        out4 = in4 ^ in6 ^ in7;
+        out5 = in0 ^ in5 ^ in7;
+        out6 = tmp0 ^ in6;
+        tmp1 = tmp0 ^ in2;
+        out0 = tmp1 ^ in3;
+        out7 = tmp1 ^ in7;
+        out1 = out0 ^ in4;
+        out3 = out0 ^ in5 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_EC(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in0 ^ in5;
+        out4 = in2 ^ in3 ^ in7;
+        out5 = in0 ^ in3 ^ in4;
+        out6 = out3 ^ in1 ^ in4;
+        out1 = out4 ^ in4;
+        out0 = out4 ^ in1 ^ in6;
+        out2 = out0 ^ out5 ^ in5;
+        out7 = out2 ^ in4 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_ED(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in4;
+        tmp1 = in3 ^ in5;
+        out4 = tmp0 ^ in3 ^ in7;
+        out3 = tmp1 ^ in0;
+        out1 = out4 ^ in1;
+        out5 = out3 ^ in4;
+        out7 = out1 ^ out5 ^ in6;
+        out2 = tmp0 ^ out7;
+        out0 = tmp1 ^ out7;
+        out6 = out2 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_EE(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in2;
+        tmp0 = in0 ^ in1;
+        out5 = in0 ^ in3;
+        tmp1 = tmp0 ^ in2;
+        out6 = tmp0 ^ in4;
+        tmp2 = tmp1 ^ out5;
+        out7 = tmp1 ^ in5;
+        out1 = tmp2 ^ out6 ^ in7;
+        out0 = tmp2 ^ in6;
+        tmp3 = out7 ^ in1;
+        out3 = tmp3 ^ in7;
+        out2 = tmp3 ^ in4 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_EF(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in2 ^ in4;
+        tmp0 = in0 ^ in5;
+        tmp1 = in4 ^ in6;
+        out5 = tmp0 ^ in3;
+        out2 = tmp0 ^ tmp1;
+        out6 = tmp1 ^ in0 ^ in1;
+        out3 = out5 ^ in2 ^ in7;
+        out7 = out3 ^ in1 ^ in3;
+        out0 = out4 ^ out6 ^ in3;
+        out1 = tmp1 ^ out0 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_F0(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in2;
+        tmp1 = in4 ^ in5;
+        out2 = tmp0 ^ in6;
+        out3 = tmp1 ^ in1;
+        tmp2 = tmp1 ^ in7;
+        out1 = out2 ^ out3 ^ in3;
+        tmp3 = tmp0 ^ tmp2;
+        out0 = tmp3 ^ in3;
+        out5 = tmp3 ^ in0;
+        out4 = out1 ^ out5 ^ in4;
+        out7 = out4 ^ in2;
+        out6 = tmp2 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_F1(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in1 ^ in6;
+        tmp0 = in3 ^ in5;
+        out3 = tmp0 ^ in1 ^ in4;
+        tmp1 = out3 ^ in2;
+        out1 = tmp1 ^ in6;
+        tmp2 = tmp1 ^ in0;
+        tmp3 = out1 ^ in5;
+        out0 = tmp2 ^ in7;
+        out6 = tmp2 ^ in4;
+        out7 = tmp3 ^ in0;
+        out5 = tmp0 ^ out0;
+        out4 = tmp3 ^ out5 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_F2(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in4 ^ in5;
+        out2 = in2 ^ in6 ^ in7;
+        tmp1 = tmp0 ^ in1;
+        tmp2 = tmp1 ^ in2;
+        out0 = tmp2 ^ in3;
+        out3 = tmp2 ^ in7;
+        out5 = out3 ^ in0 ^ in4;
+        tmp3 = tmp0 ^ out5;
+        out7 = tmp3 ^ in3;
+        out4 = tmp3 ^ out2;
+        out1 = out0 ^ out4 ^ in4;
+        out6 = tmp1 ^ out1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_F3(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in6 ^ in7;
+        tmp0 = in0 ^ in1;
+        out4 = tmp0 ^ in6;
+        tmp1 = tmp0 ^ in2;
+        out5 = tmp1 ^ in7;
+        out6 = tmp1 ^ in3;
+        out7 = out6 ^ in4;
+        out0 = out7 ^ in5;
+        out1 = out0 ^ in6;
+        out3 = out0 ^ in0 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_F4(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in0 ^ in1 ^ in2;
+        tmp0 = out2 ^ in3;
+        out4 = tmp0 ^ in4;
+        out5 = out4 ^ in5;
+        out6 = out5 ^ in6;
+        out7 = out6 ^ in7;
+        out0 = out7 ^ in0;
+        out1 = out0 ^ in1;
+        out3 = tmp0 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_F5(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in0 ^ in1;
+        tmp0 = out2 ^ in2;
+        out4 = tmp0 ^ in3;
+        out5 = out4 ^ in4;
+        out6 = out5 ^ in5;
+        out7 = out6 ^ in6;
+        out0 = out7 ^ in7;
+        out1 = out0 ^ in0;
+        out3 = tmp0 ^ out0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_F6(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in7;
+        out2 = tmp0 ^ in2;
+        out4 = out2 ^ in1 ^ in4;
+        out7 = out4 ^ in3 ^ in5;
+        out5 = out7 ^ in4 ^ in7;
+        out0 = tmp0 ^ out7 ^ in6;
+        tmp1 = out0 ^ in1;
+        out6 = out0 ^ in0 ^ in5;
+        out3 = tmp1 ^ in3;
+        out1 = tmp0 ^ tmp1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_F7(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in0 ^ in7;
+        tmp0 = out2 ^ in1;
+        out4 = tmp0 ^ in2;
+        out5 = out4 ^ in3 ^ in7;
+        out6 = out5 ^ in4;
+        out7 = out6 ^ in5;
+        out0 = out7 ^ in6;
+        out1 = out0 ^ in7;
+        out3 = tmp0 ^ out1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_F8(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in4;
+        tmp1 = in3 ^ in5;
+        tmp2 = tmp0 ^ in6;
+        out4 = tmp0 ^ tmp1;
+        out1 = tmp1 ^ in2 ^ in4;
+        out3 = tmp2 ^ in1;
+        out5 = out3 ^ in5;
+        out7 = out1 ^ out5 ^ in7;
+        out6 = tmp1 ^ out7;
+        out0 = tmp2 ^ out7;
+        out2 = out6 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_F9(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in5;
+        tmp1 = in0 ^ in6;
+        out4 = tmp0 ^ in0;
+        tmp2 = tmp1 ^ in4;
+        tmp3 = tmp1 ^ in2;
+        out5 = tmp2 ^ in1;
+        out3 = out5 ^ in3;
+        tmp4 = tmp3 ^ out3;
+        out1 = tmp4 ^ in5;
+        out0 = tmp4 ^ in0 ^ in7;
+        out6 = tmp0 ^ out0 ^ in4;
+        out7 = tmp2 ^ tmp4;
+        out2 = tmp3 ^ out6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_FA(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in1;
+        tmp1 = tmp0 ^ in2;
+        tmp2 = tmp0 ^ in5;
+        tmp3 = tmp1 ^ in7;
+        out5 = tmp2 ^ in6;
+        out6 = tmp3 ^ in6;
+        out7 = tmp3 ^ in3;
+        out3 = out6 ^ in4;
+        out2 = tmp1 ^ out5;
+        out4 = out2 ^ out3 ^ in1;
+        out0 = out4 ^ out7 ^ in5;
+        out1 = tmp2 ^ out0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_FB(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in5 ^ in6;
+        tmp0 = in0 ^ in1;
+        out4 = in0 ^ in5 ^ in7;
+        out5 = tmp0 ^ in6;
+        tmp1 = tmp0 ^ in2;
+        out6 = tmp1 ^ in7;
+        out7 = tmp1 ^ in3;
+        out0 = out7 ^ in4;
+        out1 = out0 ^ in5;
+        out3 = out0 ^ in6 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_FC(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in2;
+        tmp1 = in0 ^ in7;
+        out2 = tmp0 ^ tmp1 ^ in5;
+        out3 = tmp1 ^ in4;
+        tmp2 = out2 ^ in6;
+        out6 = tmp2 ^ in4;
+        out7 = tmp2 ^ in3;
+        out4 = out6 ^ in1 ^ in3;
+        tmp3 = out4 ^ in0;
+        out1 = tmp3 ^ in6;
+        out0 = tmp3 ^ in1 ^ in5;
+        out5 = tmp0 ^ out4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_FD(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in5;
+        tmp1 = in1 ^ in7;
+        out2 = tmp0 ^ tmp1;
+        out6 = out2 ^ in2 ^ in4;
+        tmp2 = out6 ^ in0;
+        out1 = tmp2 ^ in3;
+        out0 = tmp0 ^ out1 ^ in6;
+        out5 = out0 ^ in2;
+        tmp3 = out5 ^ in1;
+        out3 = tmp3 ^ in6;
+        out7 = tmp2 ^ tmp3;
+        out4 = tmp1 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_FE(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in2;
+        out2 = tmp0 ^ in5;
+        out3 = tmp0 ^ in4;
+        tmp1 = out3 ^ in6;
+        out4 = tmp1 ^ in5;
+        tmp2 = tmp1 ^ in1;
+        out6 = tmp2 ^ in7;
+        tmp3 = tmp2 ^ in0;
+        out0 = tmp3 ^ in3;
+        tmp4 = out0 ^ out4 ^ in7;
+        out5 = tmp4 ^ in6;
+        out7 = tmp4 ^ in2;
+        out1 = tmp3 ^ out5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_FF(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in0 ^ in5;
+        tmp0 = in4 ^ in7;
+        tmp1 = out2 ^ in2;
+        out4 = tmp1 ^ in6;
+        out7 = tmp1 ^ in1 ^ in3;
+        out1 = tmp0 ^ out7;
+        tmp2 = out1 ^ in5;
+        out6 = tmp2 ^ in3;
+        tmp3 = tmp2 ^ in7;
+        out0 = tmp3 ^ in6;
+        out3 = tmp3 ^ in1;
+        out5 = tmp0 ^ out0 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void (*gf8_muladd[])(void *out, void *in) = {
+    gf8_muladd_00, gf8_muladd_01, gf8_muladd_02, gf8_muladd_03, gf8_muladd_04,
+    gf8_muladd_05, gf8_muladd_06, gf8_muladd_07, gf8_muladd_08, gf8_muladd_09,
+    gf8_muladd_0A, gf8_muladd_0B, gf8_muladd_0C, gf8_muladd_0D, gf8_muladd_0E,
+    gf8_muladd_0F, gf8_muladd_10, gf8_muladd_11, gf8_muladd_12, gf8_muladd_13,
+    gf8_muladd_14, gf8_muladd_15, gf8_muladd_16, gf8_muladd_17, gf8_muladd_18,
+    gf8_muladd_19, gf8_muladd_1A, gf8_muladd_1B, gf8_muladd_1C, gf8_muladd_1D,
+    gf8_muladd_1E, gf8_muladd_1F, gf8_muladd_20, gf8_muladd_21, gf8_muladd_22,
+    gf8_muladd_23, gf8_muladd_24, gf8_muladd_25, gf8_muladd_26, gf8_muladd_27,
+    gf8_muladd_28, gf8_muladd_29, gf8_muladd_2A, gf8_muladd_2B, gf8_muladd_2C,
+    gf8_muladd_2D, gf8_muladd_2E, gf8_muladd_2F, gf8_muladd_30, gf8_muladd_31,
+    gf8_muladd_32, gf8_muladd_33, gf8_muladd_34, gf8_muladd_35, gf8_muladd_36,
+    gf8_muladd_37, gf8_muladd_38, gf8_muladd_39, gf8_muladd_3A, gf8_muladd_3B,
+    gf8_muladd_3C, gf8_muladd_3D, gf8_muladd_3E, gf8_muladd_3F, gf8_muladd_40,
+    gf8_muladd_41, gf8_muladd_42, gf8_muladd_43, gf8_muladd_44, gf8_muladd_45,
+    gf8_muladd_46, gf8_muladd_47, gf8_muladd_48, gf8_muladd_49, gf8_muladd_4A,
+    gf8_muladd_4B, gf8_muladd_4C, gf8_muladd_4D, gf8_muladd_4E, gf8_muladd_4F,
+    gf8_muladd_50, gf8_muladd_51, gf8_muladd_52, gf8_muladd_53, gf8_muladd_54,
+    gf8_muladd_55, gf8_muladd_56, gf8_muladd_57, gf8_muladd_58, gf8_muladd_59,
+    gf8_muladd_5A, gf8_muladd_5B, gf8_muladd_5C, gf8_muladd_5D, gf8_muladd_5E,
+    gf8_muladd_5F, gf8_muladd_60, gf8_muladd_61, gf8_muladd_62, gf8_muladd_63,
+    gf8_muladd_64, gf8_muladd_65, gf8_muladd_66, gf8_muladd_67, gf8_muladd_68,
+    gf8_muladd_69, gf8_muladd_6A, gf8_muladd_6B, gf8_muladd_6C, gf8_muladd_6D,
+    gf8_muladd_6E, gf8_muladd_6F, gf8_muladd_70, gf8_muladd_71, gf8_muladd_72,
+    gf8_muladd_73, gf8_muladd_74, gf8_muladd_75, gf8_muladd_76, gf8_muladd_77,
+    gf8_muladd_78, gf8_muladd_79, gf8_muladd_7A, gf8_muladd_7B, gf8_muladd_7C,
+    gf8_muladd_7D, gf8_muladd_7E, gf8_muladd_7F, gf8_muladd_80, gf8_muladd_81,
+    gf8_muladd_82, gf8_muladd_83, gf8_muladd_84, gf8_muladd_85, gf8_muladd_86,
+    gf8_muladd_87, gf8_muladd_88, gf8_muladd_89, gf8_muladd_8A, gf8_muladd_8B,
+    gf8_muladd_8C, gf8_muladd_8D, gf8_muladd_8E, gf8_muladd_8F, gf8_muladd_90,
+    gf8_muladd_91, gf8_muladd_92, gf8_muladd_93, gf8_muladd_94, gf8_muladd_95,
+    gf8_muladd_96, gf8_muladd_97, gf8_muladd_98, gf8_muladd_99, gf8_muladd_9A,
+    gf8_muladd_9B, gf8_muladd_9C, gf8_muladd_9D, gf8_muladd_9E, gf8_muladd_9F,
+    gf8_muladd_A0, gf8_muladd_A1, gf8_muladd_A2, gf8_muladd_A3, gf8_muladd_A4,
+    gf8_muladd_A5, gf8_muladd_A6, gf8_muladd_A7, gf8_muladd_A8, gf8_muladd_A9,
+    gf8_muladd_AA, gf8_muladd_AB, gf8_muladd_AC, gf8_muladd_AD, gf8_muladd_AE,
+    gf8_muladd_AF, gf8_muladd_B0, gf8_muladd_B1, gf8_muladd_B2, gf8_muladd_B3,
+    gf8_muladd_B4, gf8_muladd_B5, gf8_muladd_B6, gf8_muladd_B7, gf8_muladd_B8,
+    gf8_muladd_B9, gf8_muladd_BA, gf8_muladd_BB, gf8_muladd_BC, gf8_muladd_BD,
+    gf8_muladd_BE, gf8_muladd_BF, gf8_muladd_C0, gf8_muladd_C1, gf8_muladd_C2,
+    gf8_muladd_C3, gf8_muladd_C4, gf8_muladd_C5, gf8_muladd_C6, gf8_muladd_C7,
+    gf8_muladd_C8, gf8_muladd_C9, gf8_muladd_CA, gf8_muladd_CB, gf8_muladd_CC,
+    gf8_muladd_CD, gf8_muladd_CE, gf8_muladd_CF, gf8_muladd_D0, gf8_muladd_D1,
+    gf8_muladd_D2, gf8_muladd_D3, gf8_muladd_D4, gf8_muladd_D5, gf8_muladd_D6,
+    gf8_muladd_D7, gf8_muladd_D8, gf8_muladd_D9, gf8_muladd_DA, gf8_muladd_DB,
+    gf8_muladd_DC, gf8_muladd_DD, gf8_muladd_DE, gf8_muladd_DF, gf8_muladd_E0,
+    gf8_muladd_E1, gf8_muladd_E2, gf8_muladd_E3, gf8_muladd_E4, gf8_muladd_E5,
+    gf8_muladd_E6, gf8_muladd_E7, gf8_muladd_E8, gf8_muladd_E9, gf8_muladd_EA,
+    gf8_muladd_EB, gf8_muladd_EC, gf8_muladd_ED, gf8_muladd_EE, gf8_muladd_EF,
+    gf8_muladd_F0, gf8_muladd_F1, gf8_muladd_F2, gf8_muladd_F3, gf8_muladd_F4,
+    gf8_muladd_F5, gf8_muladd_F6, gf8_muladd_F7, gf8_muladd_F8, gf8_muladd_F9,
+    gf8_muladd_FA, gf8_muladd_FB, gf8_muladd_FC, gf8_muladd_FD, gf8_muladd_FE,
+    gf8_muladd_FF};
+
+static uint64_t zero[EC_METHOD_WORD_SIZE * 8] = {
+    0,
+};
+
+void
+ec_code_c_prepare(ec_gf_t *gf, uint32_t *values, uint32_t count)
+{
+    uint32_t i, last, tmp;
+
+    last = 1;
+    for (i = count; i > 0; i--) {
+        if (values[i - 1] != 0) {
+            tmp = values[i - 1];
+            values[i - 1] = ec_gf_div(gf, tmp, last);
+            last = tmp;
+        }
+    }
+}
+
+void
+ec_code_c_linear(void *dst, void *src, uint64_t offset, uint32_t *values,
+                 uint32_t count)
+{
+    src += offset;
+    gf8_muladd_00(dst, src);
+    while (--count > 0) {
+        src += EC_METHOD_CHUNK_SIZE;
+        gf8_muladd[*values](dst, src);
+        values++;
+    }
+}
+
+void
+ec_code_c_interleaved(void *dst, void **src, uint64_t offset, uint32_t *values,
+                      uint32_t count)
+{
+    uint32_t i, last, tmp;
+
+    i = 0;
+    while ((last = *values++) == 0) {
+        i++;
+    }
+    gf8_muladd_00(dst, src[i++] + offset);
+    while (i < count) {
+        tmp = *values++;
+        if (tmp != 0) {
+            gf8_muladd[last](dst, src[i] + offset);
+            last = tmp;
+        }
+        i++;
+    }
+    gf8_muladd[last](dst, zero);
+}
diff --git a/xlators/cluster/ec/src/ec-code-c.h b/xlators/cluster/ec/src/ec-code-c.h
new file mode 100644
index 00000000000..42b5a064eb8
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code-c.h
@@ -0,0 +1,27 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_CODE_C_H__
+#define __EC_CODE_C_H__
+
+#include "ec-types.h"
+
+void
+ec_code_c_prepare(ec_gf_t *gf, uint32_t *values, uint32_t count);
+
+void
+ec_code_c_linear(void *dst, void *src, uint64_t offset, uint32_t *values,
+                 uint32_t count);
+
+void
+ec_code_c_interleaved(void *dst, void **src, uint64_t offset, uint32_t *values,
+                      uint32_t count);
+
+#endif /* __EC_CODE_C_H__ */
diff --git a/xlators/cluster/ec/src/ec-code-intel.c b/xlators/cluster/ec/src/ec-code-intel.c
new file mode 100644
index 00000000000..f1c4e13e321
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code-intel.c
@@ -0,0 +1,594 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <inttypes.h>
+#include <string.h>
+#include <errno.h>
+
+#include "ec-code-intel.h"
+
+static void
+ec_code_intel_init(ec_code_intel_t *intel)
+{
+    memset(intel, 0, sizeof(ec_code_intel_t));
+}
+
+static void
+ec_code_intel_prefix(ec_code_intel_t *intel, uint8_t prefix)
+{
+    intel->prefix.data[intel->prefix.bytes++] = prefix;
+}
+
+static void
+ec_code_intel_rex(ec_code_intel_t *intel, gf_boolean_t w)
+{
+    gf_boolean_t present = _gf_false;
+
+    if (w) {
+        intel->rex.w = 1;
+        present = _gf_true;
+    }
+    if (intel->modrm.present) {
+        if (intel->modrm.reg > 7) {
+            intel->modrm.reg &= 7;
+            intel->rex.r = 1;
+            present = _gf_true;
+        }
+        if (intel->sib.present) {
+            if (intel->sib.index > 7) {
+                intel->sib.index &= 7;
+                intel->rex.x = 1;
+                present = _gf_true;
+            }
+            if (intel->sib.base > 7) {
+                intel->sib.base &= 7;
+                intel->rex.b = 1;
+                present = _gf_true;
+            }
+        } else if (intel->modrm.rm > 7) {
+            intel->modrm.rm &= 7;
+            intel->rex.b = 1;
+            present = _gf_true;
+        }
+    } else if (intel->reg > 7) {
+        intel->reg &= 7;
+        intel->rex.b = 1;
+        present = _gf_true;
+    }
+    intel->rex.present = present;
+}
+
+static void
+ec_code_intel_vex(ec_code_intel_t *intel, gf_boolean_t w, gf_boolean_t l,
+                  ec_code_vex_opcode_t opcode, ec_code_vex_prefix_t prefix,
+                  uint32_t reg)
+{
+    ec_code_intel_rex(intel, w);
+    if (((intel->rex.w == 1) || (intel->rex.x == 0) || (intel->rex.b == 0)) ||
+        ((opcode != VEX_OPCODE_NONE) && (opcode != VEX_OPCODE_0F))) {
+        intel->rex.present = _gf_false;
+
+        intel->vex.bytes = 3;
+        intel->vex.data[0] = 0xC4;
+        intel->vex.data[1] = ((intel->rex.r << 7) | (intel->rex.x << 6) |
+                              (intel->rex.b << 5) | opcode) ^
+                             0xE0;
+        intel->vex.data[2] = (intel->rex.w << 7) | ((~reg & 0x0F) << 3) |
+                             (l ? 0x04 : 0x00) | prefix;
+    } else {
+        intel->vex.bytes = 2;
+        intel->vex.data[0] = 0xC5;
+        intel->vex.data[1] = (intel->rex.r << 7) | ((~reg & 0x0F) << 3) |
+                             (l ? 0x04 : 0x00) | prefix;
+    }
+}
+
+static void
+ec_code_intel_modrm_reg(ec_code_intel_t *intel, uint32_t rm, uint32_t reg)
+{
+    intel->modrm.present = _gf_true;
+    intel->modrm.mod = 3;
+    intel->modrm.rm = rm;
+    intel->modrm.reg = reg;
+}
+
+static void
+ec_code_intel_modrm_mem(ec_code_intel_t *intel, uint32_t reg,
+                        ec_code_intel_reg_t base, ec_code_intel_reg_t index,
+                        uint32_t scale, int32_t offset)
+{
+    if (index == REG_SP) {
+        intel->invalid = _gf_true;
+        return;
+    }
+    if ((index != REG_NULL) && (scale != 1) && (scale != 2) && (scale != 4) &&
+        (scale != 8)) {
+        intel->invalid = _gf_true;
+        return;
+    }
+    scale >>= 1;
+    if (scale == 4) {
+        scale = 3;
+    }
+
+    intel->modrm.present = _gf_true;
+    intel->modrm.reg = reg;
+
+    intel->offset.value = offset;
+    if ((offset == 0) && (base != REG_BP)) {
+        intel->modrm.mod = 0;
+        intel->offset.bytes = 0;
+    } else if ((offset >= -128) && (offset <= 127)) {
+        intel->modrm.mod = 1;
+        intel->offset.bytes = 1;
+    } else {
+        intel->modrm.mod = 2;
+        intel->offset.bytes = 4;
+    }
+
+    intel->modrm.rm = base;
+    if ((index != REG_NULL) || (base == REG_SP)) {
+        intel->modrm.rm = 4;
+        intel->sib.present = _gf_true;
+        intel->sib.index = index;
+        if (index == REG_NULL) {
+            intel->sib.index = 4;
+        }
+        intel->sib.scale = scale;
+        intel->sib.base = base;
+        if (base == REG_NULL) {
+            intel->sib.base = 5;
+            intel->modrm.mod = 0;
+            intel->offset.bytes = 4;
+        }
+    } else if (base == REG_NULL) {
+        intel->modrm.mod = 0;
+        intel->modrm.rm = 5;
+        intel->offset.bytes = 4;
+    }
+}
+
+static void
+ec_code_intel_op_1(ec_code_intel_t *intel, uint8_t opcode, uint32_t reg)
+{
+    intel->reg = reg;
+    intel->opcode.bytes = 1;
+    intel->opcode.data[0] = opcode;
+}
+
+static void
+ec_code_intel_op_2(ec_code_intel_t *intel, uint8_t opcode1, uint8_t opcode2,
+                   uint32_t reg)
+{
+    intel->reg = reg;
+    intel->opcode.bytes = 2;
+    intel->opcode.data[0] = opcode1;
+    intel->opcode.data[1] = opcode2;
+}
+
+static void
+ec_code_intel_immediate_1(ec_code_intel_t *intel, uint32_t value)
+{
+    intel->immediate.bytes = 1;
+    intel->immediate.value = value;
+}
+
+static void
+ec_code_intel_immediate_2(ec_code_intel_t *intel, uint32_t value)
+{
+    intel->immediate.bytes = 2;
+    intel->immediate.value = value;
+}
+
+static void
+ec_code_intel_immediate_4(ec_code_intel_t *intel, uint32_t value)
+{
+    intel->immediate.bytes = 4;
+    intel->immediate.value = value;
+}
+
+static void
+ec_code_intel_emit(ec_code_builder_t *builder, ec_code_intel_t *intel)
+{
+    uint8_t insn[15];
+    uint32_t i, count;
+
+    if (intel->invalid) {
+        ec_code_error(builder, EINVAL);
+        return;
+    }
+
+    count = 0;
+    for (i = 0; i < intel->prefix.bytes; i++) {
+        insn[count++] = intel->prefix.data[i];
+    }
+    for (i = 0; i < intel->vex.bytes; i++) {
+        insn[count++] = intel->vex.data[i];
+    }
+    if (intel->rex.present) {
+        insn[count++] = 0x40 | (intel->rex.w << 3) | (intel->rex.r << 2) |
+                        (intel->rex.x << 1) | (intel->rex.b << 0);
+    }
+    for (i = 0; i < intel->opcode.bytes; i++) {
+        insn[count++] = intel->opcode.data[i];
+    }
+    if (intel->modrm.present) {
+        insn[count++] = (intel->modrm.mod << 6) | (intel->modrm.reg << 3) |
+                        (intel->modrm.rm << 0);
+        if (intel->sib.present) {
+            insn[count++] = (intel->sib.scale << 6) | (intel->sib.index << 3) |
+                            (intel->sib.base << 0);
+        }
+    }
+    for (i = 0; i < intel->offset.bytes; i++) {
+        insn[count++] = intel->offset.data[i];
+    }
+    for (i = 0; i < intel->immediate.bytes; i++) {
+        insn[count++] = intel->immediate.data[i];
+    }
+
+    ec_code_emit(builder, insn, count);
+}
+
+void
+ec_code_intel_op_push_r(ec_code_builder_t *builder, ec_code_intel_reg_t reg)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_op_1(&intel, 0x50 | (reg & 7), reg);
+    ec_code_intel_rex(&intel, _gf_false);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_pop_r(ec_code_builder_t *builder, ec_code_intel_reg_t reg)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_op_1(&intel, 0x58 | (reg & 7), reg);
+    ec_code_intel_rex(&intel, _gf_false);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_ret(ec_code_builder_t *builder, uint32_t size)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    if (size == 0) {
+        ec_code_intel_op_1(&intel, 0xC3, 0);
+    } else {
+        ec_code_intel_immediate_2(&intel, size);
+        ec_code_intel_op_1(&intel, 0xC2, 0);
+    }
+    ec_code_intel_rex(&intel, _gf_false);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_mov_r2r(ec_code_builder_t *builder, ec_code_intel_reg_t src,
+                         ec_code_intel_reg_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_modrm_reg(&intel, dst, src);
+    ec_code_intel_op_1(&intel, 0x89, 0);
+    ec_code_intel_rex(&intel, _gf_true);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_mov_r2m(ec_code_builder_t *builder, ec_code_intel_reg_t src,
+                         ec_code_intel_reg_t base, ec_code_intel_reg_t index,
+                         uint32_t scale, int32_t offset)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_modrm_mem(&intel, src, base, index, scale, offset);
+    ec_code_intel_op_1(&intel, 0x89, 0);
+    ec_code_intel_rex(&intel, _gf_true);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_mov_m2r(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                         ec_code_intel_reg_t index, uint32_t scale,
+                         int32_t offset, ec_code_intel_reg_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_modrm_mem(&intel, dst, base, index, scale, offset);
+    ec_code_intel_op_1(&intel, 0x8B, 0);
+    ec_code_intel_rex(&intel, _gf_true);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_xor_r2r(ec_code_builder_t *builder, ec_code_intel_reg_t src,
+                         ec_code_intel_reg_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_modrm_reg(&intel, dst, src);
+    ec_code_intel_op_1(&intel, 0x31, 0);
+    ec_code_intel_rex(&intel, _gf_true);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_xor_m2r(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                         ec_code_intel_reg_t index, uint32_t scale,
+                         int32_t offset, ec_code_intel_reg_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_modrm_mem(&intel, dst, base, index, scale, offset);
+    ec_code_intel_op_1(&intel, 0x33, 0);
+    ec_code_intel_rex(&intel, _gf_true);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_add_i2r(ec_code_builder_t *builder, int32_t value,
+                         ec_code_intel_reg_t reg)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    if ((value >= -128) && (value < 128)) {
+        ec_code_intel_modrm_reg(&intel, reg, 0);
+        ec_code_intel_op_1(&intel, 0x83, 0);
+        ec_code_intel_immediate_1(&intel, value);
+    } else {
+        if (reg == REG_AX) {
+            ec_code_intel_op_1(&intel, 0x05, reg);
+        } else {
+            ec_code_intel_modrm_reg(&intel, reg, 0);
+            ec_code_intel_op_1(&intel, 0x81, 0);
+        }
+        ec_code_intel_immediate_4(&intel, value);
+    }
+    ec_code_intel_rex(&intel, _gf_true);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_test_i2r(ec_code_builder_t *builder, uint32_t value,
+                          ec_code_intel_reg_t reg)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    if (reg == REG_AX) {
+        ec_code_intel_op_1(&intel, 0xA9, reg);
+    } else {
+        ec_code_intel_modrm_reg(&intel, reg, 0);
+        ec_code_intel_op_1(&intel, 0xF7, 0);
+    }
+    ec_code_intel_immediate_4(&intel, value);
+    ec_code_intel_rex(&intel, _gf_true);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_jne(ec_code_builder_t *builder, uint32_t address)
+{
+    ec_code_intel_t intel;
+    int32_t rel;
+
+    ec_code_intel_init(&intel);
+
+    rel = address - builder->address - 2;
+    if ((rel >= -128) && (rel < 128)) {
+        ec_code_intel_op_1(&intel, 0x75, 0);
+        ec_code_intel_immediate_1(&intel, rel);
+    } else {
+        rel -= 4;
+        ec_code_intel_op_2(&intel, 0x0F, 0x85, 0);
+        ec_code_intel_immediate_4(&intel, rel);
+    }
+    ec_code_intel_rex(&intel, _gf_false);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_mov_sse2sse(ec_code_builder_t *builder, uint32_t src,
+                             uint32_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_prefix(&intel, 0x66);
+    ec_code_intel_modrm_reg(&intel, src, dst);
+    ec_code_intel_op_2(&intel, 0x0F, 0x6F, 0);
+    ec_code_intel_rex(&intel, _gf_false);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_mov_sse2m(ec_code_builder_t *builder, uint32_t src,
+                           ec_code_intel_reg_t base, ec_code_intel_reg_t index,
+                           uint32_t scale, int32_t offset)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_prefix(&intel, 0x66);
+    ec_code_intel_modrm_mem(&intel, src, base, index, scale, offset);
+    ec_code_intel_op_2(&intel, 0x0F, 0x7F, 0);
+    ec_code_intel_rex(&intel, _gf_false);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_mov_m2sse(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                           ec_code_intel_reg_t index, uint32_t scale,
+                           int32_t offset, uint32_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_prefix(&intel, 0x66);
+    ec_code_intel_modrm_mem(&intel, dst, base, index, scale, offset);
+    ec_code_intel_op_2(&intel, 0x0F, 0x6F, 0);
+    ec_code_intel_rex(&intel, _gf_false);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_xor_sse2sse(ec_code_builder_t *builder, uint32_t src,
+                             uint32_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_prefix(&intel, 0x66);
+    ec_code_intel_modrm_reg(&intel, src, dst);
+    ec_code_intel_op_2(&intel, 0x0F, 0xEF, 0);
+    ec_code_intel_rex(&intel, _gf_false);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_xor_m2sse(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                           ec_code_intel_reg_t index, uint32_t scale,
+                           int32_t offset, uint32_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_prefix(&intel, 0x66);
+    ec_code_intel_modrm_mem(&intel, dst, base, index, scale, offset);
+    ec_code_intel_op_2(&intel, 0x0F, 0xEF, 0);
+    ec_code_intel_rex(&intel, _gf_false);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_mov_avx2avx(ec_code_builder_t *builder, uint32_t src,
+                             uint32_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_modrm_reg(&intel, src, dst);
+    ec_code_intel_op_1(&intel, 0x6F, 0);
+    ec_code_intel_vex(&intel, _gf_false, _gf_true, VEX_OPCODE_0F, VEX_PREFIX_66,
+                      VEX_REG_NONE);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_mov_avx2m(ec_code_builder_t *builder, uint32_t src,
+                           ec_code_intel_reg_t base, ec_code_intel_reg_t index,
+                           uint32_t scale, int32_t offset)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_modrm_mem(&intel, src, base, index, scale, offset);
+    ec_code_intel_op_1(&intel, 0x7F, 0);
+    ec_code_intel_vex(&intel, _gf_false, _gf_true, VEX_OPCODE_0F, VEX_PREFIX_66,
+                      VEX_REG_NONE);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_mov_m2avx(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                           ec_code_intel_reg_t index, uint32_t scale,
+                           int32_t offset, uint32_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_modrm_mem(&intel, dst, base, index, scale, offset);
+    ec_code_intel_op_1(&intel, 0x6F, 0);
+    ec_code_intel_vex(&intel, _gf_false, _gf_true, VEX_OPCODE_0F, VEX_PREFIX_66,
+                      VEX_REG_NONE);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_xor_avx2avx(ec_code_builder_t *builder, uint32_t src,
+                             uint32_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_modrm_reg(&intel, src, dst);
+    ec_code_intel_op_1(&intel, 0xEF, 0);
+    ec_code_intel_vex(&intel, _gf_false, _gf_true, VEX_OPCODE_0F, VEX_PREFIX_66,
+                      dst);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_xor_m2avx(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                           ec_code_intel_reg_t index, uint32_t scale,
+                           int32_t offset, uint32_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_modrm_mem(&intel, dst, base, index, scale, offset);
+    ec_code_intel_op_1(&intel, 0xEF, 0);
+    ec_code_intel_vex(&intel, _gf_false, _gf_true, VEX_OPCODE_0F, VEX_PREFIX_66,
+                      dst);
+
+    ec_code_intel_emit(builder, &intel);
+}
diff --git a/xlators/cluster/ec/src/ec-code-intel.h b/xlators/cluster/ec/src/ec-code-intel.h
new file mode 100644
index 00000000000..3fa4a174765
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code-intel.h
@@ -0,0 +1,191 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_CODE_INTEL_H__
+#define __EC_CODE_INTEL_H__
+
+#include "ec-code.h"
+
+#define VEX_REG_NONE 0
+
+enum _ec_code_intel_reg;
+typedef enum _ec_code_intel_reg ec_code_intel_reg_t;
+
+enum _ec_code_vex_prefix;
+typedef enum _ec_code_vex_prefix ec_code_vex_prefix_t;
+
+enum _ec_code_vex_opcode;
+typedef enum _ec_code_vex_opcode ec_code_vex_opcode_t;
+
+struct _ec_code_intel_buffer;
+typedef struct _ec_code_intel_buffer ec_code_intel_buffer_t;
+
+struct _ec_code_intel_sib;
+typedef struct _ec_code_intel_sib ec_code_intel_sib_t;
+
+struct _ec_code_intel_modrm;
+typedef struct _ec_code_intel_modrm ec_code_intel_modrm_t;
+
+struct _ec_code_intel_rex;
+typedef struct _ec_code_intel_rex ec_code_intel_rex_t;
+
+struct _ec_code_intel;
+typedef struct _ec_code_intel ec_code_intel_t;
+
+enum _ec_code_intel_reg {
+    REG_NULL = -1,
+    REG_AX,
+    REG_CX,
+    REG_DX,
+    REG_BX,
+    REG_SP,
+    REG_BP,
+    REG_SI,
+    REG_DI,
+    REG_8,
+    REG_9,
+    REG_10,
+    REG_11,
+    REG_12,
+    REG_13,
+    REG_14,
+    REG_15
+};
+
+enum _ec_code_vex_prefix {
+    VEX_PREFIX_NONE = 0,
+    VEX_PREFIX_66,
+    VEX_PREFIX_F3,
+    VEX_PREFIX_F2
+};
+
+enum _ec_code_vex_opcode {
+    VEX_OPCODE_NONE = 0,
+    VEX_OPCODE_0F,
+    VEX_OPCODE_0F_38,
+    VEX_OPCODE_0F_3A
+};
+
+struct _ec_code_intel_buffer {
+    uint32_t bytes;
+    union {
+        uint8_t data[4];
+        uint32_t value;
+    };
+};
+
+struct _ec_code_intel_sib {
+    gf_boolean_t present;
+    uint32_t base;
+    uint32_t index;
+    uint32_t scale;
+};
+
+struct _ec_code_intel_modrm {
+    gf_boolean_t present;
+    uint32_t mod;
+    uint32_t rm;
+    uint32_t reg;
+};
+
+struct _ec_code_intel_rex {
+    gf_boolean_t present;
+    uint32_t w;
+    uint32_t r;
+    uint32_t x;
+    uint32_t b;
+};
+
+struct _ec_code_intel {
+    gf_boolean_t invalid;
+    ec_code_intel_buffer_t prefix;
+    ec_code_intel_buffer_t opcode;
+    ec_code_intel_buffer_t offset;
+    ec_code_intel_buffer_t immediate;
+    ec_code_intel_buffer_t vex;
+    ec_code_intel_rex_t rex;
+    ec_code_intel_modrm_t modrm;
+    ec_code_intel_sib_t sib;
+    uint32_t reg;
+};
+
+void
+ec_code_intel_op_push_r(ec_code_builder_t *builder, ec_code_intel_reg_t reg);
+void
+ec_code_intel_op_pop_r(ec_code_builder_t *builder, ec_code_intel_reg_t reg);
+void
+ec_code_intel_op_ret(ec_code_builder_t *builder, uint32_t size);
+
+void
+ec_code_intel_op_mov_r2r(ec_code_builder_t *builder, ec_code_intel_reg_t src,
+                         ec_code_intel_reg_t dst);
+void
+ec_code_intel_op_mov_r2m(ec_code_builder_t *builder, ec_code_intel_reg_t src,
+                         ec_code_intel_reg_t base, ec_code_intel_reg_t index,
+                         uint32_t scale, int32_t offset);
+void
+ec_code_intel_op_mov_m2r(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                         ec_code_intel_reg_t index, uint32_t scale,
+                         int32_t offset, ec_code_intel_reg_t dst);
+void
+ec_code_intel_op_xor_r2r(ec_code_builder_t *builder, ec_code_intel_reg_t src,
+                         ec_code_intel_reg_t dst);
+void
+ec_code_intel_op_xor_m2r(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                         ec_code_intel_reg_t index, uint32_t scale,
+                         int32_t offset, ec_code_intel_reg_t dst);
+void
+ec_code_intel_op_add_i2r(ec_code_builder_t *builder, int32_t value,
+                         ec_code_intel_reg_t reg);
+void
+ec_code_intel_op_test_i2r(ec_code_builder_t *builder, uint32_t value,
+                          ec_code_intel_reg_t reg);
+void
+ec_code_intel_op_jne(ec_code_builder_t *builder, uint32_t address);
+
+void
+ec_code_intel_op_mov_sse2sse(ec_code_builder_t *builder, uint32_t src,
+                             uint32_t dst);
+void
+ec_code_intel_op_mov_sse2m(ec_code_builder_t *builder, uint32_t src,
+                           ec_code_intel_reg_t base, ec_code_intel_reg_t index,
+                           uint32_t scale, int32_t offset);
+void
+ec_code_intel_op_mov_m2sse(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                           ec_code_intel_reg_t index, uint32_t scale,
+                           int32_t offset, uint32_t dst);
+void
+ec_code_intel_op_xor_sse2sse(ec_code_builder_t *builder, uint32_t src,
+                             uint32_t dst);
+void
+ec_code_intel_op_xor_m2sse(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                           ec_code_intel_reg_t index, uint32_t scale,
+                           int32_t offset, uint32_t dst);
+
+void
+ec_code_intel_op_mov_avx2avx(ec_code_builder_t *builder, uint32_t src,
+                             uint32_t dst);
+void
+ec_code_intel_op_mov_avx2m(ec_code_builder_t *builder, uint32_t src,
+                           ec_code_intel_reg_t base, ec_code_intel_reg_t index,
+                           uint32_t scale, int32_t offset);
+void
+ec_code_intel_op_mov_m2avx(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                           ec_code_intel_reg_t index, uint32_t scale,
+                           int32_t offset, uint32_t dst);
+void
+ec_code_intel_op_xor_avx2avx(ec_code_builder_t *builder, uint32_t src,
+                             uint32_t dst);
+void
+ec_code_intel_op_xor_m2avx(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                           ec_code_intel_reg_t index, uint32_t scale,
+                           int32_t offset, uint32_t dst);
+
+#endif /* __EC_CODE_INTEL_H__ */
diff --git a/xlators/cluster/ec/src/ec-code-sse.c b/xlators/cluster/ec/src/ec-code-sse.c
new file mode 100644
index 00000000000..e11e7ff8400
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code-sse.c
@@ -0,0 +1,101 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <errno.h>
+
+#include "ec-code-intel.h"
+
+static void
+ec_code_sse_prolog(ec_code_builder_t *builder)
+{
+    builder->loop = builder->address;
+}
+
+static void
+ec_code_sse_epilog(ec_code_builder_t *builder)
+{
+    ec_code_intel_op_add_i2r(builder, 16, REG_DX);
+    ec_code_intel_op_add_i2r(builder, 16, REG_DI);
+    ec_code_intel_op_test_i2r(builder, builder->width - 1, REG_DX);
+    ec_code_intel_op_jne(builder, builder->loop);
+
+    ec_code_intel_op_ret(builder, 0);
+}
+
+static void
+ec_code_sse_load(ec_code_builder_t *builder, uint32_t dst, uint32_t idx,
+                 uint32_t bit)
+{
+    if (builder->linear) {
+        ec_code_intel_op_mov_m2sse(
+            builder, REG_SI, REG_DX, 1,
+            idx * builder->width * builder->bits + bit * builder->width, dst);
+    } else {
+        if (builder->base != idx) {
+            ec_code_intel_op_mov_m2r(builder, REG_SI, REG_NULL, 0, idx * 8,
+                                     REG_AX);
+            builder->base = idx;
+        }
+        ec_code_intel_op_mov_m2sse(builder, REG_AX, REG_DX, 1,
+                                   bit * builder->width, dst);
+    }
+}
+
+static void
+ec_code_sse_store(ec_code_builder_t *builder, uint32_t src, uint32_t bit)
+{
+    ec_code_intel_op_mov_sse2m(builder, src, REG_DI, REG_NULL, 0,
+                               bit * builder->width);
+}
+
+static void
+ec_code_sse_copy(ec_code_builder_t *builder, uint32_t dst, uint32_t src)
+{
+    ec_code_intel_op_mov_sse2sse(builder, src, dst);
+}
+
+static void
+ec_code_sse_xor2(ec_code_builder_t *builder, uint32_t dst, uint32_t src)
+{
+    ec_code_intel_op_xor_sse2sse(builder, src, dst);
+}
+
+static void
+ec_code_sse_xorm(ec_code_builder_t *builder, uint32_t dst, uint32_t idx,
+                 uint32_t bit)
+{
+    if (builder->linear) {
+        ec_code_intel_op_xor_m2sse(
+            builder, REG_SI, REG_DX, 1,
+            idx * builder->width * builder->bits + bit * builder->width, dst);
+    } else {
+        if (builder->base != idx) {
+            ec_code_intel_op_mov_m2r(builder, REG_SI, REG_NULL, 0, idx * 8,
+                                     REG_AX);
+            builder->base = idx;
+        }
+        ec_code_intel_op_xor_m2sse(builder, REG_AX, REG_DX, 1,
+                                   bit * builder->width, dst);
+    }
+}
+
+static char *ec_code_sse_needed_flags[] = {"sse2", NULL};
+
+ec_code_gen_t ec_code_gen_sse = {.name = "sse",
+                                 .flags = ec_code_sse_needed_flags,
+                                 .width = 16,
+                                 .prolog = ec_code_sse_prolog,
+                                 .epilog = ec_code_sse_epilog,
+                                 .load = ec_code_sse_load,
+                                 .store = ec_code_sse_store,
+                                 .copy = ec_code_sse_copy,
+                                 .xor2 = ec_code_sse_xor2,
+                                 .xor3 = NULL,
+                                 .xorm = ec_code_sse_xorm};
diff --git a/xlators/cluster/ec/src/ec-code-sse.h b/xlators/cluster/ec/src/ec-code-sse.h
new file mode 100644
index 00000000000..f1acbcf894b
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code-sse.h
@@ -0,0 +1,18 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_CODE_SSE_H__
+#define __EC_CODE_SSE_H__
+
+#include "ec-code.h"
+
+extern ec_code_gen_t ec_code_gen_sse;
+
+#endif /* __EC_CODE_SSE_H__ */
diff --git a/xlators/cluster/ec/src/ec-code-x64.c b/xlators/cluster/ec/src/ec-code-x64.c
new file mode 100644
index 00000000000..26565b4493f
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code-x64.c
@@ -0,0 +1,144 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <errno.h>
+
+#include "ec-code-intel.h"
+
+static ec_code_intel_reg_t ec_code_x64_regmap[] = {
+    REG_AX, REG_CX, REG_BP, REG_8,  REG_9, REG_10,
+    REG_11, REG_12, REG_13, REG_14, REG_15};
+
+static void
+ec_code_x64_prolog(ec_code_builder_t *builder)
+{
+    uint32_t i;
+
+    ec_code_intel_op_push_r(builder, REG_BP);
+    if (!builder->linear) {
+        ec_code_intel_op_push_r(builder, REG_BX);
+    }
+    if (builder->regs > 11) {
+        ec_code_error(builder, EINVAL);
+        return;
+    }
+    for (i = 7; i < builder->regs; i++) {
+        ec_code_intel_op_push_r(builder, ec_code_x64_regmap[i]);
+    }
+
+    builder->loop = builder->address;
+}
+
+static void
+ec_code_x64_epilog(ec_code_builder_t *builder)
+{
+    uint32_t i;
+
+    ec_code_intel_op_add_i2r(builder, 8, REG_DX);
+    ec_code_intel_op_add_i2r(builder, 8, REG_DI);
+    ec_code_intel_op_test_i2r(builder, builder->width - 1, REG_DX);
+    ec_code_intel_op_jne(builder, builder->loop);
+
+    if (builder->regs > 11) {
+        ec_code_error(builder, EINVAL);
+        return;
+    }
+    for (i = builder->regs; i > 7; i--) {
+        ec_code_intel_op_pop_r(builder, ec_code_x64_regmap[i - 1]);
+    }
+    if (!builder->linear) {
+        ec_code_intel_op_pop_r(builder, REG_BX);
+    }
+    ec_code_intel_op_pop_r(builder, REG_BP);
+    ec_code_intel_op_ret(builder, 0);
+}
+
+static void
+ec_code_x64_load(ec_code_builder_t *builder, uint32_t dst, uint32_t idx,
+                 uint32_t bit)
+{
+    dst = ec_code_x64_regmap[dst];
+
+    if (builder->linear) {
+        ec_code_intel_op_mov_m2r(
+            builder, REG_SI, REG_DX, 1,
+            idx * builder->width * builder->bits + bit * builder->width, dst);
+    } else {
+        if (builder->base != idx) {
+            ec_code_intel_op_mov_m2r(builder, REG_SI, REG_NULL, 0, idx * 8,
+                                     REG_BX);
+            builder->base = idx;
+        }
+        ec_code_intel_op_mov_m2r(builder, REG_BX, REG_DX, 1,
+                                 bit * builder->width, dst);
+    }
+}
+
+static void
+ec_code_x64_store(ec_code_builder_t *builder, uint32_t src, uint32_t bit)
+{
+    src = ec_code_x64_regmap[src];
+
+    ec_code_intel_op_mov_r2m(builder, src, REG_DI, REG_NULL, 0,
+                             bit * builder->width);
+}
+
+static void
+ec_code_x64_copy(ec_code_builder_t *builder, uint32_t dst, uint32_t src)
+{
+    dst = ec_code_x64_regmap[dst];
+    src = ec_code_x64_regmap[src];
+
+    ec_code_intel_op_mov_r2r(builder, src, dst);
+}
+
+static void
+ec_code_x64_xor2(ec_code_builder_t *builder, uint32_t dst, uint32_t src)
+{
+    dst = ec_code_x64_regmap[dst];
+    src = ec_code_x64_regmap[src];
+
+    ec_code_intel_op_xor_r2r(builder, src, dst);
+}
+
+static void
+ec_code_x64_xorm(ec_code_builder_t *builder, uint32_t dst, uint32_t idx,
+                 uint32_t bit)
+{
+    dst = ec_code_x64_regmap[dst];
+
+    if (builder->linear) {
+        ec_code_intel_op_xor_m2r(
+            builder, REG_SI, REG_DX, 1,
+            idx * builder->width * builder->bits + bit * builder->width, dst);
+    } else {
+        if (builder->base != idx) {
+            ec_code_intel_op_mov_m2r(builder, REG_SI, REG_NULL, 0, idx * 8,
+                                     REG_BX);
+            builder->base = idx;
+        }
+        ec_code_intel_op_xor_m2r(builder, REG_BX, REG_DX, 1,
+                                 bit * builder->width, dst);
+    }
+}
+
+static char *ec_code_x64_needed_flags[] = {NULL};
+
+ec_code_gen_t ec_code_gen_x64 = {.name = "x64",
+                                 .flags = ec_code_x64_needed_flags,
+                                 .width = sizeof(uint64_t),
+                                 .prolog = ec_code_x64_prolog,
+                                 .epilog = ec_code_x64_epilog,
+                                 .load = ec_code_x64_load,
+                                 .store = ec_code_x64_store,
+                                 .copy = ec_code_x64_copy,
+                                 .xor2 = ec_code_x64_xor2,
+                                 .xor3 = NULL,
+                                 .xorm = ec_code_x64_xorm};
diff --git a/xlators/cluster/ec/src/ec-code-x64.h b/xlators/cluster/ec/src/ec-code-x64.h
new file mode 100644
index 00000000000..bd8174e4bf5
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code-x64.h
@@ -0,0 +1,18 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_CODE_X64_H__
+#define __EC_CODE_X64_H__
+
+#include "ec-code.h"
+
+extern ec_code_gen_t ec_code_gen_x64;
+
+#endif /* __EC_CODE_X64_H__ */
diff --git a/xlators/cluster/ec/src/ec-code.c b/xlators/cluster/ec/src/ec-code.c
new file mode 100644
index 00000000000..03162ae05a9
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code.c
@@ -0,0 +1,1060 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <string.h>
+#include <sys/mman.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <ctype.h>
+
+#include <glusterfs/syscall.h>
+
+#include "ec-mem-types.h"
+#include "ec-code.h"
+#include "ec-messages.h"
+#include "ec-code-c.h"
+#include "ec-helpers.h"
+
+#ifdef USE_EC_DYNAMIC_X64
+#include "ec-code-x64.h"
+#endif
+
+#ifdef USE_EC_DYNAMIC_SSE
+#include "ec-code-sse.h"
+#endif
+
+#ifdef USE_EC_DYNAMIC_AVX
+#include "ec-code-avx.h"
+#endif
+
+#define EC_CODE_SIZE (1024 * 64)
+#define EC_CODE_ALIGN 4096
+
+#define EC_CODE_CHUNK_MIN_SIZE 512
+
+#define EC_PROC_BUFFER_SIZE 4096
+
+#define PROC_CPUINFO "/proc/cpuinfo"
+
+struct _ec_code_proc;
+typedef struct _ec_code_proc ec_code_proc_t;
+
+struct _ec_code_proc {
+    int32_t fd;
+    gf_boolean_t eof;
+    gf_boolean_t error;
+    gf_boolean_t skip;
+    ssize_t size;
+    ssize_t pos;
+    char buffer[EC_PROC_BUFFER_SIZE];
+};
+
+static ec_code_gen_t *ec_code_gen_table[] = {
+#ifdef USE_EC_DYNAMIC_AVX
+    &ec_code_gen_avx,
+#endif
+#ifdef USE_EC_DYNAMIC_SSE
+    &ec_code_gen_sse,
+#endif
+#ifdef USE_EC_DYNAMIC_X64
+    &ec_code_gen_x64,
+#endif
+    NULL};
+
+static void
+ec_code_arg_set(ec_code_arg_t *arg, uint32_t value)
+{
+    arg->value = value;
+}
+
+static void
+ec_code_arg_assign(ec_code_builder_t *builder, ec_code_op_t *op,
+                   ec_code_arg_t *arg, uint32_t reg)
+{
+    arg->value = reg;
+
+    if (builder->regs <= reg) {
+        builder->regs = reg + 1;
+    }
+}
+
+static void
+ec_code_arg_use(ec_code_builder_t *builder, ec_code_op_t *op,
+                ec_code_arg_t *arg, uint32_t reg)
+{
+    arg->value = reg;
+}
+
+static void
+ec_code_arg_update(ec_code_builder_t *builder, ec_code_op_t *op,
+                   ec_code_arg_t *arg, uint32_t reg)
+{
+    arg->value = reg;
+}
+
+static ec_code_op_t *
+ec_code_op_next(ec_code_builder_t *builder)
+{
+    ec_code_op_t *op;
+
+    op = &builder->ops[builder->count++];
+    memset(op, 0, sizeof(ec_code_op_t));
+
+    return op;
+}
+
+static void
+ec_code_load(ec_code_builder_t *builder, uint32_t bit, uint32_t offset)
+{
+    ec_code_op_t *op;
+
+    op = ec_code_op_next(builder);
+
+    op->op = EC_GF_OP_LOAD;
+    ec_code_arg_assign(builder, op, &op->arg1, builder->map[bit]);
+    ec_code_arg_set(&op->arg2, offset);
+    ec_code_arg_set(&op->arg3, bit);
+}
+
+static void
+ec_code_store(ec_code_builder_t *builder, uint32_t reg, uint32_t bit)
+{
+    ec_code_op_t *op;
+
+    op = ec_code_op_next(builder);
+
+    op->op = EC_GF_OP_STORE;
+    ec_code_arg_use(builder, op, &op->arg1, builder->map[reg]);
+    ec_code_arg_set(&op->arg2, 0);
+    ec_code_arg_set(&op->arg3, bit);
+}
+
+static void
+ec_code_copy(ec_code_builder_t *builder, uint32_t dst, uint32_t src)
+{
+    ec_code_op_t *op;
+
+    op = ec_code_op_next(builder);
+
+    op->op = EC_GF_OP_COPY;
+    ec_code_arg_assign(builder, op, &op->arg1, builder->map[dst]);
+    ec_code_arg_use(builder, op, &op->arg2, builder->map[src]);
+    ec_code_arg_set(&op->arg3, 0);
+}
+
+static void
+ec_code_xor2(ec_code_builder_t *builder, uint32_t dst, uint32_t src)
+{
+    ec_code_op_t *op;
+
+    op = ec_code_op_next(builder);
+
+    op->op = EC_GF_OP_XOR2;
+    ec_code_arg_update(builder, op, &op->arg1, builder->map[dst]);
+    ec_code_arg_use(builder, op, &op->arg2, builder->map[src]);
+    ec_code_arg_set(&op->arg3, 0);
+}
+
+static void
+ec_code_xor3(ec_code_builder_t *builder, uint32_t dst, uint32_t src1,
+             uint32_t src2)
+{
+    ec_code_op_t *op;
+
+    if (builder->code->gen->xor3 == NULL) {
+        ec_code_copy(builder, dst, src1);
+        ec_code_xor2(builder, dst, src2);
+
+        return;
+    }
+
+    op = ec_code_op_next(builder);
+
+    op->op = EC_GF_OP_XOR3;
+    ec_code_arg_assign(builder, op, &op->arg1, builder->map[dst]);
+    ec_code_arg_use(builder, op, &op->arg2, builder->map[src1]);
+    ec_code_arg_use(builder, op, &op->arg3, builder->map[src2]);
+}
+
+static void
+ec_code_xorm(ec_code_builder_t *builder, uint32_t bit, uint32_t offset)
+{
+    ec_code_op_t *op;
+
+    op = ec_code_op_next(builder);
+
+    op->op = EC_GF_OP_XORM;
+    ec_code_arg_update(builder, op, &op->arg1, builder->map[bit]);
+    ec_code_arg_set(&op->arg2, offset);
+    ec_code_arg_set(&op->arg3, bit);
+}
+
+static void
+ec_code_dup(ec_code_builder_t *builder, ec_gf_op_t *op)
+{
+    switch (op->op) {
+        case EC_GF_OP_COPY:
+            ec_code_copy(builder, op->arg1, op->arg2);
+            break;
+        case EC_GF_OP_XOR2:
+            ec_code_xor2(builder, op->arg1, op->arg2);
+            break;
+        case EC_GF_OP_XOR3:
+            ec_code_xor3(builder, op->arg1, op->arg2, op->arg3);
+            break;
+        default:
+            break;
+    }
+}
+
+static void
+ec_code_gf_load(ec_code_builder_t *builder, uint32_t offset)
+{
+    uint32_t i;
+
+    for (i = 0; i < builder->code->gf->bits; i++) {
+        ec_code_load(builder, i, offset);
+    }
+}
+
+static void
+ec_code_gf_load_xor(ec_code_builder_t *builder, uint32_t offset)
+{
+    uint32_t i;
+
+    for (i = 0; i < builder->code->gf->bits; i++) {
+        ec_code_xorm(builder, i, offset);
+    }
+}
+
+static void
+ec_code_gf_store(ec_code_builder_t *builder)
+{
+    uint32_t i;
+
+    for (i = 0; i < builder->code->gf->bits; i++) {
+        ec_code_store(builder, i, i);
+    }
+}
+
+static void
+ec_code_gf_clear(ec_code_builder_t *builder)
+{
+    uint32_t i;
+
+    ec_code_xor2(builder, 0, 0);
+    for (i = 0; i < builder->code->gf->bits; i++) {
+        ec_code_store(builder, 0, i);
+    }
+}
+
+static void
+ec_code_gf_mul(ec_code_builder_t *builder, uint32_t value)
+{
+    ec_gf_mul_t *mul;
+    ec_gf_op_t *op;
+    uint32_t map[EC_GF_MAX_REGS];
+    int32_t i;
+
+    mul = builder->code->gf->table[value];
+    for (op = mul->ops; op->op != EC_GF_OP_END; op++) {
+        ec_code_dup(builder, op);
+    }
+
+    for (i = 0; i < mul->regs; i++) {
+        map[i] = builder->map[mul->map[i]];
+    }
+    memcpy(builder->map, map, sizeof(uint32_t) * mul->regs);
+}
+
+static ec_code_builder_t *
+ec_code_prepare(ec_code_t *code, uint32_t count, uint32_t width,
+                gf_boolean_t linear)
+{
+    ec_code_builder_t *builder;
+    uint32_t i;
+
+    count *= code->gf->bits + code->gf->max_ops;
+    count += code->gf->bits;
+    builder = GF_MALLOC(
+        sizeof(ec_code_builder_t) + sizeof(ec_code_op_t) * count,
+        ec_mt_ec_code_builder_t);
+    if (builder == NULL) {
+        return EC_ERR(ENOMEM);
+    }
+
+    builder->address = 0;
+    builder->code = code;
+    builder->size = 0;
+    builder->count = 0;
+    builder->regs = 0;
+    builder->error = 0;
+    builder->bits = code->gf->bits;
+    builder->width = width;
+    builder->data = NULL;
+    builder->linear = linear;
+    builder->base = -1;
+
+    for (i = 0; i < EC_GF_MAX_REGS; i++) {
+        builder->map[i] = i;
+    }
+
+    return builder;
+}
+
+static size_t
+ec_code_space_size(void)
+{
+    return (sizeof(ec_code_space_t) + 15) & ~15;
+}
+
+static size_t
+ec_code_chunk_size(void)
+{
+    return (sizeof(ec_code_chunk_t) + 15) & ~15;
+}
+
+static ec_code_chunk_t *
+ec_code_chunk_from_space(ec_code_space_t *space)
+{
+    return (ec_code_chunk_t *)((uintptr_t)space + ec_code_space_size());
+}
+
+static void *
+ec_code_to_executable(ec_code_space_t *space, void *addr)
+{
+    return (void *)((uintptr_t)addr - (uintptr_t)space +
+                    (uintptr_t)space->exec);
+}
+
+static void *
+ec_code_from_executable(ec_code_space_t *space, void *addr)
+{
+    return (void *)((uintptr_t)addr - (uintptr_t)space->exec +
+                    (uintptr_t)space);
+}
+
+static void *
+ec_code_func_from_chunk(ec_code_chunk_t *chunk, void **exec)
+{
+    void *addr;
+
+    addr = (void *)((uintptr_t)chunk + ec_code_chunk_size());
+
+    *exec = ec_code_to_executable(chunk->space, addr);
+
+    return addr;
+}
+
+static ec_code_chunk_t *
+ec_code_chunk_from_func(ec_code_func_linear_t func)
+{
+    ec_code_chunk_t *chunk;
+
+    chunk = (ec_code_chunk_t *)((uintptr_t)func - ec_code_chunk_size());
+
+    return ec_code_from_executable(chunk->space, chunk);
+}
+
+static ec_code_chunk_t *
+ec_code_chunk_split(ec_code_chunk_t *chunk, size_t size)
+{
+    ec_code_chunk_t *extra;
+    ssize_t avail;
+
+    avail = chunk->size - size - ec_code_chunk_size();
+    if (avail > 0) {
+        extra = (ec_code_chunk_t *)((uintptr_t)chunk + chunk->size - avail);
+        extra->space = chunk->space;
+        extra->size = avail;
+        list_add(&extra->list, &chunk->list);
+        chunk->size = size;
+    }
+    list_del_init(&chunk->list);
+
+    return chunk;
+}
+
+static gf_boolean_t
+ec_code_chunk_touch(ec_code_chunk_t *prev, ec_code_chunk_t *next)
+{
+    uintptr_t end;
+
+    end = (uintptr_t)prev + ec_code_chunk_size() + prev->size;
+    return (end == (uintptr_t)next);
+}
+
+static ec_code_space_t *
+ec_code_space_create(ec_code_t *code, size_t size)
+{
+    char path[] = GLUSTERFS_LIBEXECDIR "/ec-code-dynamic.XXXXXX";
+    ec_code_space_t *space;
+    void *exec;
+    int32_t fd, err;
+
+    /* We need to create memory areas to store the generated dynamic code.
+     * Obviously these areas need to be written to be able to create the
+     * code and they also need to be executable to execute it.
+     *
+     * However it's a bad practice to have a memory region that is both
+     * writable *and* executable. In fact, selinux forbids this and causes
+     * attempts to do so to fail (unless specifically configured).
+     *
+     * To solve the problem we'll use two distinct memory areas mapped to
+     * the same physical storage. One of the memory areas will have write
+     * permission, and the other will have execute permission. Both areas
+     * will have the same contents. The physical storage will be a regular
+     * file that will be mmapped to both areas.
+     */
+
+    /* We need to create a temporary file as the backend storage for the
+     * memory mapped areas. */
+    /* coverity[secure_temp] mkstemp uses 0600 as the mode and is safe */
+    fd = mkstemp(path);
+    if (fd < 0) {
+        err = errno;
+        gf_msg(THIS->name, GF_LOG_ERROR, err, EC_MSG_DYN_CREATE_FAILED,
+               "Unable to create a temporary file for the ec dynamic "
+               "code");
+        space = EC_ERR(err);
+        goto done;
+    }
+    /* Once created we don't need to keep it in the file system. It will
+     * still exist until we close the last file descriptor or unmap the
+     * memory areas bound to the file. */
+    sys_unlink(path);
+
+    size = (size + EC_CODE_ALIGN - 1) & ~(EC_CODE_ALIGN - 1);
+    if (sys_ftruncate(fd, size) < 0) {
+        err = errno;
+        gf_msg(THIS->name, GF_LOG_ERROR, err, EC_MSG_DYN_CREATE_FAILED,
+               "Unable to resize the file for the ec dynamic code");
+        space = EC_ERR(err);
+        goto done_close;
+    }
+
+    /* This creates an executable memory area to be able to run the
+     * generated fragments of code. */
+    exec = mmap(NULL, size, PROT_READ | PROT_EXEC, MAP_SHARED, fd, 0);
+    if (exec == MAP_FAILED) {
+        err = errno;
+        gf_msg(THIS->name, GF_LOG_ERROR, err, EC_MSG_DYN_CREATE_FAILED,
+               "Unable to map the executable area for the ec dynamic "
+               "code");
+        space = EC_ERR(err);
+        goto done_close;
+    }
+    /* It's not important to check the return value of mlock(). If it fails
+     * everything will continue to work normally. */
+    mlock(exec, size);
+
+    /* This maps a read/write memory area to be able to create the dynamici
+     * code. */
+    space = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
+    if (space == MAP_FAILED) {
+        err = errno;
+        gf_msg(THIS->name, GF_LOG_ERROR, err, EC_MSG_DYN_CREATE_FAILED,
+               "Unable to map the writable area for the ec dynamic "
+               "code");
+        space = EC_ERR(err);
+
+        munmap(exec, size);
+
+        goto done_close;
+    }
+
+    space->exec = exec;
+    space->size = size;
+    space->code = code;
+    list_add_tail(&space->list, &code->spaces);
+    INIT_LIST_HEAD(&space->chunks);
+
+done_close:
+    /* If everything has succeeded, we already have the memory areas
+     * mapped. We don't need the file descriptor anymore because the
+     * backend storage will be there until the mmap()'d regions are
+     * unmapped. */
+    sys_close(fd);
+done:
+    return space;
+}
+
+static void
+ec_code_space_destroy(ec_code_space_t *space)
+{
+    list_del_init(&space->list);
+
+    munmap(space->exec, space->size);
+    munmap(space, space->size);
+}
+
+static void
+ec_code_chunk_merge(ec_code_chunk_t *chunk)
+{
+    ec_code_chunk_t *item, *tmp;
+
+    list_for_each_entry_safe(item, tmp, &chunk->space->chunks, list)
+    {
+        if ((uintptr_t)item > (uintptr_t)chunk) {
+            list_add_tail(&chunk->list, &item->list);
+            if (ec_code_chunk_touch(chunk, item)) {
+                chunk->size += item->size + ec_code_chunk_size();
+                list_del_init(&item->list);
+            }
+
+            goto check;
+        }
+        if (ec_code_chunk_touch(item, chunk)) {
+            item->size += chunk->size + ec_code_chunk_size();
+            list_del_init(&item->list);
+            chunk = item;
+        }
+    }
+    list_add_tail(&chunk->list, &chunk->space->chunks);
+
+check:
+    if (chunk->size ==
+        chunk->space->size - ec_code_space_size() - ec_code_chunk_size()) {
+        ec_code_space_destroy(chunk->space);
+    }
+}
+
+static ec_code_chunk_t *
+ec_code_space_alloc(ec_code_t *code, size_t size)
+{
+    ec_code_space_t *space;
+    ec_code_chunk_t *chunk;
+    size_t map_size;
+
+    /* To minimize fragmentation, we only allocate chunks of sizes multiples
+     * of EC_CODE_CHUNK_MIN_SIZE. */
+    size = ((size + ec_code_chunk_size() + EC_CODE_CHUNK_MIN_SIZE - 1) &
+            ~(EC_CODE_CHUNK_MIN_SIZE - 1)) -
+           ec_code_chunk_size();
+    list_for_each_entry(space, &code->spaces, list)
+    {
+        list_for_each_entry(chunk, &space->chunks, list)
+        {
+            if (chunk->size >= size) {
+                goto out;
+            }
+        }
+    }
+
+    map_size = EC_CODE_SIZE - ec_code_space_size() - ec_code_chunk_size();
+    if (map_size < size) {
+        map_size = size;
+    }
+    space = ec_code_space_create(code, map_size);
+    if (EC_IS_ERR(space)) {
+        return (ec_code_chunk_t *)space;
+    }
+
+    chunk = ec_code_chunk_from_space(space);
+    chunk->size = map_size - ec_code_space_size() - ec_code_chunk_size();
+    list_add(&chunk->list, &space->chunks);
+
+out:
+    chunk->space = space;
+
+    return ec_code_chunk_split(chunk, size);
+}
+
+static ec_code_chunk_t *
+ec_code_alloc(ec_code_t *code, uint32_t size)
+{
+    ec_code_chunk_t *chunk;
+
+    LOCK(&code->lock);
+
+    chunk = ec_code_space_alloc(code, size);
+
+    UNLOCK(&code->lock);
+
+    return chunk;
+}
+
+static void
+ec_code_free(ec_code_chunk_t *chunk)
+{
+    gf_lock_t *lock;
+
+    lock = &chunk->space->code->lock;
+    LOCK(lock);
+
+    ec_code_chunk_merge(chunk);
+
+    UNLOCK(lock);
+}
+
+static int32_t
+ec_code_write(ec_code_builder_t *builder)
+{
+    ec_code_gen_t *gen;
+    ec_code_op_t *op;
+    uint32_t i;
+
+    builder->error = 0;
+    builder->size = 0;
+    builder->address = 0;
+    builder->base = -1;
+
+    gen = builder->code->gen;
+    gen->prolog(builder);
+    for (i = 0; i < builder->count; i++) {
+        op = &builder->ops[i];
+        switch (op->op) {
+            case EC_GF_OP_LOAD:
+                gen->load(builder, op->arg1.value, op->arg2.value,
+                          op->arg3.value);
+                break;
+            case EC_GF_OP_STORE:
+                gen->store(builder, op->arg1.value, op->arg3.value);
+                break;
+            case EC_GF_OP_COPY:
+                gen->copy(builder, op->arg1.value, op->arg2.value);
+                break;
+            case EC_GF_OP_XOR2:
+                gen->xor2(builder, op->arg1.value, op->arg2.value);
+                break;
+            case EC_GF_OP_XOR3:
+                gen->xor3(builder, op->arg1.value, op->arg2.value,
+                          op->arg3.value);
+                break;
+            case EC_GF_OP_XORM:
+                gen->xorm(builder, op->arg1.value, op->arg2.value,
+                          op->arg3.value);
+                break;
+            default:
+                break;
+        }
+    }
+    gen->epilog(builder);
+
+    return builder->error;
+}
+
+static void *
+ec_code_compile(ec_code_builder_t *builder)
+{
+    ec_code_chunk_t *chunk;
+    void *func;
+    int32_t err;
+
+    err = ec_code_write(builder);
+    if (err != 0) {
+        return EC_ERR(err);
+    }
+
+    chunk = ec_code_alloc(builder->code, builder->size);
+    if (EC_IS_ERR(chunk)) {
+        return chunk;
+    }
+    builder->data = ec_code_func_from_chunk(chunk, &func);
+
+    err = ec_code_write(builder);
+    if (err != 0) {
+        ec_code_free(chunk);
+
+        return EC_ERR(err);
+    }
+
+    GF_FREE(builder);
+
+    return func;
+}
+
+ec_code_t *
+ec_code_create(ec_gf_t *gf, ec_code_gen_t *gen)
+{
+    ec_code_t *code;
+
+    code = GF_MALLOC(sizeof(ec_code_t), ec_mt_ec_code_t);
+    if (code == NULL) {
+        return EC_ERR(ENOMEM);
+    }
+    memset(code, 0, sizeof(ec_code_t));
+    INIT_LIST_HEAD(&code->spaces);
+    LOCK_INIT(&code->lock);
+
+    code->gf = gf;
+    code->gen = gen;
+
+    return code;
+}
+
+void
+ec_code_destroy(ec_code_t *code)
+{
+    if (!list_empty(&code->spaces)) {
+    }
+
+    LOCK_DESTROY(&code->lock);
+
+    GF_FREE(code);
+}
+
+static uint32_t
+ec_code_value_next(uint32_t *values, uint32_t count, uint32_t *offset)
+{
+    uint32_t i, next;
+
+    next = 0;
+    for (i = *offset + 1; i < count; i++) {
+        next = values[i];
+        if (next != 0) {
+            break;
+        }
+    }
+    *offset = i;
+
+    return next;
+}
+
+static void *
+ec_code_build_dynamic(ec_code_t *code, uint32_t width, uint32_t *values,
+                      uint32_t count, gf_boolean_t linear)
+{
+    ec_code_builder_t *builder;
+    uint32_t offset, val, next;
+
+    builder = ec_code_prepare(code, count, width, linear);
+    if (EC_IS_ERR(builder)) {
+        return builder;
+    }
+
+    offset = -1;
+    next = ec_code_value_next(values, count, &offset);
+    if (next != 0) {
+        ec_code_gf_load(builder, offset);
+        do {
+            val = next;
+            next = ec_code_value_next(values, count, &offset);
+            if (next != 0) {
+                ec_code_gf_mul(builder, ec_gf_div(code->gf, val, next));
+                ec_code_gf_load_xor(builder, offset);
+            }
+        } while (next != 0);
+        ec_code_gf_mul(builder, val);
+        ec_code_gf_store(builder);
+    } else {
+        ec_code_gf_clear(builder);
+    }
+
+    return ec_code_compile(builder);
+}
+
+static void *
+ec_code_build(ec_code_t *code, uint32_t width, uint32_t *values, uint32_t count,
+              gf_boolean_t linear)
+{
+    void *func;
+
+    if (code->gen != NULL) {
+        func = ec_code_build_dynamic(code, width, values, count, linear);
+        if (!EC_IS_ERR(func)) {
+            return func;
+        }
+
+        gf_msg_debug(THIS->name, GF_LOG_DEBUG,
+                     "Unable to generate dynamic code. Falling back "
+                     "to precompiled code");
+
+        /* The dynamic code generation shouldn't fail in normal
+         * conditions, but if it fails at some point, it's very
+         * probable that it will fail again, so we completely disable
+         * dynamic code generation. */
+        code->gen = NULL;
+    }
+
+    ec_code_c_prepare(code->gf, values, count);
+
+    if (linear) {
+        return ec_code_c_linear;
+    }
+
+    return ec_code_c_interleaved;
+}
+
+ec_code_func_linear_t
+ec_code_build_linear(ec_code_t *code, uint32_t width, uint32_t *values,
+                     uint32_t count)
+{
+    return (ec_code_func_linear_t)ec_code_build(code, width, values, count,
+                                                _gf_true);
+}
+
+ec_code_func_interleaved_t
+ec_code_build_interleaved(ec_code_t *code, uint32_t width, uint32_t *values,
+                          uint32_t count)
+{
+    return (ec_code_func_interleaved_t)ec_code_build(code, width, values, count,
+                                                     _gf_false);
+}
+
+void
+ec_code_release(ec_code_t *code, ec_code_func_t *func)
+{
+    if ((func->linear != ec_code_c_linear) &&
+        (func->interleaved != ec_code_c_interleaved)) {
+        ec_code_free(ec_code_chunk_from_func(func->linear));
+    }
+}
+
+void
+ec_code_error(ec_code_builder_t *builder, int32_t error)
+{
+    if (builder->error == 0) {
+        gf_msg(THIS->name, GF_LOG_ERROR, error, EC_MSG_DYN_CODEGEN_FAILED,
+               "Failed to generate dynamic code");
+        builder->error = error;
+    }
+}
+
+void
+ec_code_emit(ec_code_builder_t *builder, uint8_t *bytes, uint32_t count)
+{
+    if (builder->error != 0) {
+        return;
+    }
+
+    if (builder->data != NULL) {
+        memcpy(builder->data + builder->size, bytes, count);
+    }
+
+    builder->size += count;
+    builder->address += count;
+}
+
+static char *
+ec_code_proc_trim_left(char *text, ssize_t *length)
+{
+    ssize_t len;
+
+    for (len = *length; (len > 0) && isspace(*text); len--) {
+        text++;
+    }
+    *length = len;
+
+    return text;
+}
+
+static char *
+ec_code_proc_trim_right(char *text, ssize_t *length, char sep)
+{
+    char *last;
+    ssize_t len;
+
+    len = *length;
+
+    last = text;
+    for (len = *length; (len > 0) && (*text != sep); len--) {
+        if (!isspace(*text)) {
+            last = text + 1;
+        }
+        text++;
+    }
+    *last = 0;
+    *length = len;
+
+    return text;
+}
+
+static char *
+ec_code_proc_line_parse(ec_code_proc_t *file, ssize_t *length)
+{
+    char *text, *end;
+    ssize_t len;
+
+    len = file->size - file->pos;
+    text = ec_code_proc_trim_left(file->buffer + file->pos, &len);
+    end = ec_code_proc_trim_right(text, &len, '\n');
+    if (len == 0) {
+        if (!file->eof) {
+            if (text == file->buffer) {
+                file->size = file->pos = 0;
+                file->skip = _gf_true;
+            } else {
+                file->size = file->pos = end - text;
+                memmove(file->buffer, text, file->pos + 1);
+            }
+            len = sys_read(file->fd, file->buffer + file->pos,
+                           sizeof(file->buffer) - file->pos - 1);
+            if (len > 0) {
+                file->size += len;
+            }
+            file->error = len < 0;
+            file->eof = len <= 0;
+
+            return NULL;
+        }
+        file->size = file->pos = 0;
+    } else {
+        file->pos = end - file->buffer + 1;
+    }
+
+    *length = end - text;
+
+    if (file->skip) {
+        file->skip = _gf_false;
+        text = NULL;
+    }
+
+    return text;
+}
+
+static char *
+ec_code_proc_line(ec_code_proc_t *file, ssize_t *length)
+{
+    char *text;
+
+    text = NULL;
+    while (!file->eof) {
+        text = ec_code_proc_line_parse(file, length);
+        if (text != NULL) {
+            break;
+        }
+    }
+
+    return text;
+}
+
+static char *
+ec_code_proc_split(char *text, ssize_t *length, char sep)
+{
+    text = ec_code_proc_trim_right(text, length, sep);
+    if (*length == 0) {
+        return NULL;
+    }
+    (*length)--;
+    text++;
+
+    return ec_code_proc_trim_left(text, length);
+}
+
+static uint32_t
+ec_code_cpu_check(uint32_t idx, char *list, uint32_t count)
+{
+    ec_code_gen_t *gen;
+    char **ptr;
+    char *table[count + 1];
+    uint32_t i;
+
+    for (i = 0; i < count; i++) {
+        table[i] = list;
+        list += strlen(list) + 1;
+    }
+
+    gen = ec_code_gen_table[idx];
+    while (gen != NULL) {
+        for (ptr = gen->flags; *ptr != NULL; ptr++) {
+            for (i = 0; i < count; i++) {
+                if (strcmp(*ptr, table[i]) == 0) {
+                    break;
+                }
+            }
+            if (i >= count) {
+                gen = ec_code_gen_table[++idx];
+                break;
+            }
+        }
+        if (*ptr == NULL) {
+            break;
+        }
+    }
+
+    return idx;
+}
+
+ec_code_gen_t *
+ec_code_detect(xlator_t *xl, const char *def)
+{
+    ec_code_proc_t file;
+    ec_code_gen_t *gen = NULL;
+    char *line, *data, *list;
+    ssize_t length;
+    uint32_t count, base, select;
+
+    if (strcmp(def, "none") == 0) {
+        gf_msg(xl->name, GF_LOG_INFO, 0, EC_MSG_EXTENSION_NONE,
+               "Not using any cpu extensions");
+
+        return NULL;
+    }
+
+    file.fd = sys_open(PROC_CPUINFO, O_RDONLY, 0);
+    if (file.fd < 0) {
+        goto out;
+    }
+    file.size = file.pos = 0;
+    file.eof = file.error = file.skip = _gf_false;
+
+    select = 0;
+    if (strcmp(def, "auto") != 0) {
+        while (ec_code_gen_table[select] != NULL) {
+            if (strcmp(ec_code_gen_table[select]->name, def) == 0) {
+                break;
+            }
+            select++;
+        }
+        if (ec_code_gen_table[select] == NULL) {
+            gf_msg(xl->name, GF_LOG_WARNING, EINVAL, EC_MSG_EXTENSION_UNKNOWN,
+                   "CPU extension '%s' is not known. Not using any cpu "
+                   "extensions",
+                   def);
+
+            return NULL;
+        }
+    } else {
+        def = NULL;
+    }
+
+    while ((line = ec_code_proc_line(&file, &length)) != NULL) {
+        data = ec_code_proc_split(line, &length, ':');
+        if ((data != NULL) && (strcmp(line, "flags") == 0)) {
+            list = data;
+            count = 0;
+            while ((data != NULL) && (*data != 0)) {
+                count++;
+                data = ec_code_proc_split(data, &length, ' ');
+            }
+            base = select;
+            select = ec_code_cpu_check(select, list, count);
+            if ((base != select) && (def != NULL)) {
+                gf_msg(xl->name, GF_LOG_WARNING, ENOTSUP,
+                       EC_MSG_EXTENSION_UNSUPPORTED,
+                       "CPU extension '%s' is not supported", def);
+                def = NULL;
+            }
+        }
+    }
+
+    if (file.error) {
+        gf_msg(xl->name, GF_LOG_WARNING, 0, EC_MSG_EXTENSION_FAILED,
+               "Unable to determine supported CPU extensions. Not using any "
+               "cpu extensions");
+
+        gen = NULL;
+    } else {
+        gen = ec_code_gen_table[select];
+        if (gen == NULL) {
+            gf_msg(xl->name, GF_LOG_INFO, 0, EC_MSG_EXTENSION_NONE,
+                   "Not using any cpu extensions");
+        } else {
+            gf_msg(xl->name, GF_LOG_INFO, 0, EC_MSG_EXTENSION,
+                   "Using '%s' CPU extensions", gen->name);
+        }
+    }
+
+    sys_close(file.fd);
+
+out:
+    return gen;
+}
diff --git a/xlators/cluster/ec/src/ec-code.h b/xlators/cluster/ec/src/ec-code.h
new file mode 100644
index 00000000000..75fb35d93e3
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code.h
@@ -0,0 +1,44 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_CODE_H__
+#define __EC_CODE_H__
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/list.h>
+
+#include "ec-types.h"
+#include "ec-galois.h"
+
+ec_code_gen_t *
+ec_code_detect(xlator_t *xl, const char *def);
+
+ec_code_t *
+ec_code_create(ec_gf_t *gf, ec_code_gen_t *gen);
+
+void
+ec_code_destroy(ec_code_t *code);
+
+ec_code_func_linear_t
+ec_code_build_linear(ec_code_t *code, uint32_t width, uint32_t *values,
+                     uint32_t count);
+ec_code_func_interleaved_t
+ec_code_build_interleaved(ec_code_t *code, uint32_t width, uint32_t *values,
+                          uint32_t count);
+void
+ec_code_release(ec_code_t *code, ec_code_func_t *func);
+
+void
+ec_code_error(ec_code_builder_t *builder, int32_t error);
+
+void
+ec_code_emit(ec_code_builder_t *builder, uint8_t *bytes, uint32_t count);
+
+#endif /* __EC_CODE_H__ */
diff --git a/xlators/cluster/ec/src/ec-combine.c b/xlators/cluster/ec/src/ec-combine.c
new file mode 100644
index 00000000000..703a30e2485
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-combine.c
@@ -0,0 +1,995 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <fnmatch.h>
+
+#include "libxlator.h"
+#include <glusterfs/byte-order.h>
+
+#include "ec-types.h"
+#include "ec-helpers.h"
+#include "ec-common.h"
+#include "ec-combine.h"
+#include "ec-messages.h"
+#include <glusterfs/quota-common-utils.h>
+
+#define EC_QUOTA_PREFIX "trusted.glusterfs.quota."
+
+#define EC_MISSING_DATA ((data_t *)1ULL)
+
+struct _ec_dict_info;
+typedef struct _ec_dict_info ec_dict_info_t;
+
+struct _ec_dict_combine;
+typedef struct _ec_dict_combine ec_dict_combine_t;
+
+struct _ec_dict_info {
+    dict_t *dict;
+    int32_t count;
+};
+
+struct _ec_dict_combine {
+    ec_cbk_data_t *cbk;
+    int32_t which;
+};
+
+int32_t
+ec_combine_write(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    int valid = 0;
+
+    if (!fop || !dst || !src)
+        return 0;
+
+    switch (fop->id) {
+        case GF_FOP_REMOVEXATTR:
+        case GF_FOP_FREMOVEXATTR:
+        case GF_FOP_SETXATTR:
+        case GF_FOP_FSETXATTR:
+            return 1;
+
+        case GF_FOP_SYMLINK:
+        case GF_FOP_LINK:
+        case GF_FOP_CREATE:
+        case GF_FOP_MKNOD:
+        case GF_FOP_MKDIR:
+            valid = 3;
+            break;
+        case GF_FOP_UNLINK:
+        case GF_FOP_RMDIR:
+        case GF_FOP_SETATTR:
+        case GF_FOP_FSETATTR:
+        case GF_FOP_TRUNCATE:
+        case GF_FOP_FTRUNCATE:
+        case GF_FOP_WRITE:
+        case GF_FOP_FALLOCATE:
+        case GF_FOP_DISCARD:
+        case GF_FOP_ZEROFILL:
+            valid = 2;
+            break;
+        case GF_FOP_RENAME:
+            valid = 5;
+            break;
+        default:
+            gf_msg_callingfn(fop->xl->name, GF_LOG_WARNING, EINVAL,
+                             EC_MSG_INVALID_FOP, "Invalid fop %d", fop->id);
+            return 0;
+            break;
+    }
+
+    if (!ec_iatt_combine(fop, dst->iatt, src->iatt, valid)) {
+        gf_msg(fop->xl->name, GF_LOG_NOTICE, 0, EC_MSG_IATT_MISMATCH,
+               "Mismatching iatt in "
+               "answers of '%s'",
+               gf_fop_list[fop->id]);
+        return 0;
+    }
+    return 1;
+}
+
+void
+ec_iatt_time_merge(int64_t *dst_sec, uint32_t *dst_nsec, int64_t src_sec,
+                   uint32_t src_nsec)
+{
+    if ((*dst_sec < src_sec) ||
+        ((*dst_sec == src_sec) && (*dst_nsec < src_nsec))) {
+        *dst_sec = src_sec;
+        *dst_nsec = src_nsec;
+    }
+}
+
+static gf_boolean_t
+ec_iatt_is_trusted(ec_fop_data_t *fop, struct iatt *iatt)
+{
+    uint64_t ino;
+    int32_t i;
+
+    /* Only the top level fop will have fop->locks filled. */
+    while (fop->parent != NULL) {
+        fop = fop->parent;
+    }
+
+    /* Lookups are special requests always done without locks taken but they
+     * require to be able to identify differences between bricks. Special
+     * handling of these differences is already done in lookup specific code
+     * so we shouldn't ignore any difference here and consider all iatt
+     * structures as trusted. */
+    if (fop->id == GF_FOP_LOOKUP) {
+        return _gf_true;
+    }
+
+    /* Check if the iatt references an inode locked by the current fop */
+    for (i = 0; i < fop->lock_count; i++) {
+        ino = gfid_to_ino(fop->locks[i].lock->loc.inode->gfid);
+        if (iatt->ia_ino == ino) {
+            return _gf_true;
+        }
+    }
+
+    return _gf_false;
+}
+
+int32_t
+ec_iatt_combine(ec_fop_data_t *fop, struct iatt *dst, struct iatt *src,
+                int32_t count)
+{
+    int32_t i;
+    gf_boolean_t failed = _gf_false;
+
+    for (i = 0; i < count; i++) {
+        /* Check for basic fields. These fields must be equal always, even if
+         * the inode is not locked because in these cases the parent inode
+         * will be locked and differences in these fields require changes in
+         * the parent directory. */
+        if ((dst[i].ia_ino != src[i].ia_ino) ||
+            (((dst[i].ia_type == IA_IFBLK) || (dst[i].ia_type == IA_IFCHR)) &&
+             (dst[i].ia_rdev != src[i].ia_rdev)) ||
+            (gf_uuid_compare(dst[i].ia_gfid, src[i].ia_gfid) != 0)) {
+            failed = _gf_true;
+        }
+        /* Check for not so stable fields. These fields can change if the
+         * inode is not locked. */
+        if (!failed && ((dst[i].ia_uid != src[i].ia_uid) ||
+                        (dst[i].ia_gid != src[i].ia_gid) ||
+                        (st_mode_from_ia(dst[i].ia_prot, dst[i].ia_type) !=
+                         st_mode_from_ia(src[i].ia_prot, src[i].ia_type)))) {
+            if (ec_iatt_is_trusted(fop, dst)) {
+                /* If the iatt contains information from an inode that is
+                 * locked, these differences are real problems, so we need to
+                 * report them. Otherwise we ignore them and don't care which
+                 * data is returned. */
+                failed = _gf_true;
+            } else {
+                gf_msg_debug(fop->xl->name, 0,
+                             "Ignoring iatt differences because inode is not "
+                             "locked");
+            }
+        }
+        if (failed) {
+            gf_msg(fop->xl->name, GF_LOG_WARNING, 0, EC_MSG_IATT_COMBINE_FAIL,
+                   "Failed to combine iatt (inode: %" PRIu64 "-%" PRIu64
+                   ", "
+                   "links: %u-%u, uid: %u-%u, gid: %u-%u, "
+                   "rdev: %" PRIu64 "-%" PRIu64 ", size: %" PRIu64 "-%" PRIu64
+                   ", "
+                   "mode: %o-%o), %s",
+                   dst[i].ia_ino, src[i].ia_ino, dst[i].ia_nlink,
+                   src[i].ia_nlink, dst[i].ia_uid, src[i].ia_uid, dst[i].ia_gid,
+                   src[i].ia_gid, dst[i].ia_rdev, src[i].ia_rdev,
+                   dst[i].ia_size, src[i].ia_size,
+                   st_mode_from_ia(dst[i].ia_prot, dst[i].ia_type),
+                   st_mode_from_ia(src[i].ia_prot, dst[i].ia_type),
+                   ec_msg_str(fop));
+
+            return 0;
+        }
+    }
+
+    while (count-- > 0) {
+        dst[count].ia_blocks += src[count].ia_blocks;
+        if (dst[count].ia_blksize < src[count].ia_blksize) {
+            dst[count].ia_blksize = src[count].ia_blksize;
+        }
+
+        ec_iatt_time_merge(&dst[count].ia_atime, &dst[count].ia_atime_nsec,
+                           src[count].ia_atime, src[count].ia_atime_nsec);
+        ec_iatt_time_merge(&dst[count].ia_mtime, &dst[count].ia_mtime_nsec,
+                           src[count].ia_mtime, src[count].ia_mtime_nsec);
+        ec_iatt_time_merge(&dst[count].ia_ctime, &dst[count].ia_ctime_nsec,
+                           src[count].ia_ctime, src[count].ia_ctime_nsec);
+    }
+
+    return 1;
+}
+
+void
+ec_iatt_rebuild(ec_t *ec, struct iatt *iatt, int32_t count, int32_t answers)
+{
+    uint64_t blocks;
+
+    while (count-- > 0) {
+        blocks = iatt[count].ia_blocks * ec->fragments + answers - 1;
+        blocks /= answers;
+        iatt[count].ia_blocks = blocks;
+    }
+}
+
+gf_boolean_t
+ec_xattr_match(dict_t *dict, char *key, data_t *value, void *arg)
+{
+    if ((fnmatch(GF_XATTR_STIME_PATTERN, key, 0) == 0) ||
+        (strcmp(key, GET_LINK_COUNT) == 0) ||
+        (strcmp(key, GLUSTERFS_INODELK_COUNT) == 0) ||
+        (strcmp(key, GLUSTERFS_ENTRYLK_COUNT) == 0) ||
+        (strcmp(key, GLUSTERFS_OPEN_FD_COUNT) == 0)) {
+        return _gf_false;
+    }
+
+    return _gf_true;
+}
+
+gf_boolean_t
+ec_value_ignore(char *key)
+{
+    if ((strcmp(key, GF_CONTENT_KEY) == 0) ||
+        (strcmp(key, GF_XATTR_PATHINFO_KEY) == 0) ||
+        (strcmp(key, GF_XATTR_USER_PATHINFO_KEY) == 0) ||
+        (strcmp(key, GF_XATTR_LOCKINFO_KEY) == 0) ||
+        (strcmp(key, GLUSTERFS_OPEN_FD_COUNT) == 0) ||
+        (strcmp(key, GLUSTERFS_INODELK_COUNT) == 0) ||
+        (strcmp(key, GLUSTERFS_ENTRYLK_COUNT) == 0) ||
+        (strncmp(key, GF_XATTR_CLRLK_CMD, SLEN(GF_XATTR_CLRLK_CMD)) == 0) ||
+        (strcmp(key, DHT_IATT_IN_XDATA_KEY) == 0) ||
+        (strncmp(key, EC_QUOTA_PREFIX, SLEN(EC_QUOTA_PREFIX)) == 0) ||
+        (fnmatch(MARKER_XATTR_PREFIX ".*." XTIME, key, 0) == 0) ||
+        (fnmatch(GF_XATTR_MARKER_KEY ".*", key, 0) == 0) ||
+        (XATTR_IS_NODE_UUID(key))) {
+        return _gf_true;
+    }
+
+    return _gf_false;
+}
+
+int32_t
+ec_dict_compare(dict_t *dict1, dict_t *dict2)
+{
+    if (are_dicts_equal(dict1, dict2, ec_xattr_match, ec_value_ignore))
+        return 1;
+    return 0;
+}
+
+static uint32_t
+ec_dict_list(data_t **list, ec_cbk_data_t *cbk, int32_t which, char *key,
+             gf_boolean_t global)
+{
+    ec_t *ec = cbk->fop->xl->private;
+    ec_cbk_data_t *ans = NULL;
+    dict_t *dict = NULL;
+    data_t *data;
+    uint32_t count;
+    int32_t i;
+
+    for (i = 0; i < ec->nodes; i++) {
+        /* We initialize the list with EC_MISSING_DATA if we are
+         * returning a global list or the current subvolume belongs
+         * to the group of the accepted answer. Note that if some
+         * subvolume is known to be down before issuing the request,
+         * we won't have any answer from it, so we set here the
+         * appropriate default value. */
+        if (global || ((cbk->mask & (1ULL << i)) != 0)) {
+            list[i] = EC_MISSING_DATA;
+        } else {
+            list[i] = NULL;
+        }
+    }
+
+    count = 0;
+    list_for_each_entry(ans, &cbk->fop->answer_list, answer_list)
+    {
+        if (global || ((cbk->mask & ans->mask) != 0)) {
+            dict = (which == EC_COMBINE_XDATA) ? ans->xdata : ans->dict;
+            data = dict_get(dict, key);
+            if (data != NULL) {
+                list[ans->idx] = data;
+                count++;
+            }
+        }
+    }
+
+    return count;
+}
+
+int32_t
+ec_concat_prepare(xlator_t *xl, char **str, char **sep, char **post,
+                  const char *fmt, va_list args)
+{
+    char *tmp;
+    int32_t len;
+
+    len = gf_vasprintf(str, fmt, args);
+    if (len < 0) {
+        return -ENOMEM;
+    }
+
+    tmp = strchr(*str, '{');
+    if (tmp == NULL) {
+        goto out;
+    }
+    *tmp++ = 0;
+    *sep = tmp;
+    tmp = strchr(tmp, '}');
+    if (tmp == NULL) {
+        goto out;
+    }
+    *tmp++ = 0;
+    *post = tmp;
+
+    return 0;
+
+out:
+    gf_msg(xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_INVALID_FORMAT,
+           "Invalid concat format");
+
+    GF_FREE(*str);
+
+    return -EINVAL;
+}
+
+static int32_t
+ec_dict_data_concat(ec_cbk_data_t *cbk, int32_t which, char *key, char *new_key,
+                    const char *def, gf_boolean_t global, const char *fmt, ...)
+{
+    ec_t *ec = cbk->fop->xl->private;
+    data_t *data[ec->nodes];
+    char *str = NULL, *pre = NULL, *sep, *post;
+    dict_t *dict;
+    va_list args;
+    int32_t i, num, len, deflen, prelen, postlen, seplen, tmp;
+    int32_t err;
+
+    ec_dict_list(data, cbk, which, key, global);
+
+    va_start(args, fmt);
+    err = ec_concat_prepare(cbk->fop->xl, &pre, &sep, &post, fmt, args);
+    va_end(args);
+
+    if (err != 0) {
+        return err;
+    }
+
+    prelen = strlen(pre);
+    seplen = strlen(sep);
+    postlen = strlen(post);
+
+    deflen = 0;
+    if (def != NULL) {
+        deflen = strlen(def);
+    }
+
+    len = prelen + postlen + 1;
+    num = -1;
+    for (i = 0; i < ec->nodes; i++) {
+        if (data[i] == NULL) {
+            continue;
+        }
+        if (data[i] == EC_MISSING_DATA) {
+            if (def == NULL) {
+                continue;
+            }
+            len += deflen;
+        } else {
+            len += data[i]->len - 1;
+        }
+        if (num >= 0) {
+            len += seplen;
+        }
+        num++;
+    }
+
+    err = -ENOMEM;
+
+    str = GF_MALLOC(len, gf_common_mt_char);
+    if (str == NULL) {
+        goto out;
+    }
+
+    memcpy(str, pre, prelen);
+    len = prelen;
+    for (i = 0; i < ec->nodes; i++) {
+        if (data[i] == NULL) {
+            continue;
+        }
+        if (data[i] == EC_MISSING_DATA) {
+            if (deflen == 0) {
+                continue;
+            }
+            tmp = deflen;
+            memcpy(str + len, def, tmp);
+        } else {
+            tmp = data[i]->len - 1;
+            memcpy(str + len, data[i]->data, tmp);
+        }
+        len += tmp;
+        if (i < num) {
+            memcpy(str + len, sep, seplen);
+            len += seplen;
+        }
+    }
+    memcpy(str + len, post, postlen + 1);
+
+    dict = (which == EC_COMBINE_XDATA) ? cbk->xdata : cbk->dict;
+    if (new_key) {
+        key = new_key;
+    }
+    err = dict_set_dynstr(dict, key, str);
+    if (err != 0) {
+        goto out;
+    }
+
+    str = NULL;
+
+out:
+    GF_FREE(str);
+    GF_FREE(pre);
+
+    return err;
+}
+
+int32_t
+ec_dict_data_merge(ec_cbk_data_t *cbk, int32_t which, char *key)
+{
+    ec_t *ec = cbk->fop->xl->private;
+    data_t *data[ec->nodes];
+    dict_t *dict, *lockinfo, *tmp = NULL;
+    char *ptr = NULL;
+    int32_t i, len;
+    int32_t err;
+
+    ec_dict_list(data, cbk, which, key, _gf_false);
+
+    lockinfo = dict_new();
+    if (lockinfo == NULL) {
+        return -ENOMEM;
+    }
+
+    for (i = 0; i < ec->nodes; i++) {
+        if ((data[i] == NULL) || (data[i] == EC_MISSING_DATA)) {
+            continue;
+        }
+
+        tmp = dict_new();
+        if (tmp == NULL) {
+            err = -ENOMEM;
+
+            goto out;
+        }
+        err = dict_unserialize(data[i]->data, data[i]->len, &tmp);
+        if (err != 0) {
+            goto out;
+        }
+        if (dict_copy(tmp, lockinfo) == NULL) {
+            err = -ENOMEM;
+
+            goto out;
+        }
+
+        dict_unref(tmp);
+    }
+
+    tmp = NULL;
+
+    err = dict_allocate_and_serialize(lockinfo, (char **)&ptr,
+                                      (unsigned int *)&len);
+    if (err != 0) {
+        goto out;
+    }
+
+    dict = (which == EC_COMBINE_XDATA) ? cbk->xdata : cbk->dict;
+    err = dict_set_dynptr(dict, key, ptr, len);
+    if (err != 0) {
+        goto out;
+    }
+
+    ptr = NULL;
+
+out:
+    GF_FREE(ptr);
+    dict_unref(lockinfo);
+    if (tmp != NULL) {
+        dict_unref(tmp);
+    }
+
+    return err;
+}
+
+int32_t
+ec_dict_data_uuid(ec_cbk_data_t *cbk, int32_t which, char *key)
+{
+    ec_cbk_data_t *ans, *min;
+    dict_t *src, *dst;
+    data_t *data;
+
+    min = cbk;
+    for (ans = cbk->next; ans != NULL; ans = ans->next) {
+        if (ans->idx < min->idx) {
+            min = ans;
+        }
+    }
+
+    if (min != cbk) {
+        src = (which == EC_COMBINE_XDATA) ? min->xdata : min->dict;
+        dst = (which == EC_COMBINE_XDATA) ? cbk->xdata : cbk->dict;
+
+        data = dict_get(src, key);
+        if (data == NULL) {
+            return -ENOENT;
+        }
+        if (dict_set(dst, key, data) != 0) {
+            return -ENOMEM;
+        }
+    }
+
+    return 0;
+}
+
+int32_t
+ec_dict_data_iatt(ec_cbk_data_t *cbk, int32_t which, char *key)
+{
+    ec_t *ec = cbk->fop->xl->private;
+    data_t *data[ec->nodes];
+    dict_t *dict;
+    struct iatt *stbuf, *tmp;
+    int32_t i, ret;
+
+    ec_dict_list(data, cbk, which, key, _gf_false);
+
+    stbuf = NULL;
+    for (i = 0; i < ec->nodes; i++) {
+        if ((data[i] == NULL) || (data[i] == EC_MISSING_DATA)) {
+            continue;
+        }
+        tmp = data_to_iatt(data[i], key);
+        if (tmp == NULL) {
+            ret = -EINVAL;
+            goto out;
+        }
+        if (stbuf == NULL) {
+            stbuf = GF_MALLOC(sizeof(struct iatt), gf_common_mt_char);
+            if (stbuf == NULL) {
+                ret = -ENOMEM;
+                goto out;
+            }
+            *stbuf = *tmp;
+        } else {
+            if (!ec_iatt_combine(cbk->fop, stbuf, tmp, 1)) {
+                ret = -EINVAL;
+                goto out;
+            }
+        }
+    }
+
+    if ((stbuf != NULL) && (stbuf->ia_type == IA_IFREG)) {
+        ec_iatt_rebuild(ec, stbuf, 1, cbk->count);
+        /* TODO: not sure if an iatt could come in xdata from a fop that takes
+         *       no locks. */
+        if (!ec_get_inode_size(cbk->fop, cbk->fop->locks[0].lock->loc.inode,
+                               &stbuf->ia_size)) {
+            ret = -EINVAL;
+            goto out;
+        }
+    }
+
+    dict = (which == EC_COMBINE_XDATA) ? cbk->xdata : cbk->dict;
+    ret = dict_set_iatt(dict, key, stbuf, false);
+    if (ret >= 0) {
+        stbuf = NULL;
+    }
+
+out:
+    GF_FREE(stbuf);
+
+    return ret;
+}
+
+int32_t
+ec_dict_data_max32(ec_cbk_data_t *cbk, int32_t which, char *key)
+{
+    ec_t *ec = cbk->fop->xl->private;
+    data_t *data[ec->nodes];
+    dict_t *dict;
+    int32_t i;
+    uint32_t max, tmp;
+
+    ec_dict_list(data, cbk, which, key, _gf_false);
+
+    max = 0;
+    for (i = 0; i < ec->nodes; i++) {
+        if ((data[i] == NULL) || (data[i] == EC_MISSING_DATA)) {
+            continue;
+        }
+
+        tmp = data_to_uint32(data[i]);
+        if (max < tmp) {
+            max = tmp;
+        }
+    }
+
+    dict = (which == EC_COMBINE_XDATA) ? cbk->xdata : cbk->dict;
+    return dict_set_uint32(dict, key, max);
+}
+
+int32_t
+ec_dict_data_max64(ec_cbk_data_t *cbk, int32_t which, char *key)
+{
+    ec_t *ec = cbk->fop->xl->private;
+    data_t *data[ec->nodes];
+    dict_t *dict;
+    int32_t i;
+    uint64_t max, tmp;
+
+    ec_dict_list(data, cbk, which, key, _gf_false);
+
+    max = 0;
+    for (i = 0; i < ec->nodes; i++) {
+        if ((data[i] == NULL) || (data[i] == EC_MISSING_DATA)) {
+            continue;
+        }
+
+        tmp = data_to_uint64(data[i]);
+        if (max < tmp) {
+            max = tmp;
+        }
+    }
+
+    dict = (which == EC_COMBINE_XDATA) ? cbk->xdata : cbk->dict;
+    return dict_set_uint64(dict, key, max);
+}
+
+int32_t
+ec_dict_data_quota(ec_cbk_data_t *cbk, int32_t which, char *key)
+{
+    ec_t *ec = cbk->fop->xl->private;
+    data_t *data[ec->nodes];
+    dict_t *dict = NULL;
+    int32_t i = 0;
+    quota_meta_t size = {
+        0,
+    };
+    quota_meta_t max_size = {
+        0,
+    };
+
+    if (ec_dict_list(data, cbk, which, key, _gf_false) == 0) {
+        return 0;
+    }
+
+    /* Quota size xattr is managed outside of the control of the ec xlator.
+     * This means that it might not be updated at the same time on all
+     * bricks and we can receive slightly different values. If that's the
+     * case, we take the maximum of all received values.
+     */
+    for (i = 0; i < ec->nodes; i++) {
+        if ((data[i] == NULL) || (data[i] == EC_MISSING_DATA) ||
+            (quota_data_to_meta(data[i], &size) < 0)) {
+            continue;
+        }
+
+        if (size.size > max_size.size)
+            max_size.size = size.size;
+        if (size.file_count > max_size.file_count)
+            max_size.file_count = size.file_count;
+        if (size.dir_count > max_size.dir_count)
+            max_size.dir_count = size.dir_count;
+    }
+
+    max_size.size *= ec->fragments;
+
+    dict = (which == EC_COMBINE_XDATA) ? cbk->xdata : cbk->dict;
+    return quota_dict_set_meta(dict, key, &max_size, IA_IFDIR);
+}
+
+int32_t
+ec_dict_data_stime(ec_cbk_data_t *cbk, int32_t which, char *key)
+{
+    ec_t *ec = cbk->fop->xl->private;
+    data_t *data[ec->nodes];
+    dict_t *dict;
+    int32_t i, err;
+
+    ec_dict_list(data, cbk, which, key, _gf_false);
+
+    dict = (which == EC_COMBINE_XDATA) ? cbk->xdata : cbk->dict;
+    for (i = 0; i < ec->nodes; i++) {
+        if ((data[i] == NULL) || (data[i] == EC_MISSING_DATA)) {
+            continue;
+        }
+        err = gf_get_max_stime(cbk->fop->xl, dict, key, data[i]);
+        if (err != 0) {
+            gf_msg(cbk->fop->xl->name, GF_LOG_ERROR, -err,
+                   EC_MSG_STIME_COMBINE_FAIL, "STIME combination failed");
+
+            return err;
+        }
+    }
+
+    return 0;
+}
+
+int32_t
+ec_dict_data_combine(dict_t *dict, char *key, data_t *value, void *arg)
+{
+    ec_dict_combine_t *data = arg;
+
+    if ((strcmp(key, GF_XATTR_PATHINFO_KEY) == 0) ||
+        (strcmp(key, GF_XATTR_USER_PATHINFO_KEY) == 0)) {
+        return ec_dict_data_concat(data->cbk, data->which, key, NULL, NULL,
+                                   _gf_false, _gf_false, "(<EC:%s> { })",
+                                   data->cbk->fop->xl->name);
+    }
+
+    if (strncmp(key, GF_XATTR_CLRLK_CMD, SLEN(GF_XATTR_CLRLK_CMD)) == 0) {
+        return ec_dict_data_concat(data->cbk, data->which, key, NULL, NULL,
+                                   _gf_false, "{\n}");
+    }
+
+    if (strncmp(key, GF_XATTR_LOCKINFO_KEY, SLEN(GF_XATTR_LOCKINFO_KEY)) == 0) {
+        return ec_dict_data_merge(data->cbk, data->which, key);
+    }
+
+    if (strcmp(key, GET_LINK_COUNT) == 0) {
+        return ec_dict_data_max32(data->cbk, data->which, key);
+    }
+
+    if (strcmp(key, GLUSTERFS_OPEN_FD_COUNT) == 0) {
+        return ec_dict_data_max32(data->cbk, data->which, key);
+    }
+    if ((strcmp(key, GLUSTERFS_INODELK_COUNT) == 0) ||
+        (strcmp(key, GLUSTERFS_ENTRYLK_COUNT) == 0)) {
+        return ec_dict_data_max32(data->cbk, data->which, key);
+    }
+
+    if (strcmp(key, QUOTA_SIZE_KEY) == 0) {
+        return ec_dict_data_quota(data->cbk, data->which, key);
+    }
+    /* Ignore all other quota attributes */
+    if (strncmp(key, EC_QUOTA_PREFIX, SLEN(EC_QUOTA_PREFIX)) == 0) {
+        return 0;
+    }
+
+    if (XATTR_IS_NODE_UUID(key)) {
+        if (data->cbk->fop->int32) {
+            /* List of node uuid is requested */
+            return ec_dict_data_concat(data->cbk, data->which, key,
+                                       GF_XATTR_LIST_NODE_UUIDS_KEY, UUID0_STR,
+                                       _gf_true, "{ }");
+        } else {
+            return ec_dict_data_uuid(data->cbk, data->which, key);
+        }
+    }
+
+    if (fnmatch(GF_XATTR_STIME_PATTERN, key, FNM_NOESCAPE) == 0) {
+        return ec_dict_data_stime(data->cbk, data->which, key);
+    }
+
+    if (fnmatch(MARKER_XATTR_PREFIX ".*." XTIME, key, FNM_NOESCAPE) == 0) {
+        return ec_dict_data_max64(data->cbk, data->which, key);
+    }
+
+    if (strcmp(key, GF_PRESTAT) == 0 || strcmp(key, GF_POSTSTAT) == 0) {
+        return ec_dict_data_iatt(data->cbk, data->which, key);
+    }
+
+    return 0;
+}
+
+int32_t
+ec_dict_combine(ec_cbk_data_t *cbk, int32_t which)
+{
+    dict_t *dict = NULL;
+    ec_dict_combine_t data;
+    int32_t err = 0;
+
+    data.cbk = cbk;
+    data.which = which;
+
+    dict = (which == EC_COMBINE_XDATA) ? cbk->xdata : cbk->dict;
+    if (dict != NULL) {
+        err = dict_foreach(dict, ec_dict_data_combine, &data);
+        if (err != 0) {
+            gf_msg(cbk->fop->xl->name, GF_LOG_ERROR, -err,
+                   EC_MSG_DICT_COMBINE_FAIL, "Dictionary combination failed");
+
+            return err;
+        }
+    }
+
+    return 0;
+}
+
+int32_t
+ec_vector_compare(struct iovec *dst_vector, int32_t dst_count,
+                  struct iovec *src_vector, int32_t src_count)
+{
+    int32_t dst_size = 0, src_size = 0;
+
+    if (dst_count > 0) {
+        dst_size = iov_length(dst_vector, dst_count);
+    }
+    if (src_count > 0) {
+        src_size = iov_length(src_vector, src_count);
+    }
+
+    return (dst_size == src_size);
+}
+
+int32_t
+ec_flock_compare(struct gf_flock *dst, struct gf_flock *src)
+{
+    if ((dst->l_type != src->l_type) || (dst->l_whence != src->l_whence) ||
+        (dst->l_start != src->l_start) || (dst->l_len != src->l_len) ||
+        (dst->l_pid != src->l_pid) ||
+        !is_same_lkowner(&dst->l_owner, &src->l_owner)) {
+        return 0;
+    }
+
+    return 1;
+}
+
+void
+ec_statvfs_combine(struct statvfs *dst, struct statvfs *src)
+{
+    if (dst->f_bsize < src->f_bsize) {
+        dst->f_bsize = src->f_bsize;
+    }
+
+    if (dst->f_frsize < src->f_frsize) {
+        dst->f_blocks *= dst->f_frsize;
+        dst->f_blocks /= src->f_frsize;
+
+        dst->f_bfree *= dst->f_frsize;
+        dst->f_bfree /= src->f_frsize;
+
+        dst->f_bavail *= dst->f_frsize;
+        dst->f_bavail /= src->f_frsize;
+
+        dst->f_frsize = src->f_frsize;
+    } else if (dst->f_frsize > src->f_frsize) {
+        src->f_blocks *= src->f_frsize;
+        src->f_blocks /= dst->f_frsize;
+
+        src->f_bfree *= src->f_frsize;
+        src->f_bfree /= dst->f_frsize;
+
+        src->f_bavail *= src->f_frsize;
+        src->f_bavail /= dst->f_frsize;
+    }
+    if (dst->f_blocks > src->f_blocks) {
+        dst->f_blocks = src->f_blocks;
+    }
+    if (dst->f_bfree > src->f_bfree) {
+        dst->f_bfree = src->f_bfree;
+    }
+    if (dst->f_bavail > src->f_bavail) {
+        dst->f_bavail = src->f_bavail;
+    }
+
+    if (dst->f_files < src->f_files) {
+        dst->f_files = src->f_files;
+    }
+    if (dst->f_ffree > src->f_ffree) {
+        dst->f_ffree = src->f_ffree;
+    }
+    if (dst->f_favail > src->f_favail) {
+        dst->f_favail = src->f_favail;
+    }
+    if (dst->f_namemax > src->f_namemax) {
+        dst->f_namemax = src->f_namemax;
+    }
+
+    if (dst->f_flag != src->f_flag) {
+        gf_msg_debug(THIS->name, 0,
+                     "Mismatching file system flags "
+                     "(%lX, %lX)",
+                     dst->f_flag, src->f_flag);
+    }
+    dst->f_flag &= src->f_flag;
+}
+
+int32_t
+ec_combine_check(ec_cbk_data_t *dst, ec_cbk_data_t *src, ec_combine_f combine)
+{
+    ec_fop_data_t *fop = dst->fop;
+
+    if (dst->op_ret != src->op_ret) {
+        gf_msg_debug(fop->xl->name, 0,
+                     "Mismatching return code in "
+                     "answers of '%s': %d <-> %d",
+                     ec_fop_name(fop->id), dst->op_ret, src->op_ret);
+
+        return 0;
+    }
+    if (dst->op_ret < 0) {
+        if (dst->op_errno != src->op_errno) {
+            gf_msg_debug(fop->xl->name, 0,
+                         "Mismatching errno code in "
+                         "answers of '%s': %d <-> %d",
+                         ec_fop_name(fop->id), dst->op_errno, src->op_errno);
+
+            return 0;
+        }
+    }
+
+    if (!ec_dict_compare(dst->xdata, src->xdata)) {
+        gf_msg(fop->xl->name, GF_LOG_DEBUG, 0, EC_MSG_XDATA_MISMATCH,
+               "Mismatching xdata in answers "
+               "of '%s'",
+               ec_fop_name(fop->id));
+
+        return 0;
+    }
+
+    if ((dst->op_ret >= 0) && (combine != NULL)) {
+        return combine(fop, dst, src);
+    }
+
+    return 1;
+}
+
+void
+ec_combine(ec_cbk_data_t *newcbk, ec_combine_f combine)
+{
+    ec_fop_data_t *fop = newcbk->fop;
+    ec_cbk_data_t *cbk = NULL, *tmp = NULL;
+    struct list_head *item = NULL;
+    int32_t needed = 0;
+    char str[32];
+
+    LOCK(&fop->lock);
+
+    fop->received |= newcbk->mask;
+
+    item = fop->cbk_list.prev;
+    list_for_each_entry(cbk, &fop->cbk_list, list)
+    {
+        if (ec_combine_check(newcbk, cbk, combine)) {
+            newcbk->count += cbk->count;
+            newcbk->mask |= cbk->mask;
+
+            item = cbk->list.prev;
+            while (item != &fop->cbk_list) {
+                tmp = list_entry(item, ec_cbk_data_t, list);
+                if (tmp->count >= newcbk->count) {
+                    break;
+                }
+                item = item->prev;
+            }
+            list_del(&cbk->list);
+
+            newcbk->next = cbk;
+
+            break;
+        }
+    }
+    list_add(&newcbk->list, item);
+
+    ec_trace("ANSWER", fop, "combine=%s[%d]",
+             ec_bin(str, sizeof(str), newcbk->mask, 0), newcbk->count);
+
+    cbk = list_entry(fop->cbk_list.next, ec_cbk_data_t, list);
+    if ((fop->mask ^ fop->remaining) == fop->received) {
+        needed = fop->minimum - cbk->count;
+    }
+
+    UNLOCK(&fop->lock);
+
+    if (needed > 0) {
+        ec_dispatch_next(fop, newcbk->idx);
+    }
+}
diff --git a/xlators/cluster/ec/src/ec-combine.h b/xlators/cluster/ec/src/ec-combine.h
new file mode 100644
index 00000000000..1010cc3be26
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-combine.h
@@ -0,0 +1,44 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_COMBINE_H__
+#define __EC_COMBINE_H__
+
+#define EC_COMBINE_DICT 0
+#define EC_COMBINE_XDATA 1
+
+typedef int32_t (*ec_combine_f)(ec_fop_data_t *fop, ec_cbk_data_t *dst,
+                                ec_cbk_data_t *src);
+
+void
+ec_iatt_rebuild(ec_t *ec, struct iatt *iatt, int32_t count, int32_t answers);
+
+int32_t
+ec_iatt_combine(ec_fop_data_t *fop, struct iatt *dst, struct iatt *src,
+                int32_t count);
+int32_t
+ec_dict_compare(dict_t *dict1, dict_t *dict2);
+int32_t
+ec_vector_compare(struct iovec *dst_vector, int32_t dst_count,
+                  struct iovec *src_vector, int32_t src_count);
+int32_t
+ec_flock_compare(struct gf_flock *dst, struct gf_flock *src);
+void
+ec_statvfs_combine(struct statvfs *dst, struct statvfs *src);
+
+int32_t
+ec_dict_combine(ec_cbk_data_t *cbk, int32_t which);
+
+void
+ec_combine(ec_cbk_data_t *cbk, ec_combine_f combine);
+
+int32_t
+ec_combine_write(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src);
+#endif /* __EC_COMBINE_H__ */
diff --git a/xlators/cluster/ec/src/ec-common.c b/xlators/cluster/ec/src/ec-common.c
new file mode 100644
index 00000000000..b955efd8c2d
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-common.c
@@ -0,0 +1,3042 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/byte-order.h>
+#include <glusterfs/hashfn.h>
+
+#include "ec-mem-types.h"
+#include "ec-types.h"
+#include "ec-helpers.h"
+#include "ec-combine.h"
+#include "ec-common.h"
+#include "ec-fops.h"
+#include "ec-method.h"
+#include "ec.h"
+#include "ec-messages.h"
+
+#define EC_INVALID_INDEX UINT32_MAX
+
+void
+ec_update_fd_status(fd_t *fd, xlator_t *xl, int idx, int32_t ret_status)
+{
+    ec_fd_t *fd_ctx;
+
+    if (fd == NULL)
+        return;
+
+    LOCK(&fd->lock);
+    {
+        fd_ctx = __ec_fd_get(fd, xl);
+        if (fd_ctx) {
+            if (ret_status >= 0)
+                fd_ctx->fd_status[idx] = EC_FD_OPENED;
+            else
+                fd_ctx->fd_status[idx] = EC_FD_NOT_OPENED;
+        }
+    }
+    UNLOCK(&fd->lock);
+}
+
+static uintptr_t
+ec_fd_ctx_need_open(fd_t *fd, xlator_t *this, uintptr_t mask)
+{
+    int i = 0;
+    int count = 0;
+    ec_t *ec = NULL;
+    ec_fd_t *fd_ctx = NULL;
+    uintptr_t need_open = 0;
+
+    ec = this->private;
+
+    fd_ctx = ec_fd_get(fd, this);
+    if (!fd_ctx)
+        return count;
+
+    LOCK(&fd->lock);
+    {
+        for (i = 0; i < ec->nodes; i++) {
+            if ((fd_ctx->fd_status[i] == EC_FD_NOT_OPENED) &&
+                ((ec->xl_up & (1 << i)) != 0) && ((mask & (1 << i)) != 0)) {
+                fd_ctx->fd_status[i] = EC_FD_OPENING;
+                need_open |= (1 << i);
+                count++;
+            }
+        }
+    }
+    UNLOCK(&fd->lock);
+
+    /* If fd needs to open on minimum number of nodes
+     * then ignore fixing the fd as it has been
+     * requested from heal operation.
+     */
+    if (count >= ec->fragments) {
+        need_open = 0;
+    }
+
+    return need_open;
+}
+
+static gf_boolean_t
+ec_is_fd_fixable(fd_t *fd)
+{
+    if (!fd || !fd->inode)
+        return _gf_false;
+    else if (fd_is_anonymous(fd))
+        return _gf_false;
+    else if (gf_uuid_is_null(fd->inode->gfid))
+        return _gf_false;
+
+    return _gf_true;
+}
+
+static void
+ec_fix_open(ec_fop_data_t *fop, uintptr_t mask)
+{
+    uintptr_t need_open = 0;
+    int ret = 0;
+    int32_t flags = 0;
+    loc_t loc = {
+        0,
+    };
+
+    if (!ec_is_fd_fixable(fop->fd))
+        goto out;
+
+    /* Evaluate how many remote fd's to be opened */
+    need_open = ec_fd_ctx_need_open(fop->fd, fop->xl, mask);
+    if (need_open == 0) {
+        goto out;
+    }
+
+    loc.inode = inode_ref(fop->fd->inode);
+    gf_uuid_copy(loc.gfid, fop->fd->inode->gfid);
+    ret = loc_path(&loc, NULL);
+    if (ret < 0) {
+        goto out;
+    }
+
+    flags = fop->fd->flags & (~(O_TRUNC | O_APPEND | O_CREAT | O_EXCL));
+    if (IA_IFDIR == fop->fd->inode->ia_type) {
+        ec_opendir(fop->frame, fop->xl, need_open,
+                   EC_MINIMUM_ONE | EC_FOP_NO_PROPAGATE_ERROR, NULL, NULL,
+                   &fop->loc[0], fop->fd, NULL);
+    } else {
+        ec_open(fop->frame, fop->xl, need_open,
+                EC_MINIMUM_ONE | EC_FOP_NO_PROPAGATE_ERROR, NULL, NULL, &loc,
+                flags, fop->fd, NULL);
+    }
+
+out:
+    loc_wipe(&loc);
+}
+
+static off_t
+ec_range_end_get(off_t fl_start, uint64_t fl_size)
+{
+    if (fl_size > 0) {
+        if (fl_size >= EC_RANGE_FULL) {
+            /* Infinity */
+            fl_start = LLONG_MAX;
+        } else {
+            fl_start += fl_size - 1;
+            if (fl_start < 0) {
+                /* Overflow */
+                fl_start = LLONG_MAX;
+            }
+        }
+    }
+
+    return fl_start;
+}
+
+static gf_boolean_t
+ec_is_range_conflict(ec_lock_link_t *l1, ec_lock_link_t *l2)
+{
+    return ((l1->fl_end >= l2->fl_start) && (l2->fl_end >= l1->fl_start));
+}
+
+static gf_boolean_t
+ec_lock_conflict(ec_lock_link_t *l1, ec_lock_link_t *l2)
+{
+    ec_t *ec = l1->fop->xl->private;
+
+    /* Fops like access/stat won't have to worry what the other fops are
+     * modifying as the fop is wound only to one brick. So it can be
+     * executed in parallel*/
+    if (l1->fop->minimum == EC_MINIMUM_ONE ||
+        l2->fop->minimum == EC_MINIMUM_ONE)
+        return _gf_false;
+
+    if ((l1->fop->flags & EC_FLAG_LOCK_SHARED) &&
+        (l2->fop->flags & EC_FLAG_LOCK_SHARED))
+        return _gf_false;
+
+    if (!ec->parallel_writes) {
+        return _gf_true;
+    }
+
+    return ec_is_range_conflict(l1, l2);
+}
+
+uint32_t
+ec_select_first_by_read_policy(ec_t *ec, ec_fop_data_t *fop)
+{
+    if (ec->read_policy == EC_ROUND_ROBIN) {
+        return ec->idx;
+    } else if (ec->read_policy == EC_GFID_HASH) {
+        if (fop->use_fd) {
+            return SuperFastHash((char *)fop->fd->inode->gfid,
+                                 sizeof(fop->fd->inode->gfid)) %
+                   ec->nodes;
+        } else {
+            if (gf_uuid_is_null(fop->loc[0].gfid))
+                loc_gfid(&fop->loc[0], fop->loc[0].gfid);
+            return SuperFastHash((char *)fop->loc[0].gfid,
+                                 sizeof(fop->loc[0].gfid)) %
+                   ec->nodes;
+        }
+    }
+    return 0;
+}
+
+static gf_boolean_t
+ec_child_valid(ec_t *ec, ec_fop_data_t *fop, uint32_t idx)
+{
+    return (idx < ec->nodes) && (((fop->remaining >> idx) & 1) == 1);
+}
+
+static uint32_t
+ec_child_next(ec_t *ec, ec_fop_data_t *fop, uint32_t idx)
+{
+    while (!ec_child_valid(ec, fop, idx)) {
+        if (++idx >= ec->nodes) {
+            idx = 0;
+        }
+        if (idx == fop->first) {
+            return EC_INVALID_INDEX;
+        }
+    }
+
+    return idx;
+}
+
+int32_t
+ec_heal_report(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, uintptr_t mask, uintptr_t good,
+               uintptr_t bad, uint32_t pending, dict_t *xdata)
+{
+    if (op_ret < 0) {
+        gf_msg(this->name, GF_LOG_DEBUG, op_errno, EC_MSG_HEAL_FAIL,
+               "Heal failed");
+    } else {
+        if ((mask & ~good) != 0) {
+            gf_msg(this->name, GF_LOG_DEBUG, 0, EC_MSG_HEAL_SUCCESS,
+                   "Heal succeeded on %d/%d "
+                   "subvolumes",
+                   gf_bits_count(mask & ~(good | bad)),
+                   gf_bits_count(mask & ~good));
+        }
+    }
+
+    return 0;
+}
+
+static uintptr_t
+ec_fop_needs_name_heal(ec_fop_data_t *fop)
+{
+    ec_t *ec = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    ec_cbk_data_t *enoent_cbk = NULL;
+
+    ec = fop->xl->private;
+    if (fop->id != GF_FOP_LOOKUP)
+        return 0;
+
+    if (!fop->loc[0].name || strlen(fop->loc[0].name) == 0)
+        return 0;
+
+    list_for_each_entry(cbk, &fop->cbk_list, list)
+    {
+        if (cbk->op_ret < 0 && cbk->op_errno == ENOENT) {
+            enoent_cbk = cbk;
+            break;
+        }
+    }
+
+    if (!enoent_cbk)
+        return 0;
+
+    return ec->xl_up & ~enoent_cbk->mask;
+}
+
+int32_t
+ec_fop_needs_heal(ec_fop_data_t *fop)
+{
+    ec_t *ec = fop->xl->private;
+
+    if (fop->lock_count == 0) {
+        /*
+         * if fop->lock_count is zero that means it saw version mismatch
+         * without any locks so it can't be trusted. If we launch a heal
+         * based on this it will lead to INODELKs which will affect I/O
+         * performance. Considering self-heal-daemon and operations on
+         * the inode from client which take locks can still trigger the
+         * heal we can choose to not attempt a heal when fop->lock_count
+         * is zero.
+         */
+        return 0;
+    }
+    return (ec->xl_up & ~(fop->remaining | fop->good)) != 0;
+}
+
+void
+ec_check_status(ec_fop_data_t *fop)
+{
+    ec_t *ec = fop->xl->private;
+    int32_t partial = 0;
+    char str1[32], str2[32], str3[32], str4[32], str5[32];
+
+    if (!ec_fop_needs_name_heal(fop) && !ec_fop_needs_heal(fop)) {
+        return;
+    }
+
+    if (fop->answer && fop->answer->op_ret >= 0) {
+        if ((fop->id == GF_FOP_LOOKUP) || (fop->id == GF_FOP_STAT) ||
+            (fop->id == GF_FOP_FSTAT)) {
+            partial = fop->answer->iatt[0].ia_type == IA_IFDIR;
+        } else if (fop->id == GF_FOP_OPENDIR) {
+            partial = 1;
+        }
+    }
+
+    gf_msg(
+        fop->xl->name, GF_LOG_WARNING, 0, EC_MSG_OP_FAIL_ON_SUBVOLS,
+        "Operation failed on %d of %d subvolumes.(up=%s, mask=%s, "
+        "remaining=%s, good=%s, bad=%s,"
+        "(Least significant bit represents first client/brick of subvol), %s)",
+        gf_bits_count(ec->xl_up & ~(fop->remaining | fop->good)), ec->nodes,
+        ec_bin(str1, sizeof(str1), ec->xl_up, ec->nodes),
+        ec_bin(str2, sizeof(str2), fop->mask, ec->nodes),
+        ec_bin(str3, sizeof(str3), fop->remaining, ec->nodes),
+        ec_bin(str4, sizeof(str4), fop->good, ec->nodes),
+        ec_bin(str5, sizeof(str5), ec->xl_up & ~(fop->remaining | fop->good),
+               ec->nodes),
+        ec_msg_str(fop));
+    if (fop->use_fd) {
+        if (fop->fd != NULL) {
+            ec_fheal(NULL, fop->xl, -1, EC_MINIMUM_ONE, ec_heal_report, NULL,
+                     fop->fd, partial, NULL);
+        }
+    } else {
+        ec_heal(NULL, fop->xl, -1, EC_MINIMUM_ONE, ec_heal_report, NULL,
+                &fop->loc[0], partial, NULL);
+
+        if (fop->loc[1].inode != NULL) {
+            ec_heal(NULL, fop->xl, -1, EC_MINIMUM_ONE, ec_heal_report, NULL,
+                    &fop->loc[1], partial, NULL);
+        }
+    }
+}
+
+void
+ec_update_good(ec_fop_data_t *fop, uintptr_t good)
+{
+    fop->good = good;
+
+    /* Fops that are executed only on one brick do not have enough information
+     * to decide if healing is needed or not. */
+    if ((fop->expected != 1) && (fop->parent == NULL)) {
+        ec_check_status(fop);
+    }
+}
+
+void
+ec_lock_update_good(ec_lock_t *lock, ec_fop_data_t *fop)
+{
+    /* Fops that are executed only on one brick do not have enough information
+     * to update the global mask of good bricks. */
+    if (fop->expected == 1) {
+        return;
+    }
+
+    /* When updating the good mask of the lock, we only take into consideration
+     * those bits corresponding to the bricks where the fop has been executed.
+     * Bad bricks are removed from good_mask, but once marked as bad it's never
+     * set to good until the lock is released and reacquired */
+
+    lock->good_mask &= fop->good | fop->remaining;
+}
+
+void
+__ec_fop_set_error(ec_fop_data_t *fop, int32_t error)
+{
+    if ((error != 0) && (fop->error == 0)) {
+        fop->error = error;
+    }
+}
+
+void
+ec_fop_set_error(ec_fop_data_t *fop, int32_t error)
+{
+    LOCK(&fop->lock);
+
+    __ec_fop_set_error(fop, error);
+
+    UNLOCK(&fop->lock);
+}
+
+gf_boolean_t
+ec_cbk_set_error(ec_cbk_data_t *cbk, int32_t error, gf_boolean_t ro)
+{
+    if ((error != 0) && (cbk->op_ret >= 0)) {
+        /* If cbk->op_errno was 0, it means that the fop succeeded and this
+         * error has happened while processing the answer. If the operation was
+         * read-only, there's no problem (i.e. we simply return the generated
+         * error code). However if it caused a modification, we must return EIO
+         * to indicate that the operation has been partially executed. */
+        cbk->op_errno = ro ? error : EIO;
+        cbk->op_ret = -1;
+
+        ec_fop_set_error(cbk->fop, cbk->op_errno);
+    }
+
+    return (cbk->op_ret < 0);
+}
+
+ec_cbk_data_t *
+ec_fop_prepare_answer(ec_fop_data_t *fop, gf_boolean_t ro)
+{
+    ec_cbk_data_t *cbk;
+    int32_t err;
+
+    cbk = fop->answer;
+    if (cbk == NULL) {
+        ec_fop_set_error(fop, EIO);
+
+        return NULL;
+    }
+
+    if (cbk->op_ret < 0) {
+        ec_fop_set_error(fop, cbk->op_errno);
+    }
+
+    err = ec_dict_combine(cbk, EC_COMBINE_XDATA);
+    if (ec_cbk_set_error(cbk, -err, ro)) {
+        return NULL;
+    }
+
+    return cbk;
+}
+
+void
+ec_sleep(ec_fop_data_t *fop)
+{
+    LOCK(&fop->lock);
+
+    GF_ASSERT(fop->refs > 0);
+    fop->refs++;
+    fop->jobs++;
+
+    UNLOCK(&fop->lock);
+}
+
+int32_t
+ec_check_complete(ec_fop_data_t *fop, ec_resume_f resume)
+{
+    int32_t error = -1;
+
+    LOCK(&fop->lock);
+
+    GF_ASSERT(fop->resume == NULL);
+
+    if (--fop->jobs != 0) {
+        ec_trace("WAIT", fop, "resume=%p", resume);
+
+        fop->resume = resume;
+    } else {
+        error = fop->error;
+        fop->error = 0;
+    }
+
+    UNLOCK(&fop->lock);
+
+    return error;
+}
+
+void
+ec_resume(ec_fop_data_t *fop, int32_t error)
+{
+    ec_resume_f resume = NULL;
+
+    LOCK(&fop->lock);
+
+    __ec_fop_set_error(fop, error);
+
+    if (--fop->jobs == 0) {
+        resume = fop->resume;
+        fop->resume = NULL;
+        if (resume != NULL) {
+            ec_trace("RESUME", fop, "error=%d", error);
+
+            if (fop->error != 0) {
+                error = fop->error;
+            }
+            fop->error = 0;
+        }
+    }
+
+    UNLOCK(&fop->lock);
+
+    if (resume != NULL) {
+        resume(fop, error);
+    }
+
+    ec_fop_data_release(fop);
+}
+
+void
+ec_resume_parent(ec_fop_data_t *fop)
+{
+    ec_fop_data_t *parent;
+    int32_t error = 0;
+
+    parent = fop->parent;
+    if (parent != NULL) {
+        if ((fop->fop_flags & EC_FOP_NO_PROPAGATE_ERROR) == 0) {
+            error = fop->error;
+        }
+        ec_trace("RESUME_PARENT", fop, "error=%u", error);
+        fop->parent = NULL;
+        ec_resume(parent, error);
+    }
+}
+
+gf_boolean_t
+ec_is_recoverable_error(int32_t op_errno)
+{
+    switch (op_errno) {
+        case ENOTCONN:
+        case ESTALE:
+        case ENOENT:
+        case EBADFD: /*Opened fd but brick is disconnected*/
+        case EIO:    /*Backend-fs crash like XFS/ext4 etc*/
+            return _gf_true;
+    }
+    return _gf_false;
+}
+
+void
+ec_complete(ec_fop_data_t *fop)
+{
+    ec_cbk_data_t *cbk = NULL;
+    int32_t resume = 0, update = 0;
+    int healing_count = 0;
+
+    LOCK(&fop->lock);
+
+    ec_trace("COMPLETE", fop, "");
+
+    if (--fop->winds == 0) {
+        if (fop->answer == NULL) {
+            if (!list_empty(&fop->cbk_list)) {
+                cbk = list_entry(fop->cbk_list.next, ec_cbk_data_t, list);
+                healing_count = gf_bits_count(cbk->mask & fop->healing);
+                /* fop shouldn't be treated as success if it is not
+                 * successful on at least fop->minimum good copies*/
+                if ((cbk->count - healing_count) >= fop->minimum) {
+                    fop->answer = cbk;
+
+                    update = 1;
+                }
+            }
+
+            resume = 1;
+        }
+    }
+
+    UNLOCK(&fop->lock);
+
+    /* ec_update_good() locks inode->lock. This may cause deadlocks with
+       fop->lock when used in another order. Since ec_update_good() will not
+       be called more than once for each fop, it can be called from outside
+       the fop->lock locked region. */
+    if (update) {
+        ec_update_good(fop, cbk->mask);
+    }
+
+    if (resume) {
+        ec_resume(fop, 0);
+    }
+
+    ec_fop_data_release(fop);
+}
+
+/* There could be already granted locks sitting on the bricks, unlock for which
+ * must be wound at all costs*/
+static gf_boolean_t
+ec_must_wind(ec_fop_data_t *fop)
+{
+    if ((fop->id == GF_FOP_INODELK) || (fop->id == GF_FOP_FINODELK) ||
+        (fop->id == GF_FOP_LK)) {
+        if (fop->flock.l_type == F_UNLCK)
+            return _gf_true;
+    } else if ((fop->id == GF_FOP_ENTRYLK) || (fop->id == GF_FOP_FENTRYLK)) {
+        if (fop->entrylk_cmd == ENTRYLK_UNLOCK)
+            return _gf_true;
+    }
+
+    return _gf_false;
+}
+
+static gf_boolean_t
+ec_internal_op(ec_fop_data_t *fop)
+{
+    if (ec_must_wind(fop))
+        return _gf_true;
+    if (fop->id == GF_FOP_XATTROP)
+        return _gf_true;
+    if (fop->id == GF_FOP_FXATTROP)
+        return _gf_true;
+    if (fop->id == GF_FOP_OPEN)
+        return _gf_true;
+    return _gf_false;
+}
+
+char *
+ec_msg_str(ec_fop_data_t *fop)
+{
+    loc_t *loc1 = NULL;
+    loc_t *loc2 = NULL;
+    char gfid1[64] = {0};
+    char gfid2[64] = {0};
+    ec_fop_data_t *parent = fop->parent;
+
+    if (fop->errstr)
+        return fop->errstr;
+    if (!fop->use_fd) {
+        loc1 = &fop->loc[0];
+        loc2 = &fop->loc[1];
+
+        if (fop->id == GF_FOP_RENAME) {
+            gf_asprintf(&fop->errstr,
+                        "FOP : '%s' failed on '%s' and '%s' with gfids "
+                        "%s and %s respectively. Parent FOP: %s",
+                        ec_fop_name(fop->id), loc1->path, loc2->path,
+                        uuid_utoa_r(loc1->gfid, gfid1),
+                        uuid_utoa_r(loc2->gfid, gfid2),
+                        parent ? ec_fop_name(parent->id) : "No Parent");
+        } else {
+            gf_asprintf(
+                &fop->errstr,
+                "FOP : '%s' failed on '%s' with gfid %s. Parent FOP: %s",
+                ec_fop_name(fop->id), loc1->path,
+                uuid_utoa_r(loc1->gfid, gfid1),
+                parent ? ec_fop_name(parent->id) : "No Parent");
+        }
+    } else {
+        gf_asprintf(
+            &fop->errstr, "FOP : '%s' failed on gfid %s. Parent FOP: %s",
+            ec_fop_name(fop->id), uuid_utoa_r(fop->fd->inode->gfid, gfid1),
+            parent ? ec_fop_name(parent->id) : "No Parent");
+    }
+    return fop->errstr;
+}
+
+static void
+ec_log_insufficient_vol(ec_fop_data_t *fop, int32_t have, uint32_t need,
+                        int32_t loglevel)
+{
+    ec_t *ec = fop->xl->private;
+    char str1[32], str2[32], str3[32];
+
+    gf_msg(ec->xl->name, loglevel, 0, EC_MSG_CHILDS_INSUFFICIENT,
+           "Insufficient available children for this request: "
+           "Have : %d, Need : %u : Child UP : %s "
+           "Mask: %s, Healing : %s : %s ",
+           have, need, ec_bin(str1, sizeof(str1), ec->xl_up, ec->nodes),
+           ec_bin(str2, sizeof(str2), fop->mask, ec->nodes),
+           ec_bin(str3, sizeof(str3), fop->healing, ec->nodes),
+           ec_msg_str(fop));
+}
+
+static int32_t
+ec_child_select(ec_fop_data_t *fop)
+{
+    ec_t *ec = fop->xl->private;
+    int32_t first = 0, num = 0;
+
+    ec_fop_cleanup(fop);
+
+    fop->mask &= ec->node_mask;
+    /* Wind the fop on same subvols as parent for any internal extra fops like
+     * head/tail read in case of writev fop. Unlocks shouldn't do this because
+     * unlock should go on all subvols where lock is performed*/
+    if (fop->parent && !ec_internal_op(fop)) {
+        fop->mask &= (fop->parent->mask & ~fop->parent->healing);
+        if (ec_is_data_fop(fop->id)) {
+            fop->healing |= fop->parent->healing;
+        }
+    }
+
+    if ((fop->mask & ~ec->xl_up) != 0) {
+        gf_msg(fop->xl->name, GF_LOG_WARNING, 0, EC_MSG_OP_EXEC_UNAVAIL,
+               "Executing operation with "
+               "some subvolumes unavailable. (%" PRIXPTR "). %s ",
+               fop->mask & ~ec->xl_up, ec_msg_str(fop));
+        fop->mask &= ec->xl_up;
+    }
+
+    switch (fop->minimum) {
+        case EC_MINIMUM_ALL:
+            fop->minimum = gf_bits_count(fop->mask);
+            if (fop->minimum >= ec->fragments) {
+                break;
+            }
+        case EC_MINIMUM_MIN:
+            fop->minimum = ec->fragments;
+            break;
+        case EC_MINIMUM_ONE:
+            fop->minimum = 1;
+    }
+
+    if (ec->read_policy == EC_ROUND_ROBIN) {
+        first = ec->idx;
+        if (++first >= ec->nodes) {
+            first = 0;
+        }
+        ec->idx = first;
+    }
+
+    num = gf_bits_count(fop->mask);
+    /*Unconditionally wind on healing subvolumes*/
+    fop->mask |= fop->healing;
+    fop->remaining = fop->mask;
+    fop->received = 0;
+
+    ec_trace("SELECT", fop, "");
+
+    if ((num < fop->minimum) && (num < ec->fragments)) {
+        ec_log_insufficient_vol(fop, num, fop->minimum, GF_LOG_ERROR);
+        return 0;
+    }
+
+    if (!fop->parent && fop->lock_count &&
+        (fop->locks[0].update[EC_DATA_TXN] ||
+         fop->locks[0].update[EC_METADATA_TXN])) {
+        if (ec->quorum_count && (num < ec->quorum_count)) {
+            ec_log_insufficient_vol(fop, num, ec->quorum_count, GF_LOG_ERROR);
+            return 0;
+        }
+    }
+
+    return 1;
+}
+
+void
+ec_dispatch_next(ec_fop_data_t *fop, uint32_t idx)
+{
+    uint32_t i = EC_INVALID_INDEX;
+    ec_t *ec = fop->xl->private;
+
+    LOCK(&fop->lock);
+
+    i = ec_child_next(ec, fop, idx);
+    if (i < EC_MAX_NODES) {
+        idx = i;
+
+        fop->remaining ^= 1ULL << idx;
+
+        ec_trace("EXECUTE", fop, "idx=%d", idx);
+
+        fop->winds++;
+        fop->refs++;
+    }
+
+    UNLOCK(&fop->lock);
+
+    if (i < EC_MAX_NODES) {
+        fop->wind(ec, fop, idx);
+    }
+}
+
+void
+ec_dispatch_mask(ec_fop_data_t *fop, uintptr_t mask)
+{
+    ec_t *ec = fop->xl->private;
+    int32_t count, idx;
+
+    count = gf_bits_count(mask);
+
+    LOCK(&fop->lock);
+
+    ec_trace("EXECUTE", fop, "mask=%lX", mask);
+
+    fop->remaining ^= mask;
+
+    fop->winds += count;
+    fop->refs += count;
+
+    UNLOCK(&fop->lock);
+
+    idx = 0;
+    while (mask != 0) {
+        if ((mask & 1) != 0) {
+            fop->wind(ec, fop, idx);
+        }
+        idx++;
+        mask >>= 1;
+    }
+}
+
+void
+ec_dispatch_start(ec_fop_data_t *fop)
+{
+    fop->answer = NULL;
+    fop->good = 0;
+
+    INIT_LIST_HEAD(&fop->cbk_list);
+
+    if (fop->lock_count > 0) {
+        ec_owner_copy(fop->frame, &fop->req_frame->root->lk_owner);
+    }
+}
+
+void
+ec_dispatch_one(ec_fop_data_t *fop)
+{
+    ec_dispatch_start(fop);
+
+    if (ec_child_select(fop)) {
+        ec_sleep(fop);
+
+        fop->expected = 1;
+        fop->first = ec_select_first_by_read_policy(fop->xl->private, fop);
+
+        ec_dispatch_next(fop, fop->first);
+    }
+}
+
+gf_boolean_t
+ec_dispatch_one_retry(ec_fop_data_t *fop, ec_cbk_data_t **cbk)
+{
+    ec_cbk_data_t *tmp;
+
+    tmp = ec_fop_prepare_answer(fop, _gf_true);
+    if (cbk != NULL) {
+        *cbk = tmp;
+    }
+    if ((tmp != NULL) && (tmp->op_ret < 0) &&
+        ec_is_recoverable_error(tmp->op_errno)) {
+        GF_ASSERT(fop->mask & (1ULL << tmp->idx));
+        fop->mask ^= (1ULL << tmp->idx);
+        if (fop->mask) {
+            return _gf_true;
+        }
+    }
+
+    return _gf_false;
+}
+
+void
+ec_dispatch_inc(ec_fop_data_t *fop)
+{
+    ec_dispatch_start(fop);
+
+    if (ec_child_select(fop)) {
+        ec_sleep(fop);
+
+        fop->expected = gf_bits_count(fop->remaining);
+        fop->first = 0;
+
+        ec_dispatch_next(fop, 0);
+    }
+}
+
+void
+ec_dispatch_all(ec_fop_data_t *fop)
+{
+    ec_dispatch_start(fop);
+
+    if (ec_child_select(fop)) {
+        ec_sleep(fop);
+
+        fop->expected = gf_bits_count(fop->remaining);
+        fop->first = 0;
+
+        ec_dispatch_mask(fop, fop->remaining);
+    }
+}
+
+void
+ec_dispatch_min(ec_fop_data_t *fop)
+{
+    ec_t *ec = fop->xl->private;
+    uintptr_t mask;
+    uint32_t idx;
+    int32_t count;
+
+    ec_dispatch_start(fop);
+
+    if (ec_child_select(fop)) {
+        ec_sleep(fop);
+
+        fop->expected = count = ec->fragments;
+        fop->first = ec_select_first_by_read_policy(fop->xl->private, fop);
+        idx = fop->first - 1;
+        mask = 0;
+        while (count-- > 0) {
+            idx = ec_child_next(ec, fop, idx + 1);
+            if (idx < EC_MAX_NODES)
+                mask |= 1ULL << idx;
+        }
+
+        ec_dispatch_mask(fop, mask);
+    }
+}
+
+void
+ec_succeed_all(ec_fop_data_t *fop)
+{
+    ec_dispatch_start(fop);
+
+    if (ec_child_select(fop)) {
+        fop->expected = gf_bits_count(fop->remaining);
+        fop->first = 0;
+
+        /* Simulate a successful execution on all bricks */
+        ec_trace("SUCCEED", fop, "");
+
+        fop->good = fop->remaining;
+        fop->remaining = 0;
+    }
+}
+
+ec_lock_t *
+ec_lock_allocate(ec_fop_data_t *fop, loc_t *loc)
+{
+    ec_t *ec = fop->xl->private;
+    ec_lock_t *lock;
+    int32_t err;
+
+    if ((loc->inode == NULL) ||
+        (gf_uuid_is_null(loc->gfid) && gf_uuid_is_null(loc->inode->gfid))) {
+        gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_INVALID_INODE,
+               "Trying to lock based on an invalid "
+               "inode");
+
+        __ec_fop_set_error(fop, EINVAL);
+
+        return NULL;
+    }
+
+    lock = mem_get0(ec->lock_pool);
+    if (lock != NULL) {
+        lock->good_mask = UINTPTR_MAX;
+        INIT_LIST_HEAD(&lock->owners);
+        INIT_LIST_HEAD(&lock->waiting);
+        INIT_LIST_HEAD(&lock->frozen);
+        err = ec_loc_from_loc(fop->xl, &lock->loc, loc);
+        if (err != 0) {
+            mem_put(lock);
+            lock = NULL;
+
+            __ec_fop_set_error(fop, -err);
+        }
+    }
+
+    return lock;
+}
+
+void
+ec_lock_destroy(ec_lock_t *lock)
+{
+    loc_wipe(&lock->loc);
+    if (lock->fd != NULL) {
+        fd_unref(lock->fd);
+    }
+
+    mem_put(lock);
+}
+
+int32_t
+ec_lock_compare(ec_lock_t *lock1, ec_lock_t *lock2)
+{
+    return gf_uuid_compare(lock1->loc.gfid, lock2->loc.gfid);
+}
+
+static void
+ec_lock_insert(ec_fop_data_t *fop, ec_lock_t *lock, uint32_t flags, loc_t *base,
+               off_t fl_start, uint64_t fl_size)
+{
+    ec_lock_link_t *link;
+
+    /* This check is only prepared for up to 2 locks per fop. If more locks
+     * are needed this must be changed. */
+    if ((fop->lock_count > 0) &&
+        (ec_lock_compare(fop->locks[0].lock, lock) < 0)) {
+        fop->first_lock = fop->lock_count;
+    } else {
+        /* When the first lock is added to the current fop, request lock
+         * counts from locks xlator to be able to determine if there is
+         * contention and release the lock sooner. */
+        if (fop->xdata == NULL) {
+            fop->xdata = dict_new();
+            if (fop->xdata == NULL) {
+                ec_fop_set_error(fop, ENOMEM);
+                return;
+            }
+        }
+        if (dict_set_str(fop->xdata, GLUSTERFS_INODELK_DOM_COUNT,
+                         fop->xl->name) != 0) {
+            ec_fop_set_error(fop, ENOMEM);
+            return;
+        }
+    }
+
+    link = &fop->locks[fop->lock_count++];
+
+    link->lock = lock;
+    link->fop = fop;
+    link->update[EC_DATA_TXN] = (flags & EC_UPDATE_DATA) != 0;
+    link->update[EC_METADATA_TXN] = (flags & EC_UPDATE_META) != 0;
+    link->base = base;
+    link->fl_start = fl_start;
+    link->fl_end = ec_range_end_get(fl_start, fl_size);
+
+    lock->refs_pending++;
+}
+
+static void
+ec_lock_prepare_inode_internal(ec_fop_data_t *fop, loc_t *loc, uint32_t flags,
+                               loc_t *base, off_t fl_start, uint64_t fl_size)
+{
+    ec_lock_t *lock = NULL;
+    ec_inode_t *ctx;
+
+    if ((fop->parent != NULL) || (fop->error != 0) || (loc->inode == NULL)) {
+        return;
+    }
+
+    LOCK(&loc->inode->lock);
+
+    ctx = __ec_inode_get(loc->inode, fop->xl);
+    if (ctx == NULL) {
+        __ec_fop_set_error(fop, ENOMEM);
+
+        goto unlock;
+    }
+
+    if (ctx->inode_lock != NULL) {
+        lock = ctx->inode_lock;
+
+        /* If there's another lock, make sure that it's not the same. Otherwise
+         * do not insert it.
+         *
+         * This can only happen on renames where source and target names are
+         * in the same directory. */
+        if ((fop->lock_count > 0) && (fop->locks[0].lock == lock)) {
+            /* Combine data/meta updates */
+            fop->locks[0].update[EC_DATA_TXN] |= (flags & EC_UPDATE_DATA) != 0;
+            fop->locks[0].update[EC_METADATA_TXN] |= (flags & EC_UPDATE_META) !=
+                                                     0;
+
+            /* Only one base inode is allowed per fop, so there shouldn't be
+             * overwrites here. */
+            if (base != NULL) {
+                fop->locks[0].base = base;
+            }
+
+            goto update_query;
+        }
+
+        ec_trace("LOCK_INODELK", fop,
+                 "lock=%p, inode=%p. Lock already "
+                 "acquired",
+                 lock, loc->inode);
+
+        goto insert;
+    }
+
+    lock = ec_lock_allocate(fop, loc);
+    if (lock == NULL) {
+        goto unlock;
+    }
+
+    ec_trace("LOCK_CREATE", fop, "lock=%p", lock);
+
+    lock->flock.l_type = F_WRLCK;
+    lock->flock.l_whence = SEEK_SET;
+
+    lock->ctx = ctx;
+    ctx->inode_lock = lock;
+
+insert:
+    ec_lock_insert(fop, lock, flags, base, fl_start, fl_size);
+update_query:
+    lock->query |= (flags & EC_QUERY_INFO) != 0;
+unlock:
+    UNLOCK(&loc->inode->lock);
+}
+
+void
+ec_lock_prepare_inode(ec_fop_data_t *fop, loc_t *loc, uint32_t flags,
+                      off_t fl_start, uint64_t fl_size)
+{
+    ec_lock_prepare_inode_internal(fop, loc, flags, NULL, fl_start, fl_size);
+}
+
+void
+ec_lock_prepare_parent_inode(ec_fop_data_t *fop, loc_t *loc, loc_t *base,
+                             uint32_t flags)
+{
+    loc_t tmp;
+    int32_t err;
+
+    if (fop->error != 0) {
+        return;
+    }
+
+    err = ec_loc_parent(fop->xl, loc, &tmp);
+    if (err != 0) {
+        ec_fop_set_error(fop, -err);
+
+        return;
+    }
+
+    if ((flags & EC_INODE_SIZE) != 0) {
+        flags ^= EC_INODE_SIZE;
+    } else {
+        base = NULL;
+    }
+
+    ec_lock_prepare_inode_internal(fop, &tmp, flags, base, 0, EC_RANGE_FULL);
+
+    loc_wipe(&tmp);
+}
+
+void
+ec_lock_prepare_fd(ec_fop_data_t *fop, fd_t *fd, uint32_t flags, off_t fl_start,
+                   uint64_t fl_size)
+{
+    loc_t loc;
+    int32_t err;
+
+    if (fop->error != 0) {
+        return;
+    }
+
+    err = ec_loc_from_fd(fop->xl, &loc, fd);
+    if (err != 0) {
+        ec_fop_set_error(fop, -err);
+
+        return;
+    }
+
+    ec_lock_prepare_inode_internal(fop, &loc, flags, NULL, fl_start, fl_size);
+
+    loc_wipe(&loc);
+}
+
+gf_boolean_t
+ec_config_check(xlator_t *xl, ec_config_t *config)
+{
+    ec_t *ec;
+
+    ec = xl->private;
+    if ((config->version != EC_CONFIG_VERSION) ||
+        (config->algorithm != EC_CONFIG_ALGORITHM) ||
+        (config->gf_word_size != EC_GF_BITS) || (config->bricks != ec->nodes) ||
+        (config->redundancy != ec->redundancy) ||
+        (config->chunk_size != EC_METHOD_CHUNK_SIZE)) {
+        uint32_t data_bricks;
+
+        /* This combination of version/algorithm requires the following
+           values. Incorrect values for these fields are a sign of
+           corruption:
+
+             redundancy > 0
+             redundancy * 2 < bricks
+             gf_word_size must be a power of 2
+             chunk_size (in bits) must be a multiple of gf_word_size *
+                 (bricks - redundancy) */
+
+        data_bricks = config->bricks - config->redundancy;
+        if ((config->redundancy < 1) ||
+            (config->redundancy * 2 >= config->bricks) ||
+            !ec_is_power_of_2(config->gf_word_size) ||
+            ((config->chunk_size * 8) % (config->gf_word_size * data_bricks) !=
+             0)) {
+            gf_msg(xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_INVALID_CONFIG,
+                   "Invalid or corrupted config");
+        } else {
+            gf_msg(xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_INVALID_CONFIG,
+                   "Unsupported config "
+                   "(V=%u, A=%u, W=%u, "
+                   "N=%u, R=%u, S=%u)",
+                   config->version, config->algorithm, config->gf_word_size,
+                   config->bricks, config->redundancy, config->chunk_size);
+        }
+
+        return _gf_false;
+    }
+
+    return _gf_true;
+}
+
+gf_boolean_t
+ec_set_dirty_flag(ec_lock_link_t *link, ec_inode_t *ctx, uint64_t *dirty)
+{
+    gf_boolean_t set_dirty = _gf_false;
+
+    if (link->update[EC_DATA_TXN] && !ctx->dirty[EC_DATA_TXN]) {
+        if (!link->optimistic_changelog)
+            dirty[EC_DATA_TXN] = 1;
+    }
+
+    if (link->update[EC_METADATA_TXN] && !ctx->dirty[EC_METADATA_TXN]) {
+        if (!link->optimistic_changelog)
+            dirty[EC_METADATA_TXN] = 1;
+    }
+
+    if (dirty[EC_METADATA_TXN] || dirty[EC_DATA_TXN]) {
+        set_dirty = _gf_true;
+    }
+
+    return set_dirty;
+}
+
+int32_t
+ec_prepare_update_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *dict,
+                      dict_t *xdata)
+{
+    struct list_head list;
+    ec_fop_data_t *fop = cookie, *parent, *tmp;
+    ec_lock_link_t *parent_link = fop->data;
+    ec_lock_link_t *link = NULL;
+    ec_lock_t *lock = NULL;
+    ec_inode_t *ctx;
+    gf_boolean_t release = _gf_false;
+    uint64_t provided_flags = 0;
+    uint64_t dirty[EC_VERSION_SIZE] = {0, 0};
+    lock = parent_link->lock;
+    parent = parent_link->fop;
+    ctx = lock->ctx;
+
+    INIT_LIST_HEAD(&list);
+    provided_flags = EC_PROVIDED_FLAGS(parent_link->waiting_flags);
+
+    LOCK(&lock->loc.inode->lock);
+
+    list_for_each_entry(link, &lock->owners, owner_list)
+    {
+        if ((link->waiting_flags & provided_flags) != 0) {
+            link->waiting_flags ^= (link->waiting_flags & provided_flags);
+            if (EC_NEEDED_FLAGS(link->waiting_flags) == 0)
+                list_add_tail(&link->fop->cbk_list, &list);
+        }
+    }
+    if (op_ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, EC_MSG_SIZE_VERS_GET_FAIL,
+               "Failed to get size and version :  %s", ec_msg_str(fop));
+
+        goto unlock;
+    }
+
+    if (EC_FLAGS_HAVE(provided_flags, EC_FLAG_XATTROP)) {
+        op_errno = -ec_dict_del_array(dict, EC_XATTR_VERSION, ctx->pre_version,
+                                      EC_VERSION_SIZE);
+        if (op_errno != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                   EC_MSG_VER_XATTR_GET_FAIL, "Unable to get version xattr. %s",
+                   ec_msg_str(fop));
+            goto unlock;
+        }
+        ctx->post_version[0] += ctx->pre_version[0];
+        ctx->post_version[1] += ctx->pre_version[1];
+
+        ctx->have_version = _gf_true;
+
+        if (lock->loc.inode->ia_type == IA_IFREG ||
+            lock->loc.inode->ia_type == IA_INVAL) {
+            op_errno = -ec_dict_del_number(dict, EC_XATTR_SIZE, &ctx->pre_size);
+            if (op_errno != 0) {
+                if (lock->loc.inode->ia_type == IA_IFREG) {
+                    gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                           EC_MSG_SIZE_XATTR_GET_FAIL,
+                           "Unable to get size xattr. %s", ec_msg_str(fop));
+                    goto unlock;
+                }
+            } else {
+                ctx->post_size = ctx->pre_size;
+
+                ctx->have_size = _gf_true;
+            }
+
+            op_errno = -ec_dict_del_config(dict, EC_XATTR_CONFIG, &ctx->config);
+            if (op_errno != 0) {
+                if ((lock->loc.inode->ia_type == IA_IFREG) ||
+                    (op_errno != ENODATA)) {
+                    gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                           EC_MSG_CONFIG_XATTR_GET_FAIL,
+                           "Unable to get config xattr. %s", ec_msg_str(fop));
+
+                    goto unlock;
+                }
+            } else {
+                if (!ec_config_check(parent->xl, &ctx->config)) {
+                    gf_msg(this->name, GF_LOG_ERROR, EINVAL,
+                           EC_MSG_CONFIG_XATTR_INVALID, "Invalid config xattr");
+
+                    op_errno = EINVAL;
+
+                    goto unlock;
+                }
+                ctx->have_config = _gf_true;
+            }
+        }
+        ctx->have_info = _gf_true;
+    }
+
+    ec_set_dirty_flag(fop->data, ctx, dirty);
+    if (dirty[EC_METADATA_TXN] &&
+        (EC_FLAGS_HAVE(provided_flags, EC_FLAG_METADATA_DIRTY))) {
+        GF_ASSERT(!ctx->dirty[EC_METADATA_TXN]);
+        ctx->dirty[EC_METADATA_TXN] = 1;
+    }
+
+    if (dirty[EC_DATA_TXN] &&
+        (EC_FLAGS_HAVE(provided_flags, EC_FLAG_DATA_DIRTY))) {
+        GF_ASSERT(!ctx->dirty[EC_DATA_TXN]);
+        ctx->dirty[EC_DATA_TXN] = 1;
+    }
+    op_errno = 0;
+unlock:
+
+    lock->waiting_flags ^= provided_flags;
+
+    if (op_errno == 0) {
+        /* If the fop fails on any of the good bricks, it is important to mark
+         * it dirty and update versions right away if dirty was not set before.
+         */
+        if (lock->good_mask & ~(fop->good | fop->remaining)) {
+            release = _gf_true;
+        }
+
+        if (parent_link->update[0] && !parent_link->dirty[0]) {
+            lock->release |= release;
+        }
+
+        if (parent_link->update[1] && !parent_link->dirty[1]) {
+            lock->release |= release;
+        }
+
+        /* We don't allow the main fop to be executed on bricks that have not
+         * succeeded the initial xattrop. */
+        ec_lock_update_good(lock, fop);
+
+        /*As of now only data healing marks bricks as healing*/
+        lock->healing |= fop->healing;
+    }
+
+    UNLOCK(&lock->loc.inode->lock);
+
+    while (!list_empty(&list)) {
+        tmp = list_entry(list.next, ec_fop_data_t, cbk_list);
+        list_del_init(&tmp->cbk_list);
+
+        if (op_errno == 0) {
+            tmp->mask &= fop->good;
+
+            /*As of now only data healing marks bricks as healing*/
+            if (ec_is_data_fop(tmp->id)) {
+                tmp->healing |= fop->healing;
+            }
+        }
+
+        ec_resume(tmp, op_errno);
+    }
+
+    return 0;
+}
+
+static gf_boolean_t
+ec_set_needed_flag(ec_lock_t *lock, ec_lock_link_t *link, uint64_t flag)
+{
+    uint64_t current;
+
+    link->waiting_flags |= EC_FLAG_NEEDS(flag);
+
+    current = EC_NEEDED_FLAGS(lock->waiting_flags);
+    if (!EC_FLAGS_HAVE(current, flag)) {
+        lock->waiting_flags |= EC_FLAG_NEEDS(flag);
+        link->waiting_flags |= EC_FLAG_PROVIDES(flag);
+
+        return _gf_true;
+    }
+
+    return _gf_false;
+}
+
+static uint64_t
+ec_set_xattrop_flags_and_params(ec_lock_t *lock, ec_lock_link_t *link,
+                                uint64_t *dirty)
+{
+    uint64_t oldflags = 0;
+    uint64_t newflags = 0;
+    ec_inode_t *ctx = lock->ctx;
+
+    oldflags = EC_NEEDED_FLAGS(lock->waiting_flags);
+
+    if (lock->query && !ctx->have_info) {
+        ec_set_needed_flag(lock, link, EC_FLAG_XATTROP);
+    }
+
+    if (dirty[EC_DATA_TXN]) {
+        if (!ec_set_needed_flag(lock, link, EC_FLAG_DATA_DIRTY)) {
+            dirty[EC_DATA_TXN] = 0;
+        }
+    }
+
+    if (dirty[EC_METADATA_TXN]) {
+        if (!ec_set_needed_flag(lock, link, EC_FLAG_METADATA_DIRTY)) {
+            dirty[EC_METADATA_TXN] = 0;
+        }
+    }
+    newflags = EC_NEEDED_FLAGS(lock->waiting_flags);
+
+    return oldflags ^ newflags;
+}
+
+void
+ec_get_size_version(ec_lock_link_t *link)
+{
+    loc_t loc;
+    ec_lock_t *lock;
+    ec_inode_t *ctx;
+    ec_fop_data_t *fop;
+    dict_t *dict = NULL;
+    dict_t *xdata = NULL;
+    ec_t *ec = NULL;
+    int32_t error = 0;
+    gf_boolean_t set_dirty = _gf_false;
+    uint64_t allzero[EC_VERSION_SIZE] = {0, 0};
+    uint64_t dirty[EC_VERSION_SIZE] = {0, 0};
+    lock = link->lock;
+    ctx = lock->ctx;
+    fop = link->fop;
+    ec = fop->xl->private;
+    uint64_t changed_flags = 0;
+
+    if (ec->optimistic_changelog && !(ec->node_mask & ~link->lock->good_mask) &&
+        !ec_is_data_fop(fop->id))
+        link->optimistic_changelog = _gf_true;
+
+    memset(&loc, 0, sizeof(loc));
+
+    LOCK(&lock->loc.inode->lock);
+
+    set_dirty = ec_set_dirty_flag(link, ctx, dirty);
+
+    /* If ec metadata has already been retrieved, do not try again. */
+    if (ctx->have_info) {
+        if (ec_is_data_fop(fop->id)) {
+            fop->healing |= lock->healing;
+        }
+        if (!set_dirty)
+            goto unlock;
+    }
+
+    /* Determine if there's something we need to retrieve for the current
+     * operation. */
+    if (!set_dirty && !lock->query && (lock->loc.inode->ia_type != IA_IFREG) &&
+        (lock->loc.inode->ia_type != IA_INVAL)) {
+        goto unlock;
+    }
+
+    changed_flags = ec_set_xattrop_flags_and_params(lock, link, dirty);
+    if (link->waiting_flags) {
+        /* This fop needs to wait until all its flags are cleared which
+         * potentially can be cleared by other xattrops that are already
+         * wound*/
+        ec_sleep(fop);
+    } else {
+        GF_ASSERT(!changed_flags);
+    }
+
+unlock:
+    UNLOCK(&lock->loc.inode->lock);
+
+    if (!changed_flags)
+        goto out;
+
+    dict = dict_new();
+    if (dict == NULL) {
+        error = -ENOMEM;
+        goto out;
+    }
+
+    if (EC_FLAGS_HAVE(changed_flags, EC_FLAG_XATTROP)) {
+        /* Once we know that an xattrop will be needed,
+         * we try to get all available information in a
+         * single call. */
+        error = ec_dict_set_array(dict, EC_XATTR_VERSION, allzero,
+                                  EC_VERSION_SIZE);
+        if (error != 0) {
+            goto out;
+        }
+
+        if (lock->loc.inode->ia_type == IA_IFREG ||
+            lock->loc.inode->ia_type == IA_INVAL) {
+            error = ec_dict_set_number(dict, EC_XATTR_SIZE, 0);
+            if (error == 0) {
+                error = ec_dict_set_number(dict, EC_XATTR_CONFIG, 0);
+            }
+            if (error != 0) {
+                goto out;
+            }
+
+            xdata = dict_new();
+            if (xdata == NULL || dict_set_int32(xdata, GF_GET_SIZE, 1)) {
+                error = -ENOMEM;
+                goto out;
+            }
+        }
+    }
+
+    if (memcmp(allzero, dirty, sizeof(allzero))) {
+        error = ec_dict_set_array(dict, EC_XATTR_DIRTY, dirty, EC_VERSION_SIZE);
+        if (error != 0) {
+            goto out;
+        }
+    }
+
+    fop->frame->root->uid = 0;
+    fop->frame->root->gid = 0;
+
+    /* For normal fops, ec_[f]xattrop() must succeed on at least
+     * EC_MINIMUM_MIN bricks, however when this is called as part of a
+     * self-heal operation the mask of target bricks (fop->mask) could
+     * contain less than EC_MINIMUM_MIN bricks, causing the xattrop to
+     * always fail. Thus we always use the same minimum used for the main
+     * fop.
+     */
+    if (lock->fd == NULL) {
+        error = ec_loc_from_loc(fop->xl, &loc, &lock->loc);
+        if (error != 0) {
+            goto out;
+        }
+        if (gf_uuid_is_null(loc.pargfid)) {
+            if (loc.parent != NULL) {
+                inode_unref(loc.parent);
+                loc.parent = NULL;
+            }
+            GF_FREE((char *)loc.path);
+            loc.path = NULL;
+            loc.name = NULL;
+        }
+
+        ec_xattrop(fop->frame, fop->xl, fop->mask, fop->minimum,
+                   ec_prepare_update_cbk, link, &loc, GF_XATTROP_ADD_ARRAY64,
+                   dict, xdata);
+    } else {
+        ec_fxattrop(fop->frame, fop->xl, fop->mask, fop->minimum,
+                    ec_prepare_update_cbk, link, lock->fd,
+                    GF_XATTROP_ADD_ARRAY64, dict, xdata);
+    }
+
+    error = 0;
+
+out:
+    fop->frame->root->uid = fop->uid;
+    fop->frame->root->gid = fop->gid;
+
+    loc_wipe(&loc);
+
+    if (dict != NULL) {
+        dict_unref(dict);
+    }
+
+    if (xdata != NULL) {
+        dict_unref(xdata);
+    }
+
+    if (error != 0) {
+        ec_fop_set_error(fop, -error);
+    }
+}
+
+gf_boolean_t
+__ec_get_inode_size(ec_fop_data_t *fop, inode_t *inode, uint64_t *size)
+{
+    ec_inode_t *ctx;
+    gf_boolean_t found = _gf_false;
+
+    ctx = __ec_inode_get(inode, fop->xl);
+    if (ctx == NULL) {
+        goto out;
+    }
+
+    if (ctx->have_size) {
+        *size = ctx->post_size;
+        found = _gf_true;
+    }
+
+out:
+    return found;
+}
+
+gf_boolean_t
+ec_get_inode_size(ec_fop_data_t *fop, inode_t *inode, uint64_t *size)
+{
+    gf_boolean_t found = _gf_false;
+
+    LOCK(&inode->lock);
+    {
+        found = __ec_get_inode_size(fop, inode, size);
+    }
+    UNLOCK(&inode->lock);
+
+    return found;
+}
+
+gf_boolean_t
+__ec_set_inode_size(ec_fop_data_t *fop, inode_t *inode, uint64_t size)
+{
+    ec_inode_t *ctx;
+    gf_boolean_t found = _gf_false;
+
+    ctx = __ec_inode_get(inode, fop->xl);
+    if (ctx == NULL) {
+        goto out;
+    }
+
+    /* Normal fops always have ctx->have_size set. However self-heal calls this
+     * to prepare the inode, so ctx->have_size will be false. In this case we
+     * prepare both pre_size and post_size, and set have_size and have_info to
+     * true. */
+    if (!ctx->have_size) {
+        ctx->pre_size = size;
+        ctx->have_size = ctx->have_info = _gf_true;
+    }
+    ctx->post_size = size;
+
+    found = _gf_true;
+
+out:
+    return found;
+}
+
+gf_boolean_t
+ec_set_inode_size(ec_fop_data_t *fop, inode_t *inode, uint64_t size)
+{
+    gf_boolean_t found = _gf_false;
+
+    LOCK(&inode->lock);
+    {
+        found = __ec_set_inode_size(fop, inode, size);
+    }
+    UNLOCK(&inode->lock);
+
+    return found;
+}
+
+static void
+ec_release_stripe_cache(ec_inode_t *ctx)
+{
+    ec_stripe_list_t *stripe_cache = NULL;
+    ec_stripe_t *stripe = NULL;
+
+    stripe_cache = &ctx->stripe_cache;
+    while (!list_empty(&stripe_cache->lru)) {
+        stripe = list_first_entry(&stripe_cache->lru, ec_stripe_t, lru);
+        list_del(&stripe->lru);
+        GF_FREE(stripe);
+    }
+    stripe_cache->count = 0;
+    stripe_cache->max = 0;
+}
+
+void
+ec_clear_inode_info(ec_fop_data_t *fop, inode_t *inode)
+{
+    ec_inode_t *ctx;
+
+    LOCK(&inode->lock);
+
+    ctx = __ec_inode_get(inode, fop->xl);
+    if (ctx == NULL) {
+        goto unlock;
+    }
+
+    ec_release_stripe_cache(ctx);
+    ctx->have_info = _gf_false;
+    ctx->have_config = _gf_false;
+    ctx->have_version = _gf_false;
+    ctx->have_size = _gf_false;
+
+    memset(&ctx->config, 0, sizeof(ctx->config));
+    memset(ctx->pre_version, 0, sizeof(ctx->pre_version));
+    memset(ctx->post_version, 0, sizeof(ctx->post_version));
+    ctx->pre_size = ctx->post_size = 0;
+    memset(ctx->dirty, 0, sizeof(ctx->dirty));
+
+unlock:
+    UNLOCK(&inode->lock);
+}
+
+int32_t
+ec_get_real_size_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, inode_t *inode,
+                     struct iatt *buf, dict_t *xdata, struct iatt *postparent)
+{
+    ec_fop_data_t *fop = cookie;
+    ec_lock_link_t *link;
+
+    if (op_ret >= 0) {
+        link = fop->data;
+        link->size = buf->ia_size;
+    } else {
+        /* Prevent failure of parent fop. */
+        fop->error = 0;
+    }
+
+    return 0;
+}
+
+/* This function is used to get the trusted.ec.size xattr from a file when
+ * no lock is needed on the inode. This is only required to maintain iatt
+ * structs on fops that manipulate directory entries but do not operate
+ * directly on the inode, like link, rename, ...
+ *
+ * Any error processing this request is ignored. In the worst case, an invalid
+ * or not up to date value in the iatt could cause some cache invalidation.
+ */
+void
+ec_get_real_size(ec_lock_link_t *link)
+{
+    ec_fop_data_t *fop;
+    dict_t *xdata;
+
+    if (link->base == NULL || link->base->inode == NULL) {
+        return;
+    }
+
+    if (link->base->inode->ia_type != IA_IFREG) {
+        return;
+    }
+
+    fop = link->fop;
+
+    if (ec_get_inode_size(fop, link->base->inode, &link->size)) {
+        return;
+    }
+
+    xdata = dict_new();
+    if (xdata == NULL) {
+        return;
+    }
+    if (ec_dict_set_number(xdata, EC_XATTR_SIZE, 0) != 0) {
+        goto out;
+    }
+
+    /* Send a simple lookup. A single answer is considered ok since this value
+     * is only used to return an iatt struct related to an inode that is not
+     * locked and have not suffered any operation. */
+    ec_lookup(fop->frame, fop->xl, fop->mask, 1, ec_get_real_size_cbk, link,
+              link->base, xdata);
+
+out:
+    if (xdata != NULL) {
+        dict_unref(xdata);
+    }
+}
+
+static void
+ec_lock_update_fd(ec_lock_t *lock, ec_fop_data_t *fop)
+{
+    /* If the fop has an fd available, attach it to the lock structure to be
+     * able to do fxattrop calls instead of xattrop. */
+    if (fop->use_fd && (lock->fd == NULL)) {
+        lock->fd = __fd_ref(fop->fd);
+    }
+}
+
+static gf_boolean_t
+ec_link_has_lock_conflict(ec_lock_link_t *link, gf_boolean_t waitlist_check)
+{
+    ec_lock_link_t *trav_link = NULL;
+
+    list_for_each_entry(trav_link, &link->lock->owners, owner_list)
+    {
+        if (ec_lock_conflict(trav_link, link))
+            return _gf_true;
+    }
+
+    if (!waitlist_check)
+        return _gf_false;
+
+    list_for_each_entry(trav_link, &link->lock->waiting, wait_list)
+    {
+        if (ec_lock_conflict(trav_link, link))
+            return _gf_true;
+    }
+
+    return _gf_false;
+}
+
+static void
+ec_lock_wake_shared(ec_lock_t *lock, struct list_head *list)
+{
+    ec_fop_data_t *fop;
+    ec_lock_link_t *link;
+    gf_boolean_t conflict = _gf_false;
+
+    while (!conflict && !list_empty(&lock->waiting)) {
+        link = list_entry(lock->waiting.next, ec_lock_link_t, wait_list);
+        fop = link->fop;
+
+        /* If lock is not acquired, at most one fop can be assigned as owner.
+         * The following fops will need to wait in the lock->waiting queue
+         * until the lock has been fully acquired. */
+        conflict = !lock->acquired;
+
+        /* If the fop is not shareable, only this fop can be assigned as owner.
+         * Other fops will need to wait until this one finishes. */
+        if (ec_link_has_lock_conflict(link, _gf_false)) {
+            conflict = _gf_true;
+        }
+
+        /* If only one fop is allowed, it can be assigned as the owner of the
+         * lock only if there weren't any other owner. */
+        if (conflict && !list_empty(&lock->owners)) {
+            break;
+        }
+
+        list_move_tail(&link->wait_list, list);
+
+        list_add_tail(&link->owner_list, &lock->owners);
+        lock->refs_owners++;
+
+        ec_lock_update_fd(lock, fop);
+    }
+}
+
+static void
+ec_lock_apply(ec_lock_link_t *link)
+{
+    ec_fop_data_t *fop = link->fop;
+
+    fop->mask &= link->lock->good_mask;
+    fop->locked++;
+
+    ec_get_size_version(link);
+    ec_get_real_size(link);
+}
+
+gf_boolean_t
+ec_lock_acquire(ec_lock_link_t *link);
+
+static void
+ec_lock_resume_shared(struct list_head *list)
+{
+    ec_lock_link_t *link;
+
+    while (!list_empty(list)) {
+        link = list_entry(list->next, ec_lock_link_t, wait_list);
+        list_del_init(&link->wait_list);
+
+        if (link->lock->acquired) {
+            ec_lock_apply(link);
+            ec_lock(link->fop);
+        } else {
+            GF_ASSERT(list_empty(list));
+
+            ec_lock_acquire(link);
+        }
+
+        ec_resume(link->fop, 0);
+    }
+}
+
+void
+ec_lock_acquired(ec_lock_link_t *link)
+{
+    struct list_head list;
+    ec_lock_t *lock;
+    ec_fop_data_t *fop;
+
+    lock = link->lock;
+    fop = link->fop;
+
+    ec_trace("LOCKED", fop, "lock=%p", lock);
+
+    INIT_LIST_HEAD(&list);
+
+    LOCK(&lock->loc.inode->lock);
+
+    lock->acquired = _gf_true;
+    if (lock->contention) {
+        lock->release = _gf_true;
+        lock->contention = _gf_false;
+    }
+
+    ec_lock_update_fd(lock, fop);
+    ec_lock_wake_shared(lock, &list);
+
+    UNLOCK(&lock->loc.inode->lock);
+
+    ec_lock_apply(link);
+
+    if (fop->use_fd &&
+        (link->update[EC_DATA_TXN] || link->update[EC_METADATA_TXN])) {
+        /* Try to reopen closed fd's only if lock has succeeded. */
+        ec_fix_open(fop, lock->mask);
+    }
+
+    ec_lock_resume_shared(&list);
+}
+
+int32_t
+ec_locked(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+          int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = cookie;
+    ec_lock_link_t *link = NULL;
+    ec_lock_t *lock = NULL;
+
+    link = fop->data;
+    lock = link->lock;
+    if (op_ret >= 0) {
+        lock->mask = lock->good_mask = fop->good;
+        lock->healing = 0;
+
+        ec_lock_acquired(link);
+        ec_lock(fop->parent);
+    } else {
+        LOCK(&lock->loc.inode->lock);
+        {
+            lock->contention = _gf_false;
+        }
+        UNLOCK(&lock->loc.inode->lock);
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, EC_MSG_PREOP_LOCK_FAILED,
+               "Failed to complete preop lock");
+    }
+
+    return 0;
+}
+
+gf_boolean_t
+ec_lock_acquire(ec_lock_link_t *link)
+{
+    ec_lock_t *lock;
+    ec_fop_data_t *fop;
+    gf_lkowner_t lk_owner;
+
+    lock = link->lock;
+    fop = link->fop;
+
+    if (!lock->acquired) {
+        set_lk_owner_from_ptr(&lk_owner, lock);
+
+        ec_trace("LOCK_ACQUIRE", fop, "lock=%p, inode=%p", lock,
+                 lock->loc.inode);
+
+        lock->flock.l_type = F_WRLCK;
+        ec_inodelk(fop->frame, fop->xl, &lk_owner, -1, EC_MINIMUM_ALL,
+                   ec_locked, link, fop->xl->name, &lock->loc, F_SETLKW,
+                   &lock->flock, NULL);
+
+        return _gf_false;
+    }
+
+    ec_trace("LOCK_REUSE", fop, "lock=%p", lock);
+
+    ec_lock_acquired(link);
+
+    return _gf_true;
+}
+
+static ec_lock_link_t *
+ec_lock_timer_cancel(xlator_t *xl, ec_lock_t *lock)
+{
+    ec_lock_link_t *timer_link;
+
+    /* If we don't have any timer, there's nothing to cancel. */
+    if (lock->timer == NULL) {
+        return NULL;
+    }
+
+    /* We are trying to access a lock that has an unlock timer active.
+     * This means that the lock must be idle, i.e. no fop can be in the
+     * owner, waiting or frozen lists. It also means that the lock cannot
+     * have been marked as being released (this is done without timers).
+     * There should only be one owner reference, but it's possible that
+     * some fops are being prepared to use this lock. */
+    GF_ASSERT((lock->refs_owners == 1) && list_empty(&lock->owners) &&
+              list_empty(&lock->waiting));
+
+    /* We take the timer_link before cancelling the timer, since a
+     * successful cancellation will destroy it. It must not be NULL
+     * because it references the fop responsible for the delayed unlock
+     * that we are currently trying to cancel. */
+    timer_link = lock->timer->data;
+    GF_ASSERT(timer_link != NULL);
+
+    if (gf_timer_call_cancel(xl->ctx, lock->timer) < 0) {
+        /* It's too late to avoid the execution of the timer callback.
+         * Since we need to be sure that the callback has access to all
+         * needed resources, we cannot resume the execution of the
+         * timer fop now. This will be done in the callback. */
+        timer_link = NULL;
+    } else {
+        /* The timer has been cancelled. The fop referenced by
+         * timer_link holds the last reference. The caller is
+         * responsible to release it when not needed anymore. */
+        ec_trace("UNLOCK_CANCELLED", timer_link->fop, "lock=%p", lock);
+    }
+
+    /* We have two options here:
+     *
+     * 1. The timer has been successfully cancelled.
+     *
+     *    This is the easiest case and we can continue with the currently
+     *    acquired lock.
+     *
+     * 2. The timer callback has already been fired.
+     *
+     *    In this case we have not been able to cancel the timer before
+     *    the timer callback has been fired, but we also know that
+     *    lock->timer != NULL. This means that the timer callback is still
+     *    trying to acquire the inode mutex that we currently own. We are
+     *    safe until we release it. In this case we can safely clear
+     *    lock->timer. This will cause that the timer callback does nothing
+     *    once it acquires the mutex.
+     */
+    lock->timer = NULL;
+
+    return timer_link;
+}
+
+static gf_boolean_t
+ec_lock_assign_owner(ec_lock_link_t *link)
+{
+    ec_fop_data_t *fop;
+    ec_lock_t *lock;
+    ec_lock_link_t *timer_link = NULL;
+    gf_boolean_t assigned = _gf_false;
+
+    /* The link cannot be in any list because we have just finished preparing
+     * it. */
+    GF_ASSERT(list_empty(&link->wait_list));
+
+    fop = link->fop;
+    lock = link->lock;
+
+    LOCK(&lock->loc.inode->lock);
+
+    /* Since the link has just been prepared but it's not active yet, the
+     * refs_pending must be one at least (the ref owned by this link). */
+    GF_ASSERT(lock->refs_pending > 0);
+    /* The link is not pending any more. It will be assigned to the owner,
+     * waiting or frozen list. */
+    lock->refs_pending--;
+
+    if (lock->release) {
+        ec_trace("LOCK_QUEUE_FREEZE", fop, "lock=%p", lock);
+
+        /* When lock->release is set, we'll unlock the lock as soon as
+         * possible, meaning that we won't use a timer. */
+        GF_ASSERT(lock->timer == NULL);
+
+        /* The lock is marked to be released. We can still have owners and fops
+         * in the waiting ilist f they have been added before the lock has been
+         * marked to be released. However new fops are put into the frozen list
+         * to wait for the next unlock/lock cycle. */
+        list_add_tail(&link->wait_list, &lock->frozen);
+
+        goto unlock;
+    }
+
+    /* The lock is not marked to be released, so the frozen list should be
+     * empty. */
+    GF_ASSERT(list_empty(&lock->frozen));
+
+    timer_link = ec_lock_timer_cancel(fop->xl, lock);
+
+    if (!list_empty(&lock->owners)) {
+        /* There are other owners of this lock. We can only take ownership if
+         * the lock is already acquired and doesn't have conflict with existing
+         * owners, or waiters(to prevent starvation).
+         * Otherwise we need to wait.
+         */
+        if (!lock->acquired || ec_link_has_lock_conflict(link, _gf_true)) {
+            ec_trace("LOCK_QUEUE_WAIT", fop, "lock=%p", lock);
+
+            list_add_tail(&link->wait_list, &lock->waiting);
+
+            goto unlock;
+        }
+    }
+
+    list_add_tail(&link->owner_list, &lock->owners);
+
+    /* If timer_link is not NULL, it means that we have inherited the owner
+     * reference assigned to the timer fop. In this case we simply reuse it.
+     * Otherwise we need to increase the number of owners. */
+    if (timer_link == NULL) {
+        lock->refs_owners++;
+    }
+
+    assigned = _gf_true;
+
+unlock:
+    if (!assigned) {
+        /* We have not been able to take ownership of this lock. The fop must
+         * be put to sleep. */
+        ec_sleep(fop);
+    }
+
+    UNLOCK(&lock->loc.inode->lock);
+
+    /* If we have cancelled the timer, we need to resume the fop that was
+     * waiting for it. */
+    if (timer_link != NULL) {
+        ec_resume(timer_link->fop, 0);
+    }
+
+    return assigned;
+}
+
+static void
+ec_lock_next_owner(ec_lock_link_t *link, ec_cbk_data_t *cbk,
+                   gf_boolean_t release)
+{
+    struct list_head list;
+    ec_lock_t *lock = link->lock;
+    ec_fop_data_t *fop = link->fop;
+    ec_inode_t *ctx = lock->ctx;
+
+    INIT_LIST_HEAD(&list);
+
+    LOCK(&lock->loc.inode->lock);
+
+    ec_trace("LOCK_DONE", fop, "lock=%p", lock);
+
+    /* Current link must belong to the owner list of the lock. We don't
+     * decrement lock->refs_owners here because the inode mutex is released
+     * before ec_unlock() is called and we need to know when the last owner
+     * unlocks the lock to do proper cleanup. lock->refs_owners is used for
+     * this task. */
+    GF_ASSERT((lock->refs_owners > 0) && !list_empty(&link->owner_list));
+    list_del_init(&link->owner_list);
+
+    lock->release |= release;
+
+    if ((fop->error == 0) && (cbk != NULL) && (cbk->op_ret >= 0)) {
+        if (link->update[0]) {
+            ctx->post_version[0]++;
+        }
+        if (link->update[1]) {
+            ctx->post_version[1]++;
+        }
+        /* If the fop fails on any of the good bricks, it is important to mark
+         * it dirty and update versions right away. */
+        if (link->update[0] || link->update[1]) {
+            if (lock->good_mask & ~(fop->good | fop->remaining)) {
+                lock->release = _gf_true;
+            }
+        }
+    }
+
+    if (fop->healing) {
+        lock->healing = fop->healing & (fop->good | fop->remaining);
+    }
+    ec_lock_update_good(lock, fop);
+
+    ec_lock_wake_shared(lock, &list);
+
+    UNLOCK(&lock->loc.inode->lock);
+
+    ec_lock_resume_shared(&list);
+}
+
+void
+ec_lock(ec_fop_data_t *fop)
+{
+    ec_lock_link_t *link;
+
+    /* There is a chance that ec_resume is called on fop even before ec_sleep.
+     * Which can result in refs == 0 for fop leading to use after free in this
+     * function when it calls ec_sleep so do ec_sleep at start and ec_resume at
+     * the end of this function.*/
+    ec_sleep(fop);
+
+    while (fop->locked < fop->lock_count) {
+        /* Since there are only up to 2 locks per fop, this xor will change
+         * the order of the locks if fop->first_lock is 1. */
+        link = &fop->locks[fop->locked ^ fop->first_lock];
+
+        if (!ec_lock_assign_owner(link) || !ec_lock_acquire(link)) {
+            break;
+        }
+    }
+
+    ec_resume(fop, 0);
+}
+
+void
+ec_lock_unfreeze(ec_lock_link_t *link)
+{
+    struct list_head list;
+    ec_lock_t *lock;
+    gf_boolean_t destroy = _gf_false;
+
+    lock = link->lock;
+
+    INIT_LIST_HEAD(&list);
+
+    LOCK(&lock->loc.inode->lock);
+
+    /* The lock must be marked to be released here, since we have just released
+     * it and any attempt to assign it to more fops must have added them to the
+     * frozen list. We can only have one active reference here: the one that
+     * is processing this unfreeze. */
+    GF_ASSERT(lock->release && (lock->refs_owners == 1));
+    lock->release = _gf_false;
+    lock->refs_owners = 0;
+
+    lock->acquired = _gf_false;
+
+    /* We are unfreezing a lock. This means that the lock has already been
+     * released. In this state it shouldn't have a pending timer nor have any
+     * owner, and the waiting list should be empty. Only the frozen list can
+     * contain some fop. */
+    GF_ASSERT((lock->timer == NULL) && list_empty(&lock->waiting) &&
+              list_empty(&lock->owners));
+
+    /* We move all frozen fops to the waiting list. */
+    list_splice_init(&lock->frozen, &lock->waiting);
+
+    /* If we don't have any fop waiting nor there are any prepared fops using
+     * this lock, we can finally dispose it. */
+    destroy = list_empty(&lock->waiting) && (lock->refs_pending == 0);
+    if (destroy) {
+        ec_trace("LOCK_DESTROY", link->fop, "lock=%p", lock);
+
+        lock->ctx->inode_lock = NULL;
+    } else {
+        ec_trace("LOCK_UNFREEZE", link->fop, "lock=%p", lock);
+
+        ec_lock_wake_shared(lock, &list);
+    }
+
+    UNLOCK(&lock->loc.inode->lock);
+
+    ec_lock_resume_shared(&list);
+
+    if (destroy) {
+        ec_lock_destroy(lock);
+    }
+}
+
+int32_t
+ec_unlocked(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = cookie;
+    ec_lock_link_t *link = fop->data;
+
+    if (op_ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, EC_MSG_UNLOCK_FAILED,
+               "entry/inode unlocking failed :(%s)", ec_msg_str(link->fop));
+    } else {
+        ec_trace("UNLOCKED", link->fop, "lock=%p", link->lock);
+    }
+
+    ec_lock_unfreeze(link);
+
+    return 0;
+}
+
+void
+ec_unlock_lock(ec_lock_link_t *link)
+{
+    ec_lock_t *lock;
+    ec_fop_data_t *fop;
+    gf_lkowner_t lk_owner;
+
+    lock = link->lock;
+    fop = link->fop;
+
+    lock->unlock_now = _gf_false;
+    ec_clear_inode_info(fop, lock->loc.inode);
+
+    if ((lock->mask != 0) && lock->acquired) {
+        set_lk_owner_from_ptr(&lk_owner, lock);
+        lock->flock.l_type = F_UNLCK;
+        ec_trace("UNLOCK_INODELK", fop, "lock=%p, inode=%p", lock,
+                 lock->loc.inode);
+
+        ec_inodelk(fop->frame, fop->xl, &lk_owner, lock->mask, EC_MINIMUM_ONE,
+                   ec_unlocked, link, fop->xl->name, &lock->loc, F_SETLK,
+                   &lock->flock, NULL);
+    } else {
+        ec_lock_unfreeze(link);
+    }
+}
+
+void
+ec_inode_bad_inc(inode_t *inode, xlator_t *xl)
+{
+    ec_inode_t *ctx = NULL;
+
+    LOCK(&inode->lock);
+    {
+        ctx = __ec_inode_get(inode, xl);
+        if (ctx == NULL) {
+            goto unlock;
+        }
+        ctx->bad_version++;
+    }
+unlock:
+    UNLOCK(&inode->lock);
+}
+
+int32_t
+ec_update_size_version_done(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int32_t op_ret, int32_t op_errno, dict_t *xattr,
+                            dict_t *xdata)
+{
+    ec_fop_data_t *fop = cookie;
+    ec_lock_link_t *link;
+    ec_lock_t *lock;
+    ec_inode_t *ctx;
+
+    link = fop->data;
+    lock = link->lock;
+    ctx = lock->ctx;
+
+    if (op_ret < 0) {
+        if (link->lock->fd == NULL) {
+            ec_inode_bad_inc(link->lock->loc.inode, this);
+        } else {
+            ec_inode_bad_inc(link->lock->fd->inode, this);
+        }
+
+        gf_msg(fop->xl->name, fop_log_level(fop->id, op_errno), op_errno,
+               EC_MSG_SIZE_VERS_UPDATE_FAIL,
+               "Failed to update version and size. %s", ec_msg_str(fop));
+    } else {
+        fop->parent->good &= fop->good;
+
+        ec_lock_update_good(lock, fop);
+
+        if (ec_dict_del_array(xattr, EC_XATTR_VERSION, ctx->post_version,
+                              EC_VERSION_SIZE) == 0) {
+            ctx->pre_version[0] = ctx->post_version[0];
+            ctx->pre_version[1] = ctx->post_version[1];
+
+            ctx->have_version = _gf_true;
+        }
+        if (ec_dict_del_number(xattr, EC_XATTR_SIZE, &ctx->post_size) == 0) {
+            ctx->pre_size = ctx->post_size;
+
+            ctx->have_size = _gf_true;
+        }
+        if ((ec_dict_del_config(xdata, EC_XATTR_CONFIG, &ctx->config) == 0) &&
+            ec_config_check(fop->xl, &ctx->config)) {
+            ctx->have_config = _gf_true;
+        }
+
+        ctx->have_info = _gf_true;
+    }
+    /* If we are here because of fop's and other than unlock request,
+     * that means we are still holding a lock. That make sure
+     * lock->unlock_now can not be modified.
+     */
+    if (lock->unlock_now) {
+        ec_unlock_lock(fop->data);
+    }
+
+    return 0;
+}
+
+void
+ec_update_size_version(ec_lock_link_t *link, uint64_t *version, uint64_t size,
+                       uint64_t *dirty)
+{
+    ec_fop_data_t *fop;
+    ec_lock_t *lock;
+    ec_inode_t *ctx;
+    dict_t *dict = NULL;
+    uintptr_t update_on = 0;
+    int32_t err = -ENOMEM;
+
+    fop = link->fop;
+    lock = link->lock;
+    ctx = lock->ctx;
+
+    ec_trace("UPDATE", fop, "version=%ld/%ld, size=%ld, dirty=%ld/%ld",
+             version[0], version[1], size, dirty[0], dirty[1]);
+
+    dict = dict_new();
+    if (dict == NULL) {
+        goto out;
+    }
+
+    /* If we don't have version information or it has been modified, we
+     * update it. */
+    if (!ctx->have_version || (version[0] != 0) || (version[1] != 0)) {
+        err = ec_dict_set_array(dict, EC_XATTR_VERSION, version,
+                                EC_VERSION_SIZE);
+        if (err != 0) {
+            goto out;
+        }
+    }
+
+    if (size != 0) {
+        /* If size has been changed, we should already
+         * know the previous size of the file. */
+        GF_ASSERT(ctx->have_size);
+
+        err = ec_dict_set_number(dict, EC_XATTR_SIZE, size);
+        if (err != 0) {
+            goto out;
+        }
+    }
+
+    if (dirty[0] || dirty[1]) {
+        err = ec_dict_set_array(dict, EC_XATTR_DIRTY, dirty, EC_VERSION_SIZE);
+        if (err != 0) {
+            goto out;
+        }
+    }
+
+    /* If config information is not known, we request it now. */
+    if ((lock->loc.inode->ia_type == IA_IFREG) && !ctx->have_config) {
+        /* A failure requesting this xattr is ignored because it's not
+         * absolutely required right now. */
+        (void)ec_dict_set_number(dict, EC_XATTR_CONFIG, 0);
+    }
+
+    fop->frame->root->uid = 0;
+    fop->frame->root->gid = 0;
+
+    update_on = lock->good_mask | lock->healing;
+
+    if (link->lock->fd == NULL) {
+        ec_xattrop(fop->frame, fop->xl, update_on, EC_MINIMUM_MIN,
+                   ec_update_size_version_done, link, &link->lock->loc,
+                   GF_XATTROP_ADD_ARRAY64, dict, NULL);
+    } else {
+        ec_fxattrop(fop->frame, fop->xl, update_on, EC_MINIMUM_MIN,
+                    ec_update_size_version_done, link, link->lock->fd,
+                    GF_XATTROP_ADD_ARRAY64, dict, NULL);
+    }
+
+    fop->frame->root->uid = fop->uid;
+    fop->frame->root->gid = fop->gid;
+
+    dict_unref(dict);
+
+    return;
+
+out:
+    if (dict != NULL) {
+        dict_unref(dict);
+    }
+
+    ec_fop_set_error(fop, -err);
+
+    gf_msg(fop->xl->name, GF_LOG_ERROR, -err, EC_MSG_SIZE_VERS_UPDATE_FAIL,
+           "Unable to update version and size. %s", ec_msg_str(fop));
+
+    if (lock->unlock_now) {
+        ec_unlock_lock(fop->data);
+    }
+}
+
+gf_boolean_t
+ec_update_info(ec_lock_link_t *link)
+{
+    ec_lock_t *lock;
+    ec_inode_t *ctx;
+    uint64_t version[2] = {0, 0};
+    uint64_t dirty[2] = {0, 0};
+    uint64_t size;
+    ec_t *ec = NULL;
+    uintptr_t mask;
+
+    lock = link->lock;
+    ctx = lock->ctx;
+    ec = link->fop->xl->private;
+
+    /* pre_version[*] will be 0 if have_version is false */
+    version[EC_DATA_TXN] = ctx->post_version[EC_DATA_TXN] -
+                           ctx->pre_version[EC_DATA_TXN];
+    version[EC_METADATA_TXN] = ctx->post_version[EC_METADATA_TXN] -
+                               ctx->pre_version[EC_METADATA_TXN];
+
+    size = ctx->post_size - ctx->pre_size;
+    /* If we set the dirty flag for update fop, we have to unset it.
+     * If fop has failed on some bricks, leave the dirty as marked. */
+
+    if (lock->unlock_now) {
+        if (version[EC_DATA_TXN]) {
+            /*A data fop will have difference in post and pre version
+             *and for data fop we send writes on healing bricks also */
+            mask = lock->good_mask | lock->healing;
+        } else {
+            mask = lock->good_mask;
+        }
+        /* Ensure that nodes are up while doing final
+         * metadata update.*/
+        if (!(ec->node_mask & ~(mask)) && !(ec->node_mask & ~ec->xl_up)) {
+            if (ctx->dirty[EC_DATA_TXN] != 0) {
+                dirty[EC_DATA_TXN] = -1;
+            }
+            if (ctx->dirty[EC_METADATA_TXN] != 0) {
+                dirty[EC_METADATA_TXN] = -1;
+            }
+            /*If everything is fine and we already
+             *have version xattr set on entry, there
+             *is no need to update version again*/
+            if (ctx->pre_version[EC_DATA_TXN]) {
+                version[EC_DATA_TXN] = 0;
+            }
+            if (ctx->pre_version[EC_METADATA_TXN]) {
+                version[EC_METADATA_TXN] = 0;
+            }
+        } else {
+            link->optimistic_changelog = _gf_false;
+            ec_set_dirty_flag(link, ctx, dirty);
+        }
+        memset(ctx->dirty, 0, sizeof(ctx->dirty));
+    }
+
+    if ((version[EC_DATA_TXN] != 0) || (version[EC_METADATA_TXN] != 0) ||
+        (dirty[EC_DATA_TXN] != 0) || (dirty[EC_METADATA_TXN] != 0)) {
+        ec_update_size_version(link, version, size, dirty);
+        return _gf_true;
+    }
+
+    return _gf_false;
+}
+
+void
+ec_unlock_now(ec_lock_link_t *link)
+{
+    ec_lock_t *lock;
+    lock = link->lock;
+
+    ec_trace("UNLOCK_NOW", link->fop, "lock=%p", link->lock);
+    /*At this point, lock is not being used by any fop and
+     *can not be reused by any fop as it is going to be released.
+     *lock->unlock_now can not be modified at any other place.
+     */
+    lock->unlock_now = _gf_true;
+
+    if (!ec_update_info(link)) {
+        ec_unlock_lock(link);
+    }
+
+    ec_resume(link->fop, 0);
+}
+
+void
+ec_lock_release(ec_t *ec, inode_t *inode)
+{
+    ec_lock_t *lock;
+    ec_inode_t *ctx;
+    ec_lock_link_t *timer_link = NULL;
+
+    LOCK(&inode->lock);
+
+    ctx = __ec_inode_get(inode, ec->xl);
+    if (ctx == NULL) {
+        goto done;
+    }
+    lock = ctx->inode_lock;
+    if ((lock == NULL) || lock->release) {
+        goto done;
+    }
+
+    gf_msg_debug(ec->xl->name, 0, "Releasing inode %p due to lock contention",
+                 inode);
+
+    if (!lock->acquired) {
+        /* This happens if some bricks already got the lock while inodelk is in
+         * progress.  Set release to true after lock is acquired*/
+        lock->contention = _gf_true;
+        goto done;
+    }
+
+    /* The lock is not marked to be released, so the frozen list should be
+     * empty. */
+    GF_ASSERT(list_empty(&lock->frozen));
+
+    timer_link = ec_lock_timer_cancel(ec->xl, lock);
+
+    /* We mark the lock to be released as soon as possible. */
+    lock->release = _gf_true;
+
+done:
+    UNLOCK(&inode->lock);
+
+    /* If we have cancelled the timer, we need to start the unlock of the
+     * inode. If there was a timer but we have been unable to cancel it
+     * because it was just triggered, the timer callback will take care
+     * of releasing the inode. */
+    if (timer_link != NULL) {
+        ec_unlock_now(timer_link);
+    }
+}
+
+void
+ec_unlock_timer_add(ec_lock_link_t *link);
+
+void
+ec_unlock_timer_del(ec_lock_link_t *link)
+{
+    ec_lock_t *lock;
+    inode_t *inode;
+    gf_boolean_t now = _gf_false;
+
+    /* If we are here, it means that the timer has expired before having
+     * been cancelled. This guarantees that 'link' is still valid because
+     * the fop that contains it must be pending (if timer cancellation in
+     * ec_lock_assign_owner() fails, the fop is left sleeping).
+     *
+     * At the same time, the fop still has a reference to the lock, so
+     * it must also be valid.
+     */
+    lock = link->lock;
+
+    /* 'lock' must have a valid inode since it can only be destroyed
+     * when the lock itself is destroyed, but we have a reference to the
+     * lock to avoid this.
+     */
+    inode = lock->loc.inode;
+
+    LOCK(&inode->lock);
+
+    if (lock->timer != NULL) {
+        ec_trace("UNLOCK_DELAYED", link->fop, "lock=%p", lock);
+
+        /* The unlock timer has expired without anyone cancelling it.
+         * This means that it shouldn't have any owner, and the waiting
+         * and frozen lists should be empty.  It must have only one
+         * owner reference, but there can be fops being prepared
+         * though.
+         * */
+        GF_ASSERT(!lock->release && (lock->refs_owners == 1) &&
+                  list_empty(&lock->owners) && list_empty(&lock->waiting) &&
+                  list_empty(&lock->frozen));
+
+        gf_timer_call_cancel(link->fop->xl->ctx, lock->timer);
+        lock->timer = NULL;
+
+        /* Any fop being processed from now on, will need to wait
+         * until the next unlock/lock cycle. */
+        lock->release = now = _gf_true;
+    }
+
+    UNLOCK(&inode->lock);
+
+    if (now) {
+        ec_unlock_now(link);
+    } else {
+        /* The timer has been cancelled just after firing it but before
+         * getting here. This means that another fop has used the lock
+         * and everything should be handled as if this callback were
+         * have not been executed. However we still have an owner
+         * reference.
+         *
+         * We need to release our reference. If this is not the last
+         * reference (the most common case because another fop has
+         * taken another ref) we only need to decrement the counter.
+         * Otherwise we have been delayed enough so that the other fop
+         * has had time to acquire the reference, do its operation and
+         * release it. At the time of releasing it, the fop did found
+         * that the ref counter was > 1 (our reference), so the delayed
+         * unlock timer wasn't started. We need to start it again if we
+         * are the last reference.
+         *
+         * ec_unlock_timer_add() handles both cases.
+         */
+        ec_unlock_timer_add(link);
+
+        /* We need to resume the fop that was waiting for the delayed
+         * unlock.
+         */
+        ec_resume(link->fop, 0);
+    }
+}
+
+void
+ec_unlock_timer_cbk(void *data)
+{
+    ec_unlock_timer_del(data);
+}
+
+static gf_boolean_t
+ec_eager_lock_used(ec_t *ec, ec_fop_data_t *fop)
+{
+    /* Fops with no locks at this point mean that they are sent as sub-fops
+     * of other higher level fops. In this case we simply assume that the
+     * parent fop will take correct care of the eager lock. */
+    if (fop->lock_count == 0) {
+        return _gf_true;
+    }
+
+    /* We may have more than one lock, but this only happens in the rename
+     * fop, and both locks will reference an inode of the same type (a
+     * directory in this case), so we only need to check the first lock. */
+    if (fop->locks[0].lock->loc.inode->ia_type == IA_IFREG) {
+        return ec->eager_lock;
+    }
+
+    return ec->other_eager_lock;
+}
+
+static uint32_t
+ec_eager_lock_timeout(ec_t *ec, ec_lock_t *lock)
+{
+    if (lock->loc.inode->ia_type == IA_IFREG) {
+        return ec->eager_lock_timeout;
+    }
+
+    return ec->other_eager_lock_timeout;
+}
+
+static gf_boolean_t
+ec_lock_delay_create(ec_lock_link_t *link)
+{
+    struct timespec delay;
+    ec_fop_data_t *fop = link->fop;
+    ec_lock_t *lock = link->lock;
+
+    delay.tv_sec = ec_eager_lock_timeout(fop->xl->private, lock);
+    delay.tv_nsec = 0;
+    lock->timer = gf_timer_call_after(fop->xl->ctx, delay, ec_unlock_timer_cbk,
+                                      link);
+    if (lock->timer == NULL) {
+        gf_msg(fop->xl->name, GF_LOG_WARNING, ENOMEM,
+               EC_MSG_UNLOCK_DELAY_FAILED, "Unable to delay an unlock");
+
+        return _gf_false;
+    }
+
+    return _gf_true;
+}
+
+void
+ec_unlock_timer_add(ec_lock_link_t *link)
+{
+    ec_fop_data_t *fop = link->fop;
+    ec_lock_t *lock = link->lock;
+    gf_boolean_t now = _gf_false;
+
+    LOCK(&lock->loc.inode->lock);
+
+    /* We are trying to unlock the lock. We can have multiple scenarios here,
+     * but all of them need to have lock->timer == NULL:
+     *
+     * 1. There are other owners currently running that can call ec_unlock().
+     *
+     *    None of them can have started the timer until the last one. But this
+     *    call should be the consequence of this lastest one.
+     *
+     * 2. There are fops in the waiting or frozen lists.
+     *
+     *    These fops cannot call ec_unlock(). So we should be here.
+     *
+     * We must reach here with at least one owner reference.
+     */
+    GF_ASSERT((lock->timer == NULL) && (lock->refs_owners > 0));
+
+    /* If the fop detects that a heal is needed, we mark the lock to be
+     * released as soon as possible. */
+    lock->release |= ec_fop_needs_heal(fop);
+
+    if (lock->refs_owners > 1) {
+        ec_trace("UNLOCK_SKIP", fop, "lock=%p", lock);
+
+        /* If there are other owners we cannot do anything else with the lock.
+         * Note that the current fop has already been removed from the owners
+         * list in ec_lock_reuse(). */
+        lock->refs_owners--;
+
+        UNLOCK(&lock->loc.inode->lock);
+    } else if (lock->acquired) {
+        /* There are no other owners and the lock is acquired. If there were
+         * fops waiting, at least one of them should have been promoted to an
+         * owner, so the waiting list should be empty. */
+        GF_ASSERT(list_empty(&lock->owners) && list_empty(&lock->waiting));
+
+        ec_t *ec = fop->xl->private;
+
+        /* If everything goes as expected this fop will be put to sleep until
+         * the timer callback is executed. */
+        ec_sleep(fop);
+
+        /* If the lock needs to be released, or ec is shutting down, do not
+         * delay lock release. */
+        if (!lock->release && !ec->shutdown) {
+            ec_trace("UNLOCK_DELAY", fop, "lock=%p, release=%d", lock,
+                     lock->release);
+
+            if (!ec_lock_delay_create(link)) {
+                /* We are unable to create a new timer. We immediately release
+                 * the lock. */
+                lock->release = now = _gf_true;
+            }
+
+        } else {
+            ec_trace("UNLOCK_FORCE", fop, "lock=%p, release=%d", lock,
+                     lock->release);
+            lock->release = now = _gf_true;
+        }
+
+        UNLOCK(&lock->loc.inode->lock);
+
+        if (now) {
+            ec_unlock_now(link);
+        }
+    } else {
+        /* There are no owners and the lock is not acquired. This can only
+         * happen if a lock attempt has failed and we get to the unlock step
+         * of the fop. As in the previous case, the waiting list must be
+         * empty. */
+        GF_ASSERT(list_empty(&lock->owners) && list_empty(&lock->waiting));
+
+        /* We need to mark the lock to be released to correctly handle fops
+         * that may get in after we release the inode mutex but before
+         * ec_lock_unfreeze() is processed. */
+        lock->release = _gf_true;
+
+        UNLOCK(&lock->loc.inode->lock);
+
+        ec_lock_unfreeze(link);
+    }
+}
+
+void
+ec_unlock(ec_fop_data_t *fop)
+{
+    int32_t i;
+
+    for (i = 0; i < fop->lock_count; i++) {
+        ec_unlock_timer_add(&fop->locks[i]);
+    }
+}
+
+void
+ec_flush_size_version(ec_fop_data_t *fop)
+{
+    GF_ASSERT(fop->lock_count == 1);
+    ec_update_info(&fop->locks[0]);
+}
+
+static void
+ec_update_stripe(ec_t *ec, ec_stripe_list_t *stripe_cache, ec_stripe_t *stripe,
+                 ec_fop_data_t *fop)
+{
+    off_t base;
+
+    /* On write fops, we only update existing fragments if the write has
+     * succeeded. Otherwise, we remove them from the cache. */
+    if ((fop->id == GF_FOP_WRITE) && (fop->answer != NULL) &&
+        (fop->answer->op_ret >= 0)) {
+        base = stripe->frag_offset - fop->frag_range.first;
+        base *= ec->fragments;
+
+        /* We check if the stripe offset falls inside the real region
+         * modified by the write fop (a write request is allowed,
+         * though uncommon, to write less bytes than requested). The
+         * current write fop implementation doesn't allow partial
+         * writes of fragments, so if there's no error, we are sure
+         * that a full stripe has been completely modified or not
+         * touched at all. The value of op_ret may not be a multiple
+         * of the stripe size because it depends on the requested
+         * size by the user, so we update the stripe if the write has
+         * modified at least one byte (meaning ec has written the full
+         * stripe). */
+        if (base < fop->answer->op_ret + fop->head) {
+            memcpy(stripe->data, fop->vector[0].iov_base + base,
+                   ec->stripe_size);
+            list_move_tail(&stripe->lru, &stripe_cache->lru);
+
+            GF_ATOMIC_INC(ec->stats.stripe_cache.updates);
+        }
+    } else {
+        stripe->frag_offset = -1;
+        list_move(&stripe->lru, &stripe_cache->lru);
+
+        GF_ATOMIC_INC(ec->stats.stripe_cache.invals);
+    }
+}
+
+static void
+ec_update_cached_stripes(ec_fop_data_t *fop)
+{
+    uint64_t first;
+    uint64_t last;
+    ec_stripe_t *stripe = NULL;
+    ec_inode_t *ctx = NULL;
+    ec_stripe_list_t *stripe_cache = NULL;
+    inode_t *inode = NULL;
+    struct list_head *temp;
+    struct list_head sentinel;
+
+    first = fop->frag_range.first;
+    /* 'last' represents the first stripe not touched by the operation */
+    last = fop->frag_range.last;
+
+    /* If there are no modified stripes, we don't need to do anything
+     * else. */
+    if (last <= first) {
+        return;
+    }
+
+    if (!fop->use_fd) {
+        inode = fop->loc[0].inode;
+    } else {
+        inode = fop->fd->inode;
+    }
+
+    LOCK(&inode->lock);
+
+    ctx = __ec_inode_get(inode, fop->xl);
+    if (ctx == NULL) {
+        goto out;
+    }
+    stripe_cache = &ctx->stripe_cache;
+
+    /* Since we'll be moving elements of the list to the tail, we might
+     * end in an infinite loop. To avoid it, we insert a sentinel element
+     * into the list, so that it will be used to detect when we have
+     * traversed all existing elements once. */
+    list_add_tail(&sentinel, &stripe_cache->lru);
+    temp = stripe_cache->lru.next;
+    while (temp != &sentinel) {
+        stripe = list_entry(temp, ec_stripe_t, lru);
+        temp = temp->next;
+        if ((first <= stripe->frag_offset) && (stripe->frag_offset < last)) {
+            ec_update_stripe(fop->xl->private, stripe_cache, stripe, fop);
+        }
+    }
+    list_del(&sentinel);
+
+out:
+    UNLOCK(&inode->lock);
+}
+
+void
+ec_lock_reuse(ec_fop_data_t *fop)
+{
+    ec_cbk_data_t *cbk;
+    ec_t *ec = NULL;
+    int32_t i, count;
+    gf_boolean_t release = _gf_false;
+    ec = fop->xl->private;
+    cbk = fop->answer;
+
+    if (ec_eager_lock_used(ec, fop) && cbk != NULL) {
+        if (cbk->xdata != NULL) {
+            if ((dict_get_int32(cbk->xdata, GLUSTERFS_INODELK_COUNT, &count) ==
+                 0) &&
+                (count > 1)) {
+                release = _gf_true;
+            }
+            if (release) {
+                gf_msg_debug(fop->xl->name, 0, "Lock contention detected");
+            }
+        }
+    } else {
+        /* If eager lock is disabled or if we haven't get
+         * an answer with enough quorum, we always release
+         * the lock. */
+        release = _gf_true;
+    }
+    ec_update_cached_stripes(fop);
+
+    for (i = 0; i < fop->lock_count; i++) {
+        ec_lock_next_owner(&fop->locks[i], cbk, release);
+    }
+}
+
+void
+__ec_manager(ec_fop_data_t *fop, int32_t error)
+{
+    ec_t *ec = fop->xl->private;
+
+    do {
+        ec_trace("MANAGER", fop, "error=%d", error);
+
+        if (!ec_must_wind(fop)) {
+            if (ec->xl_up_count < ec->fragments) {
+                error = ENOTCONN;
+            }
+        }
+
+        if (error != 0) {
+            fop->error = error;
+            fop->state = -fop->state;
+        }
+
+        if ((fop->state == EC_STATE_END) || (fop->state == -EC_STATE_END)) {
+            ec_fop_data_release(fop);
+
+            break;
+        }
+
+        /* At each state, fop must not be used anywhere else and there
+         * shouldn't be any pending subfop going on. */
+        GF_ASSERT(fop->jobs == 0);
+
+        /* While the manager is running we need to avoid that subfops launched
+         * from it could finish and call ec_resume() before the fop->handler
+         * has completed. This could lead to the same manager being executed
+         * by two threads concurrently. ec_check_complete() will take care of
+         * this reference. */
+        fop->jobs = 1;
+
+        fop->state = fop->handler(fop, fop->state);
+        GF_ASSERT(fop->state >= 0);
+
+        error = ec_check_complete(fop, __ec_manager);
+    } while (error >= 0);
+}
+
+void
+ec_manager(ec_fop_data_t *fop, int32_t error)
+{
+    GF_ASSERT(fop->jobs == 0);
+    GF_ASSERT(fop->winds == 0);
+    GF_ASSERT(fop->error == 0);
+
+    if (fop->state == EC_STATE_START) {
+        fop->state = EC_STATE_INIT;
+    }
+
+    __ec_manager(fop, error);
+}
+
+gf_boolean_t
+__ec_is_last_fop(ec_t *ec)
+{
+    if ((list_empty(&ec->pending_fops)) &&
+        (GF_ATOMIC_GET(ec->async_fop_count) == 0)) {
+        return _gf_true;
+    }
+    return _gf_false;
+}
diff --git a/xlators/cluster/ec/src/ec-common.h b/xlators/cluster/ec/src/ec-common.h
new file mode 100644
index 00000000000..51493612ac6
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-common.h
@@ -0,0 +1,234 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_COMMON_H__
+#define __EC_COMMON_H__
+
+#include "glusterfs/compat-errno.h"  // for ENODATA on BSD
+#include "ec-data.h"
+
+typedef enum { EC_DATA_TXN, EC_METADATA_TXN } ec_txn_t;
+
+#define EC_FOP_HEAL -1
+#define EC_FOP_FHEAL -2
+
+#define EC_CONFIG_VERSION 0
+
+#define EC_CONFIG_ALGORITHM 0
+
+#define EC_FLAG_LOCK_SHARED 0x0001
+
+#define QUORUM_CBK(fn, fop, frame, cookie, this, op_ret, op_errno, params...)  \
+    do {                                                                       \
+        ec_t *__ec = fop->xl->private;                                         \
+        int32_t __op_ret = 0;                                                  \
+        int32_t __op_errno = 0;                                                \
+        int32_t __success_count = gf_bits_count(fop->good);                    \
+                                                                               \
+        __op_ret = op_ret;                                                     \
+        __op_errno = op_errno;                                                 \
+        if (!fop->parent && frame &&                                           \
+            (GF_CLIENT_PID_SELF_HEALD != frame->root->pid) &&                  \
+            __ec->quorum_count && (__success_count < __ec->quorum_count) &&    \
+            op_ret >= 0) {                                                     \
+            __op_ret = -1;                                                     \
+            __op_errno = EIO;                                                  \
+            gf_msg(__ec->xl->name, GF_LOG_ERROR, 0,                            \
+                   EC_MSG_CHILDS_INSUFFICIENT,                                 \
+                   "Insufficient available children for this request "         \
+                   "(have %d, need %d). %s",                                   \
+                   __success_count, __ec->quorum_count, ec_msg_str(fop));      \
+        }                                                                      \
+        fn(frame, cookie, this, __op_ret, __op_errno, params);                 \
+    } while (0)
+
+enum _ec_xattrop_flags {
+    EC_FLAG_XATTROP,
+    EC_FLAG_DATA_DIRTY,
+    EC_FLAG_METADATA_DIRTY,
+
+    /* Add any new flag here, before EC_FLAG_MAX. The maximum number of
+     * flags that can be defined is 16. */
+
+    EC_FLAG_MAX
+};
+
+/* We keep two sets of flags. One to determine what's really providing the
+ * current xattrop and the other to know what the parent fop of the xattrop
+ * needs to proceed. It might happen that a fop needs some information that
+ * is being already requested by a previous fop. The two sets are stored
+ * contiguously. */
+
+#define EC_FLAG_NEEDS(_flag) (1 << (_flag))
+#define EC_FLAG_PROVIDES(_flag) (1 << ((_flag) + EC_FLAG_MAX))
+
+#define EC_NEEDED_FLAGS(_flags) ((_flags) & ((1 << EC_FLAG_MAX) - 1))
+
+#define EC_PROVIDED_FLAGS(_flags) EC_NEEDED_FLAGS((_flags) >> EC_FLAG_MAX)
+
+#define EC_FLAGS_HAVE(_flags, _flag) (((_flags) & (1 << (_flag))) != 0)
+
+#define EC_SELFHEAL_BIT 62
+
+#define EC_MINIMUM_ONE (1 << 6)
+#define EC_MINIMUM_MIN (2 << 6)
+#define EC_MINIMUM_ALL (3 << 6)
+#define EC_FOP_NO_PROPAGATE_ERROR (1 << 8)
+#define EC_FOP_MINIMUM(_flags) ((_flags)&255)
+#define EC_FOP_FLAGS(_flags) ((_flags) & ~255)
+
+#define EC_UPDATE_DATA 1
+#define EC_UPDATE_META 2
+#define EC_QUERY_INFO 4
+#define EC_INODE_SIZE 8
+
+#define EC_STATE_START 0
+#define EC_STATE_END 0
+#define EC_STATE_INIT 1
+#define EC_STATE_LOCK 2
+#define EC_STATE_DISPATCH 3
+#define EC_STATE_PREPARE_ANSWER 4
+#define EC_STATE_REPORT 5
+#define EC_STATE_LOCK_REUSE 6
+#define EC_STATE_UNLOCK 7
+
+#define EC_STATE_DELAYED_START 100
+
+#define EC_STATE_HEAL_ENTRY_LOOKUP 200
+#define EC_STATE_HEAL_ENTRY_PREPARE 201
+#define EC_STATE_HEAL_PRE_INODELK_LOCK 202
+#define EC_STATE_HEAL_PRE_INODE_LOOKUP 203
+#define EC_STATE_HEAL_XATTRIBUTES_REMOVE 204
+#define EC_STATE_HEAL_XATTRIBUTES_SET 205
+#define EC_STATE_HEAL_ATTRIBUTES 206
+#define EC_STATE_HEAL_OPEN 207
+#define EC_STATE_HEAL_REOPEN_FD 208
+#define EC_STATE_HEAL_UNLOCK 209
+#define EC_STATE_HEAL_UNLOCK_ENTRY 210
+#define EC_STATE_HEAL_DATA_LOCK 211
+#define EC_STATE_HEAL_DATA_COPY 212
+#define EC_STATE_HEAL_DATA_UNLOCK 213
+#define EC_STATE_HEAL_POST_INODELK_LOCK 214
+#define EC_STATE_HEAL_POST_INODE_LOOKUP 215
+#define EC_STATE_HEAL_SETATTR 216
+#define EC_STATE_HEAL_POST_INODELK_UNLOCK 217
+#define EC_STATE_HEAL_DISPATCH 218
+
+/* Value to cover the full range of a file */
+#define EC_RANGE_FULL ((uint64_t)LLONG_MAX + 1)
+
+gf_boolean_t
+ec_dispatch_one_retry(ec_fop_data_t *fop, ec_cbk_data_t **cbk);
+void
+ec_dispatch_next(ec_fop_data_t *fop, uint32_t idx);
+
+void
+ec_complete(ec_fop_data_t *fop);
+
+void
+ec_update_good(ec_fop_data_t *fop, uintptr_t good);
+
+void
+ec_fop_set_error(ec_fop_data_t *fop, int32_t error);
+
+void
+__ec_fop_set_error(ec_fop_data_t *fop, int32_t error);
+
+ec_cbk_data_t *
+ec_fop_prepare_answer(ec_fop_data_t *fop, gf_boolean_t ro);
+
+gf_boolean_t
+ec_cbk_set_error(ec_cbk_data_t *cbk, int32_t error, gf_boolean_t ro);
+
+void
+ec_lock_prepare_inode(ec_fop_data_t *fop, loc_t *loc, uint32_t flags,
+                      off_t fl_start, uint64_t fl_size);
+void
+ec_lock_prepare_parent_inode(ec_fop_data_t *fop, loc_t *loc, loc_t *base,
+                             uint32_t flags);
+void
+ec_lock_prepare_fd(ec_fop_data_t *fop, fd_t *fd, uint32_t flags, off_t fl_start,
+                   uint64_t fl_size);
+void
+ec_lock(ec_fop_data_t *fop);
+void
+ec_lock_reuse(ec_fop_data_t *fop);
+void
+ec_unlock(ec_fop_data_t *fop);
+void
+ec_lock_release(ec_t *ec, inode_t *inode);
+
+gf_boolean_t
+ec_get_inode_size(ec_fop_data_t *fop, inode_t *inode, uint64_t *size);
+gf_boolean_t
+__ec_get_inode_size(ec_fop_data_t *fop, inode_t *inode, uint64_t *size);
+gf_boolean_t
+ec_set_inode_size(ec_fop_data_t *fop, inode_t *inode, uint64_t size);
+gf_boolean_t
+__ec_set_inode_size(ec_fop_data_t *fop, inode_t *inode, uint64_t size);
+void
+ec_clear_inode_info(ec_fop_data_t *fop, inode_t *inode);
+
+void
+ec_flush_size_version(ec_fop_data_t *fop);
+
+void
+ec_dispatch_all(ec_fop_data_t *fop);
+void
+ec_dispatch_inc(ec_fop_data_t *fop);
+void
+ec_dispatch_min(ec_fop_data_t *fop);
+void
+ec_dispatch_one(ec_fop_data_t *fop);
+
+void
+ec_succeed_all(ec_fop_data_t *fop);
+
+void
+ec_sleep(ec_fop_data_t *fop);
+void
+ec_resume(ec_fop_data_t *fop, int32_t error);
+void
+ec_resume_parent(ec_fop_data_t *fop);
+
+void
+ec_manager(ec_fop_data_t *fop, int32_t error);
+gf_boolean_t
+ec_is_recoverable_error(int32_t op_errno);
+void
+ec_handle_healers_done(ec_fop_data_t *fop);
+
+int32_t
+ec_heal_inspect(call_frame_t *frame, ec_t *ec, inode_t *inode,
+                unsigned char *locked_on, gf_boolean_t self_locked,
+                gf_boolean_t thorough, ec_heal_need_t *need_heal);
+int32_t
+ec_get_heal_info(xlator_t *this, loc_t *loc, dict_t **dict);
+
+int32_t
+ec_lock_unlocked(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata);
+
+void
+ec_update_fd_status(fd_t *fd, xlator_t *xl, int child_index,
+                    int32_t ret_status);
+gf_boolean_t
+ec_is_entry_healing(ec_fop_data_t *fop);
+void
+ec_set_entry_healing(ec_fop_data_t *fop);
+void
+ec_reset_entry_healing(ec_fop_data_t *fop);
+char *
+ec_msg_str(ec_fop_data_t *fop);
+gf_boolean_t
+__ec_is_last_fop(ec_t *ec);
+void
+ec_lock_update_good(ec_lock_t *lock, ec_fop_data_t *fop);
+#endif /* __EC_COMMON_H__ */
diff --git a/xlators/cluster/ec/src/ec-data.c b/xlators/cluster/ec/src/ec-data.c
new file mode 100644
index 00000000000..06388833546
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-data.c
@@ -0,0 +1,288 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "ec-helpers.h"
+#include "ec-common.h"
+#include "ec-data.h"
+#include "ec-messages.h"
+
+ec_cbk_data_t *
+ec_cbk_data_allocate(call_frame_t *frame, xlator_t *this, ec_fop_data_t *fop,
+                     int32_t id, int32_t idx, int32_t op_ret, int32_t op_errno)
+{
+    ec_cbk_data_t *cbk;
+    ec_t *ec = this->private;
+
+    if (fop->xl != this) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, EC_MSG_XLATOR_MISMATCH,
+               "Mismatching xlators between request "
+               "and answer (req=%s, ans=%s).",
+               fop->xl->name, this->name);
+
+        return NULL;
+    }
+    if (fop->frame != frame) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, EC_MSG_FRAME_MISMATCH,
+               "Mismatching frames between request "
+               "and answer (req=%p, ans=%p).",
+               fop->frame, frame);
+
+        return NULL;
+    }
+    if (fop->id != id) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, EC_MSG_FOP_MISMATCH,
+               "Mismatching fops between request "
+               "and answer (req=%d, ans=%d).",
+               fop->id, id);
+
+        return NULL;
+    }
+
+    cbk = mem_get0(ec->cbk_pool);
+    if (cbk == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+               "Failed to allocate memory for an "
+               "answer.");
+        return NULL;
+    }
+
+    cbk->fop = fop;
+    cbk->idx = idx;
+    cbk->mask = 1ULL << idx;
+    cbk->count = 1;
+    cbk->op_ret = op_ret;
+    cbk->op_errno = op_errno;
+    INIT_LIST_HEAD(&cbk->entries.list);
+
+    LOCK(&fop->lock);
+
+    list_add_tail(&cbk->answer_list, &fop->answer_list);
+
+    UNLOCK(&fop->lock);
+
+    return cbk;
+}
+
+void
+ec_cbk_data_destroy(ec_cbk_data_t *cbk)
+{
+    if (cbk->xdata != NULL) {
+        dict_unref(cbk->xdata);
+    }
+    if (cbk->dict != NULL) {
+        dict_unref(cbk->dict);
+    }
+    if (cbk->inode != NULL) {
+        inode_unref(cbk->inode);
+    }
+    if (cbk->fd != NULL) {
+        fd_unref(cbk->fd);
+    }
+    if (cbk->buffers != NULL) {
+        iobref_unref(cbk->buffers);
+    }
+    GF_FREE(cbk->vector);
+    gf_dirent_free(&cbk->entries);
+    GF_FREE(cbk->str);
+
+    mem_put(cbk);
+}
+
+ec_fop_data_t *
+ec_fop_data_allocate(call_frame_t *frame, xlator_t *this, int32_t id,
+                     uint32_t flags, uintptr_t target, uint32_t fop_flags,
+                     ec_wind_f wind, ec_handler_f handler, ec_cbk_t cbks,
+                     void *data)
+{
+    ec_fop_data_t *fop, *parent;
+    ec_t *ec = this->private;
+
+    fop = mem_get0(ec->fop_pool);
+    if (fop == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+               "Failed to allocate memory for a "
+               "request.");
+
+        return NULL;
+    }
+
+    INIT_LIST_HEAD(&fop->cbk_list);
+    INIT_LIST_HEAD(&fop->healer);
+    INIT_LIST_HEAD(&fop->answer_list);
+    INIT_LIST_HEAD(&fop->pending_list);
+    INIT_LIST_HEAD(&fop->locks[0].owner_list);
+    INIT_LIST_HEAD(&fop->locks[0].wait_list);
+    INIT_LIST_HEAD(&fop->locks[1].owner_list);
+    INIT_LIST_HEAD(&fop->locks[1].wait_list);
+
+    fop->xl = this;
+    fop->req_frame = frame;
+
+    /* fops need a private frame to be able to execute some postop operations
+     * even if the original fop has completed and reported back to the upper
+     * xlator and it has destroyed the base frame.
+     *
+     * TODO: minimize usage of private frames. Reuse req_frame as much as
+     *       possible.
+     */
+    if (frame != NULL) {
+        fop->frame = copy_frame(frame);
+    } else {
+        fop->frame = create_frame(this, this->ctx->pool);
+    }
+    if (fop->frame == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+               "Failed to create a private frame "
+               "for a request");
+
+        mem_put(fop);
+
+        return NULL;
+    }
+    fop->id = id;
+    fop->refs = 1;
+
+    fop->flags = flags;
+    fop->minimum = EC_FOP_MINIMUM(fop_flags);
+    fop->fop_flags = EC_FOP_FLAGS(fop_flags);
+    fop->mask = target;
+
+    fop->wind = wind;
+    fop->handler = handler;
+    fop->cbks = cbks;
+    fop->data = data;
+
+    fop->uid = fop->frame->root->uid;
+    fop->gid = fop->frame->root->gid;
+
+    LOCK_INIT(&fop->lock);
+
+    fop->frame->local = fop;
+
+    if (frame != NULL) {
+        parent = frame->local;
+        if (parent != NULL) {
+            ec_sleep(parent);
+        }
+
+        fop->parent = parent;
+    }
+
+    LOCK(&ec->lock);
+
+    list_add_tail(&fop->pending_list, &ec->pending_fops);
+
+    UNLOCK(&ec->lock);
+
+    return fop;
+}
+
+void
+ec_fop_data_acquire(ec_fop_data_t *fop)
+{
+    LOCK(&fop->lock);
+
+    ec_trace("ACQUIRE", fop, "");
+
+    fop->refs++;
+
+    UNLOCK(&fop->lock);
+}
+
+static void
+ec_handle_last_pending_fop_completion(ec_fop_data_t *fop, gf_boolean_t *notify)
+{
+    ec_t *ec = fop->xl->private;
+
+    *notify = _gf_false;
+
+    if (!list_empty(&fop->pending_list)) {
+        LOCK(&ec->lock);
+        {
+            list_del_init(&fop->pending_list);
+            *notify = __ec_is_last_fop(ec);
+        }
+        UNLOCK(&ec->lock);
+    }
+}
+
+void
+ec_fop_cleanup(ec_fop_data_t *fop)
+{
+    ec_cbk_data_t *cbk, *tmp;
+
+    list_for_each_entry_safe(cbk, tmp, &fop->answer_list, answer_list)
+    {
+        list_del_init(&cbk->answer_list);
+
+        ec_cbk_data_destroy(cbk);
+    }
+    INIT_LIST_HEAD(&fop->cbk_list);
+
+    fop->answer = NULL;
+}
+
+void
+ec_fop_data_release(ec_fop_data_t *fop)
+{
+    ec_t *ec = NULL;
+    int32_t refs;
+    gf_boolean_t notify = _gf_false;
+
+    LOCK(&fop->lock);
+
+    ec_trace("RELEASE", fop, "");
+
+    GF_ASSERT(fop->refs > 0);
+    refs = --fop->refs;
+
+    UNLOCK(&fop->lock);
+
+    if (refs == 0) {
+        fop->frame->local = NULL;
+        STACK_DESTROY(fop->frame->root);
+
+        LOCK_DESTROY(&fop->lock);
+
+        if (fop->xdata != NULL) {
+            dict_unref(fop->xdata);
+        }
+        if (fop->dict != NULL) {
+            dict_unref(fop->dict);
+        }
+        if (fop->inode != NULL) {
+            inode_unref(fop->inode);
+        }
+        if (fop->fd != NULL) {
+            fd_unref(fop->fd);
+        }
+        if (fop->buffers != NULL) {
+            iobref_unref(fop->buffers);
+        }
+        GF_FREE(fop->vector);
+        GF_FREE(fop->str[0]);
+        GF_FREE(fop->str[1]);
+        loc_wipe(&fop->loc[0]);
+        loc_wipe(&fop->loc[1]);
+        GF_FREE(fop->errstr);
+
+        ec_resume_parent(fop);
+
+        ec_fop_cleanup(fop);
+
+        ec = fop->xl->private;
+        ec_handle_last_pending_fop_completion(fop, &notify);
+        ec_handle_healers_done(fop);
+        mem_put(fop);
+        if (notify) {
+            ec_pending_fops_completed(ec);
+        }
+    }
+}
diff --git a/xlators/cluster/ec/src/ec-data.h b/xlators/cluster/ec/src/ec-data.h
new file mode 100644
index 00000000000..c8a74ffe1ed
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-data.h
@@ -0,0 +1,35 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_DATA_H__
+#define __EC_DATA_H__
+
+#include "ec-types.h"
+
+ec_cbk_data_t *
+ec_cbk_data_allocate(call_frame_t *frame, xlator_t *this, ec_fop_data_t *fop,
+                     int32_t id, int32_t idx, int32_t op_ret, int32_t op_errno);
+ec_fop_data_t *
+ec_fop_data_allocate(call_frame_t *frame, xlator_t *this, int32_t id,
+                     uint32_t flags, uintptr_t target, uint32_t fop_flags,
+                     ec_wind_f wind, ec_handler_f handler, ec_cbk_t cbks,
+                     void *data);
+void
+ec_fop_data_acquire(ec_fop_data_t *fop);
+void
+ec_fop_data_release(ec_fop_data_t *fop);
+
+void
+ec_fop_cleanup(ec_fop_data_t *fop);
+
+void
+ec_pending_fops_completed(ec_t *ec);
+
+#endif /* __EC_DATA_H__ */
diff --git a/xlators/cluster/ec/src/ec-dir-read.c b/xlators/cluster/ec/src/ec-dir-read.c
new file mode 100644
index 00000000000..f71dcfac293
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-dir-read.c
@@ -0,0 +1,647 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "ec.h"
+#include "ec-messages.h"
+#include "ec-helpers.h"
+#include "ec-common.h"
+#include "ec-combine.h"
+#include "ec-fops.h"
+
+/****************************************************************
+ *
+ * File Operation: opendir
+ *
+ ***************************************************************/
+
+int32_t
+ec_combine_opendir(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    if (dst->fd != src->fd) {
+        gf_msg(fop->xl->name, GF_LOG_NOTICE, 0, EC_MSG_FD_MISMATCH,
+               "Mismatching fd in answers "
+               "of 'GF_FOP_OPENDIR': %p <-> %p",
+               dst->fd, src->fd);
+
+        return 0;
+    }
+
+    return 1;
+}
+
+int32_t
+ec_opendir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_OPENDIR, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            if (fd != NULL) {
+                cbk->fd = fd_ref(fd);
+                if (cbk->fd == NULL) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0,
+                           EC_MSG_FILE_DESC_REF_FAIL,
+                           "Failed to reference a "
+                           "file descriptor.");
+
+                    goto out;
+                }
+            }
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, ec_combine_opendir);
+
+        ec_update_fd_status(fd, this, idx, op_ret);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_opendir(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_opendir_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->opendir,
+                      &fop->loc[0], fop->fd, fop->xdata);
+}
+
+int32_t
+ec_manager_opendir(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+    ec_fd_t *ctx;
+    int32_t err;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            LOCK(&fop->fd->lock);
+
+            ctx = __ec_fd_get(fop->fd, fop->xl);
+            if (ctx == NULL) {
+                UNLOCK(&fop->fd->lock);
+
+                fop->error = ENOMEM;
+
+                return EC_STATE_REPORT;
+            }
+            if (!ctx->loc.inode) {
+                err = ec_loc_from_loc(fop->xl, &ctx->loc, &fop->loc[0]);
+                if (err != 0) {
+                    UNLOCK(&fop->fd->lock);
+
+                    fop->error = -err;
+
+                    return EC_STATE_REPORT;
+                }
+            }
+
+            UNLOCK(&fop->fd->lock);
+
+            /* Fall through */
+
+        case EC_STATE_LOCK:
+            ec_lock_prepare_inode(fop, &fop->loc[0], EC_QUERY_INFO, 0,
+                                  EC_RANGE_FULL);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_true);
+            if (cbk != NULL) {
+                /* Save which subvolumes successfully opened the directory.
+                 * If ctx->open is 0, it means that readdir cannot be
+                 * processed in this directory.
+                 */
+                LOCK(&fop->fd->lock);
+
+                ctx = __ec_fd_get(fop->fd, fop->xl);
+                if (ctx != NULL) {
+                    ctx->open |= cbk->mask;
+                }
+
+                UNLOCK(&fop->fd->lock);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.opendir != NULL) {
+                fop->cbks.opendir(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                  cbk->op_errno, cbk->fd, cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.opendir != NULL) {
+                fop->cbks.opendir(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                  NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_opendir(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_opendir_cbk_t func, void *data, loc_t *loc,
+           fd_t *fd, dict_t *xdata)
+{
+    ec_cbk_t callback = {.opendir = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(OPENDIR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_OPENDIR, EC_FLAG_LOCK_SHARED,
+                               target, fop_flags, ec_wind_opendir,
+                               ec_manager_opendir, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
+
+/* Returns -1 if client_id is invalid else index of child subvol in xl_list */
+int
+ec_deitransform(xlator_t *this, off_t offset)
+{
+    int idx = -1;
+    int client_id = -1;
+    ec_t *ec = this->private;
+    char id[32] = {0};
+    int err;
+
+    client_id = gf_deitransform(this, offset);
+    sprintf(id, "%d", client_id);
+    err = dict_get_int32(ec->leaf_to_subvolid, id, &idx);
+    if (err < 0) {
+        idx = err;
+        goto out;
+    }
+
+out:
+    if (idx < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, EC_MSG_INVALID_REQUEST,
+               "Invalid index %d in readdirp request", client_id);
+        idx = -EINVAL;
+    }
+    return idx;
+}
+
+/* FOP: readdir */
+
+void
+ec_adjust_readdirp(ec_t *ec, int32_t idx, gf_dirent_t *entries)
+{
+    gf_dirent_t *entry;
+
+    list_for_each_entry(entry, &entries->list, list)
+    {
+        if (!entry->inode)
+            continue;
+
+        if (entry->d_stat.ia_type == IA_IFREG) {
+            if ((entry->dict == NULL) ||
+                (ec_dict_del_number(entry->dict, EC_XATTR_SIZE,
+                                    &entry->d_stat.ia_size) != 0)) {
+                inode_unref(entry->inode);
+                entry->inode = NULL;
+            } else {
+                ec_iatt_rebuild(ec, &entry->d_stat, 1, 1);
+            }
+        }
+    }
+}
+
+int32_t
+ec_common_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, gf_dirent_t *entries,
+                      dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, fop->id, idx, op_ret,
+                               op_errno);
+    if (cbk) {
+        if (xdata)
+            cbk->xdata = dict_ref(xdata);
+        if (cbk->op_ret >= 0)
+            list_splice_init(&entries->list, &cbk->entries.list);
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_readdir(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_common_readdir_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->readdir,
+                      fop->fd, fop->size, fop->offset, fop->xdata);
+}
+
+int32_t
+ec_manager_readdir(ec_fop_data_t *fop, int32_t state)
+{
+    ec_fd_t *ctx = NULL;
+    ec_cbk_data_t *cbk = NULL;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            /* Return error if opendir has not been successfully called on
+             * any subvolume. */
+            ctx = ec_fd_get(fop->fd, fop->xl);
+            if (ctx == NULL) {
+                fop->error = ENOMEM;
+            } else if (ctx->open == 0) {
+                fop->error = EBADFD;
+            }
+
+            if (fop->error) {
+                gf_msg(fop->xl->name, GF_LOG_ERROR, fop->error,
+                       EC_MSG_INVALID_REQUEST, "EC is not winding readdir: %s",
+                       ec_msg_str(fop));
+                return EC_STATE_REPORT;
+            }
+
+            if (fop->id == GF_FOP_READDIRP) {
+                int32_t err;
+
+                if (fop->xdata == NULL) {
+                    fop->xdata = dict_new();
+                    if (fop->xdata == NULL) {
+                        fop->error = ENOMEM;
+
+                        return EC_STATE_REPORT;
+                    }
+                }
+
+                err = dict_set_uint64(fop->xdata, EC_XATTR_SIZE, 0);
+                if (err != 0) {
+                    fop->error = -err;
+
+                    return EC_STATE_REPORT;
+                }
+            }
+
+            if (fop->offset != 0) {
+                /* Non-zero offset is irrecoverable error as the offset may not
+                 * be valid on other bricks*/
+                int32_t idx = -1;
+
+                idx = ec_deitransform(fop->xl, fop->offset);
+
+                if (idx < 0) {
+                    fop->error = -idx;
+                    return EC_STATE_REPORT;
+                }
+                fop->mask &= 1ULL << idx;
+            } else {
+                ec_lock_prepare_fd(fop, fop->fd, EC_QUERY_INFO, 0,
+                                   EC_RANGE_FULL);
+                ec_lock(fop);
+            }
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_one(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            if (ec_dispatch_one_retry(fop, &cbk)) {
+                return EC_STATE_DISPATCH;
+            }
+
+            if ((cbk != NULL) && (cbk->op_ret > 0) &&
+                (fop->id == GF_FOP_READDIRP)) {
+                ec_adjust_readdirp(fop->xl->private, cbk->idx, &cbk->entries);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+            GF_ASSERT(cbk);
+            if (fop->id == GF_FOP_READDIR) {
+                if (fop->cbks.readdir != NULL) {
+                    fop->cbks.readdir(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                      cbk->op_errno, &cbk->entries, cbk->xdata);
+                }
+            } else {
+                if (fop->cbks.readdirp != NULL) {
+                    fop->cbks.readdirp(fop->req_frame, fop, fop->xl,
+                                       cbk->op_ret, cbk->op_errno,
+                                       &cbk->entries, cbk->xdata);
+                }
+            }
+            if (fop->offset == 0)
+                return EC_STATE_LOCK_REUSE;
+            else
+                return EC_STATE_END;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            if (fop->id == GF_FOP_READDIR) {
+                if (fop->cbks.readdir != NULL) {
+                    fop->cbks.readdir(fop->req_frame, fop, fop->xl, -1,
+                                      fop->error, NULL, NULL);
+                }
+            } else {
+                if (fop->cbks.readdirp != NULL) {
+                    fop->cbks.readdirp(fop->req_frame, fop, fop->xl, -1,
+                                       fop->error, NULL, NULL);
+                }
+            }
+            if (fop->offset == 0)
+                return EC_STATE_LOCK_REUSE;
+            else
+                return EC_STATE_END;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            GF_ASSERT(fop->offset == 0);
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            GF_ASSERT(fop->offset == 0);
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_readdir(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_readdir_cbk_t func, void *data, fd_t *fd,
+           size_t size, off_t offset, dict_t *xdata)
+{
+    ec_cbk_t callback = {.readdir = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(READDIR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_READDIR, EC_FLAG_LOCK_SHARED,
+                               target, fop_flags, ec_wind_readdir,
+                               ec_manager_readdir, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->size = size;
+    fop->offset = offset;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
+
+/* FOP: readdirp */
+
+void
+ec_wind_readdirp(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_common_readdir_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->readdirp,
+                      fop->fd, fop->size, fop->offset, fop->xdata);
+}
+
+void
+ec_readdirp(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_readdirp_cbk_t func, void *data, fd_t *fd,
+            size_t size, off_t offset, dict_t *xdata)
+{
+    ec_cbk_t callback = {.readdirp = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(READDIRP) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(
+        frame, this, GF_FOP_READDIRP, EC_FLAG_LOCK_SHARED, target, fop_flags,
+        ec_wind_readdirp, ec_manager_readdir, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->size = size;
+    fop->offset = offset;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
diff --git a/xlators/cluster/ec/src/ec-dir-write.c b/xlators/cluster/ec/src/ec-dir-write.c
new file mode 100644
index 00000000000..53d27d895c3
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-dir-write.c
@@ -0,0 +1,1487 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "ec.h"
+#include "ec-messages.h"
+#include "ec-helpers.h"
+#include "ec-common.h"
+#include "ec-combine.h"
+#include "ec-method.h"
+#include "ec-fops.h"
+
+int
+ec_dir_write_cbk(call_frame_t *frame, xlator_t *this, void *cookie, int op_ret,
+                 int op_errno, struct iatt *poststat, struct iatt *preparent,
+                 struct iatt *postparent, struct iatt *preparent2,
+                 struct iatt *postparent2, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int i = 0;
+    int idx = 0;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+    idx = (long)cookie;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, fop->id, idx, op_ret,
+                               op_errno);
+    if (!cbk)
+        goto out;
+
+    if (xdata)
+        cbk->xdata = dict_ref(xdata);
+
+    if (op_ret < 0)
+        goto out;
+
+    if (poststat)
+        cbk->iatt[i++] = *poststat;
+
+    if (preparent)
+        cbk->iatt[i++] = *preparent;
+
+    if (postparent)
+        cbk->iatt[i++] = *postparent;
+
+    if (preparent2)
+        cbk->iatt[i++] = *preparent2;
+
+    if (postparent2)
+        cbk->iatt[i++] = *postparent2;
+
+out:
+    if (cbk)
+        ec_combine(cbk, ec_combine_write);
+
+    if (fop)
+        ec_complete(fop);
+    return 0;
+}
+
+/* FOP: create */
+
+int32_t
+ec_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, fd_t *fd, inode_t *inode, struct iatt *buf,
+              struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    return ec_dir_write_cbk(frame, this, cookie, op_ret, op_errno, buf,
+                            preparent, postparent, NULL, NULL, xdata);
+}
+
+void
+ec_wind_create(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_create_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->create,
+                      &fop->loc[0], fop->int32, fop->mode[0], fop->mode[1],
+                      fop->fd, fop->xdata);
+}
+
+int32_t
+ec_manager_create(ec_fop_data_t *fop, int32_t state)
+{
+    ec_config_t config;
+    ec_t *ec;
+    ec_cbk_data_t *cbk;
+    ec_fd_t *ctx;
+    uint64_t version[2] = {0, 0};
+    int32_t err;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            LOCK(&fop->fd->lock);
+
+            ctx = __ec_fd_get(fop->fd, fop->xl);
+            if (ctx == NULL) {
+                UNLOCK(&fop->fd->lock);
+
+                fop->error = ENOMEM;
+
+                return EC_STATE_REPORT;
+            }
+            err = ec_loc_from_loc(fop->xl, &ctx->loc, &fop->loc[0]);
+            if (err != 0) {
+                UNLOCK(&fop->fd->lock);
+
+                fop->error = -err;
+
+                return EC_STATE_REPORT;
+            }
+
+            ctx->flags = fop->int32;
+
+            UNLOCK(&fop->fd->lock);
+
+            if (fop->xdata == NULL) {
+                fop->xdata = dict_new();
+                if (fop->xdata == NULL) {
+                    fop->error = ENOMEM;
+
+                    return EC_STATE_REPORT;
+                }
+            }
+
+            ec = fop->xl->private;
+
+            config.version = EC_CONFIG_VERSION;
+            config.algorithm = EC_CONFIG_ALGORITHM;
+            config.gf_word_size = EC_GF_BITS;
+            config.bricks = ec->nodes;
+            config.redundancy = ec->redundancy;
+            config.chunk_size = EC_METHOD_CHUNK_SIZE;
+
+            err = ec_dict_set_config(fop->xdata, EC_XATTR_CONFIG, &config);
+            if (err != 0) {
+                fop->error = -err;
+
+                return EC_STATE_REPORT;
+            }
+            err = ec_dict_set_array(fop->xdata, EC_XATTR_VERSION, version,
+                                    EC_VERSION_SIZE);
+            if (err != 0) {
+                fop->error = -err;
+
+                return EC_STATE_REPORT;
+            }
+            err = ec_dict_set_number(fop->xdata, EC_XATTR_SIZE, 0);
+            if (err != 0) {
+                fop->error = -err;
+
+                return EC_STATE_REPORT;
+            }
+
+            /* We need to write to specific offsets on the bricks, so we
+             * need to remove O_APPEND from flags (if present) */
+            fop->int32 &= ~O_APPEND;
+
+            /* Fall through */
+
+        case EC_STATE_LOCK:
+            ec_lock_prepare_parent_inode(fop, &fop->loc[0], NULL,
+                                         EC_UPDATE_DATA | EC_UPDATE_META);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                int32_t err;
+
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 3, cbk->count);
+
+                err = ec_loc_update(fop->xl, &fop->loc[0], cbk->inode,
+                                    &cbk->iatt[0]);
+                if (!ec_cbk_set_error(cbk, -err, _gf_false)) {
+                    LOCK(&fop->fd->lock);
+
+                    ctx = __ec_fd_get(fop->fd, fop->xl);
+                    if (ctx != NULL) {
+                        ctx->open |= cbk->mask;
+                    }
+
+                    UNLOCK(&fop->fd->lock);
+                }
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.create != NULL) {
+                QUORUM_CBK(fop->cbks.create, fop, fop->req_frame, fop, fop->xl,
+                           cbk->op_ret, cbk->op_errno, fop->fd,
+                           fop->loc[0].inode, &cbk->iatt[0], &cbk->iatt[1],
+                           &cbk->iatt[2], cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.create != NULL) {
+                fop->cbks.create(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                 NULL, NULL, NULL, NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_create(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_create_cbk_t func, void *data, loc_t *loc,
+          int32_t flags, mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
+{
+    ec_cbk_t callback = {.create = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(CREATE) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_CREATE, 0, target, fop_flags,
+                               ec_wind_create, ec_manager_create, callback,
+                               data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->int32 = flags;
+    fop->mode[0] = mode;
+    fop->mode[1] = umask;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: link */
+
+int32_t
+ec_link_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, inode_t *inode, struct iatt *buf,
+            struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    return ec_dir_write_cbk(frame, this, cookie, op_ret, op_errno, buf,
+                            preparent, postparent, NULL, NULL, xdata);
+}
+
+void
+ec_wind_link(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_link_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->link,
+                      &fop->loc[0], &fop->loc[1], fop->xdata);
+}
+
+int32_t
+ec_manager_link(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            ec_lock_prepare_parent_inode(
+                fop, &fop->loc[1], &fop->loc[0],
+                EC_UPDATE_DATA | EC_UPDATE_META | EC_INODE_SIZE);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                int32_t err;
+
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 3, cbk->count);
+
+                if (cbk->iatt[0].ia_type == IA_IFREG) {
+                    cbk->iatt[0].ia_size = fop->locks[0].size;
+                }
+
+                err = ec_loc_update(fop->xl, &fop->loc[0], cbk->inode,
+                                    &cbk->iatt[0]);
+                ec_cbk_set_error(cbk, -err, _gf_false);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.link != NULL) {
+                QUORUM_CBK(fop->cbks.link, fop, fop->req_frame, fop, fop->xl,
+                           cbk->op_ret, cbk->op_errno, fop->loc[0].inode,
+                           &cbk->iatt[0], &cbk->iatt[1], &cbk->iatt[2],
+                           cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.link != NULL) {
+                fop->cbks.link(fop->req_frame, fop, fop->xl, -1, fop->error,
+                               NULL, NULL, NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_link(call_frame_t *frame, xlator_t *this, uintptr_t target,
+        uint32_t fop_flags, fop_link_cbk_t func, void *data, loc_t *oldloc,
+        loc_t *newloc, dict_t *xdata)
+{
+    ec_cbk_t callback = {.link = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(LINK) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_LINK, 0, target, fop_flags,
+                               ec_wind_link, ec_manager_link, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    if (oldloc != NULL) {
+        if (loc_copy(&fop->loc[0], oldloc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (newloc != NULL) {
+        if (loc_copy(&fop->loc[1], newloc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: mkdir */
+
+int32_t
+ec_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, inode_t *inode, struct iatt *buf,
+             struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    return ec_dir_write_cbk(frame, this, cookie, op_ret, op_errno, buf,
+                            preparent, postparent, NULL, NULL, xdata);
+}
+
+void
+ec_wind_mkdir(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_mkdir_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->mkdir,
+                      &fop->loc[0], fop->mode[0], fop->mode[1], fop->xdata);
+}
+
+int32_t
+ec_manager_mkdir(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+    uint64_t version[2] = {0, 0};
+    int32_t err;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            if (fop->xdata == NULL) {
+                fop->xdata = dict_new();
+                if (fop->xdata == NULL) {
+                    fop->error = ENOMEM;
+
+                    return EC_STATE_REPORT;
+                }
+            }
+
+            err = ec_dict_set_array(fop->xdata, EC_XATTR_VERSION, version,
+                                    EC_VERSION_SIZE);
+            if (err != 0) {
+                fop->error = -err;
+                return EC_STATE_REPORT;
+            }
+
+            /* Fall through */
+
+        case EC_STATE_LOCK:
+            ec_lock_prepare_parent_inode(fop, &fop->loc[0], NULL,
+                                         EC_UPDATE_DATA | EC_UPDATE_META);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                int32_t err;
+
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 3, cbk->count);
+
+                err = ec_loc_update(fop->xl, &fop->loc[0], cbk->inode,
+                                    &cbk->iatt[0]);
+                ec_cbk_set_error(cbk, -err, _gf_false);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.mkdir != NULL) {
+                QUORUM_CBK(fop->cbks.mkdir, fop, fop->req_frame, fop, fop->xl,
+                           cbk->op_ret, cbk->op_errno, fop->loc[0].inode,
+                           &cbk->iatt[0], &cbk->iatt[1], &cbk->iatt[2],
+                           cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            cbk = fop->answer;
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.mkdir != NULL) {
+                fop->cbks.mkdir(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                NULL, NULL, NULL, NULL,
+                                ((cbk) ? cbk->xdata : NULL));
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_mkdir(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_mkdir_cbk_t func, void *data, loc_t *loc,
+         mode_t mode, mode_t umask, dict_t *xdata)
+{
+    ec_cbk_t callback = {.mkdir = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(MKDIR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_MKDIR, 0, target, fop_flags,
+                               ec_wind_mkdir, ec_manager_mkdir, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->mode[0] = mode;
+    fop->mode[1] = umask;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: mknod */
+
+int32_t
+ec_mknod_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, inode_t *inode, struct iatt *buf,
+             struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    return ec_dir_write_cbk(frame, this, cookie, op_ret, op_errno, buf,
+                            preparent, postparent, NULL, NULL, xdata);
+}
+
+void
+ec_wind_mknod(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_mknod_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->mknod,
+                      &fop->loc[0], fop->mode[0], fop->dev, fop->mode[1],
+                      fop->xdata);
+}
+
+int32_t
+ec_manager_mknod(ec_fop_data_t *fop, int32_t state)
+{
+    ec_config_t config;
+    ec_t *ec;
+    ec_cbk_data_t *cbk;
+    uint64_t version[2] = {0, 0};
+
+    switch (state) {
+        case EC_STATE_INIT:
+            if (S_ISREG(fop->mode[0])) {
+                int32_t err;
+
+                if (fop->xdata == NULL) {
+                    fop->xdata = dict_new();
+                    if (fop->xdata == NULL) {
+                        fop->error = ENOMEM;
+
+                        return EC_STATE_REPORT;
+                    }
+                }
+
+                ec = fop->xl->private;
+
+                config.version = EC_CONFIG_VERSION;
+                config.algorithm = EC_CONFIG_ALGORITHM;
+                config.gf_word_size = EC_GF_BITS;
+                config.bricks = ec->nodes;
+                config.redundancy = ec->redundancy;
+                config.chunk_size = EC_METHOD_CHUNK_SIZE;
+
+                err = ec_dict_set_config(fop->xdata, EC_XATTR_CONFIG, &config);
+                if (err != 0) {
+                    fop->error = -err;
+
+                    return EC_STATE_REPORT;
+                }
+                err = ec_dict_set_array(fop->xdata, EC_XATTR_VERSION, version,
+                                        EC_VERSION_SIZE);
+                if (err != 0) {
+                    fop->error = -err;
+
+                    return EC_STATE_REPORT;
+                }
+                err = ec_dict_set_number(fop->xdata, EC_XATTR_SIZE, 0);
+                if (err != 0) {
+                    fop->error = -err;
+
+                    return EC_STATE_REPORT;
+                }
+            }
+
+            /* Fall through */
+
+        case EC_STATE_LOCK:
+            ec_lock_prepare_parent_inode(fop, &fop->loc[0], NULL,
+                                         EC_UPDATE_DATA | EC_UPDATE_META);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                int32_t err;
+
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 3, cbk->count);
+
+                err = ec_loc_update(fop->xl, &fop->loc[0], cbk->inode,
+                                    &cbk->iatt[0]);
+                ec_cbk_set_error(cbk, -err, _gf_false);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.mknod != NULL) {
+                QUORUM_CBK(fop->cbks.mknod, fop, fop->req_frame, fop, fop->xl,
+                           cbk->op_ret, cbk->op_errno, fop->loc[0].inode,
+                           &cbk->iatt[0], &cbk->iatt[1], &cbk->iatt[2],
+                           cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.mknod != NULL) {
+                fop->cbks.mknod(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                NULL, NULL, NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_mknod(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_mknod_cbk_t func, void *data, loc_t *loc,
+         mode_t mode, dev_t rdev, mode_t umask, dict_t *xdata)
+{
+    ec_cbk_t callback = {.mknod = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(MKNOD) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_MKNOD, 0, target, fop_flags,
+                               ec_wind_mknod, ec_manager_mknod, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->mode[0] = mode;
+    fop->dev = rdev;
+    fop->mode[1] = umask;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: rename */
+
+int32_t
+ec_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct iatt *buf, struct iatt *preoldparent,
+              struct iatt *postoldparent, struct iatt *prenewparent,
+              struct iatt *postnewparent, dict_t *xdata)
+{
+    return ec_dir_write_cbk(frame, this, cookie, op_ret, op_errno, buf,
+                            preoldparent, postoldparent, prenewparent,
+                            postnewparent, xdata);
+}
+
+void
+ec_wind_rename(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_rename_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->rename,
+                      &fop->loc[0], &fop->loc[1], fop->xdata);
+}
+
+int32_t
+ec_manager_rename(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            ec_lock_prepare_parent_inode(
+                fop, &fop->loc[0], &fop->loc[0],
+                EC_UPDATE_DATA | EC_UPDATE_META | EC_INODE_SIZE);
+            ec_lock_prepare_parent_inode(fop, &fop->loc[1], NULL,
+                                         EC_UPDATE_DATA | EC_UPDATE_META);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 5, cbk->count);
+
+                if (cbk->iatt[0].ia_type == IA_IFREG) {
+                    cbk->iatt[0].ia_size = fop->locks[0].size;
+                }
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.rename != NULL) {
+                QUORUM_CBK(fop->cbks.rename, fop, fop->req_frame, fop, fop->xl,
+                           cbk->op_ret, cbk->op_errno, &cbk->iatt[0],
+                           &cbk->iatt[1], &cbk->iatt[2], &cbk->iatt[3],
+                           &cbk->iatt[4], cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.rename != NULL) {
+                fop->cbks.rename(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                 NULL, NULL, NULL, NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_rename(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_rename_cbk_t func, void *data, loc_t *oldloc,
+          loc_t *newloc, dict_t *xdata)
+{
+    ec_cbk_t callback = {.rename = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(RENAME) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_RENAME, 0, target, fop_flags,
+                               ec_wind_rename, ec_manager_rename, callback,
+                               data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    if (oldloc != NULL) {
+        if (loc_copy(&fop->loc[0], oldloc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (newloc != NULL) {
+        if (loc_copy(&fop->loc[1], newloc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: rmdir */
+
+int32_t
+ec_rmdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, struct iatt *preparent, struct iatt *postparent,
+             dict_t *xdata)
+{
+    return ec_dir_write_cbk(frame, this, cookie, op_ret, op_errno, NULL,
+                            preparent, postparent, NULL, NULL, xdata);
+}
+
+void
+ec_wind_rmdir(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_rmdir_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->rmdir,
+                      &fop->loc[0], fop->int32, fop->xdata);
+}
+
+int32_t
+ec_manager_rmdir(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            ec_lock_prepare_parent_inode(fop, &fop->loc[0], NULL,
+                                         EC_UPDATE_DATA | EC_UPDATE_META);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            ec_fop_prepare_answer(fop, _gf_false);
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.rmdir != NULL) {
+                QUORUM_CBK(fop->cbks.rmdir, fop, fop->req_frame, fop, fop->xl,
+                           cbk->op_ret, cbk->op_errno, &cbk->iatt[0],
+                           &cbk->iatt[1], cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.rmdir != NULL) {
+                fop->cbks.rmdir(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_rmdir(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_rmdir_cbk_t func, void *data, loc_t *loc,
+         int xflags, dict_t *xdata)
+{
+    ec_cbk_t callback = {.rmdir = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(RMDIR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_RMDIR, 0, target, fop_flags,
+                               ec_wind_rmdir, ec_manager_rmdir, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->int32 = xflags;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: symlink */
+
+int32_t
+ec_symlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, inode_t *inode,
+               struct iatt *buf, struct iatt *preparent,
+               struct iatt *postparent, dict_t *xdata)
+{
+    return ec_dir_write_cbk(frame, this, cookie, op_ret, op_errno, buf,
+                            preparent, postparent, NULL, NULL, xdata);
+}
+
+void
+ec_wind_symlink(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_symlink_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->symlink,
+                      fop->str[0], &fop->loc[0], fop->mode[0], fop->xdata);
+}
+
+int32_t
+ec_manager_symlink(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            ec_lock_prepare_parent_inode(fop, &fop->loc[0], NULL,
+                                         EC_UPDATE_DATA | EC_UPDATE_META);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                int32_t err;
+
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 3, cbk->count);
+
+                err = ec_loc_update(fop->xl, &fop->loc[0], cbk->inode,
+                                    &cbk->iatt[0]);
+                ec_cbk_set_error(cbk, -err, _gf_false);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.symlink != NULL) {
+                QUORUM_CBK(fop->cbks.symlink, fop, fop->req_frame, fop, fop->xl,
+                           cbk->op_ret, cbk->op_errno, fop->loc[0].inode,
+                           &cbk->iatt[0], &cbk->iatt[1], &cbk->iatt[2],
+                           cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.symlink != NULL) {
+                fop->cbks.symlink(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                  NULL, NULL, NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_symlink(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_symlink_cbk_t func, void *data,
+           const char *linkname, loc_t *loc, mode_t umask, dict_t *xdata)
+{
+    ec_cbk_t callback = {.symlink = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(SYMLINK) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_SYMLINK, 0, target,
+                               fop_flags, ec_wind_symlink, ec_manager_symlink,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->mode[0] = umask;
+
+    if (linkname != NULL) {
+        fop->str[0] = gf_strdup(linkname);
+        if (fop->str[0] == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a string.");
+
+            goto out;
+        }
+    }
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: unlink */
+
+int32_t
+ec_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct iatt *preparent, struct iatt *postparent,
+              dict_t *xdata)
+{
+    return ec_dir_write_cbk(frame, this, cookie, op_ret, op_errno, NULL,
+                            preparent, postparent, NULL, NULL, xdata);
+}
+
+void
+ec_wind_unlink(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_unlink_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->unlink,
+                      &fop->loc[0], fop->int32, fop->xdata);
+}
+
+int32_t
+ec_manager_unlink(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            ec_lock_prepare_parent_inode(fop, &fop->loc[0], NULL,
+                                         EC_UPDATE_DATA | EC_UPDATE_META);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            ec_fop_prepare_answer(fop, _gf_false);
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.unlink != NULL) {
+                QUORUM_CBK(fop->cbks.unlink, fop, fop->req_frame, fop, fop->xl,
+                           cbk->op_ret, cbk->op_errno, &cbk->iatt[0],
+                           &cbk->iatt[1], cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.unlink != NULL) {
+                fop->cbks.unlink(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                 NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_unlink(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_unlink_cbk_t func, void *data, loc_t *loc,
+          int xflags, dict_t *xdata)
+{
+    ec_cbk_t callback = {.unlink = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(UNLINK) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_UNLINK, 0, target, fop_flags,
+                               ec_wind_unlink, ec_manager_unlink, callback,
+                               data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->int32 = xflags;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL);
+    }
+}
diff --git a/xlators/cluster/ec/src/ec-fops.h b/xlators/cluster/ec/src/ec-fops.h
new file mode 100644
index 00000000000..07edf8a7fec
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-fops.h
@@ -0,0 +1,254 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_FOPS_H__
+#define __EC_FOPS_H__
+
+#include <glusterfs/xlator.h>
+
+#include "ec-types.h"
+#include "ec-common.h"
+
+void
+ec_access(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_access_cbk_t func, void *data, loc_t *loc,
+          int32_t mask, dict_t *xdata);
+
+void
+ec_create(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_create_cbk_t func, void *data, loc_t *loc,
+          int32_t flags, mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata);
+
+void
+ec_entrylk(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_entrylk_cbk_t func, void *data,
+           const char *volume, loc_t *loc, const char *basename,
+           entrylk_cmd cmd, entrylk_type type, dict_t *xdata);
+
+void
+ec_fentrylk(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_fentrylk_cbk_t func, void *data,
+            const char *volume, fd_t *fd, const char *basename, entrylk_cmd cmd,
+            entrylk_type type, dict_t *xdata);
+
+void
+ec_flush(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_flush_cbk_t func, void *data, fd_t *fd,
+         dict_t *xdata);
+
+void
+ec_fsync(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_fsync_cbk_t func, void *data, fd_t *fd,
+         int32_t datasync, dict_t *xdata);
+
+void
+ec_fsyncdir(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_fsyncdir_cbk_t func, void *data, fd_t *fd,
+            int32_t datasync, dict_t *xdata);
+
+void
+ec_getxattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_getxattr_cbk_t func, void *data, loc_t *loc,
+            const char *name, dict_t *xdata);
+
+void
+ec_fgetxattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+             uint32_t fop_flags, fop_fgetxattr_cbk_t func, void *data, fd_t *fd,
+             const char *name, dict_t *xdata);
+
+void
+ec_heal(call_frame_t *frame, xlator_t *this, uintptr_t target,
+        uint32_t fop_flags, fop_heal_cbk_t func, void *data, loc_t *loc,
+        int32_t partial, dict_t *xdata);
+
+void
+ec_fheal(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_fheal_cbk_t func, void *data, fd_t *fd,
+         int32_t partial, dict_t *xdata);
+
+void
+ec_inodelk(call_frame_t *frame, xlator_t *this, gf_lkowner_t *owner,
+           uintptr_t target, uint32_t fop_flags, fop_inodelk_cbk_t func,
+           void *data, const char *volume, loc_t *loc, int32_t cmd,
+           struct gf_flock *flock, dict_t *xdata);
+
+void
+ec_finodelk(call_frame_t *frame, xlator_t *this, gf_lkowner_t *owner,
+            uintptr_t target, uint32_t fop_flags, fop_finodelk_cbk_t func,
+            void *data, const char *volume, fd_t *fd, int32_t cmd,
+            struct gf_flock *flock, dict_t *xdata);
+
+void
+ec_link(call_frame_t *frame, xlator_t *this, uintptr_t target,
+        uint32_t fop_flags, fop_link_cbk_t func, void *data, loc_t *oldloc,
+        loc_t *newloc, dict_t *xdata);
+
+void
+ec_lk(call_frame_t *frame, xlator_t *this, uintptr_t target, uint32_t fop_flags,
+      fop_lk_cbk_t func, void *data, fd_t *fd, int32_t cmd,
+      struct gf_flock *flock, dict_t *xdata);
+
+void
+ec_lookup(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_lookup_cbk_t func, void *data, loc_t *loc,
+          dict_t *xdata);
+
+void
+ec_mkdir(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_mkdir_cbk_t func, void *data, loc_t *loc,
+         mode_t mode, mode_t umask, dict_t *xdata);
+
+void
+ec_mknod(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_mknod_cbk_t func, void *data, loc_t *loc,
+         mode_t mode, dev_t rdev, mode_t umask, dict_t *xdata);
+
+void
+ec_open(call_frame_t *frame, xlator_t *this, uintptr_t target,
+        uint32_t fop_flags, fop_open_cbk_t func, void *data, loc_t *loc,
+        int32_t flags, fd_t *fd, dict_t *xdata);
+
+void
+ec_opendir(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_opendir_cbk_t func, void *data, loc_t *loc,
+           fd_t *fd, dict_t *xdata);
+
+void
+ec_readdir(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_readdir_cbk_t func, void *data, fd_t *fd,
+           size_t size, off_t offset, dict_t *xdata);
+
+void
+ec_readdirp(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_readdirp_cbk_t func, void *data, fd_t *fd,
+            size_t size, off_t offset, dict_t *xdata);
+
+void
+ec_readlink(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_readlink_cbk_t func, void *data, loc_t *loc,
+            size_t size, dict_t *xdata);
+
+void
+ec_readv(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_readv_cbk_t func, void *data, fd_t *fd,
+         size_t size, off_t offset, uint32_t flags, dict_t *xdata);
+
+void
+ec_removexattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+               uint32_t fop_flags, fop_removexattr_cbk_t func, void *data,
+               loc_t *loc, const char *name, dict_t *xdata);
+
+void
+ec_fremovexattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+                uint32_t fop_flags, fop_fremovexattr_cbk_t func, void *data,
+                fd_t *fd, const char *name, dict_t *xdata);
+
+void
+ec_rename(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_rename_cbk_t func, void *data, loc_t *oldloc,
+          loc_t *newloc, dict_t *xdata);
+
+void
+ec_rmdir(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_rmdir_cbk_t func, void *data, loc_t *loc,
+         int xflags, dict_t *xdata);
+
+void
+ec_setattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_setattr_cbk_t func, void *data, loc_t *loc,
+           struct iatt *stbuf, int32_t valid, dict_t *xdata);
+
+void
+ec_fsetattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_fsetattr_cbk_t func, void *data, fd_t *fd,
+            struct iatt *stbuf, int32_t valid, dict_t *xdata);
+
+void
+ec_setxattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_setxattr_cbk_t func, void *data, loc_t *loc,
+            dict_t *dict, int32_t flags, dict_t *xdata);
+
+void
+ec_fsetxattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+             uint32_t fop_flags, fop_fsetxattr_cbk_t func, void *data, fd_t *fd,
+             dict_t *dict, int32_t flags, dict_t *xdata);
+
+void
+ec_stat(call_frame_t *frame, xlator_t *this, uintptr_t target,
+        uint32_t fop_flags, fop_stat_cbk_t func, void *data, loc_t *loc,
+        dict_t *xdata);
+
+void
+ec_fstat(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_fstat_cbk_t func, void *data, fd_t *fd,
+         dict_t *xdata);
+
+void
+ec_statfs(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_statfs_cbk_t func, void *data, loc_t *loc,
+          dict_t *xdata);
+
+void
+ec_symlink(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_symlink_cbk_t func, void *data,
+           const char *linkname, loc_t *loc, mode_t umask, dict_t *xdata);
+
+void
+ec_fallocate(call_frame_t *frame, xlator_t *this, uintptr_t target,
+             uint32_t fop_flags, fop_fallocate_cbk_t func, void *data, fd_t *fd,
+             int32_t mode, off_t offset, size_t len, dict_t *xdata);
+
+void
+ec_discard(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_discard_cbk_t func, void *data, fd_t *fd,
+           off_t offset, size_t len, dict_t *xdata);
+
+void
+ec_truncate(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_truncate_cbk_t func, void *data, loc_t *loc,
+            off_t offset, dict_t *xdata);
+
+void
+ec_ftruncate(call_frame_t *frame, xlator_t *this, uintptr_t target,
+             uint32_t fop_flags, fop_ftruncate_cbk_t func, void *data, fd_t *fd,
+             off_t offset, dict_t *xdata);
+
+void
+ec_unlink(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_unlink_cbk_t func, void *data, loc_t *loc,
+          int xflags, dict_t *xdata);
+
+void
+ec_writev(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_writev_cbk_t func, void *data, fd_t *fd,
+          struct iovec *vector, int32_t count, off_t offset, uint32_t flags,
+          struct iobref *iobref, dict_t *xdata);
+
+void
+ec_xattrop(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_xattrop_cbk_t func, void *data, loc_t *loc,
+           gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata);
+
+void
+ec_fxattrop(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_fxattrop_cbk_t func, void *data, fd_t *fd,
+            gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata);
+
+void
+ec_seek(call_frame_t *frame, xlator_t *this, uintptr_t target,
+        uint32_t fop_flags, fop_seek_cbk_t func, void *data, fd_t *fd,
+        off_t offset, gf_seek_what_t what, dict_t *xdata);
+
+void
+ec_ipc(call_frame_t *frame, xlator_t *this, uintptr_t target,
+       uint32_t fop_flags, fop_ipc_cbk_t func, void *data, int32_t op,
+       dict_t *xdata);
+
+#endif /* __EC_FOPS_H__ */
diff --git a/xlators/cluster/ec/src/ec-galois.c b/xlators/cluster/ec/src/ec-galois.c
new file mode 100644
index 00000000000..6e4990c71f5
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-galois.c
@@ -0,0 +1,183 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <string.h>
+
+#include "ec-mem-types.h"
+#include "ec-gf8.h"
+#include "ec-helpers.h"
+
+static ec_gf_t *
+ec_gf_alloc(uint32_t bits, uint32_t mod)
+{
+    ec_gf_t *gf;
+
+    gf = GF_MALLOC(sizeof(ec_gf_t), ec_mt_ec_gf_t);
+    if (gf == NULL) {
+        goto failed;
+    }
+
+    gf->bits = bits;
+    gf->size = 1 << bits;
+    gf->mod = mod;
+
+    gf->log = GF_MALLOC(sizeof(uint32_t) * (gf->size * 2 - 1),
+                        gf_common_mt_int);
+    if (gf->log == NULL) {
+        goto failed_gf;
+    }
+    gf->pow = GF_MALLOC(sizeof(uint32_t) * (gf->size * 2 - 1),
+                        gf_common_mt_int);
+    if (gf->pow == NULL) {
+        goto failed_log;
+    }
+
+    return gf;
+
+failed_log:
+    GF_FREE(gf->log);
+failed_gf:
+    GF_FREE(gf);
+failed:
+    return EC_ERR(ENOMEM);
+}
+
+static void
+ec_gf_init_tables(ec_gf_t *gf)
+{
+    uint32_t i, tmp;
+
+    memset(gf->log, -1, sizeof(uint32_t) * gf->size);
+
+    gf->pow[0] = 1;
+    gf->log[0] = gf->size;
+    gf->log[1] = 0;
+    for (i = 1; i < gf->size; i++) {
+        tmp = gf->pow[i - 1] << 1;
+        if (tmp >= gf->size) {
+            tmp ^= gf->mod;
+        }
+        gf->pow[i + gf->size - 1] = gf->pow[i] = tmp;
+        gf->log[tmp + gf->size - 1] = gf->log[tmp] = i;
+    }
+}
+
+ec_gf_t *
+ec_gf_prepare(uint32_t bits, uint32_t mod)
+{
+    ec_gf_mul_t **tbl;
+    ec_gf_t *gf;
+    uint32_t i, j;
+
+    if (bits != 8) {
+        return EC_ERR(EINVAL);
+    }
+
+    tbl = ec_gf8_mul;
+    if (mod == 0) {
+        mod = 0x11d;
+    }
+
+    gf = ec_gf_alloc(bits, mod);
+    if (EC_IS_ERR(gf)) {
+        return gf;
+    }
+    ec_gf_init_tables(gf);
+
+    gf->table = tbl;
+    gf->min_ops = bits * bits;
+    gf->max_ops = 0;
+    gf->avg_ops = 0;
+    for (i = 1; i < gf->size; i++) {
+        for (j = 0; tbl[i]->ops[j].op != EC_GF_OP_END; j++) {
+        }
+        if (gf->max_ops < j) {
+            gf->max_ops = j;
+        }
+        if (gf->min_ops > j) {
+            gf->min_ops = j;
+        }
+        gf->avg_ops += j;
+    }
+    gf->avg_ops /= gf->size;
+
+    return gf;
+}
+
+void
+ec_gf_destroy(ec_gf_t *gf)
+{
+    GF_FREE(gf->pow);
+    GF_FREE(gf->log);
+    GF_FREE(gf);
+}
+
+uint32_t
+ec_gf_add(ec_gf_t *gf, uint32_t a, uint32_t b)
+{
+    if ((a >= gf->size) || (b >= gf->size)) {
+        return gf->size;
+    }
+
+    return a ^ b;
+}
+
+uint32_t
+ec_gf_mul(ec_gf_t *gf, uint32_t a, uint32_t b)
+{
+    if ((a >= gf->size) || (b >= gf->size)) {
+        return gf->size;
+    }
+
+    if ((a != 0) && (b != 0)) {
+        return gf->pow[gf->log[a] + gf->log[b]];
+    }
+
+    return 0;
+}
+
+uint32_t
+ec_gf_div(ec_gf_t *gf, uint32_t a, uint32_t b)
+{
+    if ((a >= gf->size) || (b >= gf->size)) {
+        return gf->size;
+    }
+
+    if (b != 0) {
+        if (a != 0) {
+            return gf->pow[gf->size - 1 + gf->log[a] - gf->log[b]];
+        }
+
+        return 0;
+    }
+
+    return gf->size;
+}
+
+uint32_t
+ec_gf_exp(ec_gf_t *gf, uint32_t a, uint32_t b)
+{
+    uint32_t r;
+
+    if ((a >= gf->size) || ((a == 0) && (b == 0))) {
+        return gf->size;
+    }
+
+    r = 1;
+    while (b != 0) {
+        if ((b & 1) != 0) {
+            r = ec_gf_mul(gf, r, a);
+        }
+        a = ec_gf_mul(gf, a, a);
+        b >>= 1;
+    }
+
+    return r;
+}
diff --git a/xlators/cluster/ec/src/ec-galois.h b/xlators/cluster/ec/src/ec-galois.h
new file mode 100644
index 00000000000..ed55d53e419
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-galois.h
@@ -0,0 +1,32 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_GALOIS_H__
+#define __EC_GALOIS_H__
+
+#include <inttypes.h>
+
+#include "ec-types.h"
+
+ec_gf_t *
+ec_gf_prepare(uint32_t bits, uint32_t mod);
+void
+ec_gf_destroy(ec_gf_t *gf);
+
+uint32_t
+ec_gf_add(ec_gf_t *gf, uint32_t a, uint32_t b);
+uint32_t
+ec_gf_mul(ec_gf_t *gf, uint32_t a, uint32_t b);
+uint32_t
+ec_gf_div(ec_gf_t *gf, uint32_t a, uint32_t b);
+uint32_t
+ec_gf_exp(ec_gf_t *gf, uint32_t a, uint32_t b);
+
+#endif /* __EC_GALOIS_H__ */
diff --git a/xlators/cluster/ec/src/ec-generic.c b/xlators/cluster/ec/src/ec-generic.c
new file mode 100644
index 00000000000..884deb93669
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-generic.c
@@ -0,0 +1,1591 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/byte-order.h>
+
+#include "ec.h"
+#include "ec-messages.h"
+#include "ec-helpers.h"
+#include "ec-common.h"
+#include "ec-combine.h"
+#include "ec-fops.h"
+
+/* FOP: flush */
+
+int32_t
+ec_flush_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_FLUSH, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_flush(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_flush_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->flush, fop->fd,
+                      fop->xdata);
+}
+
+int32_t
+ec_manager_flush(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            ec_lock_prepare_fd(fop, fop->fd, 0, 0, EC_RANGE_FULL);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_flush_size_version(fop);
+
+            return EC_STATE_DELAYED_START;
+
+        case EC_STATE_DELAYED_START:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            ec_fop_prepare_answer(fop, _gf_false);
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.flush != NULL) {
+                fop->cbks.flush(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                cbk->op_errno, cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DELAYED_START:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.flush != NULL) {
+                fop->cbks.flush(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+static int32_t
+ec_validate_fd(fd_t *fd, xlator_t *xl)
+{
+    uint64_t iversion = 0;
+    uint64_t fversion = 0;
+    ec_inode_t *inode_ctx = NULL;
+    ec_fd_t *fd_ctx = NULL;
+
+    LOCK(&fd->lock);
+    {
+        fd_ctx = __ec_fd_get(fd, xl);
+        if (fd_ctx) {
+            fversion = fd_ctx->bad_version;
+        }
+    }
+    UNLOCK(&fd->lock);
+
+    LOCK(&fd->inode->lock);
+    {
+        inode_ctx = __ec_inode_get(fd->inode, xl);
+        if (inode_ctx) {
+            iversion = inode_ctx->bad_version;
+        }
+    }
+    UNLOCK(&fd->inode->lock);
+    if (fversion < iversion) {
+        return EBADF;
+    }
+    return 0;
+}
+
+void
+ec_flush(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_flush_cbk_t func, void *data, fd_t *fd,
+         dict_t *xdata)
+{
+    ec_cbk_t callback = {.flush = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FLUSH) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    if (fd) {
+        error = ec_validate_fd(fd, this);
+        if (error) {
+            gf_msg(this->name, GF_LOG_ERROR, EBADF, EC_MSG_FD_BAD,
+                   "Failing %s on %s", gf_fop_list[GF_FOP_FLUSH],
+                   fd->inode ? uuid_utoa(fd->inode->gfid) : "");
+            goto out;
+        }
+    }
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FLUSH, 0, target, fop_flags,
+                               ec_wind_flush, ec_manager_flush, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
+
+/* FOP: fsync */
+
+int32_t
+ec_combine_fsync(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    if (!ec_iatt_combine(fop, dst->iatt, src->iatt, 2)) {
+        gf_msg(fop->xl->name, GF_LOG_NOTICE, 0, EC_MSG_IATT_MISMATCH,
+               "Mismatching iatt in "
+               "answers of 'GF_FOP_FSYNC'");
+
+        return 0;
+    }
+
+    return 1;
+}
+
+int32_t
+ec_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, struct iatt *prebuf, struct iatt *postbuf,
+             dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_FSYNC, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            if (prebuf != NULL) {
+                cbk->iatt[0] = *prebuf;
+            }
+            if (postbuf != NULL) {
+                cbk->iatt[1] = *postbuf;
+            }
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, ec_combine_fsync);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_fsync(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_fsync_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->fsync, fop->fd,
+                      fop->int32, fop->xdata);
+}
+
+int32_t
+ec_manager_fsync(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            ec_lock_prepare_fd(fop, fop->fd, EC_QUERY_INFO, 0, EC_RANGE_FULL);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_flush_size_version(fop);
+
+            return EC_STATE_DELAYED_START;
+
+        case EC_STATE_DELAYED_START:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 2, cbk->count);
+
+                /* This shouldn't fail because we have the inode locked. */
+                GF_ASSERT(ec_get_inode_size(fop, fop->fd->inode,
+                                            &cbk->iatt[0].ia_size));
+                cbk->iatt[1].ia_size = cbk->iatt[0].ia_size;
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.fsync != NULL) {
+                fop->cbks.fsync(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                cbk->op_errno, &cbk->iatt[0], &cbk->iatt[1],
+                                cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+        case -EC_STATE_DELAYED_START:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.fsync != NULL) {
+                fop->cbks.fsync(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_fsync(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_fsync_cbk_t func, void *data, fd_t *fd,
+         int32_t datasync, dict_t *xdata)
+{
+    ec_cbk_t callback = {.fsync = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FSYNC) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    if (fd) {
+        error = ec_validate_fd(fd, this);
+        if (error) {
+            gf_msg(this->name, GF_LOG_ERROR, EBADF, EC_MSG_FD_BAD,
+                   "Failing %s on %s", gf_fop_list[GF_FOP_FSYNC],
+                   fd->inode ? uuid_utoa(fd->inode->gfid) : "");
+            goto out;
+        }
+    }
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FSYNC, 0, target, fop_flags,
+                               ec_wind_fsync, ec_manager_fsync, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->int32 = datasync;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: fsyncdir */
+
+int32_t
+ec_fsyncdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_FSYNCDIR, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_fsyncdir(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_fsyncdir_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->fsyncdir,
+                      fop->fd, fop->int32, fop->xdata);
+}
+
+int32_t
+ec_manager_fsyncdir(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            ec_lock_prepare_fd(fop, fop->fd, 0, 0, EC_RANGE_FULL);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_flush_size_version(fop);
+
+            return EC_STATE_DELAYED_START;
+
+        case EC_STATE_DELAYED_START:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            ec_fop_prepare_answer(fop, _gf_false);
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.fsyncdir != NULL) {
+                fop->cbks.fsyncdir(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                   cbk->op_errno, cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+        case -EC_STATE_DELAYED_START:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.fsyncdir != NULL) {
+                fop->cbks.fsyncdir(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                   NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_fsyncdir(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_fsyncdir_cbk_t func, void *data, fd_t *fd,
+            int32_t datasync, dict_t *xdata)
+{
+    ec_cbk_t callback = {.fsyncdir = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FSYNCDIR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FSYNCDIR, 0, target,
+                               fop_flags, ec_wind_fsyncdir, ec_manager_fsyncdir,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->int32 = datasync;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
+
+/* FOP: lookup */
+
+void
+ec_lookup_rebuild(ec_t *ec, ec_fop_data_t *fop, ec_cbk_data_t *cbk)
+{
+    ec_inode_t *ctx = NULL;
+    uint64_t size = 0;
+    int32_t have_size = 0, err;
+
+    if (cbk->op_ret < 0) {
+        return;
+    }
+
+    ec_dict_del_array(cbk->xdata, EC_XATTR_VERSION, cbk->version,
+                      EC_VERSION_SIZE);
+
+    err = ec_loc_update(fop->xl, &fop->loc[0], cbk->inode, &cbk->iatt[0]);
+    if (ec_cbk_set_error(cbk, -err, _gf_true)) {
+        return;
+    }
+
+    LOCK(&cbk->inode->lock);
+
+    ctx = __ec_inode_get(cbk->inode, fop->xl);
+    if (ctx != NULL) {
+        if (ctx->have_version) {
+            cbk->version[0] = ctx->post_version[0];
+            cbk->version[1] = ctx->post_version[1];
+        }
+        if (ctx->have_size) {
+            size = ctx->post_size;
+            have_size = 1;
+        }
+    }
+
+    UNLOCK(&cbk->inode->lock);
+
+    if (cbk->iatt[0].ia_type == IA_IFREG) {
+        cbk->size = cbk->iatt[0].ia_size;
+        ec_dict_del_number(cbk->xdata, EC_XATTR_SIZE, &cbk->iatt[0].ia_size);
+        if (have_size) {
+            cbk->iatt[0].ia_size = size;
+        }
+    }
+}
+
+int32_t
+ec_combine_lookup(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    if (!ec_iatt_combine(fop, dst->iatt, src->iatt, 2)) {
+        gf_msg(fop->xl->name, GF_LOG_DEBUG, 0, EC_MSG_IATT_MISMATCH,
+               "Mismatching iatt in "
+               "answers of 'GF_FOP_LOOKUP'");
+
+        return 0;
+    }
+
+    return 1;
+}
+
+int32_t
+ec_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, inode_t *inode, struct iatt *buf, dict_t *xdata,
+              struct iatt *postparent)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+    uint64_t dirty[2] = {0};
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_LOOKUP, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            if (inode != NULL) {
+                cbk->inode = inode_ref(inode);
+                if (cbk->inode == NULL) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_INODE_REF_FAIL,
+                           "Failed to reference an inode.");
+
+                    goto out;
+                }
+            }
+            if (buf != NULL) {
+                cbk->iatt[0] = *buf;
+            }
+            if (postparent != NULL) {
+                cbk->iatt[1] = *postparent;
+            }
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+            ec_dict_del_array(xdata, EC_XATTR_DIRTY, dirty, EC_VERSION_SIZE);
+        }
+
+        ec_combine(cbk, ec_combine_lookup);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_lookup(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_lookup_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->lookup,
+                      &fop->loc[0], fop->xdata);
+}
+
+int32_t
+ec_manager_lookup(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+    int32_t err;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            if (fop->xdata == NULL) {
+                fop->xdata = dict_new();
+                if (fop->xdata == NULL) {
+                    gf_msg(fop->xl->name, GF_LOG_ERROR, ENOMEM,
+                           EC_MSG_LOOKUP_REQ_PREP_FAIL,
+                           "Unable to prepare "
+                           "lookup request");
+
+                    fop->error = ENOMEM;
+
+                    return EC_STATE_REPORT;
+                }
+            } else {
+                /*TODO: To be handled once we have 'syndromes' */
+                dict_del(fop->xdata, GF_CONTENT_KEY);
+            }
+            err = dict_set_uint64(fop->xdata, EC_XATTR_SIZE, 0);
+            if (err == 0) {
+                err = dict_set_uint64(fop->xdata, EC_XATTR_VERSION, 0);
+            }
+            if (err == 0) {
+                err = dict_set_uint64(fop->xdata, EC_XATTR_DIRTY, 0);
+            }
+            if (err != 0) {
+                gf_msg(fop->xl->name, GF_LOG_ERROR, -err,
+                       EC_MSG_LOOKUP_REQ_PREP_FAIL,
+                       "Unable to prepare lookup "
+                       "request");
+
+                fop->error = -err;
+
+                return EC_STATE_REPORT;
+            }
+
+            /* Fall through */
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            /*
+             * Lookup happens without any lock, so there is a chance that it
+             * will have answers before modification happened and after
+             * modification happened in the same response. So choose the next
+             * best answer when the answers don't match for EC_MINIMUM_MIN
+             */
+
+            if (!fop->answer && !list_empty(&fop->cbk_list)) {
+                fop->answer = list_entry(fop->cbk_list.next, ec_cbk_data_t,
+                                         list);
+            }
+
+            cbk = ec_fop_prepare_answer(fop, _gf_true);
+            if (cbk != NULL) {
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 2, cbk->count);
+
+                ec_lookup_rebuild(fop->xl->private, fop, cbk);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.lookup != NULL) {
+                fop->cbks.lookup(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                 cbk->op_errno, cbk->inode, &cbk->iatt[0],
+                                 cbk->xdata, &cbk->iatt[1]);
+            }
+
+            return EC_STATE_END;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.lookup != NULL) {
+                fop->cbks.lookup(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                 NULL, NULL, NULL, NULL);
+            }
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_lookup(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_lookup_cbk_t func, void *data, loc_t *loc,
+          dict_t *xdata)
+{
+    ec_cbk_t callback = {.lookup = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(LOOKUP) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_LOOKUP, EC_FLAG_LOCK_SHARED,
+                               target, fop_flags, ec_wind_lookup,
+                               ec_manager_lookup, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        /* Do not log failures here as a memory problem would have already
+         * been logged by the corresponding alloc functions */
+        if (fop->xdata == NULL)
+            goto out;
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: statfs */
+
+int32_t
+ec_combine_statfs(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    ec_statvfs_combine(&dst->statvfs, &src->statvfs);
+
+    return 1;
+}
+
+int32_t
+ec_statfs_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct statvfs *buf, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_STATFS, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            if (buf != NULL) {
+                cbk->statvfs = *buf;
+            }
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, ec_combine_statfs);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_statfs(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_statfs_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->statfs,
+                      &fop->loc[0], fop->xdata);
+}
+
+int32_t
+ec_manager_statfs(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk = NULL;
+    gf_boolean_t deem_statfs_enabled = _gf_false;
+    int32_t err = 0;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_true);
+            if (cbk != NULL) {
+                ec_t *ec = fop->xl->private;
+
+                if (cbk->xdata) {
+                    err = dict_get_int8(cbk->xdata, "quota-deem-statfs",
+                                        (int8_t *)&deem_statfs_enabled);
+                    if (err != -ENOENT) {
+                        ec_cbk_set_error(cbk, -err, _gf_true);
+                    }
+                }
+
+                if (err != 0 || deem_statfs_enabled == _gf_false) {
+                    cbk->statvfs.f_blocks *= ec->fragments;
+                    cbk->statvfs.f_bfree *= ec->fragments;
+                    cbk->statvfs.f_bavail *= ec->fragments;
+                }
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.statfs != NULL) {
+                fop->cbks.statfs(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                 cbk->op_errno, &cbk->statvfs, cbk->xdata);
+            }
+
+            return EC_STATE_END;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.statfs != NULL) {
+                fop->cbks.statfs(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                 NULL, NULL);
+            }
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_statfs(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_statfs_cbk_t func, void *data, loc_t *loc,
+          dict_t *xdata)
+{
+    ec_cbk_t callback = {.statfs = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(STATFS) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_STATFS, EC_FLAG_LOCK_SHARED,
+                               target, fop_flags, ec_wind_statfs,
+                               ec_manager_statfs, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
+
+/* FOP: xattrop */
+
+int32_t
+ec_combine_xattrop(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    if (!ec_dict_compare(dst->dict, src->dict)) {
+        gf_msg(fop->xl->name, GF_LOG_DEBUG, 0, EC_MSG_DICT_MISMATCH,
+               "Mismatching dictionary in "
+               "answers of 'GF_FOP_XATTROP'");
+
+        return 0;
+    }
+
+    return 1;
+}
+
+int32_t
+ec_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, dict_t *xattr, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_lock_link_t *link = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    uint64_t dirty[2] = {0};
+    data_t *data;
+    uint64_t *version;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, fop->id, idx, op_ret,
+                               op_errno);
+    if (!cbk)
+        goto out;
+
+    if (op_ret >= 0) {
+        cbk->dict = dict_ref(xattr);
+
+        data = dict_get(cbk->dict, EC_XATTR_VERSION);
+        if ((data != NULL) && (data->len >= sizeof(uint64_t))) {
+            version = (uint64_t *)data->data;
+
+            if (((ntoh64(version[0]) >> EC_SELFHEAL_BIT) & 1) != 0) {
+                LOCK(&fop->lock);
+
+                fop->healing |= 1ULL << idx;
+
+                UNLOCK(&fop->lock);
+            }
+        }
+
+        ec_dict_del_array(xattr, EC_XATTR_DIRTY, dirty, EC_VERSION_SIZE);
+        link = fop->data;
+        if (link) {
+            /*Keep a note of if the dirty is already set or not*/
+            link->dirty[0] |= (dirty[0] != 0);
+            link->dirty[1] |= (dirty[1] != 0);
+        }
+    }
+
+    if (xdata)
+        cbk->xdata = dict_ref(xdata);
+
+    ec_combine(cbk, ec_combine_xattrop);
+
+out:
+    if (fop)
+        ec_complete(fop);
+
+    return 0;
+}
+
+void
+ec_wind_xattrop(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_xattrop_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->xattrop,
+                      &fop->loc[0], fop->xattrop_flags, fop->dict, fop->xdata);
+}
+
+int32_t
+ec_manager_xattrop(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            if (fop->fd == NULL) {
+                ec_lock_prepare_inode(fop, &fop->loc[0], EC_UPDATE_META, 0,
+                                      EC_RANGE_FULL);
+            } else {
+                ec_lock_prepare_fd(fop, fop->fd, EC_UPDATE_META, 0,
+                                   EC_RANGE_FULL);
+            }
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                int32_t err;
+
+                err = ec_dict_combine(cbk, EC_COMBINE_DICT);
+                ec_cbk_set_error(cbk, -err, _gf_false);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->id == GF_FOP_XATTROP) {
+                if (fop->cbks.xattrop != NULL) {
+                    fop->cbks.xattrop(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                      cbk->op_errno, cbk->dict, cbk->xdata);
+                }
+            } else {
+                if (fop->cbks.fxattrop != NULL) {
+                    fop->cbks.fxattrop(fop->req_frame, fop, fop->xl,
+                                       cbk->op_ret, cbk->op_errno, cbk->dict,
+                                       cbk->xdata);
+                }
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->id == GF_FOP_XATTROP) {
+                if (fop->cbks.xattrop != NULL) {
+                    fop->cbks.xattrop(fop->req_frame, fop, fop->xl, -1,
+                                      fop->error, NULL, NULL);
+                }
+            } else {
+                if (fop->cbks.fxattrop != NULL) {
+                    fop->cbks.fxattrop(fop->req_frame, fop, fop->xl, -1,
+                                       fop->error, NULL, NULL);
+                }
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_xattrop(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_xattrop_cbk_t func, void *data, loc_t *loc,
+           gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+    ec_cbk_t callback = {.xattrop = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(XATTROP) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_XATTROP, 0, target,
+                               fop_flags, ec_wind_xattrop, ec_manager_xattrop,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->xattrop_flags = optype;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xattr != NULL) {
+        fop->dict = dict_ref(xattr);
+        if (fop->dict == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
+
+void
+ec_wind_fxattrop(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_xattrop_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->fxattrop,
+                      fop->fd, fop->xattrop_flags, fop->dict, fop->xdata);
+}
+
+void
+ec_fxattrop(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_fxattrop_cbk_t func, void *data, fd_t *fd,
+            gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+    ec_cbk_t callback = {.fxattrop = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FXATTROP) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FXATTROP, 0, target,
+                               fop_flags, ec_wind_fxattrop, ec_manager_xattrop,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->xattrop_flags = optype;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xattr != NULL) {
+        fop->dict = dict_ref(xattr);
+        if (fop->dict == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
+
+/* FOP: IPC */
+
+int32_t
+ec_ipc_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+           int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_IPC, idx, op_ret,
+                               op_errno);
+
+    if (cbk != NULL) {
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+        }
+
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_ipc(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_ipc_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->ipc, fop->int32,
+                      fop->xdata);
+}
+
+int32_t
+ec_manager_ipc(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            ec_fop_prepare_answer(fop, _gf_true);
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+            if (fop->cbks.ipc != NULL) {
+                fop->cbks.ipc(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                              cbk->op_errno, cbk->xdata);
+            }
+
+            return EC_STATE_END;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.ipc != NULL) {
+                fop->cbks.ipc(fop->req_frame, fop, fop->xl, -1, fop->error,
+                              NULL);
+            }
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_ipc(call_frame_t *frame, xlator_t *this, uintptr_t target,
+       uint32_t fop_flags, fop_ipc_cbk_t func, void *data, int32_t op,
+       dict_t *xdata)
+{
+    ec_cbk_t callback = {.ipc = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(IPC) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_IPC, 0, target, fop_flags,
+                               ec_wind_ipc, ec_manager_ipc, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+    }
+    fop->int32 = op;
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
diff --git a/xlators/cluster/ec/src/ec-gf8.c b/xlators/cluster/ec/src/ec-gf8.c
new file mode 100644
index 00000000000..039adae5929
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-gf8.c
@@ -0,0 +1,5882 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "ec-gf8.h"
+
+static ec_gf_op_t ec_gf8_mul_00_ops[] = {{EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_00 = {0,
+                                    {
+                                        0,
+                                    },
+                                    ec_gf8_mul_00_ops};
+
+static ec_gf_op_t ec_gf8_mul_01_ops[] = {{EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_01 = {8,
+                                    {
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                    },
+                                    ec_gf8_mul_01_ops};
+
+static ec_gf_op_t ec_gf8_mul_02_ops[] = {{EC_GF_OP_XOR2, 1, 7, 0},
+                                         {EC_GF_OP_XOR2, 2, 7, 0},
+                                         {EC_GF_OP_XOR2, 3, 7, 0},
+                                         {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_02 = {8,
+                                    {
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                    },
+                                    ec_gf8_mul_02_ops};
+
+static ec_gf_op_t ec_gf8_mul_03_ops[] = {
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_COPY, 8, 3, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 4, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_03 = {9,
+                                    {
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        8,
+                                    },
+                                    ec_gf8_mul_03_ops};
+
+static ec_gf_op_t ec_gf8_mul_04_ops[] = {
+    {EC_GF_OP_XOR3, 8, 6, 7}, {EC_GF_OP_XOR2, 2, 8, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 1, 8, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_04 = {9,
+                                    {
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        8,
+                                    },
+                                    ec_gf8_mul_04_ops};
+
+static ec_gf_op_t ec_gf8_mul_05_ops[] = {
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_05 = {8,
+                                    {
+                                        0,
+                                        1,
+                                        2,
+                                        6,
+                                        7,
+                                        3,
+                                        4,
+                                        5,
+                                    },
+                                    ec_gf8_mul_05_ops};
+
+static ec_gf_op_t ec_gf8_mul_06_ops[] = {
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_COPY, 8, 2, 0},
+    {EC_GF_OP_XOR2, 8, 3, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_06 = {9,
+                                    {
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        8,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                    },
+                                    ec_gf8_mul_06_ops};
+
+static ec_gf_op_t ec_gf8_mul_07_ops[] = {
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_07 = {8,
+                                    {
+                                        6,
+                                        0,
+                                        1,
+                                        3,
+                                        2,
+                                        4,
+                                        5,
+                                        7,
+                                    },
+                                    ec_gf8_mul_07_ops};
+
+static ec_gf_op_t ec_gf8_mul_08_ops[] = {
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR3, 8, 6, 7},
+    {EC_GF_OP_XOR2, 1, 8, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 2, 8, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_08 = {9,
+                                    {
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        8,
+                                    },
+                                    ec_gf8_mul_08_ops};
+
+static ec_gf_op_t ec_gf8_mul_09_ops[] = {
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_09 = {8,
+                                    {
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        5,
+                                        6,
+                                        7,
+                                        4,
+                                    },
+                                    ec_gf8_mul_09_ops};
+
+static ec_gf_op_t ec_gf8_mul_0A_ops[] = {
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_0A = {8,
+                                    {
+                                        5,
+                                        0,
+                                        1,
+                                        2,
+                                        6,
+                                        7,
+                                        3,
+                                        4,
+                                    },
+                                    ec_gf8_mul_0A_ops};
+
+static ec_gf_op_t ec_gf8_mul_0B_ops[] = {
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_COPY, 9, 3, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_COPY, 8, 5, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR3, 3, 8, 6}, {EC_GF_OP_XOR2, 1, 9, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_0B = {10,
+                                    {
+                                        7,
+                                        1,
+                                        5,
+                                        2,
+                                        4,
+                                        3,
+                                        0,
+                                        6,
+                                        8,
+                                        9,
+                                    },
+                                    ec_gf8_mul_0B_ops};
+
+static ec_gf_op_t ec_gf8_mul_0C_ops[] = {
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_COPY, 8, 1, 0},
+    {EC_GF_OP_XOR2, 8, 2, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_0C = {9,
+                                    {
+                                        5,
+                                        7,
+                                        0,
+                                        1,
+                                        8,
+                                        2,
+                                        3,
+                                        4,
+                                        6,
+                                    },
+                                    ec_gf8_mul_0C_ops};
+
+static ec_gf_op_t ec_gf8_mul_0D_ops[] = {
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR3, 8, 2, 4}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 1, 8, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR3, 2, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_0D = {9,
+                                    {
+                                        5,
+                                        6,
+                                        7,
+                                        3,
+                                        1,
+                                        0,
+                                        2,
+                                        4,
+                                        8,
+                                    },
+                                    ec_gf8_mul_0D_ops};
+
+static ec_gf_op_t ec_gf8_mul_0E_ops[] = {
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_0E = {8,
+                                    {
+                                        7,
+                                        0,
+                                        6,
+                                        1,
+                                        3,
+                                        2,
+                                        4,
+                                        5,
+                                    },
+                                    ec_gf8_mul_0E_ops};
+
+static ec_gf_op_t ec_gf8_mul_0F_ops[] = {
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_0F = {8,
+                                    {
+                                        1,
+                                        0,
+                                        5,
+                                        6,
+                                        7,
+                                        2,
+                                        3,
+                                        4,
+                                    },
+                                    ec_gf8_mul_0F_ops};
+
+static ec_gf_op_t ec_gf8_mul_10_ops[] = {
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_10 = {8,
+                                    {
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                    },
+                                    ec_gf8_mul_10_ops};
+
+static ec_gf_op_t ec_gf8_mul_11_ops[] = {
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_11 = {8,
+                                    {
+                                        4,
+                                        1,
+                                        2,
+                                        6,
+                                        0,
+                                        5,
+                                        7,
+                                        3,
+                                    },
+                                    ec_gf8_mul_11_ops};
+
+static ec_gf_op_t ec_gf8_mul_12_ops[] = {
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_12 = {8,
+                                    {
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        5,
+                                        6,
+                                        4,
+                                    },
+                                    ec_gf8_mul_12_ops};
+
+static ec_gf_op_t ec_gf8_mul_13_ops[] = {
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR3, 8, 3, 7},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 6, 8, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 0, 8, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_13 = {9,
+                                    {
+                                        4,
+                                        5,
+                                        2,
+                                        6,
+                                        0,
+                                        1,
+                                        7,
+                                        3,
+                                        8,
+                                    },
+                                    ec_gf8_mul_13_ops};
+
+static ec_gf_op_t ec_gf8_mul_14_ops[] = {
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_14 = {8,
+                                    {
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        4,
+                                        5,
+                                        3,
+                                    },
+                                    ec_gf8_mul_14_ops};
+
+static ec_gf_op_t ec_gf8_mul_15_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR3, 5, 8, 7},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_15 = {9,
+                                    {
+                                        0,
+                                        1,
+                                        2,
+                                        4,
+                                        7,
+                                        6,
+                                        5,
+                                        3,
+                                        8,
+                                    },
+                                    ec_gf8_mul_15_ops};
+
+static ec_gf_op_t ec_gf8_mul_16_ops[] = {
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_16 = {8,
+                                    {
+                                        6,
+                                        7,
+                                        4,
+                                        1,
+                                        2,
+                                        3,
+                                        5,
+                                        0,
+                                    },
+                                    ec_gf8_mul_16_ops};
+
+static ec_gf_op_t ec_gf8_mul_17_ops[] = {
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_17 = {8,
+                                    {
+                                        5,
+                                        7,
+                                        0,
+                                        1,
+                                        3,
+                                        2,
+                                        4,
+                                        6,
+                                    },
+                                    ec_gf8_mul_17_ops};
+
+static ec_gf_op_t ec_gf8_mul_18_ops[] = {
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 6, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_18 = {9,
+                                    {
+                                        4,
+                                        5,
+                                        7,
+                                        6,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        8,
+                                    },
+                                    ec_gf8_mul_18_ops};
+
+static ec_gf_op_t ec_gf8_mul_19_ops[] = {
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_19 = {8,
+                                    {
+                                        0,
+                                        5,
+                                        2,
+                                        6,
+                                        7,
+                                        1,
+                                        3,
+                                        4,
+                                    },
+                                    ec_gf8_mul_19_ops};
+
+static ec_gf_op_t ec_gf8_mul_1A_ops[] = {
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_1A = {8,
+                                    {
+                                        7,
+                                        0,
+                                        4,
+                                        5,
+                                        3,
+                                        1,
+                                        2,
+                                        6,
+                                    },
+                                    ec_gf8_mul_1A_ops};
+
+static ec_gf_op_t ec_gf8_mul_1B_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_1B = {8,
+                                    {
+                                        7,
+                                        4,
+                                        5,
+                                        6,
+                                        3,
+                                        1,
+                                        2,
+                                        0,
+                                    },
+                                    ec_gf8_mul_1B_ops};
+
+static ec_gf_op_t ec_gf8_mul_1C_ops[] = {
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_1C = {8,
+                                    {
+                                        5,
+                                        4,
+                                        3,
+                                        0,
+                                        1,
+                                        7,
+                                        2,
+                                        6,
+                                    },
+                                    ec_gf8_mul_1C_ops};
+
+static ec_gf_op_t ec_gf8_mul_1D_ops[] = {
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR3, 8, 4, 2},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 5, 8, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_1D = {9,
+                                    {
+                                        0,
+                                        7,
+                                        5,
+                                        8,
+                                        2,
+                                        3,
+                                        4,
+                                        1,
+                                        6,
+                                    },
+                                    ec_gf8_mul_1D_ops};
+
+static ec_gf_op_t ec_gf8_mul_1E_ops[] = {
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_1E = {8,
+                                    {
+                                        4,
+                                        7,
+                                        5,
+                                        1,
+                                        6,
+                                        0,
+                                        2,
+                                        3,
+                                    },
+                                    ec_gf8_mul_1E_ops};
+
+static ec_gf_op_t ec_gf8_mul_1F_ops[] = {
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR3, 8, 3, 7},
+    {EC_GF_OP_XOR2, 4, 8, 0}, {EC_GF_OP_XOR2, 1, 8, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_1F = {9,
+                                    {
+                                        1,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        3,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_1F_ops};
+
+static ec_gf_op_t ec_gf8_mul_20_ops[] = {
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_20 = {8,
+                                    {
+                                        7,
+                                        4,
+                                        5,
+                                        6,
+                                        3,
+                                        0,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_20_ops};
+
+static ec_gf_op_t ec_gf8_mul_21_ops[] = {
+    {EC_GF_OP_COPY, 9, 0, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR3, 8, 7, 5}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 3, 8, 0}, {EC_GF_OP_XOR2, 2, 8, 0},
+    {EC_GF_OP_XOR2, 4, 9, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_21 = {10,
+                                    {
+                                        0,
+                                        1,
+                                        2,
+                                        7,
+                                        5,
+                                        4,
+                                        3,
+                                        6,
+                                        8,
+                                        9,
+                                    },
+                                    ec_gf8_mul_21_ops};
+
+static ec_gf_op_t ec_gf8_mul_22_ops[] = {
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_22 = {8,
+                                    {
+                                        3,
+                                        0,
+                                        5,
+                                        2,
+                                        6,
+                                        4,
+                                        1,
+                                        7,
+                                    },
+                                    ec_gf8_mul_22_ops};
+
+static ec_gf_op_t ec_gf8_mul_23_ops[] = {
+    {EC_GF_OP_COPY, 8, 2, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 3, 8, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_23 = {9,
+                                    {
+                                        0,
+                                        4,
+                                        3,
+                                        2,
+                                        5,
+                                        6,
+                                        1,
+                                        8,
+                                        7,
+                                    },
+                                    ec_gf8_mul_23_ops};
+
+static ec_gf_op_t ec_gf8_mul_24_ops[] = {
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_24 = {8,
+                                    {
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        4,
+                                        5,
+                                        3,
+                                    },
+                                    ec_gf8_mul_24_ops};
+
+static ec_gf_op_t ec_gf8_mul_25_ops[] = {
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_25 = {8,
+                                    {
+                                        2,
+                                        7,
+                                        0,
+                                        1,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                    },
+                                    ec_gf8_mul_25_ops};
+
+static ec_gf_op_t ec_gf8_mul_26_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_26 = {8,
+                                    {
+                                        3,
+                                        4,
+                                        1,
+                                        2,
+                                        0,
+                                        5,
+                                        6,
+                                        7,
+                                    },
+                                    ec_gf8_mul_26_ops};
+
+static ec_gf_op_t ec_gf8_mul_27_ops[] = {
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_27 = {8,
+                                    {
+                                        3,
+                                        0,
+                                        1,
+                                        2,
+                                        6,
+                                        7,
+                                        4,
+                                        5,
+                                    },
+                                    ec_gf8_mul_27_ops};
+
+static ec_gf_op_t ec_gf8_mul_28_ops[] = {
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_28 = {8,
+                                    {
+                                        5,
+                                        6,
+                                        3,
+                                        0,
+                                        1,
+                                        2,
+                                        4,
+                                        7,
+                                    },
+                                    ec_gf8_mul_28_ops};
+
+static ec_gf_op_t ec_gf8_mul_29_ops[] = {
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_29 = {8,
+                                    {
+                                        4,
+                                        6,
+                                        3,
+                                        5,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_29_ops};
+
+static ec_gf_op_t ec_gf8_mul_2A_ops[] = {
+    {EC_GF_OP_COPY, 8, 1, 0}, {EC_GF_OP_XOR2, 8, 0, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR3, 6, 8, 4}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_2A = {9,
+                                    {
+                                        3,
+                                        4,
+                                        7,
+                                        2,
+                                        6,
+                                        5,
+                                        1,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_2A_ops};
+
+static ec_gf_op_t ec_gf8_mul_2B_ops[] = {
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_2B = {8,
+                                    {
+                                        3,
+                                        4,
+                                        7,
+                                        5,
+                                        6,
+                                        0,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_2B_ops};
+
+static ec_gf_op_t ec_gf8_mul_2C_ops[] = {
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_2C = {8,
+                                    {
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        2,
+                                        3,
+                                        4,
+                                        1,
+                                    },
+                                    ec_gf8_mul_2C_ops};
+
+static ec_gf_op_t ec_gf8_mul_2D_ops[] = {
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR3, 8, 4, 6},
+    {EC_GF_OP_XOR2, 5, 8, 0}, {EC_GF_OP_XOR2, 7, 8, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_2D = {9,
+                                    {
+                                        7,
+                                        0,
+                                        3,
+                                        5,
+                                        1,
+                                        4,
+                                        2,
+                                        6,
+                                        8,
+                                    },
+                                    ec_gf8_mul_2D_ops};
+
+static ec_gf_op_t ec_gf8_mul_2E_ops[] = {
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_COPY, 8, 4, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 8, 7, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 2, 8, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 6, 8, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_2E = {9,
+                                    {
+                                        5,
+                                        0,
+                                        7,
+                                        3,
+                                        2,
+                                        6,
+                                        4,
+                                        1,
+                                        8,
+                                    },
+                                    ec_gf8_mul_2E_ops};
+
+static ec_gf_op_t ec_gf8_mul_2F_ops[] = {
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR3, 8, 7, 6}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 3, 8, 0},
+    {EC_GF_OP_XOR2, 2, 8, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_2F = {9,
+                                    {
+                                        6,
+                                        3,
+                                        2,
+                                        5,
+                                        7,
+                                        0,
+                                        1,
+                                        4,
+                                        8,
+                                    },
+                                    ec_gf8_mul_2F_ops};
+
+static ec_gf_op_t ec_gf8_mul_30_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 8, 1, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR3, 6, 8, 7},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_30 = {9,
+                                    {
+                                        3,
+                                        4,
+                                        7,
+                                        5,
+                                        0,
+                                        6,
+                                        1,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_30_ops};
+
+static ec_gf_op_t ec_gf8_mul_31_ops[] = {
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_31 = {8,
+                                    {
+                                        7,
+                                        1,
+                                        4,
+                                        5,
+                                        6,
+                                        0,
+                                        2,
+                                        3,
+                                    },
+                                    ec_gf8_mul_31_ops};
+
+static ec_gf_op_t ec_gf8_mul_32_ops[] = {
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_32 = {8,
+                                    {
+                                        3,
+                                        4,
+                                        6,
+                                        7,
+                                        5,
+                                        0,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_32_ops};
+
+static ec_gf_op_t ec_gf8_mul_33_ops[] = {
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_33 = {8,
+                                    {
+                                        5,
+                                        4,
+                                        3,
+                                        0,
+                                        2,
+                                        1,
+                                        6,
+                                        7,
+                                    },
+                                    ec_gf8_mul_33_ops};
+
+static ec_gf_op_t ec_gf8_mul_34_ops[] = {
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_34 = {8,
+                                    {
+                                        7,
+                                        5,
+                                        3,
+                                        0,
+                                        2,
+                                        4,
+                                        1,
+                                        6,
+                                    },
+                                    ec_gf8_mul_34_ops};
+
+static ec_gf_op_t ec_gf8_mul_35_ops[] = {
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_35 = {8,
+                                    {
+                                        6,
+                                        7,
+                                        5,
+                                        4,
+                                        2,
+                                        0,
+                                        1,
+                                        3,
+                                    },
+                                    ec_gf8_mul_35_ops};
+
+static ec_gf_op_t ec_gf8_mul_36_ops[] = {
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_36 = {8,
+                                    {
+                                        6,
+                                        7,
+                                        4,
+                                        1,
+                                        2,
+                                        3,
+                                        0,
+                                        5,
+                                    },
+                                    ec_gf8_mul_36_ops};
+
+static ec_gf_op_t ec_gf8_mul_37_ops[] = {
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR3, 8, 0, 1},
+    {EC_GF_OP_XOR2, 3, 8, 0}, {EC_GF_OP_XOR2, 7, 8, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_37 = {9,
+                                    {
+                                        6,
+                                        7,
+                                        2,
+                                        1,
+                                        0,
+                                        3,
+                                        4,
+                                        5,
+                                        8,
+                                    },
+                                    ec_gf8_mul_37_ops};
+
+static ec_gf_op_t ec_gf8_mul_38_ops[] = {
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR3, 8, 6, 7},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 0, 8, 0}, {EC_GF_OP_XOR2, 4, 8, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_38 = {9,
+                                    {
+                                        4,
+                                        5,
+                                        6,
+                                        3,
+                                        0,
+                                        1,
+                                        7,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_38_ops};
+
+static ec_gf_op_t ec_gf8_mul_39_ops[] = {
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_39 = {8,
+                                    {
+                                        1,
+                                        6,
+                                        3,
+                                        0,
+                                        5,
+                                        2,
+                                        4,
+                                        7,
+                                    },
+                                    ec_gf8_mul_39_ops};
+
+static ec_gf_op_t ec_gf8_mul_3A_ops[] = {
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_3A = {8,
+                                    {
+                                        3,
+                                        4,
+                                        7,
+                                        0,
+                                        5,
+                                        6,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_3A_ops};
+
+static ec_gf_op_t ec_gf8_mul_3B_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR3, 8, 7, 3}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 1, 8, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_3B = {9,
+                                    {
+                                        3,
+                                        0,
+                                        1,
+                                        7,
+                                        6,
+                                        2,
+                                        4,
+                                        8,
+                                        5,
+                                    },
+                                    ec_gf8_mul_3B_ops};
+
+static ec_gf_op_t ec_gf8_mul_3C_ops[] = {
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_3C = {8,
+                                    {
+                                        3,
+                                        6,
+                                        4,
+                                        1,
+                                        7,
+                                        2,
+                                        0,
+                                        5,
+                                    },
+                                    ec_gf8_mul_3C_ops};
+
+static ec_gf_op_t ec_gf8_mul_3D_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_3D = {8,
+                                    {
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                    },
+                                    ec_gf8_mul_3D_ops};
+
+static ec_gf_op_t ec_gf8_mul_3E_ops[] = {
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_3E = {8,
+                                    {
+                                        6,
+                                        1,
+                                        2,
+                                        7,
+                                        0,
+                                        3,
+                                        5,
+                                        4,
+                                    },
+                                    ec_gf8_mul_3E_ops};
+
+static ec_gf_op_t ec_gf8_mul_3F_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0},  {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_COPY, 10, 4, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0},  {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0},  {EC_GF_OP_COPY, 9, 2, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0},  {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0},  {EC_GF_OP_XOR3, 4, 9, 7},
+    {EC_GF_OP_XOR2, 3, 4, 0},  {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0},  {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 3, 10, 0}, {EC_GF_OP_XOR2, 5, 8, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_3F = {11,
+                                    {
+                                        1,
+                                        7,
+                                        6,
+                                        2,
+                                        4,
+                                        3,
+                                        5,
+                                        0,
+                                        8,
+                                        9,
+                                        10,
+                                    },
+                                    ec_gf8_mul_3F_ops};
+
+static ec_gf_op_t ec_gf8_mul_40_ops[] = {
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR3, 8, 7, 6},
+    {EC_GF_OP_XOR2, 1, 8, 0}, {EC_GF_OP_XOR2, 5, 8, 0},
+    {EC_GF_OP_XOR2, 4, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_40 = {9,
+                                    {
+                                        5,
+                                        7,
+                                        4,
+                                        6,
+                                        2,
+                                        3,
+                                        0,
+                                        1,
+                                        8,
+                                    },
+                                    ec_gf8_mul_40_ops};
+
+static ec_gf_op_t ec_gf8_mul_41_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 8, 4, 0},
+    {EC_GF_OP_XOR2, 8, 5, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_41 = {9,
+                                    {
+                                        0,
+                                        7,
+                                        6,
+                                        5,
+                                        3,
+                                        4,
+                                        8,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_41_ops};
+
+static ec_gf_op_t ec_gf8_mul_42_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 8, 3, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 7, 8, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_42 = {9,
+                                    {
+                                        2,
+                                        7,
+                                        1,
+                                        6,
+                                        4,
+                                        3,
+                                        0,
+                                        5,
+                                        8,
+                                    },
+                                    ec_gf8_mul_42_ops};
+
+static ec_gf_op_t ec_gf8_mul_43_ops[] = {
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_43 = {8,
+                                    {
+                                        2,
+                                        6,
+                                        4,
+                                        1,
+                                        7,
+                                        3,
+                                        0,
+                                        5,
+                                    },
+                                    ec_gf8_mul_43_ops};
+
+static ec_gf_op_t ec_gf8_mul_44_ops[] = {
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_44 = {8,
+                                    {
+                                        2,
+                                        3,
+                                        4,
+                                        1,
+                                        6,
+                                        5,
+                                        0,
+                                        7,
+                                    },
+                                    ec_gf8_mul_44_ops};
+
+static ec_gf_op_t ec_gf8_mul_45_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_45 = {8,
+                                    {
+                                        2,
+                                        3,
+                                        0,
+                                        1,
+                                        7,
+                                        4,
+                                        5,
+                                        6,
+                                    },
+                                    ec_gf8_mul_45_ops};
+
+static ec_gf_op_t ec_gf8_mul_46_ops[] = {
+    {EC_GF_OP_XOR3, 8, 2, 4}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 8, 0, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 1, 8, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_46 = {9,
+                                    {
+                                        2,
+                                        0,
+                                        1,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        8,
+                                    },
+                                    ec_gf8_mul_46_ops};
+
+static ec_gf_op_t ec_gf8_mul_47_ops[] = {
+    {EC_GF_OP_XOR3, 8, 0, 1}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 3, 8, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 4, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_47 = {9,
+                                    {
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        8,
+                                    },
+                                    ec_gf8_mul_47_ops};
+
+static ec_gf_op_t ec_gf8_mul_48_ops[] = {
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_48 = {8,
+                                    {
+                                        4,
+                                        5,
+                                        6,
+                                        0,
+                                        1,
+                                        3,
+                                        7,
+                                        2,
+                                    },
+                                    ec_gf8_mul_48_ops};
+
+static ec_gf_op_t ec_gf8_mul_49_ops[] = {
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR3, 8, 0, 6},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 7, 8, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR3, 1, 8, 5},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_49 = {9,
+                                    {
+                                        7,
+                                        2,
+                                        4,
+                                        0,
+                                        3,
+                                        5,
+                                        1,
+                                        6,
+                                        8,
+                                    },
+                                    ec_gf8_mul_49_ops};
+
+static ec_gf_op_t ec_gf8_mul_4A_ops[] = {
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_4A = {8,
+                                    {
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        3,
+                                        4,
+                                        2,
+                                    },
+                                    ec_gf8_mul_4A_ops};
+
+static ec_gf_op_t ec_gf8_mul_4B_ops[] = {
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR3, 8, 3, 7}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 4, 8, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 5, 8, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_4B = {9,
+                                    {
+                                        5,
+                                        3,
+                                        6,
+                                        7,
+                                        0,
+                                        2,
+                                        4,
+                                        1,
+                                        8,
+                                    },
+                                    ec_gf8_mul_4B_ops};
+
+static ec_gf_op_t ec_gf8_mul_4C_ops[] = {
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_4C = {8,
+                                    {
+                                        5,
+                                        3,
+                                        4,
+                                        7,
+                                        0,
+                                        6,
+                                        2,
+                                        1,
+                                    },
+                                    ec_gf8_mul_4C_ops};
+
+static ec_gf_op_t ec_gf8_mul_4D_ops[] = {
+    {EC_GF_OP_COPY, 8, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR3, 9, 3, 1},
+    {EC_GF_OP_XOR2, 5, 9, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR3, 0, 8, 2},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_4D = {10,
+                                    {
+                                        0,
+                                        9,
+                                        3,
+                                        5,
+                                        6,
+                                        4,
+                                        7,
+                                        1,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_4D_ops};
+
+static ec_gf_op_t ec_gf8_mul_4E_ops[] = {
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_4E = {8,
+                                    {
+                                        2,
+                                        3,
+                                        0,
+                                        1,
+                                        5,
+                                        6,
+                                        7,
+                                        4,
+                                    },
+                                    ec_gf8_mul_4E_ops};
+
+static ec_gf_op_t ec_gf8_mul_4F_ops[] = {
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_4F = {8,
+                                    {
+                                        0,
+                                        3,
+                                        5,
+                                        6,
+                                        1,
+                                        2,
+                                        7,
+                                        4,
+                                    },
+                                    ec_gf8_mul_4F_ops};
+
+static ec_gf_op_t ec_gf8_mul_50_ops[] = {
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_50 = {8,
+                                    {
+                                        4,
+                                        5,
+                                        7,
+                                        3,
+                                        0,
+                                        1,
+                                        2,
+                                        6,
+                                    },
+                                    ec_gf8_mul_50_ops};
+
+static ec_gf_op_t ec_gf8_mul_51_ops[] = {
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_51 = {8,
+                                    {
+                                        0,
+                                        1,
+                                        7,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                    },
+                                    ec_gf8_mul_51_ops};
+
+static ec_gf_op_t ec_gf8_mul_52_ops[] = {
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_COPY, 9, 4, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR3, 3, 5, 8},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 2, 9, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_52 = {10,
+                                    {
+                                        2,
+                                        3,
+                                        1,
+                                        4,
+                                        6,
+                                        7,
+                                        0,
+                                        5,
+                                        8,
+                                        9,
+                                    },
+                                    ec_gf8_mul_52_ops};
+
+static ec_gf_op_t ec_gf8_mul_53_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_53 = {8,
+                                    {
+                                        2,
+                                        0,
+                                        1,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        3,
+                                    },
+                                    ec_gf8_mul_53_ops};
+
+static ec_gf_op_t ec_gf8_mul_54_ops[] = {
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_54 = {8,
+                                    {
+                                        7,
+                                        3,
+                                        0,
+                                        4,
+                                        2,
+                                        6,
+                                        5,
+                                        1,
+                                    },
+                                    ec_gf8_mul_54_ops};
+
+static ec_gf_op_t ec_gf8_mul_55_ops[] = {
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_55 = {8,
+                                    {
+                                        1,
+                                        5,
+                                        6,
+                                        4,
+                                        3,
+                                        7,
+                                        2,
+                                        0,
+                                    },
+                                    ec_gf8_mul_55_ops};
+
+static ec_gf_op_t ec_gf8_mul_56_ops[] = {
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_56 = {8,
+                                    {
+                                        2,
+                                        3,
+                                        0,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        1,
+                                    },
+                                    ec_gf8_mul_56_ops};
+
+static ec_gf_op_t ec_gf8_mul_57_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_57 = {8,
+                                    {
+                                        2,
+                                        3,
+                                        0,
+                                        1,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                    },
+                                    ec_gf8_mul_57_ops};
+
+static ec_gf_op_t ec_gf8_mul_58_ops[] = {
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_58 = {8,
+                                    {
+                                        4,
+                                        3,
+                                        2,
+                                        7,
+                                        0,
+                                        1,
+                                        5,
+                                        6,
+                                    },
+                                    ec_gf8_mul_58_ops};
+
+static ec_gf_op_t ec_gf8_mul_59_ops[] = {
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_59 = {8,
+                                    {
+                                        7,
+                                        3,
+                                        5,
+                                        6,
+                                        1,
+                                        2,
+                                        0,
+                                        4,
+                                    },
+                                    ec_gf8_mul_59_ops};
+
+static ec_gf_op_t ec_gf8_mul_5A_ops[] = {
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_5A = {8,
+                                    {
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        5,
+                                        4,
+                                    },
+                                    ec_gf8_mul_5A_ops};
+
+static ec_gf_op_t ec_gf8_mul_5B_ops[] = {
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_5B = {8,
+                                    {
+                                        6,
+                                        0,
+                                        7,
+                                        5,
+                                        2,
+                                        1,
+                                        3,
+                                        4,
+                                    },
+                                    ec_gf8_mul_5B_ops};
+
+static ec_gf_op_t ec_gf8_mul_5C_ops[] = {
+    {EC_GF_OP_COPY, 8, 3, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 2, 8, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_5C = {9,
+                                    {
+                                        7,
+                                        5,
+                                        2,
+                                        4,
+                                        1,
+                                        0,
+                                        6,
+                                        3,
+                                        8,
+                                    },
+                                    ec_gf8_mul_5C_ops};
+
+static ec_gf_op_t ec_gf8_mul_5D_ops[] = {
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_5D = {8,
+                                    {
+                                        1,
+                                        3,
+                                        5,
+                                        4,
+                                        6,
+                                        7,
+                                        2,
+                                        0,
+                                    },
+                                    ec_gf8_mul_5D_ops};
+
+static ec_gf_op_t ec_gf8_mul_5E_ops[] = {
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_5E = {8,
+                                    {
+                                        4,
+                                        3,
+                                        6,
+                                        2,
+                                        5,
+                                        7,
+                                        0,
+                                        1,
+                                    },
+                                    ec_gf8_mul_5E_ops};
+
+static ec_gf_op_t ec_gf8_mul_5F_ops[] = {
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_5F = {8,
+                                    {
+                                        6,
+                                        1,
+                                        3,
+                                        4,
+                                        5,
+                                        7,
+                                        2,
+                                        0,
+                                    },
+                                    ec_gf8_mul_5F_ops};
+
+static ec_gf_op_t ec_gf8_mul_60_ops[] = {
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_60 = {8,
+                                    {
+                                        2,
+                                        3,
+                                        4,
+                                        7,
+                                        5,
+                                        6,
+                                        0,
+                                        1,
+                                    },
+                                    ec_gf8_mul_60_ops};
+
+static ec_gf_op_t ec_gf8_mul_61_ops[] = {
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_61 = {8,
+                                    {
+                                        0,
+                                        5,
+                                        6,
+                                        7,
+                                        4,
+                                        2,
+                                        1,
+                                        3,
+                                    },
+                                    ec_gf8_mul_61_ops};
+
+static ec_gf_op_t ec_gf8_mul_62_ops[] = {
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_62 = {8,
+                                    {
+                                        2,
+                                        0,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        1,
+                                    },
+                                    ec_gf8_mul_62_ops};
+
+static ec_gf_op_t ec_gf8_mul_63_ops[] = {
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_63 = {8,
+                                    {
+                                        3,
+                                        4,
+                                        6,
+                                        5,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_63_ops};
+
+static ec_gf_op_t ec_gf8_mul_64_ops[] = {
+    {EC_GF_OP_COPY, 8, 1, 0}, {EC_GF_OP_XOR2, 8, 0, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 8, 7, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_64 = {9,
+                                    {
+                                        2,
+                                        3,
+                                        4,
+                                        6,
+                                        5,
+                                        7,
+                                        8,
+                                        1,
+                                        0,
+                                    },
+                                    ec_gf8_mul_64_ops};
+
+static ec_gf_op_t ec_gf8_mul_65_ops[] = {
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_65 = {8,
+                                    {
+                                        2,
+                                        5,
+                                        1,
+                                        3,
+                                        4,
+                                        0,
+                                        6,
+                                        7,
+                                    },
+                                    ec_gf8_mul_65_ops};
+
+static ec_gf_op_t ec_gf8_mul_66_ops[] = {
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_66 = {8,
+                                    {
+                                        2,
+                                        3,
+                                        1,
+                                        4,
+                                        5,
+                                        7,
+                                        0,
+                                        6,
+                                    },
+                                    ec_gf8_mul_66_ops};
+
+static ec_gf_op_t ec_gf8_mul_67_ops[] = {
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_67 = {8,
+                                    {
+                                        2,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        3,
+                                        1,
+                                        0,
+                                    },
+                                    ec_gf8_mul_67_ops};
+
+static ec_gf_op_t ec_gf8_mul_68_ops[] = {
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_68 = {8,
+                                    {
+                                        5,
+                                        7,
+                                        2,
+                                        3,
+                                        0,
+                                        6,
+                                        4,
+                                        1,
+                                    },
+                                    ec_gf8_mul_68_ops};
+
+static ec_gf_op_t ec_gf8_mul_69_ops[] = {
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_69 = {8,
+                                    {
+                                        0,
+                                        1,
+                                        3,
+                                        2,
+                                        4,
+                                        5,
+                                        7,
+                                        6,
+                                    },
+                                    ec_gf8_mul_69_ops};
+
+static ec_gf_op_t ec_gf8_mul_6A_ops[] = {
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_6A = {8,
+                                    {
+                                        5,
+                                        7,
+                                        4,
+                                        6,
+                                        1,
+                                        2,
+                                        0,
+                                        3,
+                                    },
+                                    ec_gf8_mul_6A_ops};
+
+static ec_gf_op_t ec_gf8_mul_6B_ops[] = {
+    {EC_GF_OP_COPY, 8, 1, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 0, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_6B = {9,
+                                    {
+                                        6,
+                                        7,
+                                        2,
+                                        0,
+                                        3,
+                                        1,
+                                        5,
+                                        4,
+                                        8,
+                                    },
+                                    ec_gf8_mul_6B_ops};
+
+static ec_gf_op_t ec_gf8_mul_6C_ops[] = {
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_6C = {8,
+                                    {
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                    },
+                                    ec_gf8_mul_6C_ops};
+
+static ec_gf_op_t ec_gf8_mul_6D_ops[] = {
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR3, 8, 3, 4}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 0, 8, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 6, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_6D = {9,
+                                    {
+                                        3,
+                                        6,
+                                        7,
+                                        0,
+                                        4,
+                                        5,
+                                        1,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_6D_ops};
+
+static ec_gf_op_t ec_gf8_mul_6E_ops[] = {
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_6E = {8,
+                                    {
+                                        5,
+                                        6,
+                                        3,
+                                        1,
+                                        7,
+                                        2,
+                                        0,
+                                        4,
+                                    },
+                                    ec_gf8_mul_6E_ops};
+
+static ec_gf_op_t ec_gf8_mul_6F_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR3, 0, 8, 7}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_6F = {9,
+                                    {
+                                        2,
+                                        6,
+                                        3,
+                                        7,
+                                        0,
+                                        1,
+                                        4,
+                                        5,
+                                        8,
+                                    },
+                                    ec_gf8_mul_6F_ops};
+
+static ec_gf_op_t ec_gf8_mul_70_ops[] = {
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_70 = {8,
+                                    {
+                                        3,
+                                        4,
+                                        5,
+                                        2,
+                                        6,
+                                        0,
+                                        1,
+                                        7,
+                                    },
+                                    ec_gf8_mul_70_ops};
+
+static ec_gf_op_t ec_gf8_mul_71_ops[] = {
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_71 = {8,
+                                    {
+                                        4,
+                                        7,
+                                        5,
+                                        3,
+                                        6,
+                                        0,
+                                        2,
+                                        1,
+                                    },
+                                    ec_gf8_mul_71_ops};
+
+static ec_gf_op_t ec_gf8_mul_72_ops[] = {
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_72 = {8,
+                                    {
+                                        0,
+                                        5,
+                                        2,
+                                        7,
+                                        4,
+                                        1,
+                                        3,
+                                        6,
+                                    },
+                                    ec_gf8_mul_72_ops};
+
+static ec_gf_op_t ec_gf8_mul_73_ops[] = {
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_73 = {8,
+                                    {
+                                        6,
+                                        0,
+                                        1,
+                                        7,
+                                        4,
+                                        5,
+                                        2,
+                                        3,
+                                    },
+                                    ec_gf8_mul_73_ops};
+
+static ec_gf_op_t ec_gf8_mul_74_ops[] = {
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_74 = {8,
+                                    {
+                                        3,
+                                        2,
+                                        1,
+                                        0,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                    },
+                                    ec_gf8_mul_74_ops};
+
+static ec_gf_op_t ec_gf8_mul_75_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_75 = {8,
+                                    {
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                    },
+                                    ec_gf8_mul_75_ops};
+
+static ec_gf_op_t ec_gf8_mul_76_ops[] = {
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR3, 8, 6, 2},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 0, 8, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_76 = {9,
+                                    {
+                                        2,
+                                        3,
+                                        0,
+                                        6,
+                                        5,
+                                        1,
+                                        7,
+                                        8,
+                                        4,
+                                    },
+                                    ec_gf8_mul_76_ops};
+
+static ec_gf_op_t ec_gf8_mul_77_ops[] = {
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_77 = {8,
+                                    {
+                                        7,
+                                        4,
+                                        3,
+                                        6,
+                                        0,
+                                        1,
+                                        5,
+                                        2,
+                                    },
+                                    ec_gf8_mul_77_ops};
+
+static ec_gf_op_t ec_gf8_mul_78_ops[] = {
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR3, 8, 0, 2},
+    {EC_GF_OP_XOR2, 4, 8, 0}, {EC_GF_OP_XOR2, 1, 8, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_78 = {9,
+                                    {
+                                        4,
+                                        7,
+                                        3,
+                                        2,
+                                        5,
+                                        1,
+                                        6,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_78_ops};
+
+static ec_gf_op_t ec_gf8_mul_79_ops[] = {
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR3, 8, 4, 7},
+    {EC_GF_OP_XOR2, 0, 8, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 6, 8, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_79 = {9,
+                                    {
+                                        4,
+                                        5,
+                                        7,
+                                        3,
+                                        1,
+                                        6,
+                                        2,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_79_ops};
+
+static ec_gf_op_t ec_gf8_mul_7A_ops[] = {
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_7A = {8,
+                                    {
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                    },
+                                    ec_gf8_mul_7A_ops};
+
+static ec_gf_op_t ec_gf8_mul_7B_ops[] = {
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR3, 8, 5, 3},
+    {EC_GF_OP_XOR2, 8, 0, 0}, {EC_GF_OP_COPY, 9, 4, 0},
+    {EC_GF_OP_XOR2, 8, 2, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 4, 8, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR3, 4, 1, 9},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_7B = {10,
+                                    {
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        8,
+                                        5,
+                                        6,
+                                        0,
+                                        7,
+                                        9,
+                                    },
+                                    ec_gf8_mul_7B_ops};
+
+static ec_gf_op_t ec_gf8_mul_7C_ops[] = {
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_7C = {8,
+                                    {
+                                        2,
+                                        4,
+                                        1,
+                                        6,
+                                        3,
+                                        5,
+                                        7,
+                                        0,
+                                    },
+                                    ec_gf8_mul_7C_ops};
+
+static ec_gf_op_t ec_gf8_mul_7D_ops[] = {
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_7D = {8,
+                                    {
+                                        1,
+                                        0,
+                                        3,
+                                        5,
+                                        6,
+                                        7,
+                                        2,
+                                        4,
+                                    },
+                                    ec_gf8_mul_7D_ops};
+
+static ec_gf_op_t ec_gf8_mul_7E_ops[] = {
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_COPY, 8, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR3, 6, 2, 7},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 6, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_7E = {9,
+                                    {
+                                        5,
+                                        1,
+                                        2,
+                                        0,
+                                        7,
+                                        3,
+                                        4,
+                                        6,
+                                        8,
+                                    },
+                                    ec_gf8_mul_7E_ops};
+
+static ec_gf_op_t ec_gf8_mul_7F_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR3, 9, 7, 5}, {EC_GF_OP_XOR2, 2, 9, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 6, 9, 0},
+    {EC_GF_OP_XOR3, 9, 6, 4}, {EC_GF_OP_XOR2, 7, 9, 0},
+    {EC_GF_OP_XOR2, 3, 9, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 7, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_7F = {10,
+                                    {
+                                        4,
+                                        1,
+                                        0,
+                                        5,
+                                        6,
+                                        7,
+                                        2,
+                                        3,
+                                        8,
+                                        9,
+                                    },
+                                    ec_gf8_mul_7F_ops};
+
+static ec_gf_op_t ec_gf8_mul_80_ops[] = {
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_80 = {8,
+                                    {
+                                        7,
+                                        5,
+                                        6,
+                                        4,
+                                        1,
+                                        2,
+                                        3,
+                                        0,
+                                    },
+                                    ec_gf8_mul_80_ops};
+
+static ec_gf_op_t ec_gf8_mul_81_ops[] = {
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_81 = {8,
+                                    {
+                                        2,
+                                        7,
+                                        4,
+                                        1,
+                                        5,
+                                        6,
+                                        3,
+                                        0,
+                                    },
+                                    ec_gf8_mul_81_ops};
+
+static ec_gf_op_t ec_gf8_mul_82_ops[] = {
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_COPY, 8, 6, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR3, 5, 8, 7}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_82 = {9,
+                                    {
+                                        6,
+                                        2,
+                                        7,
+                                        5,
+                                        1,
+                                        3,
+                                        4,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_82_ops};
+
+static ec_gf_op_t ec_gf8_mul_83_ops[] = {
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_83 = {8,
+                                    {
+                                        3,
+                                        5,
+                                        6,
+                                        7,
+                                        1,
+                                        2,
+                                        4,
+                                        0,
+                                    },
+                                    ec_gf8_mul_83_ops};
+
+static ec_gf_op_t ec_gf8_mul_84_ops[] = {
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_84 = {8,
+                                    {
+                                        7,
+                                        6,
+                                        0,
+                                        4,
+                                        1,
+                                        5,
+                                        3,
+                                        2,
+                                    },
+                                    ec_gf8_mul_84_ops};
+
+static ec_gf_op_t ec_gf8_mul_85_ops[] = {
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_85 = {8,
+                                    {
+                                        7,
+                                        6,
+                                        0,
+                                        3,
+                                        2,
+                                        4,
+                                        5,
+                                        1,
+                                    },
+                                    ec_gf8_mul_85_ops};
+
+static ec_gf_op_t ec_gf8_mul_86_ops[] = {
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_86 = {8,
+                                    {
+                                        1,
+                                        2,
+                                        6,
+                                        4,
+                                        5,
+                                        7,
+                                        3,
+                                        0,
+                                    },
+                                    ec_gf8_mul_86_ops};
+
+static ec_gf_op_t ec_gf8_mul_87_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_COPY, 8, 1, 0},
+    {EC_GF_OP_XOR2, 8, 6, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR3, 5, 8, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 2, 8, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_87 = {9,
+                                    {
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        7,
+                                        6,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_87_ops};
+
+static ec_gf_op_t ec_gf8_mul_88_ops[] = {
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_88 = {8,
+                                    {
+                                        6,
+                                        7,
+                                        3,
+                                        1,
+                                        2,
+                                        4,
+                                        5,
+                                        0,
+                                    },
+                                    ec_gf8_mul_88_ops};
+
+static ec_gf_op_t ec_gf8_mul_89_ops[] = {
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR3, 8, 5, 2},
+    {EC_GF_OP_XOR2, 4, 8, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 0, 8, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_89 = {9,
+                                    {
+                                        2,
+                                        1,
+                                        6,
+                                        5,
+                                        7,
+                                        3,
+                                        4,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_89_ops};
+
+static ec_gf_op_t ec_gf8_mul_8A_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_8A = {8,
+                                    {
+                                        1,
+                                        2,
+                                        3,
+                                        0,
+                                        6,
+                                        7,
+                                        4,
+                                        5,
+                                    },
+                                    ec_gf8_mul_8A_ops};
+
+static ec_gf_op_t ec_gf8_mul_8B_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_8B = {8,
+                                    {
+                                        6,
+                                        1,
+                                        2,
+                                        3,
+                                        5,
+                                        7,
+                                        4,
+                                        0,
+                                    },
+                                    ec_gf8_mul_8B_ops};
+
+static ec_gf_op_t ec_gf8_mul_8C_ops[] = {
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_8C = {8,
+                                    {
+                                        1,
+                                        2,
+                                        0,
+                                        7,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                    },
+                                    ec_gf8_mul_8C_ops};
+
+static ec_gf_op_t ec_gf8_mul_8D_ops[] = {
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_8D = {8,
+                                    {
+                                        7,
+                                        1,
+                                        3,
+                                        2,
+                                        4,
+                                        5,
+                                        0,
+                                        6,
+                                    },
+                                    ec_gf8_mul_8D_ops};
+
+static ec_gf_op_t ec_gf8_mul_8E_ops[] = {{EC_GF_OP_XOR2, 2, 0, 0},
+                                         {EC_GF_OP_XOR2, 3, 0, 0},
+                                         {EC_GF_OP_XOR2, 4, 0, 0},
+                                         {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_8E = {8,
+                                    {
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                    },
+                                    ec_gf8_mul_8E_ops};
+
+static ec_gf_op_t ec_gf8_mul_8F_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_8F = {8,
+                                    {
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                    },
+                                    ec_gf8_mul_8F_ops};
+
+static ec_gf_op_t ec_gf8_mul_90_ops[] = {
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_90 = {8,
+                                    {
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        3,
+                                        2,
+                                    },
+                                    ec_gf8_mul_90_ops};
+
+static ec_gf_op_t ec_gf8_mul_91_ops[] = {
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_COPY, 9, 1, 0}, {EC_GF_OP_COPY, 8, 3, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 7, 9, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR3, 5, 8, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_91 = {10,
+                                    {
+                                        2,
+                                        3,
+                                        1,
+                                        4,
+                                        0,
+                                        6,
+                                        7,
+                                        5,
+                                        8,
+                                        9,
+                                    },
+                                    ec_gf8_mul_91_ops};
+
+static ec_gf_op_t ec_gf8_mul_92_ops[] = {
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_92 = {8,
+                                    {
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        5,
+                                        4,
+                                    },
+                                    ec_gf8_mul_92_ops};
+
+static ec_gf_op_t ec_gf8_mul_93_ops[] = {
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_93 = {8,
+                                    {
+                                        6,
+                                        4,
+                                        5,
+                                        1,
+                                        7,
+                                        2,
+                                        3,
+                                        0,
+                                    },
+                                    ec_gf8_mul_93_ops};
+
+static ec_gf_op_t ec_gf8_mul_94_ops[] = {
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_94 = {8,
+                                    {
+                                        7,
+                                        5,
+                                        0,
+                                        2,
+                                        6,
+                                        1,
+                                        3,
+                                        4,
+                                    },
+                                    ec_gf8_mul_94_ops};
+
+static ec_gf_op_t ec_gf8_mul_95_ops[] = {
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_95 = {8,
+                                    {
+                                        7,
+                                        6,
+                                        1,
+                                        3,
+                                        0,
+                                        4,
+                                        5,
+                                        2,
+                                    },
+                                    ec_gf8_mul_95_ops};
+
+static ec_gf_op_t ec_gf8_mul_96_ops[] = {
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR3, 8, 0, 4}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 7, 8, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 8, 3, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 1, 8, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 5, 8, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_96 = {9,
+                                    {
+                                        4,
+                                        0,
+                                        1,
+                                        6,
+                                        7,
+                                        2,
+                                        3,
+                                        5,
+                                        8,
+                                    },
+                                    ec_gf8_mul_96_ops};
+
+static ec_gf_op_t ec_gf8_mul_97_ops[] = {
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_COPY, 8, 2, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 8, 6, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 1, 8, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 5, 8, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_97 = {9,
+                                    {
+                                        4,
+                                        5,
+                                        3,
+                                        6,
+                                        7,
+                                        1,
+                                        2,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_97_ops};
+
+static ec_gf_op_t ec_gf8_mul_98_ops[] = {
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_98 = {8,
+                                    {
+                                        4,
+                                        2,
+                                        3,
+                                        6,
+                                        7,
+                                        5,
+                                        1,
+                                        0,
+                                    },
+                                    ec_gf8_mul_98_ops};
+
+static ec_gf_op_t ec_gf8_mul_99_ops[] = {
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_99 = {8,
+                                    {
+                                        6,
+                                        5,
+                                        3,
+                                        7,
+                                        0,
+                                        1,
+                                        4,
+                                        2,
+                                    },
+                                    ec_gf8_mul_99_ops};
+
+static ec_gf_op_t ec_gf8_mul_9A_ops[] = {
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR3, 8, 4, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 7, 8, 0}, {EC_GF_OP_XOR2, 1, 8, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_9A = {9,
+                                    {
+                                        6,
+                                        3,
+                                        4,
+                                        0,
+                                        5,
+                                        1,
+                                        2,
+                                        7,
+                                        8,
+                                    },
+                                    ec_gf8_mul_9A_ops};
+
+static ec_gf_op_t ec_gf8_mul_9B_ops[] = {
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_COPY, 9, 5, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR3, 8, 3, 2}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 3, 9, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_9B = {10,
+                                    {
+                                        4,
+                                        5,
+                                        8,
+                                        6,
+                                        7,
+                                        1,
+                                        2,
+                                        0,
+                                        3,
+                                        9,
+                                    },
+                                    ec_gf8_mul_9B_ops};
+
+static ec_gf_op_t ec_gf8_mul_9C_ops[] = {
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_9C = {8,
+                                    {
+                                        3,
+                                        2,
+                                        1,
+                                        0,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                    },
+                                    ec_gf8_mul_9C_ops};
+
+static ec_gf_op_t ec_gf8_mul_9D_ops[] = {
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_9D = {8,
+                                    {
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        7,
+                                        4,
+                                        5,
+                                        6,
+                                    },
+                                    ec_gf8_mul_9D_ops};
+
+static ec_gf_op_t ec_gf8_mul_9E_ops[] = {
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_COPY, 8, 7, 0},
+    {EC_GF_OP_XOR2, 8, 5, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 0, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_9E = {9,
+                                    {
+                                        4,
+                                        5,
+                                        3,
+                                        8,
+                                        6,
+                                        0,
+                                        2,
+                                        7,
+                                        1,
+                                    },
+                                    ec_gf8_mul_9E_ops};
+
+static ec_gf_op_t ec_gf8_mul_9F_ops[] = {
+    {EC_GF_OP_XOR3, 8, 1, 2}, {EC_GF_OP_XOR2, 8, 3, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 6, 8, 0}, {EC_GF_OP_XOR2, 5, 8, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_9F = {9,
+                                    {
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        8,
+                                    },
+                                    ec_gf8_mul_9F_ops};
+
+static ec_gf_op_t ec_gf8_mul_A0_ops[] = {
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_A0 = {8,
+                                    {
+                                        3,
+                                        1,
+                                        6,
+                                        7,
+                                        5,
+                                        2,
+                                        4,
+                                        0,
+                                    },
+                                    ec_gf8_mul_A0_ops};
+
+static ec_gf_op_t ec_gf8_mul_A1_ops[] = {
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR3, 8, 0, 6},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 7, 8, 0},
+    {EC_GF_OP_XOR2, 3, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_A1 = {9,
+                                    {
+                                        7,
+                                        4,
+                                        1,
+                                        5,
+                                        6,
+                                        0,
+                                        2,
+                                        3,
+                                        8,
+                                    },
+                                    ec_gf8_mul_A1_ops};
+
+static ec_gf_op_t ec_gf8_mul_A2_ops[] = {
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_A2 = {8,
+                                    {
+                                        7,
+                                        0,
+                                        6,
+                                        3,
+                                        2,
+                                        1,
+                                        4,
+                                        5,
+                                    },
+                                    ec_gf8_mul_A2_ops};
+
+static ec_gf_op_t ec_gf8_mul_A3_ops[] = {
+    {EC_GF_OP_COPY, 8, 2, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 3, 8, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_A3 = {9,
+                                    {
+                                        3,
+                                        7,
+                                        2,
+                                        6,
+                                        1,
+                                        4,
+                                        0,
+                                        5,
+                                        8,
+                                    },
+                                    ec_gf8_mul_A3_ops};
+
+static ec_gf_op_t ec_gf8_mul_A4_ops[] = {
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_A4 = {8,
+                                    {
+                                        5,
+                                        6,
+                                        7,
+                                        2,
+                                        4,
+                                        3,
+                                        0,
+                                        1,
+                                    },
+                                    ec_gf8_mul_A4_ops};
+
+static ec_gf_op_t ec_gf8_mul_A5_ops[] = {
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR3, 8, 5, 6}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 4, 8, 0}, {EC_GF_OP_XOR2, 7, 8, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_A5 = {9,
+                                    {
+                                        1,
+                                        4,
+                                        2,
+                                        5,
+                                        6,
+                                        7,
+                                        3,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_A5_ops};
+
+static ec_gf_op_t ec_gf8_mul_A6_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_A6 = {8,
+                                    {
+                                        1,
+                                        2,
+                                        0,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                    },
+                                    ec_gf8_mul_A6_ops};
+
+static ec_gf_op_t ec_gf8_mul_A7_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_A7 = {8,
+                                    {
+                                        0,
+                                        1,
+                                        2,
+                                        5,
+                                        6,
+                                        7,
+                                        3,
+                                        4,
+                                    },
+                                    ec_gf8_mul_A7_ops};
+
+static ec_gf_op_t ec_gf8_mul_A8_ops[] = {
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 8, 1, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_COPY, 9, 4, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 8, 3, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 2, 9, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_A8 = {10,
+                                    {
+                                        1,
+                                        7,
+                                        5,
+                                        8,
+                                        6,
+                                        3,
+                                        4,
+                                        0,
+                                        2,
+                                        9,
+                                    },
+                                    ec_gf8_mul_A8_ops};
+
+static ec_gf_op_t ec_gf8_mul_A9_ops[] = {
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_A9 = {8,
+                                    {
+                                        3,
+                                        7,
+                                        6,
+                                        1,
+                                        2,
+                                        0,
+                                        4,
+                                        5,
+                                    },
+                                    ec_gf8_mul_A9_ops};
+
+static ec_gf_op_t ec_gf8_mul_AA_ops[] = {
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_AA = {8,
+                                    {
+                                        0,
+                                        4,
+                                        5,
+                                        3,
+                                        6,
+                                        7,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_AA_ops};
+
+static ec_gf_op_t ec_gf8_mul_AB_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_COPY, 9, 6, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 8, 7, 0}, {EC_GF_OP_XOR2, 3, 8, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR3, 3, 9, 7},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_AB = {10,
+                                    {
+                                        2,
+                                        3,
+                                        8,
+                                        0,
+                                        5,
+                                        6,
+                                        1,
+                                        4,
+                                        7,
+                                        9,
+                                    },
+                                    ec_gf8_mul_AB_ops};
+
+static ec_gf_op_t ec_gf8_mul_AC_ops[] = {
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_AC = {8,
+                                    {
+                                        3,
+                                        2,
+                                        1,
+                                        0,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                    },
+                                    ec_gf8_mul_AC_ops};
+
+static ec_gf_op_t ec_gf8_mul_AD_ops[] = {
+    {EC_GF_OP_XOR3, 8, 1, 2}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 4, 8, 0},
+    {EC_GF_OP_XOR2, 5, 8, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_AD = {9,
+                                    {
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_AD_ops};
+
+static ec_gf_op_t ec_gf8_mul_AE_ops[] = {
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_COPY, 8, 5, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 4, 8, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_AE = {9,
+                                    {
+                                        7,
+                                        0,
+                                        5,
+                                        6,
+                                        3,
+                                        4,
+                                        1,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_AE_ops};
+
+static ec_gf_op_t ec_gf8_mul_AF_ops[] = {
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_AF = {8,
+                                    {
+                                        0,
+                                        1,
+                                        2,
+                                        7,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                    },
+                                    ec_gf8_mul_AF_ops};
+
+static ec_gf_op_t ec_gf8_mul_B0_ops[] = {
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_B0 = {8,
+                                    {
+                                        4,
+                                        0,
+                                        7,
+                                        2,
+                                        3,
+                                        1,
+                                        6,
+                                        5,
+                                    },
+                                    ec_gf8_mul_B0_ops};
+
+static ec_gf_op_t ec_gf8_mul_B1_ops[] = {
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_COPY, 8, 4, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR3, 5, 8, 1}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_B1 = {9,
+                                    {
+                                        2,
+                                        6,
+                                        4,
+                                        7,
+                                        0,
+                                        1,
+                                        3,
+                                        5,
+                                        8,
+                                    },
+                                    ec_gf8_mul_B1_ops};
+
+static ec_gf_op_t ec_gf8_mul_B2_ops[] = {
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR3, 8, 4, 5},
+    {EC_GF_OP_XOR2, 2, 8, 0}, {EC_GF_OP_XOR2, 8, 1, 0},
+    {EC_GF_OP_XOR2, 7, 8, 0}, {EC_GF_OP_XOR2, 3, 8, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_B2 = {9,
+                                    {
+                                        0,
+                                        7,
+                                        4,
+                                        5,
+                                        6,
+                                        1,
+                                        2,
+                                        3,
+                                        8,
+                                    },
+                                    ec_gf8_mul_B2_ops};
+
+static ec_gf_op_t ec_gf8_mul_B3_ops[] = {
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_COPY, 9, 5, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR3, 8, 6, 4},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 8, 5, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 7, 8, 0},
+    {EC_GF_OP_XOR2, 0, 8, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR3, 1, 9, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_B3 = {10,
+                                    {
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        1,
+                                        6,
+                                        0,
+                                        7,
+                                        8,
+                                        9,
+                                    },
+                                    ec_gf8_mul_B3_ops};
+
+static ec_gf_op_t ec_gf8_mul_B4_ops[] = {
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_B4 = {8,
+                                    {
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                    },
+                                    ec_gf8_mul_B4_ops};
+
+static ec_gf_op_t ec_gf8_mul_B5_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_COPY, 8, 6, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR3, 4, 8, 3}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_B5 = {9,
+                                    {
+                                        3,
+                                        4,
+                                        0,
+                                        7,
+                                        1,
+                                        5,
+                                        6,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_B5_ops};
+
+static ec_gf_op_t ec_gf8_mul_B6_ops[] = {
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_B6 = {8,
+                                    {
+                                        5,
+                                        3,
+                                        6,
+                                        4,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_B6_ops};
+
+static ec_gf_op_t ec_gf8_mul_B7_ops[] = {
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_B7 = {8,
+                                    {
+                                        5,
+                                        0,
+                                        1,
+                                        4,
+                                        2,
+                                        6,
+                                        7,
+                                        3,
+                                    },
+                                    ec_gf8_mul_B7_ops};
+
+static ec_gf_op_t ec_gf8_mul_B8_ops[] = {
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_B8 = {8,
+                                    {
+                                        6,
+                                        4,
+                                        5,
+                                        1,
+                                        2,
+                                        0,
+                                        7,
+                                        3,
+                                    },
+                                    ec_gf8_mul_B8_ops};
+
+static ec_gf_op_t ec_gf8_mul_B9_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR3, 0, 8, 2}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_B9 = {9,
+                                    {
+                                        6,
+                                        7,
+                                        0,
+                                        2,
+                                        1,
+                                        4,
+                                        5,
+                                        3,
+                                        8,
+                                    },
+                                    ec_gf8_mul_B9_ops};
+
+static ec_gf_op_t ec_gf8_mul_BA_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_BA = {8,
+                                    {
+                                        1,
+                                        2,
+                                        4,
+                                        3,
+                                        5,
+                                        6,
+                                        0,
+                                        7,
+                                    },
+                                    ec_gf8_mul_BA_ops};
+
+static ec_gf_op_t ec_gf8_mul_BB_ops[] = {
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_COPY, 8, 3, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 8, 5, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 8, 7, 0}, {EC_GF_OP_XOR2, 2, 8, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_BB = {9,
+                                    {
+                                        7,
+                                        2,
+                                        1,
+                                        8,
+                                        3,
+                                        5,
+                                        6,
+                                        4,
+                                        0,
+                                    },
+                                    ec_gf8_mul_BB_ops};
+
+static ec_gf_op_t ec_gf8_mul_BC_ops[] = {
+    {EC_GF_OP_COPY, 8, 1, 0}, {EC_GF_OP_XOR2, 8, 2, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 7, 8, 0}, {EC_GF_OP_XOR3, 2, 8, 4},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_BC = {9,
+                                    {
+                                        2,
+                                        6,
+                                        3,
+                                        4,
+                                        5,
+                                        1,
+                                        7,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_BC_ops};
+
+static ec_gf_op_t ec_gf8_mul_BD_ops[] = {
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_BD = {8,
+                                    {
+                                        4,
+                                        5,
+                                        0,
+                                        2,
+                                        7,
+                                        1,
+                                        6,
+                                        3,
+                                    },
+                                    ec_gf8_mul_BD_ops};
+
+static ec_gf_op_t ec_gf8_mul_BE_ops[] = {
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_BE = {8,
+                                    {
+                                        0,
+                                        6,
+                                        7,
+                                        4,
+                                        5,
+                                        1,
+                                        3,
+                                        2,
+                                    },
+                                    ec_gf8_mul_BE_ops};
+
+static ec_gf_op_t ec_gf8_mul_BF_ops[] = {
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_BF = {8,
+                                    {
+                                        5,
+                                        6,
+                                        1,
+                                        7,
+                                        3,
+                                        0,
+                                        2,
+                                        4,
+                                    },
+                                    ec_gf8_mul_BF_ops};
+
+static ec_gf_op_t ec_gf8_mul_C0_ops[] = {
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_C0 = {8,
+                                    {
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        7,
+                                        5,
+                                        6,
+                                        0,
+                                    },
+                                    ec_gf8_mul_C0_ops};
+
+static ec_gf_op_t ec_gf8_mul_C1_ops[] = {
+    {EC_GF_OP_XOR3, 8, 1, 2}, {EC_GF_OP_XOR2, 8, 3, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 6, 8, 0}, {EC_GF_OP_XOR2, 5, 8, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_C1 = {9,
+                                    {
+                                        5,
+                                        6,
+                                        7,
+                                        4,
+                                        1,
+                                        2,
+                                        3,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_C1_ops};
+
+static ec_gf_op_t ec_gf8_mul_C2_ops[] = {
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_C2 = {8,
+                                    {
+                                        7,
+                                        6,
+                                        3,
+                                        0,
+                                        1,
+                                        4,
+                                        5,
+                                        2,
+                                    },
+                                    ec_gf8_mul_C2_ops};
+
+static ec_gf_op_t ec_gf8_mul_C3_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR3, 0, 2, 6}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR3, 9, 1, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 7, 9, 0},
+    {EC_GF_OP_XOR2, 3, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_C3 = {10,
+                                    {
+                                        5,
+                                        6,
+                                        4,
+                                        7,
+                                        1,
+                                        2,
+                                        3,
+                                        0,
+                                        8,
+                                        9,
+                                    },
+                                    ec_gf8_mul_C3_ops};
+
+static ec_gf_op_t ec_gf8_mul_C4_ops[] = {
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_C4 = {8,
+                                    {
+                                        0,
+                                        2,
+                                        1,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                    },
+                                    ec_gf8_mul_C4_ops};
+
+static ec_gf_op_t ec_gf8_mul_C5_ops[] = {
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_C5 = {8,
+                                    {
+                                        4,
+                                        3,
+                                        5,
+                                        7,
+                                        6,
+                                        2,
+                                        0,
+                                        1,
+                                    },
+                                    ec_gf8_mul_C5_ops};
+
+static ec_gf_op_t ec_gf8_mul_C6_ops[] = {
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_COPY, 8, 4, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR3, 9, 5, 4},
+    {EC_GF_OP_XOR2, 6, 9, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 7, 9, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 6, 8, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_C6 = {10,
+                                    {
+                                        6,
+                                        3,
+                                        0,
+                                        4,
+                                        5,
+                                        7,
+                                        2,
+                                        1,
+                                        8,
+                                        9,
+                                    },
+                                    ec_gf8_mul_C6_ops};
+
+static ec_gf_op_t ec_gf8_mul_C7_ops[] = {
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_C7 = {8,
+                                    {
+                                        7,
+                                        0,
+                                        6,
+                                        2,
+                                        5,
+                                        3,
+                                        4,
+                                        1,
+                                    },
+                                    ec_gf8_mul_C7_ops};
+
+static ec_gf_op_t ec_gf8_mul_C8_ops[] = {
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_C8 = {8,
+                                    {
+                                        1,
+                                        3,
+                                        2,
+                                        4,
+                                        6,
+                                        7,
+                                        5,
+                                        0,
+                                    },
+                                    ec_gf8_mul_C8_ops};
+
+static ec_gf_op_t ec_gf8_mul_C9_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_C9 = {8,
+                                    {
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                    },
+                                    ec_gf8_mul_C9_ops};
+
+static ec_gf_op_t ec_gf8_mul_CA_ops[] = {
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_CA = {8,
+                                    {
+                                        1,
+                                        2,
+                                        5,
+                                        7,
+                                        3,
+                                        4,
+                                        0,
+                                        6,
+                                    },
+                                    ec_gf8_mul_CA_ops};
+
+static ec_gf_op_t ec_gf8_mul_CB_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_CB = {8,
+                                    {
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        7,
+                                        6,
+                                        0,
+                                        1,
+                                    },
+                                    ec_gf8_mul_CB_ops};
+
+static ec_gf_op_t ec_gf8_mul_CC_ops[] = {
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_CC = {8,
+                                    {
+                                        2,
+                                        7,
+                                        1,
+                                        0,
+                                        5,
+                                        6,
+                                        3,
+                                        4,
+                                    },
+                                    ec_gf8_mul_CC_ops};
+
+static ec_gf_op_t ec_gf8_mul_CD_ops[] = {
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_CD = {8,
+                                    {
+                                        0,
+                                        6,
+                                        1,
+                                        2,
+                                        7,
+                                        3,
+                                        4,
+                                        5,
+                                    },
+                                    ec_gf8_mul_CD_ops};
+
+static ec_gf_op_t ec_gf8_mul_CE_ops[] = {
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_COPY, 8, 7, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR3, 3, 6, 8},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR3, 8, 2, 3},
+    {EC_GF_OP_XOR2, 1, 8, 0}, {EC_GF_OP_XOR2, 4, 8, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_CE = {9,
+                                    {
+                                        5,
+                                        7,
+                                        3,
+                                        0,
+                                        2,
+                                        6,
+                                        4,
+                                        1,
+                                        8,
+                                    },
+                                    ec_gf8_mul_CE_ops};
+
+static ec_gf_op_t ec_gf8_mul_CF_ops[] = {
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_CF = {8,
+                                    {
+                                        3,
+                                        6,
+                                        7,
+                                        0,
+                                        2,
+                                        4,
+                                        5,
+                                        1,
+                                    },
+                                    ec_gf8_mul_CF_ops};
+
+static ec_gf_op_t ec_gf8_mul_D0_ops[] = {
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_D0 = {8,
+                                    {
+                                        5,
+                                        6,
+                                        7,
+                                        2,
+                                        0,
+                                        3,
+                                        1,
+                                        4,
+                                    },
+                                    ec_gf8_mul_D0_ops};
+
+static ec_gf_op_t ec_gf8_mul_D1_ops[] = {
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR3, 8, 6, 0},
+    {EC_GF_OP_XOR2, 4, 8, 0}, {EC_GF_OP_XOR2, 1, 8, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_D1 = {9,
+                                    {
+                                        5,
+                                        6,
+                                        3,
+                                        2,
+                                        0,
+                                        7,
+                                        4,
+                                        1,
+                                        8,
+                                    },
+                                    ec_gf8_mul_D1_ops};
+
+static ec_gf_op_t ec_gf8_mul_D2_ops[] = {
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_D2 = {8,
+                                    {
+                                        7,
+                                        0,
+                                        2,
+                                        1,
+                                        3,
+                                        4,
+                                        6,
+                                        5,
+                                    },
+                                    ec_gf8_mul_D2_ops};
+
+static ec_gf_op_t ec_gf8_mul_D3_ops[] = {
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_COPY, 8, 4, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 8, 6, 0}, {EC_GF_OP_XOR2, 3, 8, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_D3 = {9,
+                                    {
+                                        0,
+                                        3,
+                                        2,
+                                        8,
+                                        4,
+                                        6,
+                                        7,
+                                        1,
+                                        5,
+                                    },
+                                    ec_gf8_mul_D3_ops};
+
+static ec_gf_op_t ec_gf8_mul_D4_ops[] = {
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_COPY, 8, 1, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR3, 1, 7, 8},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_D4 = {9,
+                                    {
+                                        4,
+                                        1,
+                                        7,
+                                        5,
+                                        0,
+                                        6,
+                                        3,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_D4_ops};
+
+static ec_gf_op_t ec_gf8_mul_D5_ops[] = {
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_D5 = {8,
+                                    {
+                                        6,
+                                        7,
+                                        4,
+                                        5,
+                                        2,
+                                        3,
+                                        1,
+                                        0,
+                                    },
+                                    ec_gf8_mul_D5_ops};
+
+static ec_gf_op_t ec_gf8_mul_D6_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 7, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_D6 = {9,
+                                    {
+                                        0,
+                                        6,
+                                        2,
+                                        7,
+                                        1,
+                                        3,
+                                        4,
+                                        5,
+                                        8,
+                                    },
+                                    ec_gf8_mul_D6_ops};
+
+static ec_gf_op_t ec_gf8_mul_D7_ops[] = {
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR3, 8, 3, 5}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 0, 8, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR3, 6, 7, 8}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_D7 = {9,
+                                    {
+                                        3,
+                                        4,
+                                        6,
+                                        5,
+                                        0,
+                                        7,
+                                        1,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_D7_ops};
+
+static ec_gf_op_t ec_gf8_mul_D8_ops[] = {
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_D8 = {8,
+                                    {
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                    },
+                                    ec_gf8_mul_D8_ops};
+
+static ec_gf_op_t ec_gf8_mul_D9_ops[] = {
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_D9 = {8,
+                                    {
+                                        1,
+                                        2,
+                                        6,
+                                        7,
+                                        4,
+                                        5,
+                                        0,
+                                        3,
+                                    },
+                                    ec_gf8_mul_D9_ops};
+
+static ec_gf_op_t ec_gf8_mul_DA_ops[] = {
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR3, 8, 2, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 1, 8, 0},
+    {EC_GF_OP_XOR2, 5, 8, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_DA = {9,
+                                    {
+                                        2,
+                                        5,
+                                        7,
+                                        1,
+                                        0,
+                                        4,
+                                        3,
+                                        6,
+                                        8,
+                                    },
+                                    ec_gf8_mul_DA_ops};
+
+static ec_gf_op_t ec_gf8_mul_DB_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 8, 4, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 3, 8, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_DB = {9,
+                                    {
+                                        7,
+                                        5,
+                                        6,
+                                        2,
+                                        3,
+                                        4,
+                                        1,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_DB_ops};
+
+static ec_gf_op_t ec_gf8_mul_DC_ops[] = {
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_DC = {8,
+                                    {
+                                        4,
+                                        5,
+                                        2,
+                                        6,
+                                        7,
+                                        1,
+                                        0,
+                                        3,
+                                    },
+                                    ec_gf8_mul_DC_ops};
+
+static ec_gf_op_t ec_gf8_mul_DD_ops[] = {
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_DD = {8,
+                                    {
+                                        1,
+                                        2,
+                                        3,
+                                        6,
+                                        7,
+                                        0,
+                                        4,
+                                        5,
+                                    },
+                                    ec_gf8_mul_DD_ops};
+
+static ec_gf_op_t ec_gf8_mul_DE_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_DE = {8,
+                                    {
+                                        0,
+                                        5,
+                                        2,
+                                        6,
+                                        7,
+                                        1,
+                                        3,
+                                        4,
+                                    },
+                                    ec_gf8_mul_DE_ops};
+
+static ec_gf_op_t ec_gf8_mul_DF_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 8, 3, 0},
+    {EC_GF_OP_COPY, 9, 0, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 8, 7, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 5, 8, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR3, 1, 9, 2}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_DF = {10,
+                                    {
+                                        7,
+                                        2,
+                                        8,
+                                        4,
+                                        3,
+                                        1,
+                                        0,
+                                        6,
+                                        5,
+                                        9,
+                                    },
+                                    ec_gf8_mul_DF_ops};
+
+static ec_gf_op_t ec_gf8_mul_E0_ops[] = {
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_E0 = {8,
+                                    {
+                                        2,
+                                        3,
+                                        4,
+                                        7,
+                                        5,
+                                        6,
+                                        0,
+                                        1,
+                                    },
+                                    ec_gf8_mul_E0_ops};
+
+static ec_gf_op_t ec_gf8_mul_E1_ops[] = {
+    {EC_GF_OP_COPY, 8, 1, 0}, {EC_GF_OP_XOR2, 8, 7, 0},
+    {EC_GF_OP_XOR2, 3, 8, 0}, {EC_GF_OP_XOR3, 9, 5, 3},
+    {EC_GF_OP_XOR2, 0, 9, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 4, 9, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 2, 8, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_E1 = {10,
+                                    {
+                                        0,
+                                        7,
+                                        1,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        2,
+                                        8,
+                                        9,
+                                    },
+                                    ec_gf8_mul_E1_ops};
+
+static ec_gf_op_t ec_gf8_mul_E2_ops[] = {
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_E2 = {8,
+                                    {
+                                        2,
+                                        3,
+                                        7,
+                                        1,
+                                        5,
+                                        6,
+                                        0,
+                                        4,
+                                    },
+                                    ec_gf8_mul_E2_ops};
+
+static ec_gf_op_t ec_gf8_mul_E3_ops[] = {
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR3, 8, 2, 7}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 0, 8, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR3, 6, 8, 4},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_E3 = {9,
+                                    {
+                                        5,
+                                        4,
+                                        7,
+                                        2,
+                                        1,
+                                        3,
+                                        6,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_E3_ops};
+
+static ec_gf_op_t ec_gf8_mul_E4_ops[] = {
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_E4 = {8,
+                                    {
+                                        7,
+                                        0,
+                                        1,
+                                        6,
+                                        3,
+                                        4,
+                                        2,
+                                        5,
+                                    },
+                                    ec_gf8_mul_E4_ops};
+
+static ec_gf_op_t ec_gf8_mul_E5_ops[] = {
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 3, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_E5 = {9,
+                                    {
+                                        4,
+                                        5,
+                                        3,
+                                        6,
+                                        7,
+                                        1,
+                                        0,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_E5_ops};
+
+static ec_gf_op_t ec_gf8_mul_E6_ops[] = {
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_E6 = {8,
+                                    {
+                                        5,
+                                        4,
+                                        3,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_E6_ops};
+
+static ec_gf_op_t ec_gf8_mul_E7_ops[] = {
+    {EC_GF_OP_COPY, 8, 6, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR3, 9, 0, 6}, {EC_GF_OP_XOR2, 4, 9, 0},
+    {EC_GF_OP_XOR2, 5, 9, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 7, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_E7 = {10,
+                                    {
+                                        1,
+                                        4,
+                                        3,
+                                        6,
+                                        7,
+                                        5,
+                                        2,
+                                        0,
+                                        8,
+                                        9,
+                                    },
+                                    ec_gf8_mul_E7_ops};
+
+static ec_gf_op_t ec_gf8_mul_E8_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_E8 = {8,
+                                    {
+                                        1,
+                                        4,
+                                        2,
+                                        7,
+                                        3,
+                                        0,
+                                        5,
+                                        6,
+                                    },
+                                    ec_gf8_mul_E8_ops};
+
+static ec_gf_op_t ec_gf8_mul_E9_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_COPY, 8, 1, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR3, 1, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_E9 = {9,
+                                    {
+                                        6,
+                                        2,
+                                        0,
+                                        3,
+                                        4,
+                                        1,
+                                        5,
+                                        7,
+                                        8,
+                                    },
+                                    ec_gf8_mul_E9_ops};
+
+static ec_gf_op_t ec_gf8_mul_EA_ops[] = {
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_EA = {8,
+                                    {
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_EA_ops};
+
+static ec_gf_op_t ec_gf8_mul_EB_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_EB = {8,
+                                    {
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_EB_ops};
+
+static ec_gf_op_t ec_gf8_mul_EC_ops[] = {
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR3, 8, 4, 0}, {EC_GF_OP_XOR2, 1, 8, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 3, 8, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_EC = {9,
+                                    {
+                                        7,
+                                        4,
+                                        3,
+                                        0,
+                                        2,
+                                        5,
+                                        1,
+                                        6,
+                                        8,
+                                    },
+                                    ec_gf8_mul_EC_ops};
+
+static ec_gf_op_t ec_gf8_mul_ED_ops[] = {
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_ED = {8,
+                                    {
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        4,
+                                        3,
+                                        2,
+                                    },
+                                    ec_gf8_mul_ED_ops};
+
+static ec_gf_op_t ec_gf8_mul_EE_ops[] = {
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR3, 8, 2, 3},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 4, 8, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 8, 5, 0},
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 1, 8, 0}, {EC_GF_OP_XOR2, 7, 8, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_EE = {9,
+                                    {
+                                        6,
+                                        4,
+                                        5,
+                                        7,
+                                        2,
+                                        3,
+                                        0,
+                                        1,
+                                        8,
+                                    },
+                                    ec_gf8_mul_EE_ops};
+
+static ec_gf_op_t ec_gf8_mul_EF_ops[] = {
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_COPY, 8, 0, 0},
+    {EC_GF_OP_XOR2, 8, 2, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 7, 8, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 6, 8, 0},
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_EF = {9,
+                                    {
+                                        6,
+                                        4,
+                                        5,
+                                        7,
+                                        2,
+                                        0,
+                                        3,
+                                        1,
+                                        8,
+                                    },
+                                    ec_gf8_mul_EF_ops};
+
+static ec_gf_op_t ec_gf8_mul_F0_ops[] = {
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR3, 8, 3, 6},
+    {EC_GF_OP_XOR2, 5, 8, 0}, {EC_GF_OP_XOR2, 8, 4, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 7, 8, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 1, 8, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_F0 = {9,
+                                    {
+                                        3,
+                                        4,
+                                        6,
+                                        1,
+                                        2,
+                                        0,
+                                        5,
+                                        7,
+                                        8,
+                                    },
+                                    ec_gf8_mul_F0_ops};
+
+static ec_gf_op_t ec_gf8_mul_F1_ops[] = {
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_COPY, 8, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_COPY, 9, 2, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 9, 0, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 7, 9, 0}, {EC_GF_OP_XOR2, 4, 9, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR3, 9, 8, 7},
+    {EC_GF_OP_XOR2, 1, 9, 0}, {EC_GF_OP_XOR2, 5, 9, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_F1 = {10,
+                                    {
+                                        7,
+                                        2,
+                                        6,
+                                        3,
+                                        5,
+                                        1,
+                                        4,
+                                        0,
+                                        8,
+                                        9,
+                                    },
+                                    ec_gf8_mul_F1_ops};
+
+static ec_gf_op_t ec_gf8_mul_F2_ops[] = {
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR3, 8, 6, 4},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 3, 8, 0}, {EC_GF_OP_XOR2, 5, 8, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_F2 = {9,
+                                    {
+                                        1,
+                                        0,
+                                        6,
+                                        7,
+                                        4,
+                                        5,
+                                        2,
+                                        3,
+                                        8,
+                                    },
+                                    ec_gf8_mul_F2_ops};
+
+static ec_gf_op_t ec_gf8_mul_F3_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_F3 = {8,
+                                    {
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                    },
+                                    ec_gf8_mul_F3_ops};
+
+static ec_gf_op_t ec_gf8_mul_F4_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_F4 = {8,
+                                    {
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                    },
+                                    ec_gf8_mul_F4_ops};
+
+static ec_gf_op_t ec_gf8_mul_F5_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_F5 = {8,
+                                    {
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                    },
+                                    ec_gf8_mul_F5_ops};
+
+static ec_gf_op_t ec_gf8_mul_F6_ops[] = {
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_COPY, 8, 3, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_COPY, 9, 3, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 9, 4, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 6, 9, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR3, 7, 8, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_F6 = {10,
+                                    {
+                                        0,
+                                        6,
+                                        2,
+                                        7,
+                                        4,
+                                        3,
+                                        5,
+                                        9,
+                                        1,
+                                        8,
+                                    },
+                                    ec_gf8_mul_F6_ops};
+
+static ec_gf_op_t ec_gf8_mul_F7_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_F7 = {8,
+                                    {
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                    },
+                                    ec_gf8_mul_F7_ops};
+
+static ec_gf_op_t ec_gf8_mul_F8_ops[] = {
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_F8 = {8,
+                                    {
+                                        6,
+                                        2,
+                                        0,
+                                        1,
+                                        4,
+                                        5,
+                                        3,
+                                        7,
+                                    },
+                                    ec_gf8_mul_F8_ops};
+
+static ec_gf_op_t ec_gf8_mul_F9_ops[] = {
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR3, 8, 7, 1}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 4, 8, 0}, {EC_GF_OP_XOR2, 5, 8, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_F9 = {9,
+                                    {
+                                        4,
+                                        1,
+                                        7,
+                                        6,
+                                        0,
+                                        3,
+                                        5,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_F9_ops};
+
+static ec_gf_op_t ec_gf8_mul_FA_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_FA = {8,
+                                    {
+                                        0,
+                                        1,
+                                        2,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        3,
+                                    },
+                                    ec_gf8_mul_FA_ops};
+
+static ec_gf_op_t ec_gf8_mul_FB_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_FB = {8,
+                                    {
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                    },
+                                    ec_gf8_mul_FB_ops};
+
+static ec_gf_op_t ec_gf8_mul_FC_ops[] = {
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_COPY, 9, 3, 0},
+    {EC_GF_OP_XOR3, 8, 5, 7}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 8, 3, 0}, {EC_GF_OP_XOR2, 2, 8, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR3, 0, 9, 2}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_FC = {10,
+                                    {
+                                        5,
+                                        6,
+                                        3,
+                                        7,
+                                        1,
+                                        8,
+                                        0,
+                                        4,
+                                        2,
+                                        9,
+                                    },
+                                    ec_gf8_mul_FC_ops};
+
+static ec_gf_op_t ec_gf8_mul_FD_ops[] = {
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_COPY, 8, 7, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR3, 1, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_FD = {9,
+                                    {
+                                        5,
+                                        3,
+                                        7,
+                                        6,
+                                        1,
+                                        2,
+                                        4,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_FD_ops};
+
+static ec_gf_op_t ec_gf8_mul_FE_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_COPY, 8, 2, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 8, 5, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 7, 8, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_FE = {9,
+                                    {
+                                        3,
+                                        4,
+                                        8,
+                                        2,
+                                        5,
+                                        0,
+                                        6,
+                                        1,
+                                        7,
+                                    },
+                                    ec_gf8_mul_FE_ops};
+
+static ec_gf_op_t ec_gf8_mul_FF_ops[] = {
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_COPY, 9, 0, 0},
+    {EC_GF_OP_COPY, 8, 4, 0}, {EC_GF_OP_XOR2, 9, 1, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 9, 4, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 3, 9, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR3, 3, 8, 5}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_FF = {10,
+                                    {
+                                        6,
+                                        5,
+                                        0,
+                                        1,
+                                        2,
+                                        4,
+                                        9,
+                                        3,
+                                        7,
+                                        8,
+                                    },
+                                    ec_gf8_mul_FF_ops};
+
+ec_gf_mul_t *ec_gf8_mul[] = {
+    &ec_gf8_mul_00, &ec_gf8_mul_01, &ec_gf8_mul_02, &ec_gf8_mul_03,
+    &ec_gf8_mul_04, &ec_gf8_mul_05, &ec_gf8_mul_06, &ec_gf8_mul_07,
+    &ec_gf8_mul_08, &ec_gf8_mul_09, &ec_gf8_mul_0A, &ec_gf8_mul_0B,
+    &ec_gf8_mul_0C, &ec_gf8_mul_0D, &ec_gf8_mul_0E, &ec_gf8_mul_0F,
+    &ec_gf8_mul_10, &ec_gf8_mul_11, &ec_gf8_mul_12, &ec_gf8_mul_13,
+    &ec_gf8_mul_14, &ec_gf8_mul_15, &ec_gf8_mul_16, &ec_gf8_mul_17,
+    &ec_gf8_mul_18, &ec_gf8_mul_19, &ec_gf8_mul_1A, &ec_gf8_mul_1B,
+    &ec_gf8_mul_1C, &ec_gf8_mul_1D, &ec_gf8_mul_1E, &ec_gf8_mul_1F,
+    &ec_gf8_mul_20, &ec_gf8_mul_21, &ec_gf8_mul_22, &ec_gf8_mul_23,
+    &ec_gf8_mul_24, &ec_gf8_mul_25, &ec_gf8_mul_26, &ec_gf8_mul_27,
+    &ec_gf8_mul_28, &ec_gf8_mul_29, &ec_gf8_mul_2A, &ec_gf8_mul_2B,
+    &ec_gf8_mul_2C, &ec_gf8_mul_2D, &ec_gf8_mul_2E, &ec_gf8_mul_2F,
+    &ec_gf8_mul_30, &ec_gf8_mul_31, &ec_gf8_mul_32, &ec_gf8_mul_33,
+    &ec_gf8_mul_34, &ec_gf8_mul_35, &ec_gf8_mul_36, &ec_gf8_mul_37,
+    &ec_gf8_mul_38, &ec_gf8_mul_39, &ec_gf8_mul_3A, &ec_gf8_mul_3B,
+    &ec_gf8_mul_3C, &ec_gf8_mul_3D, &ec_gf8_mul_3E, &ec_gf8_mul_3F,
+    &ec_gf8_mul_40, &ec_gf8_mul_41, &ec_gf8_mul_42, &ec_gf8_mul_43,
+    &ec_gf8_mul_44, &ec_gf8_mul_45, &ec_gf8_mul_46, &ec_gf8_mul_47,
+    &ec_gf8_mul_48, &ec_gf8_mul_49, &ec_gf8_mul_4A, &ec_gf8_mul_4B,
+    &ec_gf8_mul_4C, &ec_gf8_mul_4D, &ec_gf8_mul_4E, &ec_gf8_mul_4F,
+    &ec_gf8_mul_50, &ec_gf8_mul_51, &ec_gf8_mul_52, &ec_gf8_mul_53,
+    &ec_gf8_mul_54, &ec_gf8_mul_55, &ec_gf8_mul_56, &ec_gf8_mul_57,
+    &ec_gf8_mul_58, &ec_gf8_mul_59, &ec_gf8_mul_5A, &ec_gf8_mul_5B,
+    &ec_gf8_mul_5C, &ec_gf8_mul_5D, &ec_gf8_mul_5E, &ec_gf8_mul_5F,
+    &ec_gf8_mul_60, &ec_gf8_mul_61, &ec_gf8_mul_62, &ec_gf8_mul_63,
+    &ec_gf8_mul_64, &ec_gf8_mul_65, &ec_gf8_mul_66, &ec_gf8_mul_67,
+    &ec_gf8_mul_68, &ec_gf8_mul_69, &ec_gf8_mul_6A, &ec_gf8_mul_6B,
+    &ec_gf8_mul_6C, &ec_gf8_mul_6D, &ec_gf8_mul_6E, &ec_gf8_mul_6F,
+    &ec_gf8_mul_70, &ec_gf8_mul_71, &ec_gf8_mul_72, &ec_gf8_mul_73,
+    &ec_gf8_mul_74, &ec_gf8_mul_75, &ec_gf8_mul_76, &ec_gf8_mul_77,
+    &ec_gf8_mul_78, &ec_gf8_mul_79, &ec_gf8_mul_7A, &ec_gf8_mul_7B,
+    &ec_gf8_mul_7C, &ec_gf8_mul_7D, &ec_gf8_mul_7E, &ec_gf8_mul_7F,
+    &ec_gf8_mul_80, &ec_gf8_mul_81, &ec_gf8_mul_82, &ec_gf8_mul_83,
+    &ec_gf8_mul_84, &ec_gf8_mul_85, &ec_gf8_mul_86, &ec_gf8_mul_87,
+    &ec_gf8_mul_88, &ec_gf8_mul_89, &ec_gf8_mul_8A, &ec_gf8_mul_8B,
+    &ec_gf8_mul_8C, &ec_gf8_mul_8D, &ec_gf8_mul_8E, &ec_gf8_mul_8F,
+    &ec_gf8_mul_90, &ec_gf8_mul_91, &ec_gf8_mul_92, &ec_gf8_mul_93,
+    &ec_gf8_mul_94, &ec_gf8_mul_95, &ec_gf8_mul_96, &ec_gf8_mul_97,
+    &ec_gf8_mul_98, &ec_gf8_mul_99, &ec_gf8_mul_9A, &ec_gf8_mul_9B,
+    &ec_gf8_mul_9C, &ec_gf8_mul_9D, &ec_gf8_mul_9E, &ec_gf8_mul_9F,
+    &ec_gf8_mul_A0, &ec_gf8_mul_A1, &ec_gf8_mul_A2, &ec_gf8_mul_A3,
+    &ec_gf8_mul_A4, &ec_gf8_mul_A5, &ec_gf8_mul_A6, &ec_gf8_mul_A7,
+    &ec_gf8_mul_A8, &ec_gf8_mul_A9, &ec_gf8_mul_AA, &ec_gf8_mul_AB,
+    &ec_gf8_mul_AC, &ec_gf8_mul_AD, &ec_gf8_mul_AE, &ec_gf8_mul_AF,
+    &ec_gf8_mul_B0, &ec_gf8_mul_B1, &ec_gf8_mul_B2, &ec_gf8_mul_B3,
+    &ec_gf8_mul_B4, &ec_gf8_mul_B5, &ec_gf8_mul_B6, &ec_gf8_mul_B7,
+    &ec_gf8_mul_B8, &ec_gf8_mul_B9, &ec_gf8_mul_BA, &ec_gf8_mul_BB,
+    &ec_gf8_mul_BC, &ec_gf8_mul_BD, &ec_gf8_mul_BE, &ec_gf8_mul_BF,
+    &ec_gf8_mul_C0, &ec_gf8_mul_C1, &ec_gf8_mul_C2, &ec_gf8_mul_C3,
+    &ec_gf8_mul_C4, &ec_gf8_mul_C5, &ec_gf8_mul_C6, &ec_gf8_mul_C7,
+    &ec_gf8_mul_C8, &ec_gf8_mul_C9, &ec_gf8_mul_CA, &ec_gf8_mul_CB,
+    &ec_gf8_mul_CC, &ec_gf8_mul_CD, &ec_gf8_mul_CE, &ec_gf8_mul_CF,
+    &ec_gf8_mul_D0, &ec_gf8_mul_D1, &ec_gf8_mul_D2, &ec_gf8_mul_D3,
+    &ec_gf8_mul_D4, &ec_gf8_mul_D5, &ec_gf8_mul_D6, &ec_gf8_mul_D7,
+    &ec_gf8_mul_D8, &ec_gf8_mul_D9, &ec_gf8_mul_DA, &ec_gf8_mul_DB,
+    &ec_gf8_mul_DC, &ec_gf8_mul_DD, &ec_gf8_mul_DE, &ec_gf8_mul_DF,
+    &ec_gf8_mul_E0, &ec_gf8_mul_E1, &ec_gf8_mul_E2, &ec_gf8_mul_E3,
+    &ec_gf8_mul_E4, &ec_gf8_mul_E5, &ec_gf8_mul_E6, &ec_gf8_mul_E7,
+    &ec_gf8_mul_E8, &ec_gf8_mul_E9, &ec_gf8_mul_EA, &ec_gf8_mul_EB,
+    &ec_gf8_mul_EC, &ec_gf8_mul_ED, &ec_gf8_mul_EE, &ec_gf8_mul_EF,
+    &ec_gf8_mul_F0, &ec_gf8_mul_F1, &ec_gf8_mul_F2, &ec_gf8_mul_F3,
+    &ec_gf8_mul_F4, &ec_gf8_mul_F5, &ec_gf8_mul_F6, &ec_gf8_mul_F7,
+    &ec_gf8_mul_F8, &ec_gf8_mul_F9, &ec_gf8_mul_FA, &ec_gf8_mul_FB,
+    &ec_gf8_mul_FC, &ec_gf8_mul_FD, &ec_gf8_mul_FE, &ec_gf8_mul_FF};
diff --git a/xlators/cluster/ec/src/ec-gf8.h b/xlators/cluster/ec/src/ec-gf8.h
new file mode 100644
index 00000000000..4aca91127fc
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-gf8.h
@@ -0,0 +1,18 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_GF8_H__
+#define __EC_GF8_H__
+
+#include "ec-galois.h"
+
+extern ec_gf_mul_t *ec_gf8_mul[];
+
+#endif /* __EC_GF8_H__ */
diff --git a/xlators/cluster/ec/src/ec-heal.c b/xlators/cluster/ec/src/ec-heal.c
new file mode 100644
index 00000000000..7d991f04aac
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-heal.c
@@ -0,0 +1,3367 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/defaults.h>
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/byte-order.h>
+#include <glusterfs/syncop.h>
+#include <glusterfs/syncop-utils.h>
+#include <glusterfs/cluster-syncop.h>
+
+#include "ec.h"
+#include "ec-types.h"
+#include "ec-messages.h"
+#include "ec-helpers.h"
+#include "ec-common.h"
+#include "ec-combine.h"
+#include "ec-method.h"
+#include "ec-fops.h"
+#include "ec-heald.h"
+
+#define EC_COUNT(array, max)                                                   \
+    ({                                                                         \
+        int __i;                                                               \
+        int __res = 0;                                                         \
+        for (__i = 0; __i < max; __i++)                                        \
+            if (array[__i])                                                    \
+                __res++;                                                       \
+        __res;                                                                 \
+    })
+#define EC_INTERSECT(dst, src1, src2, max)                                     \
+    ({                                                                         \
+        int __i;                                                               \
+        for (__i = 0; __i < max; __i++)                                        \
+            dst[__i] = src1[__i] && src2[__i];                                 \
+    })
+#define EC_ADJUST_SOURCE(source, sources, max)                                 \
+    ({                                                                         \
+        int __i;                                                               \
+        if (sources[source] == 0) {                                            \
+            source = -1;                                                       \
+            for (__i = 0; __i < max; __i++)                                    \
+                if (sources[__i])                                              \
+                    source = __i;                                              \
+        }                                                                      \
+    })
+#define IA_EQUAL(f, s, field)                                                  \
+    (memcmp(&(f.ia_##field), &(s.ia_##field), sizeof(s.ia_##field)) == 0)
+#define EC_REPLIES_ALLOC(replies, numsubvols)                                  \
+    do {                                                                       \
+        int __i = 0;                                                           \
+        replies = alloca0(numsubvols * sizeof(*replies));                      \
+        for (__i = 0; __i < numsubvols; __i++)                                 \
+            INIT_LIST_HEAD(&replies[__i].entries.list);                        \
+    } while (0)
+
+struct ec_name_data {
+    call_frame_t *frame;
+    unsigned char *participants;
+    unsigned char *failed_on;
+    unsigned char *gfidless;
+    unsigned char *enoent;
+    unsigned char *same;
+    char *name;
+    inode_t *parent;
+    default_args_cbk_t *replies;
+    uint32_t heal_pending;
+};
+
+static char *ec_ignore_xattrs[] = {GF_SELINUX_XATTR_KEY, QUOTA_SIZE_KEY, NULL};
+
+static gf_boolean_t
+ec_ignorable_key_match(dict_t *dict, char *key, data_t *val, void *mdata)
+{
+    int i = 0;
+
+    if (!key)
+        goto out;
+
+    if (strncmp(key, EC_XATTR_PREFIX, SLEN(EC_XATTR_PREFIX)) == 0)
+        return _gf_true;
+
+    for (i = 0; ec_ignore_xattrs[i]; i++) {
+        if (!strcmp(key, ec_ignore_xattrs[i]))
+            return _gf_true;
+    }
+
+out:
+    return _gf_false;
+}
+
+static gf_boolean_t
+ec_sh_key_match(dict_t *dict, char *key, data_t *val, void *mdata)
+{
+    return !ec_ignorable_key_match(dict, key, val, mdata);
+}
+/* FOP: heal */
+
+void
+ec_set_entry_healing(ec_fop_data_t *fop)
+{
+    ec_inode_t *ctx = NULL;
+    loc_t *loc = NULL;
+
+    if (!fop)
+        return;
+
+    loc = &fop->loc[0];
+    LOCK(&loc->inode->lock);
+    {
+        ctx = __ec_inode_get(loc->inode, fop->xl);
+        if (ctx) {
+            ctx->heal_count += 1;
+        }
+    }
+    UNLOCK(&loc->inode->lock);
+}
+
+void
+ec_reset_entry_healing(ec_fop_data_t *fop)
+{
+    ec_inode_t *ctx = NULL;
+    loc_t *loc = NULL;
+    int32_t heal_count = 0;
+    if (!fop)
+        return;
+
+    loc = &fop->loc[0];
+    LOCK(&loc->inode->lock);
+    {
+        ctx = __ec_inode_get(loc->inode, fop->xl);
+        if (ctx) {
+            ctx->heal_count += -1;
+            heal_count = ctx->heal_count;
+        }
+    }
+    UNLOCK(&loc->inode->lock);
+    GF_ASSERT(heal_count >= 0);
+}
+
+uintptr_t
+ec_heal_check(ec_fop_data_t *fop, uintptr_t *pgood)
+{
+    ec_cbk_data_t *cbk;
+    uintptr_t mask[2] = {0, 0};
+
+    list_for_each_entry(cbk, &fop->cbk_list, list)
+    {
+        mask[cbk->op_ret >= 0] |= cbk->mask;
+    }
+
+    if (pgood != NULL) {
+        *pgood = mask[1];
+    }
+
+    return mask[0];
+}
+
+void
+ec_heal_update(ec_fop_data_t *fop, int32_t is_open)
+{
+    ec_heal_t *heal = fop->data;
+    uintptr_t good, bad;
+
+    bad = ec_heal_check(fop, &good);
+
+    LOCK(&heal->lock);
+
+    heal->bad &= ~bad;
+    if (is_open) {
+        heal->open |= good;
+    }
+
+    UNLOCK(&heal->lock);
+
+    fop->error = 0;
+}
+
+void
+ec_heal_avoid(ec_fop_data_t *fop)
+{
+    ec_heal_t *heal = fop->data;
+    uintptr_t bad;
+
+    bad = ec_heal_check(fop, NULL);
+
+    LOCK(&heal->lock);
+
+    heal->good &= ~bad;
+
+    UNLOCK(&heal->lock);
+}
+
+int32_t
+ec_heal_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = cookie;
+    ec_heal_t *heal = fop->data;
+
+    if (op_ret >= 0) {
+        GF_ASSERT(
+            ec_set_inode_size(heal->fop, heal->fd->inode, heal->total_size));
+    }
+
+    return 0;
+}
+
+void
+ec_heal_lock(ec_heal_t *heal, int32_t type, fd_t *fd, loc_t *loc, off_t offset,
+             size_t size)
+{
+    struct gf_flock flock;
+    fop_inodelk_cbk_t cbk = NULL;
+
+    flock.l_type = type;
+    flock.l_whence = SEEK_SET;
+    flock.l_start = offset;
+    flock.l_len = size;
+    flock.l_pid = 0;
+    flock.l_owner.len = 0;
+
+    if (type == F_UNLCK) {
+        /* Remove inode size information before unlocking it. */
+        if (fd == NULL) {
+            ec_clear_inode_info(heal->fop, heal->loc.inode);
+        } else {
+            ec_clear_inode_info(heal->fop, heal->fd->inode);
+        }
+        cbk = ec_lock_unlocked;
+    } else {
+        /* Otherwise use the callback to update size information. */
+        cbk = ec_heal_lock_cbk;
+    }
+
+    if (fd != NULL) {
+        ec_finodelk(heal->fop->frame, heal->xl,
+                    &heal->fop->frame->root->lk_owner, heal->fop->mask,
+                    EC_MINIMUM_ALL, cbk, heal, heal->xl->name, fd, F_SETLKW,
+                    &flock, NULL);
+    } else {
+        ec_inodelk(heal->fop->frame, heal->xl,
+                   &heal->fop->frame->root->lk_owner, heal->fop->mask,
+                   EC_MINIMUM_ALL, cbk, heal, heal->xl->name, loc, F_SETLKW,
+                   &flock, NULL);
+    }
+}
+
+void
+ec_heal_inodelk(ec_heal_t *heal, int32_t type, int32_t use_fd, off_t offset,
+                size_t size)
+{
+    ec_heal_lock(heal, type, use_fd ? heal->fd : NULL, &heal->loc, offset,
+                 size);
+}
+
+int32_t
+ec_heal_xattr_clean(dict_t *dict, char *key, data_t *data, void *arg)
+{
+    dict_t *base = arg;
+
+    if (ec_ignorable_key_match(NULL, key, NULL, NULL)) {
+        dict_del(dict, key);
+        return 0;
+    }
+
+    if (dict_get(base, key) != NULL)
+        dict_del(dict, key);
+
+    return 0;
+}
+
+/********************************************************************
+ * ec_wind_xattrop_parallel:
+ *              Helper function to update the extended attributes
+ *    in parallel.
+ *
+ *******************************************************************/
+void
+ec_wind_xattrop_parallel(call_frame_t *frame, xlator_t *subvol, int child_index,
+                         loc_t *loc, gf_xattrop_flags_t flags, dict_t **dict,
+                         dict_t *xdata)
+{
+    gf_msg_debug("EC", 0, "WIND: on child %d ", child_index);
+    STACK_WIND_COOKIE(
+        frame, cluster_xattrop_cbk, (void *)(uintptr_t)child_index, subvol,
+        subvol->fops->xattrop, loc, flags, dict[child_index], xdata);
+}
+
+int32_t
+ec_heal_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                   struct iatt *postbuf, dict_t *xdata)
+{
+    ec_fop_data_t *fop = cookie;
+    ec_heal_t *heal = fop->data;
+
+    ec_trace("WRITE_CBK", cookie, "ret=%d, errno=%d", op_ret, op_errno);
+
+    gf_msg_debug(fop->xl->name, 0,
+                 "%s: write op_ret %d, op_errno %s"
+                 " at %" PRIu64,
+                 uuid_utoa(heal->fd->inode->gfid), op_ret, strerror(op_errno),
+                 heal->offset);
+
+    ec_heal_update(cookie, 0);
+
+    return 0;
+}
+
+int32_t
+ec_heal_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct iovec *vector,
+                  int32_t count, struct iatt *stbuf, struct iobref *iobref,
+                  dict_t *xdata)
+{
+    ec_fop_data_t *fop = cookie;
+    ec_heal_t *heal = fop->data;
+
+    ec_trace("READ_CBK", fop, "ret=%d, errno=%d", op_ret, op_errno);
+
+    ec_heal_avoid(fop);
+
+    if (op_ret > 0) {
+        gf_msg_debug(fop->xl->name, 0,
+                     "%s: read succeeded, proceeding "
+                     "to write at %" PRIu64,
+                     uuid_utoa(heal->fd->inode->gfid), heal->offset);
+        ec_writev(heal->fop->frame, heal->xl, heal->bad, EC_MINIMUM_ONE,
+                  ec_heal_writev_cbk, heal, heal->fd, vector, count,
+                  heal->offset, 0, iobref, NULL);
+    } else {
+        if (op_ret < 0) {
+            gf_msg_debug(fop->xl->name, 0,
+                         "%s: read failed %s, failing "
+                         "to heal block at %" PRIu64,
+                         uuid_utoa(heal->fd->inode->gfid), strerror(op_errno),
+                         heal->offset);
+            heal->bad = 0;
+        }
+        heal->done = 1;
+    }
+
+    return 0;
+}
+
+void
+ec_heal_data_block(ec_heal_t *heal)
+{
+    ec_trace("DATA", heal->fop, "good=%lX, bad=%lX", heal->good, heal->bad);
+
+    if ((heal->good != 0) && (heal->bad != 0) &&
+        (heal->iatt.ia_type == IA_IFREG)) {
+        ec_readv(heal->fop->frame, heal->xl, heal->good, EC_MINIMUM_MIN,
+                 ec_heal_readv_cbk, heal, heal->fd, heal->size, heal->offset, 0,
+                 NULL);
+    }
+}
+
+/* FOP: fheal */
+
+void
+ec_fheal(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_fheal_cbk_t func, void *data, fd_t *fd,
+         int32_t partial, dict_t *xdata)
+{
+    ec_fd_t *ctx = ec_fd_get(fd, this);
+
+    if (ctx != NULL) {
+        gf_msg_trace("ec", 0, "FHEAL ctx: flags=%X, open=%" PRIXPTR, ctx->flags,
+                     ctx->open);
+        ec_heal(frame, this, target, fop_flags, func, data, &ctx->loc, partial,
+                xdata);
+    }
+}
+
+/* Common heal code */
+void
+ec_mask_to_char_array(uintptr_t mask, unsigned char *array, int numsubvols)
+{
+    int i = 0;
+
+    for (i = 0; i < numsubvols; i++)
+        array[i] = ((mask >> i) & 1);
+}
+
+uintptr_t
+ec_char_array_to_mask(unsigned char *array, int numsubvols)
+{
+    int i = 0;
+    uintptr_t mask = 0;
+
+    if (array == NULL)
+        goto out;
+
+    for (i = 0; i < numsubvols; i++)
+        if (array[i])
+            mask |= (1ULL << i);
+out:
+    return mask;
+}
+
+int
+ec_heal_entry_find_direction(ec_t *ec, default_args_cbk_t *replies,
+                             uint64_t *versions, uint64_t *dirty,
+                             unsigned char *sources,
+                             unsigned char *healed_sinks)
+{
+    uint64_t xattr[EC_VERSION_SIZE] = {0};
+    int source = -1;
+    uint64_t max_version = 0;
+    int ret = 0;
+    int i = 0;
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (!replies[i].valid)
+            continue;
+
+        if (replies[i].op_ret == -1)
+            continue;
+
+        if (source == -1)
+            source = i;
+
+        ret = ec_dict_get_array(replies[i].xdata, EC_XATTR_VERSION, xattr,
+                                EC_VERSION_SIZE);
+        if (ret == 0) {
+            versions[i] = xattr[EC_DATA_TXN];
+            if (max_version < versions[i]) {
+                max_version = versions[i];
+                source = i;
+            }
+        }
+
+        memset(xattr, 0, sizeof(xattr));
+        ret = ec_dict_get_array(replies[i].xdata, EC_XATTR_DIRTY, xattr,
+                                EC_VERSION_SIZE);
+        if (ret == 0) {
+            dirty[i] = xattr[EC_DATA_TXN];
+        }
+    }
+
+    if (source < 0)
+        goto out;
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (!replies[i].valid)
+            continue;
+
+        if (replies[i].op_ret == -1)
+            continue;
+
+        if (versions[i] == versions[source])
+            sources[i] = 1;
+        else
+            healed_sinks[i] = 1;
+    }
+
+out:
+    return source;
+}
+
+int
+ec_adjust_versions(call_frame_t *frame, ec_t *ec, ec_txn_t type, inode_t *inode,
+                   int source, unsigned char *sources,
+                   unsigned char *healed_sinks, uint64_t *versions,
+                   uint64_t *dirty)
+{
+    int i = 0;
+    int ret = 0;
+    int call_count = 0;
+    dict_t **xattr = NULL;
+    int op_ret = 0;
+    loc_t loc = {0};
+    gf_boolean_t erase_dirty = _gf_false;
+    uint64_t *versions_xattr = NULL;
+    uint64_t *dirty_xattr = NULL;
+    uint64_t allzero[2] = {0};
+    unsigned char *on = NULL;
+    unsigned char *output = NULL;
+    default_args_cbk_t *replies = NULL;
+
+    /* Allocate the required memory */
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+    on = alloca0(ec->nodes);
+    output = alloca0(ec->nodes);
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    xattr = GF_CALLOC(ec->nodes, sizeof(*xattr), gf_common_mt_pointer);
+    if (!xattr) {
+        op_ret = -ENOMEM;
+        goto out;
+    }
+    for (i = 0; i < ec->nodes; i++) {
+        xattr[i] = dict_new();
+        if (!xattr[i]) {
+            op_ret = -ENOMEM;
+            goto out;
+        }
+    }
+
+    /* dirty xattr represents if the file/dir needs heal. Unless all the
+     * copies are healed, don't erase it */
+    if (EC_COUNT(sources, ec->nodes) + EC_COUNT(healed_sinks, ec->nodes) ==
+        ec->nodes)
+        erase_dirty = _gf_true;
+    else
+        op_ret = -ENOTCONN;
+
+    /* Populate the xattr array */
+    for (i = 0; i < ec->nodes; i++) {
+        if (!sources[i] && !healed_sinks[i])
+            continue;
+        versions_xattr = GF_CALLOC(EC_VERSION_SIZE, sizeof(*versions_xattr),
+                                   gf_common_mt_pointer);
+        if (!versions_xattr) {
+            op_ret = -ENOMEM;
+            continue;
+        }
+
+        versions_xattr[type] = hton64(versions[source] - versions[i]);
+        ret = dict_set_bin(xattr[i], EC_XATTR_VERSION, versions_xattr,
+                           (sizeof(*versions_xattr) * EC_VERSION_SIZE));
+        if (ret < 0) {
+            op_ret = -ENOMEM;
+            continue;
+        }
+
+        if (erase_dirty) {
+            dirty_xattr = GF_CALLOC(EC_VERSION_SIZE, sizeof(*dirty_xattr),
+                                    gf_common_mt_pointer);
+            if (!dirty_xattr) {
+                op_ret = -ENOMEM;
+                continue;
+            }
+
+            dirty_xattr[type] = hton64(-dirty[i]);
+            ret = dict_set_bin(xattr[i], EC_XATTR_DIRTY, dirty_xattr,
+                               (sizeof(*dirty_xattr) * EC_VERSION_SIZE));
+            if (ret < 0) {
+                op_ret = -ENOMEM;
+                continue;
+            }
+        }
+
+        if (memcmp(versions_xattr, allzero,
+                   (sizeof(*versions_xattr) * EC_VERSION_SIZE)) == 0) {
+            if (!erase_dirty) {
+                continue;
+            }
+
+            if (memcmp(dirty_xattr, allzero,
+                       (sizeof(*dirty_xattr) * EC_VERSION_SIZE)) == 0) {
+                continue;
+            }
+        }
+
+        on[i] = 1;
+        call_count++;
+    }
+
+    /* Update the bricks with xattr */
+    if (call_count) {
+        PARALLEL_FOP_ONLIST(ec->xl_list, on, ec->nodes, replies, frame,
+                            ec_wind_xattrop_parallel, &loc,
+                            GF_XATTROP_ADD_ARRAY64, xattr, NULL);
+        ret = cluster_fop_success_fill(replies, ec->nodes, output);
+    }
+
+    if (ret < call_count) {
+        op_ret = -ENOTCONN;
+        goto out;
+    }
+
+out:
+    /* Cleanup */
+    if (xattr) {
+        for (i = 0; i < ec->nodes; i++) {
+            if (xattr[i])
+                dict_unref(xattr[i]);
+        }
+        GF_FREE(xattr);
+    }
+    cluster_replies_wipe(replies, ec->nodes);
+    loc_wipe(&loc);
+    return op_ret;
+}
+
+int
+ec_heal_metadata_find_direction(ec_t *ec, default_args_cbk_t *replies,
+                                uint64_t *versions, uint64_t *dirty,
+                                unsigned char *sources,
+                                unsigned char *healed_sinks)
+{
+    uint64_t xattr[EC_VERSION_SIZE] = {0};
+    uint64_t max_version = 0;
+    int same_count = 0;
+    int max_same_count = 0;
+    int same_source = -1;
+    int ret = 0;
+    int i = 0;
+    int j = 0;
+    int *groups = NULL;
+    struct iatt source_ia = {0};
+    struct iatt child_ia = {0};
+
+    groups = alloca0(ec->nodes * sizeof(*groups));
+    for (i = 0; i < ec->nodes; i++)
+        groups[i] = -1;
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (!replies[i].valid)
+            continue;
+        if (replies[i].op_ret < 0)
+            continue;
+        ret = ec_dict_get_array(replies[i].xdata, EC_XATTR_VERSION, xattr,
+                                EC_VERSION_SIZE);
+        if (ret == 0) {
+            versions[i] = xattr[EC_METADATA_TXN];
+        }
+
+        memset(xattr, 0, sizeof(xattr));
+        ret = ec_dict_get_array(replies[i].xdata, EC_XATTR_DIRTY, xattr,
+                                EC_VERSION_SIZE);
+        if (ret == 0) {
+            dirty[i] = xattr[EC_METADATA_TXN];
+        }
+        if (groups[i] >= 0) /*Already part of group*/
+            continue;
+        groups[i] = i;
+        same_count = 1;
+        source_ia = replies[i].stat;
+        for (j = i + 1; j < ec->nodes; j++) {
+            if (!replies[j].valid || replies[j].op_ret < 0)
+                continue;
+            child_ia = replies[j].stat;
+            if (!IA_EQUAL(source_ia, child_ia, gfid) ||
+                !IA_EQUAL(source_ia, child_ia, type) ||
+                !IA_EQUAL(source_ia, child_ia, prot) ||
+                !IA_EQUAL(source_ia, child_ia, uid) ||
+                !IA_EQUAL(source_ia, child_ia, gid))
+                continue;
+            if (!are_dicts_equal(replies[i].xdata, replies[j].xdata,
+                                 ec_sh_key_match, NULL))
+                continue;
+            groups[j] = i;
+            same_count++;
+        }
+
+        if (max_same_count < same_count) {
+            max_same_count = same_count;
+            same_source = i;
+        }
+    }
+
+    if (max_same_count < ec->fragments) {
+        ret = -EIO;
+        goto out;
+    }
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (groups[i] == groups[same_source])
+            sources[i] = 1;
+        else if (replies[i].valid && replies[i].op_ret >= 0)
+            healed_sinks[i] = 1;
+    }
+    for (i = 0; i < ec->nodes; i++) {
+        if (sources[i] && (versions[i] > max_version)) {
+            same_source = i;
+            max_version = versions[i];
+        }
+    }
+    ret = same_source;
+out:
+    return ret;
+}
+
+int
+__ec_heal_metadata_prepare(call_frame_t *frame, ec_t *ec, inode_t *inode,
+                           unsigned char *locked_on,
+                           default_args_cbk_t *replies, uint64_t *versions,
+                           uint64_t *dirty, unsigned char *sources,
+                           unsigned char *healed_sinks)
+{
+    loc_t loc = {0};
+    unsigned char *output = NULL;
+    unsigned char *lookup_on = NULL;
+    int ret = 0;
+    int source = 0;
+    default_args_cbk_t *greplies = NULL;
+    int i = 0;
+    EC_REPLIES_ALLOC(greplies, ec->nodes);
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+    output = alloca0(ec->nodes);
+    lookup_on = alloca0(ec->nodes);
+    ret = cluster_lookup(ec->xl_list, locked_on, ec->nodes, replies, output,
+                         frame, ec->xl, &loc, NULL);
+    if (ret <= ec->fragments) {
+        ret = -ENOTCONN;
+        goto out;
+    }
+
+    memcpy(lookup_on, output, ec->nodes);
+    /*Use getxattr to get the filtered xattrs which filter internal xattrs*/
+    ret = cluster_getxattr(ec->xl_list, lookup_on, ec->nodes, greplies, output,
+                           frame, ec->xl, &loc, NULL, NULL);
+    for (i = 0; i < ec->nodes; i++) {
+        if (lookup_on[i] && !output[i]) {
+            replies[i].valid = 0;
+            continue;
+        }
+        if (replies[i].xdata) {
+            dict_unref(replies[i].xdata);
+            replies[i].xdata = NULL;
+            if (greplies[i].xattr)
+                replies[i].xdata = dict_ref(greplies[i].xattr);
+        }
+    }
+
+    source = ec_heal_metadata_find_direction(ec, replies, versions, dirty,
+                                             sources, healed_sinks);
+    if (source < 0) {
+        ret = -EIO;
+        goto out;
+    }
+    ret = source;
+out:
+    cluster_replies_wipe(greplies, ec->nodes);
+    loc_wipe(&loc);
+    return ret;
+}
+
+/* Metadata heal */
+int
+__ec_removexattr_sinks(call_frame_t *frame, ec_t *ec, inode_t *inode,
+                       int source, unsigned char *sources,
+                       unsigned char *healed_sinks, default_args_cbk_t *replies)
+{
+    int i = 0;
+    int ret = 0;
+    loc_t loc = {0};
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (i == source)
+            continue;
+        if (!sources[i] && !healed_sinks[i])
+            continue;
+        ret = dict_foreach(replies[i].xdata, ec_heal_xattr_clean,
+                           replies[source].xdata);
+        if (ret < 0) {
+            sources[i] = 0;
+            healed_sinks[i] = 0;
+            continue;
+        }
+
+        if (replies[i].xdata->count == 0) {
+            continue;
+        } else if (sources[i]) {
+            /* This can happen if setxattr/removexattr succeeds on
+             * the bricks but fails to update the version. This
+             * will make sure that the xattrs are made equal after
+             * heal*/
+            sources[i] = 0;
+            healed_sinks[i] = 1;
+        }
+
+        ret = syncop_removexattr(ec->xl_list[i], &loc, "", replies[i].xdata,
+                                 NULL);
+        if (ret < 0)
+            healed_sinks[i] = 0;
+    }
+
+    loc_wipe(&loc);
+    if (EC_COUNT(healed_sinks, ec->nodes) == 0)
+        return -ENOTCONN;
+    return 0;
+}
+
+int
+__ec_heal_metadata(call_frame_t *frame, ec_t *ec, inode_t *inode,
+                   unsigned char *locked_on, unsigned char *sources,
+                   unsigned char *healed_sinks)
+{
+    loc_t loc = {0};
+    int ret = 0;
+    int source = 0;
+    default_args_cbk_t *replies = NULL;
+    default_args_cbk_t *sreplies = NULL;
+    uint64_t *versions = NULL;
+    uint64_t *dirty = NULL;
+    unsigned char *output = NULL;
+    dict_t *source_dict = NULL;
+    struct iatt source_buf = {0};
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    EC_REPLIES_ALLOC(sreplies, ec->nodes);
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+    output = alloca0(ec->nodes);
+    versions = alloca0(ec->nodes * sizeof(*versions));
+    dirty = alloca0(ec->nodes * sizeof(*dirty));
+    source = __ec_heal_metadata_prepare(frame, ec, inode, locked_on, replies,
+                                        versions, dirty, sources, healed_sinks);
+    if (source < 0) {
+        ret = -EIO;
+        goto out;
+    }
+
+    if ((EC_COUNT(sources, ec->nodes) == ec->nodes) ||
+        (EC_COUNT(healed_sinks, ec->nodes) == 0)) {
+        ret = 0;
+        goto erase_dirty;
+    }
+
+    source_buf = replies[source].stat;
+    ret = cluster_setattr(ec->xl_list, healed_sinks, ec->nodes, sreplies,
+                          output, frame, ec->xl, &loc, &source_buf,
+                          GF_SET_ATTR_MODE | GF_SET_ATTR_UID | GF_SET_ATTR_GID,
+                          NULL);
+    /*In case the operation fails on some of the subvols*/
+    memcpy(healed_sinks, output, ec->nodes);
+    if (EC_COUNT(healed_sinks, ec->nodes) == 0) {
+        ret = -ENOTCONN;
+        goto out;
+    }
+
+    ret = __ec_removexattr_sinks(frame, ec, inode, source, sources,
+                                 healed_sinks, replies);
+    if (ret < 0)
+        goto out;
+
+    source_dict = dict_ref(replies[source].xdata);
+    if (dict_foreach_match(source_dict, ec_ignorable_key_match, NULL,
+                           dict_remove_foreach_fn, NULL) == -1) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    ret = cluster_setxattr(ec->xl_list, healed_sinks, ec->nodes, replies,
+                           output, frame, ec->xl, &loc, source_dict, 0, NULL);
+
+    EC_INTERSECT(healed_sinks, healed_sinks, output, ec->nodes);
+    if (EC_COUNT(healed_sinks, ec->nodes) == 0) {
+        ret = -ENOTCONN;
+        goto out;
+    }
+
+erase_dirty:
+    ret = ec_adjust_versions(frame, ec, EC_METADATA_TXN, inode, source, sources,
+                             healed_sinks, versions, dirty);
+out:
+    if (source_dict)
+        dict_unref(source_dict);
+
+    loc_wipe(&loc);
+    cluster_replies_wipe(replies, ec->nodes);
+    cluster_replies_wipe(sreplies, ec->nodes);
+    return ret;
+}
+
+int
+ec_heal_metadata(call_frame_t *frame, ec_t *ec, inode_t *inode,
+                 unsigned char *sources, unsigned char *healed_sinks)
+{
+    unsigned char *locked_on = NULL;
+    unsigned char *up_subvols = NULL;
+    unsigned char *output = NULL;
+    int ret = 0;
+    default_args_cbk_t *replies = NULL;
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    locked_on = alloca0(ec->nodes);
+    output = alloca0(ec->nodes);
+    up_subvols = alloca0(ec->nodes);
+    ec_mask_to_char_array(ec->xl_up, up_subvols, ec->nodes);
+    ret = cluster_inodelk(ec->xl_list, up_subvols, ec->nodes, replies,
+                          locked_on, frame, ec->xl, ec->xl->name, inode, 0, 0);
+    {
+        if (ret <= ec->fragments) {
+            gf_msg_debug(ec->xl->name, 0,
+                         "%s: Skipping heal "
+                         "as only %d number of subvolumes could "
+                         "be locked",
+                         uuid_utoa(inode->gfid), ret);
+            ret = -ENOTCONN;
+            goto unlock;
+        }
+        ret = __ec_heal_metadata(frame, ec, inode, locked_on, sources,
+                                 healed_sinks);
+    }
+unlock:
+    cluster_uninodelk(ec->xl_list, locked_on, ec->nodes, replies, output, frame,
+                      ec->xl, ec->xl->name, inode, 0, 0);
+    cluster_replies_wipe(replies, ec->nodes);
+    return ret;
+}
+
+/*entry heal*/
+int
+__ec_heal_entry_prepare(call_frame_t *frame, ec_t *ec, inode_t *inode,
+                        unsigned char *locked_on, uint64_t *versions,
+                        uint64_t *dirty, unsigned char *sources,
+                        unsigned char *healed_sinks)
+{
+    loc_t loc = {0};
+    int source = 0;
+    int ret = 0;
+    default_args_cbk_t *replies = NULL;
+    unsigned char *output = NULL;
+    dict_t *xdata = NULL;
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+    xdata = dict_new();
+    if (!xdata) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    if (dict_set_uint64(xdata, EC_XATTR_VERSION, 0) ||
+        dict_set_uint64(xdata, EC_XATTR_DIRTY, 0)) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    output = alloca0(ec->nodes);
+    ret = cluster_lookup(ec->xl_list, locked_on, ec->nodes, replies, output,
+                         frame, ec->xl, &loc, xdata);
+    if (ret <= ec->fragments) {
+        ret = -ENOTCONN;
+        goto out;
+    }
+
+    source = ec_heal_entry_find_direction(ec, replies, versions, dirty, sources,
+                                          healed_sinks);
+    if (source < 0) {
+        ret = -EIO;
+        goto out;
+    }
+    ret = source;
+out:
+    if (xdata)
+        dict_unref(xdata);
+    loc_wipe(&loc);
+    cluster_replies_wipe(replies, ec->nodes);
+    return ret;
+}
+int32_t
+ec_set_new_entry_dirty(ec_t *ec, loc_t *loc, struct iatt *ia,
+                       call_frame_t *frame, xlator_t *this, unsigned char *on)
+{
+    dict_t *xattr = NULL;
+    int32_t ret = -1;
+    default_args_cbk_t *replies = NULL;
+    unsigned char *output = NULL;
+    uint64_t dirty[EC_VERSION_SIZE] = {1, 1};
+    loc_t newloc = {0};
+
+    /*Symlinks don't have any data to be healed*/
+    if (ia->ia_type == IA_IFLNK)
+        dirty[EC_DATA_TXN] = 0;
+
+    newloc.inode = inode_ref(loc->inode);
+    gf_uuid_copy(newloc.gfid, ia->ia_gfid);
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    output = alloca0(ec->nodes);
+    xattr = dict_new();
+    if (!xattr) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    ret = ec_dict_set_array(xattr, EC_XATTR_DIRTY, dirty, EC_VERSION_SIZE);
+    if (ret)
+        goto out;
+
+    ret = cluster_xattrop(ec->xl_list, on, ec->nodes, replies, output, frame,
+                          ec->xl, &newloc, GF_XATTROP_ADD_ARRAY64, xattr, NULL);
+
+    if (ret < ec->fragments) {
+        ret = -ENOTCONN;
+        goto out;
+    }
+
+out:
+    if (xattr)
+        dict_unref(xattr);
+    cluster_replies_wipe(replies, ec->nodes);
+    loc_wipe(&newloc);
+    return ret;
+}
+
+/*Name heal*/
+int
+ec_delete_stale_name(dict_t *gfid_db, char *key, data_t *d, void *data)
+{
+    struct ec_name_data *name_data = data;
+    struct iatt *ia = NULL;
+    ec_t *ec = NULL;
+    loc_t loc = {0};
+    unsigned char *same = data_to_bin(d);
+    default_args_cbk_t *replies = NULL;
+    unsigned char *output = NULL;
+    int ret = 0;
+    int estale_count = 0;
+    int i = 0;
+    call_frame_t *frame = name_data->frame;
+    uuid_t gfid;
+
+    ec = name_data->frame->this->private;
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    if (EC_COUNT(same, ec->nodes) >= ec->fragments) {
+        ret = 0;
+        goto out;
+    }
+
+    loc.parent = inode_ref(name_data->parent);
+    loc.inode = inode_new(name_data->parent->table);
+    if (!loc.inode) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    gf_uuid_parse(key, gfid);
+    gf_uuid_copy(loc.pargfid, name_data->parent->gfid);
+    loc.name = name_data->name;
+    output = alloca0(ec->nodes);
+    ret = cluster_lookup(ec->xl_list, name_data->participants, ec->nodes,
+                         replies, output, name_data->frame, ec->xl, &loc, NULL);
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (!replies[i].valid)
+            continue;
+        if (replies[i].op_ret == -1) {
+            if (replies[i].op_errno == ESTALE || replies[i].op_errno == ENOENT)
+                estale_count++;
+            else
+                name_data->participants[i] = 0;
+        } else if (gf_uuid_compare(gfid, replies[i].stat.ia_gfid)) {
+            estale_count++;
+            gf_msg_debug(ec->xl->name, 0, "%s/%s: different gfid as %s",
+                         uuid_utoa(name_data->parent->gfid), name_data->name,
+                         key);
+        }
+    }
+
+    if (estale_count <= ec->redundancy) {
+        /* We have at least ec->fragments number of fragments, so the
+         * file is recoverable, so don't delete it*/
+
+        /* Please note that the lookup call above could fail with
+         * ENOTCONN on all subvoumes and still this branch will be
+         * true, but in those cases conservatively we decide to not
+         * delete the file until we are sure*/
+        ret = 0;
+        goto out;
+    }
+
+    /*Noway to recover, delete the name*/
+    loc_wipe(&loc);
+    loc.parent = inode_ref(name_data->parent);
+    gf_uuid_copy(loc.pargfid, loc.parent->gfid);
+    loc.name = name_data->name;
+    for (i = 0; i < ec->nodes; i++) {
+        if (same[i] && replies[i].valid && (replies[i].op_ret == 0)) {
+            ia = &replies[i].stat;
+            break;
+        }
+    }
+
+    if (!ia) {
+        ret = -ENOTCONN;
+        goto out;
+    }
+
+    if (IA_ISDIR(ia->ia_type)) {
+        ret = cluster_rmdir(ec->xl_list, same, ec->nodes, replies, output,
+                            frame, ec->xl, &loc, 1, NULL);
+        gf_msg_debug(ec->xl->name, 0,
+                     "cluster rmdir succeeded on %d "
+                     "nodes",
+                     ret);
+    } else {
+        ret = cluster_unlink(ec->xl_list, same, ec->nodes, replies, output,
+                             frame, ec->xl, &loc, 0, NULL);
+        gf_msg_debug(ec->xl->name, 0,
+                     "cluster unlink succeeded on %d "
+                     "nodes",
+                     ret);
+    }
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (output[i]) {
+            same[i] = 0;
+            name_data->enoent[i] = 1;
+        } else {
+            /*op failed*/
+            if (same[i])
+                name_data->participants[i] = 0;
+        }
+    }
+    ret = 0;
+    /*This will help in making decisions about creating names*/
+    dict_del(gfid_db, key);
+out:
+    if (ret < 0) {
+        gf_msg_debug(ec->xl->name, 0, "%s/%s: heal failed %s",
+                     uuid_utoa(name_data->parent->gfid), name_data->name,
+                     strerror(-ret));
+    }
+    cluster_replies_wipe(replies, ec->nodes);
+    loc_wipe(&loc);
+    return ret;
+}
+
+int
+ec_delete_stale_names(call_frame_t *frame, ec_t *ec, inode_t *parent,
+                      char *name, default_args_cbk_t *replies, dict_t *gfid_db,
+                      unsigned char *enoent, unsigned char *gfidless,
+                      unsigned char *participants)
+{
+    struct ec_name_data name_data = {0};
+
+    name_data.enoent = enoent;
+    name_data.gfidless = gfidless;
+    name_data.participants = participants;
+    name_data.name = name;
+    name_data.parent = parent;
+    name_data.frame = frame;
+    name_data.replies = replies;
+    return dict_foreach(gfid_db, ec_delete_stale_name, &name_data);
+}
+
+int
+_assign_same(dict_t *dict, char *key, data_t *value, void *data)
+{
+    struct ec_name_data *name_data = data;
+
+    name_data->same = data_to_bin(value);
+    return 0;
+}
+
+int
+ec_create_name(call_frame_t *frame, ec_t *ec, inode_t *parent, char *name,
+               default_args_cbk_t *lookup_replies, dict_t *gfid_db,
+               unsigned char *enoent, unsigned char *participants)
+{
+    int ret = 0;
+    int i = 0;
+    struct ec_name_data name_data = {0};
+    struct iatt *ia = NULL;
+    unsigned char *output = 0;
+    unsigned char *output1 = 0;
+    unsigned char *on = NULL;
+    default_args_cbk_t *replies = NULL;
+    loc_t loc = {0};
+    loc_t srcloc = {0};
+    unsigned char *link = NULL;
+    unsigned char *create = NULL;
+    dict_t *xdata = NULL;
+    char *linkname = NULL;
+    ec_config_t config;
+
+    /* There should be just one gfid key */
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    if (gfid_db->count != 1) {
+        ret = -EINVAL;
+        goto out;
+    }
+
+    ret = dict_foreach(gfid_db, _assign_same, &name_data);
+    if (ret < 0)
+        goto out;
+    /*There should at least be one valid success reply with gfid*/
+    for (i = 0; i < ec->nodes; i++)
+        if (name_data.same[i])
+            break;
+
+    if (i == ec->nodes) {
+        ret = -EINVAL;
+        goto out;
+    }
+
+    ia = &lookup_replies[i].stat;
+    xdata = dict_new();
+    loc.parent = inode_ref(parent);
+    gf_uuid_copy(loc.pargfid, parent->gfid);
+    loc.inode = inode_new(parent->table);
+    if (loc.inode)
+        srcloc.inode = inode_ref(loc.inode);
+    gf_uuid_copy(srcloc.gfid, ia->ia_gfid);
+    if (!loc.inode || !xdata ||
+        dict_set_static_bin(xdata, "gfid-req", ia->ia_gfid,
+                            sizeof(ia->ia_gfid))) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    loc.name = name;
+    link = alloca0(ec->nodes);
+    create = alloca0(ec->nodes);
+    on = alloca0(ec->nodes);
+    output = alloca0(ec->nodes);
+    output1 = alloca0(ec->nodes);
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (!lookup_replies[i].valid)
+            continue;
+        if (lookup_replies[i].op_ret)
+            continue;
+        on[i] = 1;
+    }
+    switch (ia->ia_type) {
+        case IA_IFDIR:
+            ec_set_new_entry_dirty(ec, &loc, ia, frame, ec->xl, on);
+            (void)cluster_mkdir(
+                ec->xl_list, enoent, ec->nodes, replies, output, frame, ec->xl,
+                &loc, st_mode_from_ia(ia->ia_prot, ia->ia_type), 0, xdata);
+            break;
+
+        case IA_IFLNK:
+            /*Check for hard links and create/link*/
+            ret = cluster_lookup(ec->xl_list, enoent, ec->nodes, replies,
+                                 output, frame, ec->xl, &srcloc, NULL);
+            for (i = 0; i < ec->nodes; i++) {
+                if (output[i]) {
+                    link[i] = 1;
+                } else {
+                    if (replies[i].op_errno == ENOENT ||
+                        replies[i].op_errno == ESTALE) {
+                        create[i] = 1;
+                    }
+                }
+            }
+
+            if (EC_COUNT(link, ec->nodes)) {
+                cluster_link(ec->xl_list, link, ec->nodes, replies, output1,
+                             frame, ec->xl, &srcloc, &loc, NULL);
+            }
+
+            if (EC_COUNT(create, ec->nodes)) {
+                cluster_readlink(ec->xl_list, name_data.same, ec->nodes,
+                                 replies, output, frame, ec->xl, &srcloc, 4096,
+                                 NULL);
+                if (EC_COUNT(output, ec->nodes) == 0) {
+                    ret = -ENOTCONN;
+                    goto out;
+                }
+
+                for (i = 0; i < ec->nodes; i++) {
+                    if (output[i])
+                        break;
+                }
+                linkname = alloca0(strlen(replies[i].buf) + 1);
+                strcpy(linkname, replies[i].buf);
+                ec_set_new_entry_dirty(ec, &loc, ia, frame, ec->xl, on);
+                cluster_symlink(ec->xl_list, create, ec->nodes, replies, output,
+                                frame, ec->xl, linkname, &loc, 0, xdata);
+            }
+            for (i = 0; i < ec->nodes; i++)
+                if (output1[i])
+                    output[i] = 1;
+            break;
+        case IA_IFREG:
+            ec_set_new_entry_dirty(ec, &loc, ia, frame, ec->xl, on);
+            config.version = EC_CONFIG_VERSION;
+            config.algorithm = EC_CONFIG_ALGORITHM;
+            config.gf_word_size = EC_GF_BITS;
+            config.bricks = ec->nodes;
+            config.redundancy = ec->redundancy;
+            config.chunk_size = EC_METHOD_CHUNK_SIZE;
+
+            ret = ec_dict_set_config(xdata, EC_XATTR_CONFIG, &config);
+            if (ret != 0) {
+                goto out;
+            }
+
+            /* Fall through */
+
+        default:
+            ret = dict_set_int32(xdata, GLUSTERFS_INTERNAL_FOP_KEY, 1);
+            if (ret)
+                goto out;
+            ret = cluster_mknod(
+                ec->xl_list, enoent, ec->nodes, replies, output, frame, ec->xl,
+                &loc, st_mode_from_ia(ia->ia_prot, ia->ia_type),
+                makedev(ia_major(ia->ia_rdev), ia_minor(ia->ia_rdev)), 0,
+                xdata);
+            break;
+    }
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (enoent[i] && !output[i])
+            participants[i] = 0;
+    }
+
+    ret = 0;
+out:
+    if (ret < 0)
+        gf_msg_debug(ec->xl->name, 0, "%s/%s: heal failed %s",
+                     uuid_utoa(parent->gfid), name, strerror(-ret));
+    cluster_replies_wipe(replies, ec->nodes);
+    loc_wipe(&loc);
+    loc_wipe(&srcloc);
+    if (xdata)
+        dict_unref(xdata);
+    return ret;
+}
+
+int
+__ec_heal_name(call_frame_t *frame, ec_t *ec, inode_t *parent, char *name,
+               unsigned char *participants)
+{
+    unsigned char *output = NULL;
+    unsigned char *enoent = NULL;
+    default_args_cbk_t *replies = NULL;
+    dict_t *xdata = NULL;
+    dict_t *gfid_db = NULL;
+    int ret = 0;
+    loc_t loc = {0};
+    int i = 0;
+    struct iatt *ia = NULL;
+    char gfid[64] = {0};
+    unsigned char *same = NULL;
+    unsigned char *gfidless = NULL;
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    loc.parent = inode_ref(parent);
+    loc.inode = inode_new(parent->table);
+    gf_uuid_copy(loc.pargfid, parent->gfid);
+    loc.name = name;
+    xdata = dict_new();
+    gfid_db = dict_new();
+    if (!xdata || !gfid_db || !loc.inode) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    ret = dict_set_int32(xdata, GF_GFIDLESS_LOOKUP, 1);
+    if (ret) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    output = alloca0(ec->nodes);
+    gfidless = alloca0(ec->nodes);
+    enoent = alloca0(ec->nodes);
+    ret = cluster_lookup(ec->xl_list, participants, ec->nodes, replies, output,
+                         frame, ec->xl, &loc, NULL);
+    for (i = 0; i < ec->nodes; i++) {
+        if (!replies[i].valid)
+            continue;
+
+        if (replies[i].op_ret == -1) {
+            /*If ESTALE comes here, that means parent dir is not
+             * present, nothing to do there, so reset participants
+             * for that brick*/
+            if (replies[i].op_errno == ENOENT)
+                enoent[i] = 1;
+            else
+                participants[i] = 0;
+            continue;
+        }
+        ia = &replies[i].stat;
+        if (gf_uuid_is_null(ia->ia_gfid)) {
+            if (IA_ISDIR(ia->ia_type) || ia->ia_size == 0)
+                gfidless[i] = 1;
+            else
+                participants[i] = 0;
+        } else {
+            uuid_utoa_r(ia->ia_gfid, gfid);
+            ret = dict_get_bin(gfid_db, gfid, (void **)&same);
+            if (ret < 0) {
+                same = alloca0(ec->nodes);
+            }
+            same[i] = 1;
+            if (ret < 0) {
+                ret = dict_set_static_bin(gfid_db, gfid, same, ec->nodes);
+            }
+            if (ret < 0)
+                goto out;
+        }
+    }
+
+    ret = ec_delete_stale_names(frame, ec, parent, name, replies, gfid_db,
+                                enoent, gfidless, participants);
+
+    if (gfid_db->count == 0) {
+        /* All entries seem to be stale entries and deleted,
+         * nothing more to do.*/
+        goto out;
+    }
+
+    if (gfid_db->count > 1) {
+        gf_msg(ec->xl->name, GF_LOG_INFO, 0, EC_MSG_HEAL_FAIL,
+               "%s/%s: Not able to heal", uuid_utoa(parent->gfid), name);
+        memset(participants, 0, ec->nodes);
+        goto out;
+    }
+
+    EC_INTERSECT(enoent, enoent, participants, ec->nodes);
+    if (EC_COUNT(enoent, ec->nodes) == 0) {
+        ret = 0;
+        goto out;
+    }
+
+    ret = ec_create_name(frame, ec, parent, name, replies, gfid_db, enoent,
+                         participants);
+    if (ret >= 0) {
+        /* If ec_create_name() succeeded we return 1 to indicate that a new
+         * file has been created and it will need to be healed. */
+        ret = 1;
+    }
+out:
+    cluster_replies_wipe(replies, ec->nodes);
+    loc_wipe(&loc);
+    if (xdata)
+        dict_unref(xdata);
+    if (gfid_db)
+        dict_unref(gfid_db);
+    return ret;
+}
+
+int
+ec_heal_name(call_frame_t *frame, ec_t *ec, inode_t *parent, char *name,
+             unsigned char *participants)
+{
+    int ret = 0;
+    default_args_cbk_t *replies = NULL;
+    unsigned char *output = NULL;
+    unsigned char *locked_on = NULL;
+    loc_t loc = {0};
+
+    loc.parent = inode_ref(parent);
+    loc.name = name;
+    loc.inode = inode_new(parent->table);
+    if (!loc.inode) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    output = alloca0(ec->nodes);
+    locked_on = alloca0(ec->nodes);
+    ret = cluster_inodelk(ec->xl_list, participants, ec->nodes, replies,
+                          locked_on, frame, ec->xl, ec->xl->name, parent, 0, 0);
+    {
+        if (ret <= ec->fragments) {
+            gf_msg_debug(ec->xl->name, 0,
+                         "%s/%s: Skipping "
+                         "heal as only %d number of subvolumes could "
+                         "be locked",
+                         uuid_utoa(parent->gfid), name, ret);
+            ret = -ENOTCONN;
+            goto unlock;
+        }
+        EC_INTERSECT(participants, participants, locked_on, ec->nodes);
+        ret = __ec_heal_name(frame, ec, parent, name, participants);
+    }
+unlock:
+    cluster_uninodelk(ec->xl_list, locked_on, ec->nodes, replies, output, frame,
+                      ec->xl, ec->xl->name, parent, 0, 0);
+out:
+    cluster_replies_wipe(replies, ec->nodes);
+    loc_wipe(&loc);
+    return ret;
+}
+
+int
+ec_name_heal_handler(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent,
+                     void *data)
+{
+    struct ec_name_data *name_data = data;
+    xlator_t *this = THIS;
+    ec_t *ec = this->private;
+    unsigned char *name_on = alloca0(ec->nodes);
+    int i = 0;
+    int ret = 0;
+
+    if (ec->shutdown) {
+        gf_msg_debug(this->name, 0,
+                     "Cancelling directory heal "
+                     "because EC is stopping.");
+        return -ENOTCONN;
+    }
+
+    memcpy(name_on, name_data->participants, ec->nodes);
+    ret = ec_heal_name(name_data->frame, ec, parent->inode, entry->d_name,
+                       name_on);
+
+    if (ret < 0) {
+        memset(name_on, 0, ec->nodes);
+    } else {
+        name_data->heal_pending += ret;
+    }
+
+    for (i = 0; i < ec->nodes; i++)
+        if (name_data->participants[i] && !name_on[i])
+            name_data->failed_on[i] = 1;
+
+    return 0;
+}
+
+int
+ec_heal_names(call_frame_t *frame, ec_t *ec, inode_t *inode,
+              unsigned char *participants, uint32_t *pending)
+{
+    int i = 0;
+    int j = 0;
+    loc_t loc = {0};
+    struct ec_name_data name_data = {0};
+    int ret = 0;
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+    name_data.frame = frame;
+    name_data.participants = participants;
+    name_data.failed_on = alloca0(ec->nodes);
+    name_data.heal_pending = 0;
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (!participants[i])
+            continue;
+        ret = syncop_dir_scan(ec->xl_list[i], &loc, GF_CLIENT_PID_SELF_HEALD,
+                              &name_data, ec_name_heal_handler);
+        if (ret < 0) {
+            break;
+        }
+        for (j = 0; j < ec->nodes; j++)
+            if (name_data.failed_on[j])
+                participants[j] = 0;
+
+        if (EC_COUNT(participants, ec->nodes) <= ec->fragments) {
+            ret = -ENOTCONN;
+            break;
+        }
+    }
+    *pending += name_data.heal_pending;
+
+    loc_wipe(&loc);
+    return ret;
+}
+
+int
+__ec_heal_entry(call_frame_t *frame, ec_t *ec, inode_t *inode,
+                unsigned char *heal_on, unsigned char *sources,
+                unsigned char *healed_sinks, uint32_t *pending)
+{
+    unsigned char *locked_on = NULL;
+    unsigned char *output = NULL;
+    uint64_t *versions = NULL;
+    uint64_t *dirty = NULL;
+    unsigned char *participants = NULL;
+    default_args_cbk_t *replies = NULL;
+    int ret = 0;
+    int source = 0;
+    int i = 0;
+
+    locked_on = alloca0(ec->nodes);
+    output = alloca0(ec->nodes);
+    versions = alloca0(ec->nodes * sizeof(*versions));
+    dirty = alloca0(ec->nodes * sizeof(*dirty));
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    ret = cluster_inodelk(ec->xl_list, heal_on, ec->nodes, replies, locked_on,
+                          frame, ec->xl, ec->xl->name, inode, 0, 0);
+    {
+        if (ret <= ec->fragments) {
+            gf_msg_debug(ec->xl->name, 0,
+                         "%s: Skipping heal "
+                         "as only %d number of subvolumes could "
+                         "be locked",
+                         uuid_utoa(inode->gfid), ret);
+            ret = -ENOTCONN;
+            goto unlock;
+        }
+        ret = __ec_heal_entry_prepare(frame, ec, inode, locked_on, versions,
+                                      dirty, sources, healed_sinks);
+        source = ret;
+    }
+unlock:
+    cluster_uninodelk(ec->xl_list, locked_on, ec->nodes, replies, output, frame,
+                      ec->xl, ec->xl->name, inode, 0, 0);
+    if (ret < 0)
+        goto out;
+
+    participants = alloca0(ec->nodes);
+    for (i = 0; i < ec->nodes; i++) {
+        if (sources[i] || healed_sinks[i])
+            participants[i] = 1;
+    }
+    ret = ec_heal_names(frame, ec, inode, participants, pending);
+
+    if (EC_COUNT(participants, ec->nodes) <= ec->fragments)
+        goto out;
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (!participants[i]) {
+            sources[i] = 0;
+            healed_sinks[i] = 0;
+        }
+    }
+
+    ec_adjust_versions(frame, ec, EC_DATA_TXN, inode, source, sources,
+                       healed_sinks, versions, dirty);
+out:
+    cluster_replies_wipe(replies, ec->nodes);
+    return ret;
+}
+
+int
+ec_heal_entry(call_frame_t *frame, ec_t *ec, inode_t *inode,
+              unsigned char *sources, unsigned char *healed_sinks,
+              uint32_t *pending)
+{
+    unsigned char *locked_on = NULL;
+    unsigned char *up_subvols = NULL;
+    unsigned char *output = NULL;
+    char selfheal_domain[1024] = {0};
+    int ret = 0;
+    default_args_cbk_t *replies = NULL;
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    locked_on = alloca0(ec->nodes);
+    output = alloca0(ec->nodes);
+    up_subvols = alloca0(ec->nodes);
+
+    sprintf(selfheal_domain, "%s:self-heal", ec->xl->name);
+    ec_mask_to_char_array(ec->xl_up, up_subvols, ec->nodes);
+    /*If other processes are already doing the heal, don't block*/
+    ret = cluster_tiebreaker_inodelk(ec->xl_list, up_subvols, ec->nodes,
+                                     replies, locked_on, frame, ec->xl,
+                                     selfheal_domain, inode, 0, 0);
+    {
+        if (ret <= ec->fragments) {
+            gf_msg_debug(ec->xl->name, 0,
+                         "%s: Skipping heal "
+                         "as only %d number of subvolumes could "
+                         "be locked",
+                         uuid_utoa(inode->gfid), ret);
+            ret = -ENOTCONN;
+            goto unlock;
+        }
+        ret = __ec_heal_entry(frame, ec, inode, locked_on, sources,
+                              healed_sinks, pending);
+    }
+unlock:
+    cluster_uninodelk(ec->xl_list, locked_on, ec->nodes, replies, output, frame,
+                      ec->xl, selfheal_domain, inode, 0, 0);
+    cluster_replies_wipe(replies, ec->nodes);
+    return ret;
+}
+
+/*Find direction for data heal and heal info*/
+int
+ec_heal_data_find_direction(ec_t *ec, default_args_cbk_t *replies,
+                            uint64_t *data_versions, uint64_t *dirty,
+                            uint64_t *size, unsigned char *sources,
+                            unsigned char *healed_sinks,
+                            gf_boolean_t check_ondisksize, int which)
+{
+    uint64_t xattr[EC_VERSION_SIZE] = {0};
+    char version_size[128] = {0};
+    dict_t *version_size_db = NULL;
+    unsigned char *same = NULL;
+    int max_same_count = 0;
+    int source = 0;
+    int i = 0;
+    int ret = 0;
+    dict_t *dict = NULL;
+    uint64_t source_size = 0;
+
+    version_size_db = dict_new();
+    if (!version_size_db) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (!replies[i].valid)
+            continue;
+        if (replies[i].op_ret < 0)
+            continue;
+        dict = (which == EC_COMBINE_XDATA) ? replies[i].xdata
+                                           : replies[i].xattr;
+
+        ret = ec_dict_get_array(dict, EC_XATTR_VERSION, xattr, EC_VERSION_SIZE);
+        if (ret == 0) {
+            data_versions[i] = xattr[EC_DATA_TXN];
+        }
+
+        memset(xattr, 0, sizeof(xattr));
+        ret = ec_dict_get_array(dict, EC_XATTR_DIRTY, xattr, EC_VERSION_SIZE);
+        if (ret == 0) {
+            dirty[i] = xattr[EC_DATA_TXN];
+        }
+        ret = ec_dict_del_number(dict, EC_XATTR_SIZE, &size[i]);
+        /*Build a db of same metadata and data version and size*/
+        snprintf(version_size, sizeof(version_size), "%" PRIu64 "-%" PRIu64,
+                 data_versions[i], size[i]);
+
+        ret = dict_get_bin(version_size_db, version_size, (void **)&same);
+        if (ret < 0) {
+            same = alloca0(ec->nodes);
+        }
+
+        same[i] = 1;
+        if (max_same_count < EC_COUNT(same, ec->nodes)) {
+            max_same_count = EC_COUNT(same, ec->nodes);
+            source = i;
+        }
+
+        if (ret < 0) {
+            ret = dict_set_static_bin(version_size_db, version_size, same,
+                                      ec->nodes);
+        }
+
+        if (ret < 0) {
+            ret = -ENOMEM;
+            goto out;
+        }
+    }
+    /* If we don't have ec->fragments number of same version,size it is not
+     * recoverable*/
+    if (max_same_count < ec->fragments) {
+        ret = -EIO;
+        goto out;
+    } else {
+        snprintf(version_size, sizeof(version_size), "%" PRIu64 "-%" PRIu64,
+                 data_versions[source], size[source]);
+
+        ret = dict_get_bin(version_size_db, version_size, (void **)&same);
+        if (ret < 0)
+            goto out;
+        memcpy(sources, same, ec->nodes);
+        for (i = 0; i < ec->nodes; i++) {
+            if (replies[i].valid && (replies[i].op_ret == 0) && !sources[i])
+                healed_sinks[i] = 1;
+        }
+    }
+
+    /* There could be files with versions, size same but on disk ia_size
+     * could be different because of disk crashes, mark them as sinks as
+     * well*/
+
+    if (check_ondisksize) {
+        source_size = size[source];
+        ec_adjust_size_up(ec, &source_size, _gf_true);
+
+        for (i = 0; i < ec->nodes; i++) {
+            if (sources[i]) {
+                if (replies[i].stat.ia_size != source_size) {
+                    sources[i] = 0;
+                    healed_sinks[i] = 1;
+                    max_same_count--;
+                } else {
+                    source = i;
+                }
+            }
+        }
+        if (max_same_count < ec->fragments) {
+            ret = -EIO;
+            goto out;
+        }
+    }
+
+    ret = source;
+out:
+    if (version_size_db)
+        dict_unref(version_size_db);
+    return ret;
+}
+
+int
+__ec_heal_data_prepare(call_frame_t *frame, ec_t *ec, fd_t *fd,
+                       unsigned char *locked_on, uint64_t *versions,
+                       uint64_t *dirty, uint64_t *size, unsigned char *sources,
+                       unsigned char *healed_sinks, unsigned char *trim,
+                       struct iatt *stbuf)
+{
+    default_args_cbk_t *replies = NULL;
+    default_args_cbk_t *fstat_replies = NULL;
+    unsigned char *output = NULL;
+    unsigned char *fstat_output = NULL;
+    dict_t *xattrs = NULL;
+    uint64_t zero_array[2] = {0};
+    int source = 0;
+    int ret = 0;
+    uint64_t zero_value = 0;
+    int i = 0;
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    EC_REPLIES_ALLOC(fstat_replies, ec->nodes);
+    output = alloca0(ec->nodes);
+    fstat_output = alloca0(ec->nodes);
+    xattrs = dict_new();
+    if (!xattrs ||
+        dict_set_static_bin(xattrs, EC_XATTR_VERSION, zero_array,
+                            sizeof(zero_array)) ||
+        dict_set_static_bin(xattrs, EC_XATTR_DIRTY, zero_array,
+                            sizeof(zero_array)) ||
+        dict_set_static_bin(xattrs, EC_XATTR_SIZE, &zero_value,
+                            sizeof(zero_value))) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    ret = cluster_fxattrop(ec->xl_list, locked_on, ec->nodes, replies, output,
+                           frame, ec->xl, fd, GF_XATTROP_ADD_ARRAY64, xattrs,
+                           NULL);
+
+    ret = cluster_fstat(ec->xl_list, locked_on, ec->nodes, fstat_replies,
+                        fstat_output, frame, ec->xl, fd, NULL);
+
+    for (i = 0; i < ec->nodes; i++) {
+        output[i] = output[i] && fstat_output[i];
+        replies[i].valid = output[i];
+        if (output[i])
+            replies[i].stat = fstat_replies[i].stat;
+    }
+
+    if (EC_COUNT(output, ec->nodes) <= ec->fragments) {
+        ret = -ENOTCONN;
+        goto out;
+    }
+
+    source = ec_heal_data_find_direction(ec, replies, versions, dirty, size,
+                                         sources, healed_sinks, _gf_true,
+                                         EC_COMBINE_DICT);
+    ret = source;
+    if (ret < 0)
+        goto out;
+
+    if (stbuf)
+        *stbuf = replies[source].stat;
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (healed_sinks[i]) {
+            if (replies[i].stat.ia_size)
+                trim[i] = 1;
+        }
+    }
+
+    if (EC_COUNT(sources, ec->nodes) < ec->fragments) {
+        ret = -ENOTCONN;
+        goto out;
+    }
+
+    ret = source;
+out:
+    if (xattrs)
+        dict_unref(xattrs);
+    cluster_replies_wipe(replies, ec->nodes);
+    cluster_replies_wipe(fstat_replies, ec->nodes);
+    if (ret < 0) {
+        gf_msg_debug(ec->xl->name, 0, "%s: heal failed %s",
+                     uuid_utoa(fd->inode->gfid), strerror(-ret));
+    } else {
+        gf_msg_debug(ec->xl->name, 0,
+                     "%s: sources: %d, sinks: "
+                     "%d",
+                     uuid_utoa(fd->inode->gfid), EC_COUNT(sources, ec->nodes),
+                     EC_COUNT(healed_sinks, ec->nodes));
+    }
+    return ret;
+}
+
+int
+__ec_heal_mark_sinks(call_frame_t *frame, ec_t *ec, fd_t *fd,
+                     uint64_t *versions, unsigned char *healed_sinks)
+{
+    int i = 0;
+    int ret = 0;
+    unsigned char *mark = NULL;
+    dict_t *xattrs = NULL;
+    default_args_cbk_t *replies = NULL;
+    unsigned char *output = NULL;
+    uint64_t versions_xattr[2] = {0};
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    xattrs = dict_new();
+    if (!xattrs) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    mark = alloca0(ec->nodes);
+    for (i = 0; i < ec->nodes; i++) {
+        if (!healed_sinks[i])
+            continue;
+        if ((versions[i] >> EC_SELFHEAL_BIT) & 1)
+            continue;
+        mark[i] = 1;
+    }
+
+    if (EC_COUNT(mark, ec->nodes) == 0)
+        return 0;
+
+    versions_xattr[EC_DATA_TXN] = hton64(1ULL << EC_SELFHEAL_BIT);
+    if (dict_set_static_bin(xattrs, EC_XATTR_VERSION, versions_xattr,
+                            sizeof(versions_xattr))) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    output = alloca0(ec->nodes);
+    ret = cluster_fxattrop(ec->xl_list, mark, ec->nodes, replies, output, frame,
+                           ec->xl, fd, GF_XATTROP_ADD_ARRAY64, xattrs, NULL);
+    for (i = 0; i < ec->nodes; i++) {
+        if (!output[i]) {
+            if (mark[i])
+                healed_sinks[i] = 0;
+            continue;
+        }
+        versions[i] |= (1ULL << EC_SELFHEAL_BIT);
+    }
+
+    if (EC_COUNT(healed_sinks, ec->nodes) == 0) {
+        ret = -ENOTCONN;
+        goto out;
+    }
+    ret = 0;
+
+out:
+    cluster_replies_wipe(replies, ec->nodes);
+    if (xattrs)
+        dict_unref(xattrs);
+    if (ret < 0)
+        gf_msg_debug(ec->xl->name, 0, "%s: heal failed %s",
+                     uuid_utoa(fd->inode->gfid), strerror(-ret));
+    return ret;
+}
+
+int32_t
+ec_manager_heal_block(ec_fop_data_t *fop, int32_t state)
+{
+    ec_heal_t *heal = fop->data;
+    heal->fop = fop;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            ec_owner_set(fop->frame, fop->frame->root);
+
+            ec_heal_inodelk(heal, F_WRLCK, 1, 0, 0);
+
+            return EC_STATE_HEAL_DATA_COPY;
+
+        case EC_STATE_HEAL_DATA_COPY:
+            gf_msg_debug(fop->xl->name, 0, "%s: read/write starting",
+                         uuid_utoa(heal->fd->inode->gfid));
+            ec_heal_data_block(heal);
+
+            return EC_STATE_HEAL_DATA_UNLOCK;
+
+        case -EC_STATE_HEAL_DATA_COPY:
+        case -EC_STATE_HEAL_DATA_UNLOCK:
+        case EC_STATE_HEAL_DATA_UNLOCK:
+            ec_heal_inodelk(heal, F_UNLCK, 1, 0, 0);
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            if (fop->cbks.heal) {
+                fop->cbks.heal(fop->req_frame, fop->data, fop->xl, 0, 0,
+                               (heal->good | heal->bad), heal->good, heal->bad,
+                               0, NULL);
+            }
+
+            return EC_STATE_END;
+        case -EC_STATE_REPORT:
+            if (fop->cbks.heal) {
+                fop->cbks.heal(fop->req_frame, fop->data, fop->xl, -1,
+                               fop->error, 0, 0, 0, 0, NULL);
+            }
+
+            return EC_STATE_END;
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, 0, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+/*Takes lock */
+void
+ec_heal_block(call_frame_t *frame, xlator_t *this, uintptr_t target,
+              uint32_t fop_flags, fop_heal_cbk_t func, ec_heal_t *heal)
+{
+    ec_cbk_t callback = {.heal = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(HEAL) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, EC_FOP_HEAL, 0, target, fop_flags,
+                               NULL, ec_manager_heal_block, callback, heal);
+    if (fop == NULL)
+        goto out;
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, heal, this, -1, error, 0, 0, 0, 0, NULL);
+    }
+}
+
+int32_t
+ec_heal_block_done(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, uintptr_t mask,
+                   uintptr_t good, uintptr_t bad, uint32_t pending,
+                   dict_t *xdata)
+{
+    ec_heal_t *heal = cookie;
+
+    if (heal->fop) {
+        heal->fop->heal = NULL;
+    }
+    heal->fop = NULL;
+    heal->error = op_ret < 0 ? op_errno : 0;
+    syncbarrier_wake(heal->data);
+    return 0;
+}
+
+int
+ec_sync_heal_block(call_frame_t *frame, xlator_t *this, ec_heal_t *heal)
+{
+    ec_heal_block(frame, this, heal->bad | heal->good, EC_MINIMUM_ONE,
+                  ec_heal_block_done, heal);
+    syncbarrier_wait(heal->data, 1);
+    if (heal->error != 0) {
+        return -heal->error;
+    }
+    if (heal->bad == 0)
+        return -ENOTCONN;
+    return 0;
+}
+
+int
+ec_rebuild_data(call_frame_t *frame, ec_t *ec, fd_t *fd, uint64_t size,
+                unsigned char *sources, unsigned char *healed_sinks)
+{
+    ec_heal_t *heal = NULL;
+    int ret = 0;
+    syncbarrier_t barrier;
+
+    if (syncbarrier_init(&barrier))
+        return -ENOMEM;
+
+    heal = alloca0(sizeof(*heal));
+    heal->fd = fd_ref(fd);
+    heal->xl = ec->xl;
+    heal->data = &barrier;
+    ec_adjust_size_up(ec, &size, _gf_false);
+    heal->total_size = size;
+    heal->size = (128 * GF_UNIT_KB * (ec->self_heal_window_size));
+    /* We need to adjust the size to a multiple of the stripe size of the
+     * volume. Otherwise writes would need to fill gaps (head and/or tail)
+     * with existent data from the bad bricks. This could be garbage on a
+     * damaged file or it could fail if there aren't enough bricks. */
+    heal->size -= heal->size % ec->stripe_size;
+    heal->bad = ec_char_array_to_mask(healed_sinks, ec->nodes);
+    heal->good = ec_char_array_to_mask(sources, ec->nodes);
+    heal->iatt.ia_type = IA_IFREG;
+    LOCK_INIT(&heal->lock);
+
+    for (heal->offset = 0; (heal->offset < size) && !heal->done;
+         heal->offset += heal->size) {
+        /* We immediately abort any heal if a shutdown request has been
+         * received to avoid delays. The healing of this file will be
+         * restarted by another SHD or other client that accesses the
+         * file. */
+        if (ec->shutdown) {
+            gf_msg_debug(ec->xl->name, 0,
+                         "Cancelling heal because "
+                         "EC is stopping.");
+            ret = -ENOTCONN;
+            break;
+        }
+
+        gf_msg_debug(ec->xl->name, 0,
+                     "%s: sources: %d, sinks: "
+                     "%d, offset: %" PRIu64 " bsize: %" PRIu64,
+                     uuid_utoa(fd->inode->gfid), EC_COUNT(sources, ec->nodes),
+                     EC_COUNT(healed_sinks, ec->nodes), heal->offset,
+                     heal->size);
+        ret = ec_sync_heal_block(frame, ec->xl, heal);
+        if (ret < 0)
+            break;
+    }
+    memset(healed_sinks, 0, ec->nodes);
+    ec_mask_to_char_array(heal->bad, healed_sinks, ec->nodes);
+    fd_unref(heal->fd);
+    LOCK_DESTROY(&heal->lock);
+    syncbarrier_destroy(heal->data);
+    if (ret < 0)
+        gf_msg_debug(ec->xl->name, 0, "%s: heal failed %s",
+                     uuid_utoa(fd->inode->gfid), strerror(-ret));
+    return ret;
+}
+
+int
+__ec_heal_trim_sinks(call_frame_t *frame, ec_t *ec, fd_t *fd,
+                     unsigned char *healed_sinks, unsigned char *trim,
+                     uint64_t size)
+{
+    default_args_cbk_t *replies = NULL;
+    unsigned char *output = NULL;
+    int ret = 0;
+    int i = 0;
+    off_t trim_offset = 0;
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    output = alloca0(ec->nodes);
+
+    if (EC_COUNT(trim, ec->nodes) == 0) {
+        ret = 0;
+        goto out;
+    }
+    trim_offset = size;
+    ec_adjust_offset_up(ec, &trim_offset, _gf_true);
+    ret = cluster_ftruncate(ec->xl_list, trim, ec->nodes, replies, output,
+                            frame, ec->xl, fd, trim_offset, NULL);
+    for (i = 0; i < ec->nodes; i++) {
+        if (!output[i] && trim[i])
+            healed_sinks[i] = 0;
+    }
+
+    if (EC_COUNT(healed_sinks, ec->nodes) == 0) {
+        ret = -ENOTCONN;
+        goto out;
+    }
+
+out:
+    cluster_replies_wipe(replies, ec->nodes);
+    if (ret < 0)
+        gf_msg_debug(ec->xl->name, 0, "%s: heal failed %s",
+                     uuid_utoa(fd->inode->gfid), strerror(-ret));
+    return ret;
+}
+
+int
+ec_data_undo_pending(call_frame_t *frame, ec_t *ec, fd_t *fd, dict_t *xattr,
+                     uint64_t *versions, uint64_t *dirty, uint64_t *size,
+                     int source, gf_boolean_t erase_dirty, int idx)
+{
+    uint64_t versions_xattr[2] = {0};
+    uint64_t dirty_xattr[2] = {0};
+    uint64_t allzero[2] = {0};
+    uint64_t size_xattr = 0;
+    int ret = 0;
+
+    versions_xattr[EC_DATA_TXN] = hton64(versions[source] - versions[idx]);
+    ret = dict_set_static_bin(xattr, EC_XATTR_VERSION, versions_xattr,
+                              sizeof(versions_xattr));
+    if (ret < 0)
+        goto out;
+
+    size_xattr = hton64(size[source] - size[idx]);
+    ret = dict_set_static_bin(xattr, EC_XATTR_SIZE, &size_xattr,
+                              sizeof(size_xattr));
+    if (ret < 0)
+        goto out;
+
+    if (erase_dirty) {
+        dirty_xattr[EC_DATA_TXN] = hton64(-dirty[idx]);
+        ret = dict_set_static_bin(xattr, EC_XATTR_DIRTY, dirty_xattr,
+                                  sizeof(dirty_xattr));
+        if (ret < 0)
+            goto out;
+    }
+
+    if ((memcmp(versions_xattr, allzero, sizeof(allzero)) == 0) &&
+        (memcmp(dirty_xattr, allzero, sizeof(allzero)) == 0) &&
+        (size_xattr == 0)) {
+        ret = 0;
+        goto out;
+    }
+
+    ret = syncop_fxattrop(ec->xl_list[idx], fd, GF_XATTROP_ADD_ARRAY64, xattr,
+                          NULL, NULL, NULL);
+out:
+    return ret;
+}
+
+int
+__ec_fd_data_adjust_versions(call_frame_t *frame, ec_t *ec, fd_t *fd,
+                             unsigned char *sources,
+                             unsigned char *healed_sinks, uint64_t *versions,
+                             uint64_t *dirty, uint64_t *size)
+{
+    dict_t *xattr = NULL;
+    int i = 0;
+    int ret = 0;
+    int op_ret = 0;
+    int source = -1;
+    gf_boolean_t erase_dirty = _gf_false;
+
+    xattr = dict_new();
+    if (!xattr) {
+        op_ret = -ENOMEM;
+        goto out;
+    }
+
+    /* dirty xattr represents if the file needs heal. Unless all the
+     * copies are healed, don't erase it */
+    if (EC_COUNT(sources, ec->nodes) + EC_COUNT(healed_sinks, ec->nodes) ==
+        ec->nodes)
+        erase_dirty = _gf_true;
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (sources[i]) {
+            source = i;
+            break;
+        }
+    }
+
+    if (source == -1) {
+        op_ret = -ENOTCONN;
+        goto out;
+    }
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (healed_sinks[i]) {
+            ret = ec_data_undo_pending(frame, ec, fd, xattr, versions, dirty,
+                                       size, source, erase_dirty, i);
+            if (ret < 0)
+                goto out;
+        }
+    }
+
+    if (!erase_dirty)
+        goto out;
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (sources[i]) {
+            ret = ec_data_undo_pending(frame, ec, fd, xattr, versions, dirty,
+                                       size, source, erase_dirty, i);
+            if (ret < 0)
+                continue;
+        }
+    }
+out:
+    if (xattr)
+        dict_unref(xattr);
+    return op_ret;
+}
+
+int
+ec_restore_time_and_adjust_versions(call_frame_t *frame, ec_t *ec, fd_t *fd,
+                                    unsigned char *sources,
+                                    unsigned char *healed_sinks,
+                                    uint64_t *versions, uint64_t *dirty,
+                                    uint64_t *size)
+{
+    unsigned char *locked_on = NULL;
+    unsigned char *participants = NULL;
+    unsigned char *output = NULL;
+    default_args_cbk_t *replies = NULL;
+    unsigned char *postsh_sources = NULL;
+    unsigned char *postsh_healed_sinks = NULL;
+    unsigned char *postsh_trim = NULL;
+    uint64_t *postsh_versions = NULL;
+    uint64_t *postsh_dirty = NULL;
+    uint64_t *postsh_size = NULL;
+    int ret = 0;
+    int i = 0;
+    struct iatt source_buf = {0};
+    loc_t loc = {0};
+
+    locked_on = alloca0(ec->nodes);
+    output = alloca0(ec->nodes);
+    participants = alloca0(ec->nodes);
+    postsh_sources = alloca0(ec->nodes);
+    postsh_healed_sinks = alloca0(ec->nodes);
+    postsh_trim = alloca0(ec->nodes);
+    postsh_versions = alloca0(ec->nodes * sizeof(*postsh_versions));
+    postsh_dirty = alloca0(ec->nodes * sizeof(*postsh_dirty));
+    postsh_size = alloca0(ec->nodes * sizeof(*postsh_size));
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (healed_sinks[i] || sources[i])
+            participants[i] = 1;
+    }
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    ret = cluster_inodelk(ec->xl_list, participants, ec->nodes, replies,
+                          locked_on, frame, ec->xl, ec->xl->name, fd->inode, 0,
+                          0);
+    {
+        if (ret <= ec->fragments) {
+            gf_msg_debug(ec->xl->name, 0,
+                         "%s: Skipping heal "
+                         "as only %d number of subvolumes could "
+                         "be locked",
+                         uuid_utoa(fd->inode->gfid), ret);
+            ret = -ENOTCONN;
+            goto unlock;
+        }
+
+        ret = __ec_heal_data_prepare(frame, ec, fd, locked_on, postsh_versions,
+                                     postsh_dirty, postsh_size, postsh_sources,
+                                     postsh_healed_sinks, postsh_trim,
+                                     &source_buf);
+        if (ret < 0)
+            goto unlock;
+
+        loc.inode = inode_ref(fd->inode);
+        gf_uuid_copy(loc.gfid, fd->inode->gfid);
+        ret = cluster_setattr(
+            ec->xl_list, healed_sinks, ec->nodes, replies, output, frame,
+            ec->xl, &loc, &source_buf,
+            GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME | GF_SET_ATTR_CTIME, NULL);
+        EC_INTERSECT(healed_sinks, healed_sinks, output, ec->nodes);
+        if (EC_COUNT(healed_sinks, ec->nodes) == 0) {
+            ret = -ENOTCONN;
+            goto unlock;
+        }
+        ret = __ec_fd_data_adjust_versions(frame, ec, fd, sources, healed_sinks,
+                                           versions, dirty, size);
+    }
+unlock:
+    cluster_uninodelk(ec->xl_list, locked_on, ec->nodes, replies, output, frame,
+                      ec->xl, ec->xl->name, fd->inode, 0, 0);
+    cluster_replies_wipe(replies, ec->nodes);
+    loc_wipe(&loc);
+    return ret;
+}
+
+int
+__ec_heal_data(call_frame_t *frame, ec_t *ec, fd_t *fd, unsigned char *heal_on,
+               unsigned char *sources, unsigned char *healed_sinks)
+{
+    unsigned char *locked_on = NULL;
+    unsigned char *output = NULL;
+    uint64_t *versions = NULL;
+    uint64_t *dirty = NULL;
+    uint64_t *size = NULL;
+    unsigned char *trim = NULL;
+    default_args_cbk_t *replies = NULL;
+    int ret = 0;
+    int source = 0;
+
+    locked_on = alloca0(ec->nodes);
+    output = alloca0(ec->nodes);
+    trim = alloca0(ec->nodes);
+    versions = alloca0(ec->nodes * sizeof(*versions));
+    dirty = alloca0(ec->nodes * sizeof(*dirty));
+    size = alloca0(ec->nodes * sizeof(*size));
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    ret = cluster_inodelk(ec->xl_list, heal_on, ec->nodes, replies, locked_on,
+                          frame, ec->xl, ec->xl->name, fd->inode, 0, 0);
+    {
+        if (ret <= ec->fragments) {
+            gf_msg_debug(ec->xl->name, 0,
+                         "%s: Skipping heal "
+                         "as only %d number of subvolumes could "
+                         "be locked",
+                         uuid_utoa(fd->inode->gfid), ret);
+            ret = -ENOTCONN;
+            goto unlock;
+        }
+
+        ret = __ec_heal_data_prepare(frame, ec, fd, locked_on, versions, dirty,
+                                     size, sources, healed_sinks, trim, NULL);
+        if (ret < 0)
+            goto unlock;
+
+        if (EC_COUNT(healed_sinks, ec->nodes) == 0) {
+            ret = __ec_fd_data_adjust_versions(
+                frame, ec, fd, sources, healed_sinks, versions, dirty, size);
+            goto unlock;
+        }
+
+        source = ret;
+        ret = __ec_heal_mark_sinks(frame, ec, fd, versions, healed_sinks);
+        if (ret < 0)
+            goto unlock;
+
+        ret = __ec_heal_trim_sinks(frame, ec, fd, healed_sinks, trim,
+                                   size[source]);
+    }
+unlock:
+    cluster_uninodelk(ec->xl_list, locked_on, ec->nodes, replies, output, frame,
+                      ec->xl, ec->xl->name, fd->inode, 0, 0);
+    if (ret < 0)
+        goto out;
+
+    if (EC_COUNT(healed_sinks, ec->nodes) == 0)
+        goto out;
+
+    gf_msg_debug(ec->xl->name, 0,
+                 "%s: sources: %d, sinks: "
+                 "%d",
+                 uuid_utoa(fd->inode->gfid), EC_COUNT(sources, ec->nodes),
+                 EC_COUNT(healed_sinks, ec->nodes));
+
+    ret = ec_rebuild_data(frame, ec, fd, size[source], sources, healed_sinks);
+    if (ret < 0)
+        goto out;
+
+    ret = ec_restore_time_and_adjust_versions(
+        frame, ec, fd, sources, healed_sinks, versions, dirty, size);
+out:
+    cluster_replies_wipe(replies, ec->nodes);
+    return ret;
+}
+
+int
+ec_heal_data(call_frame_t *frame, ec_t *ec, gf_boolean_t block, inode_t *inode,
+             unsigned char *sources, unsigned char *healed_sinks)
+{
+    unsigned char *locked_on = NULL;
+    unsigned char *up_subvols = NULL;
+    unsigned char *output = NULL;
+    default_args_cbk_t *replies = NULL;
+    fd_t *fd = NULL;
+    loc_t loc = {0};
+    char selfheal_domain[1024] = {0};
+    int ret = 0;
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+
+    locked_on = alloca0(ec->nodes);
+    output = alloca0(ec->nodes);
+    up_subvols = alloca0(ec->nodes);
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+
+    fd = fd_create(inode, 0);
+    if (!fd) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    ec_mask_to_char_array(ec->xl_up, up_subvols, ec->nodes);
+
+    ret = cluster_open(ec->xl_list, up_subvols, ec->nodes, replies, output,
+                       frame, ec->xl, &loc, O_RDWR | O_LARGEFILE, fd, NULL);
+    if (ret <= ec->fragments) {
+        ret = -ENOTCONN;
+        goto out;
+    }
+
+    fd_bind(fd);
+    sprintf(selfheal_domain, "%s:self-heal", ec->xl->name);
+    /*If other processes are already doing the heal, don't block*/
+    if (block) {
+        ret = cluster_inodelk(ec->xl_list, output, ec->nodes, replies,
+                              locked_on, frame, ec->xl, selfheal_domain, inode,
+                              0, 0);
+    } else {
+        ret = cluster_tiebreaker_inodelk(ec->xl_list, output, ec->nodes,
+                                         replies, locked_on, frame, ec->xl,
+                                         selfheal_domain, inode, 0, 0);
+    }
+    {
+        if (ret <= ec->fragments) {
+            gf_msg_debug(ec->xl->name, 0,
+                         "%s: Skipping heal "
+                         "as only %d number of subvolumes could "
+                         "be locked",
+                         uuid_utoa(inode->gfid), ret);
+            ret = -ENOTCONN;
+            goto unlock;
+        }
+        ret = __ec_heal_data(frame, ec, fd, locked_on, sources, healed_sinks);
+    }
+unlock:
+    cluster_uninodelk(ec->xl_list, locked_on, ec->nodes, replies, output, frame,
+                      ec->xl, selfheal_domain, inode, 0, 0);
+out:
+    if (fd)
+        fd_unref(fd);
+    loc_wipe(&loc);
+    cluster_replies_wipe(replies, ec->nodes);
+    return ret;
+}
+
+int
+ec_heal_purge_stale_index(call_frame_t *frame, ec_t *ec, inode_t *inode)
+{
+    int i = 0;
+    int ret = 0;
+    dict_t **xattr = NULL;
+    loc_t loc = {0};
+    uint64_t dirty_xattr[EC_VERSION_SIZE] = {0};
+    unsigned char *on = NULL;
+    default_args_cbk_t *replies = NULL;
+    dict_t *dict = NULL;
+
+    /* Allocate the required memory */
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+    on = alloca0(ec->nodes);
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    xattr = GF_CALLOC(ec->nodes, sizeof(*xattr), gf_common_mt_pointer);
+    if (!xattr) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    dict = dict_new();
+    if (!dict) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    for (i = 0; i < ec->nodes; i++) {
+        xattr[i] = dict;
+        on[i] = 1;
+    }
+    ret = dict_set_static_bin(dict, EC_XATTR_DIRTY, dirty_xattr,
+                              (sizeof(*dirty_xattr) * EC_VERSION_SIZE));
+    if (ret < 0) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    PARALLEL_FOP_ONLIST(ec->xl_list, on, ec->nodes, replies, frame,
+                        ec_wind_xattrop_parallel, &loc, GF_XATTROP_ADD_ARRAY64,
+                        xattr, NULL);
+out:
+    if (dict) {
+        dict_unref(dict);
+    }
+    if (xattr) {
+        GF_FREE(xattr);
+    }
+    cluster_replies_wipe(replies, ec->nodes);
+    loc_wipe(&loc);
+    return ret;
+}
+
+void
+ec_heal_do(xlator_t *this, void *data, loc_t *loc, int32_t partial)
+{
+    call_frame_t *frame = NULL;
+    unsigned char *participants = NULL;
+    unsigned char *msources = NULL;
+    unsigned char *mhealed_sinks = NULL;
+    unsigned char *sources = NULL;
+    unsigned char *healed_sinks = NULL;
+    ec_t *ec = NULL;
+    int ret = 0;
+    int op_ret = 0;
+    int op_errno = 0;
+    intptr_t mgood = 0;
+    intptr_t mbad = 0;
+    intptr_t good = 0;
+    intptr_t bad = 0;
+    uint32_t pending = 0;
+    ec_fop_data_t *fop = data;
+    gf_boolean_t blocking = _gf_false;
+    ec_heal_need_t need_heal = EC_HEAL_NONEED;
+    unsigned char *up_subvols = NULL;
+    char up_bricks[32];
+
+    ec = this->private;
+
+    /* If it is heal request from getxattr, complete the heal and then
+     * unwind, if it is ec_heal with NULL as frame then no need to block
+     * the heal as the caller doesn't care about its completion. In case
+     * of heald whichever gets tiebreaking inodelk will take care of the
+     * heal, so no need to block*/
+    if (fop->req_frame && !ec->shd.iamshd)
+        blocking = _gf_true;
+
+    frame = create_frame(this, this->ctx->pool);
+    if (!frame)
+        goto out;
+
+    ec_owner_set(frame, frame->root);
+    /*Do heal as root*/
+    frame->root->uid = 0;
+    frame->root->gid = 0;
+    /*Mark the fops as internal*/
+    frame->root->pid = GF_CLIENT_PID_SELF_HEALD;
+    participants = alloca0(ec->nodes);
+    ec_mask_to_char_array(ec->xl_up, participants, ec->nodes);
+
+    up_subvols = alloca0(ec->nodes);
+    ec_mask_to_char_array(ec->xl_up, up_subvols, ec->nodes);
+
+    if (loc->name && strlen(loc->name)) {
+        ret = ec_heal_name(frame, ec, loc->parent, (char *)loc->name,
+                           participants);
+        if (ret >= 0) {
+            gf_msg_debug(this->name, 0,
+                         "%s: name heal "
+                         "successful on %" PRIXPTR,
+                         loc->path,
+                         ec_char_array_to_mask(participants, ec->nodes));
+        } else {
+            gf_msg_debug(
+                this->name, 0,
+                "%s: name heal "
+                "failed. ret = %d, subvolumes up = %s",
+                loc->path, ret,
+                ec_bin(up_bricks, sizeof(up_bricks), ec->xl_up, ec->nodes));
+        }
+    }
+
+    /* Mount triggers heal only when it detects that it must need heal, shd
+     * triggers heals periodically which need not be thorough*/
+    if (ec->shd.iamshd && (ret <= 0)) {
+        ec_heal_inspect(frame, ec, loc->inode, up_subvols, _gf_false, _gf_false,
+                        &need_heal);
+
+        if (need_heal == EC_HEAL_PURGE_INDEX) {
+            gf_msg(ec->xl->name, GF_LOG_INFO, 0, EC_MSG_HEAL_FAIL,
+                   "Index entry needs to be purged for: %s ",
+                   uuid_utoa(loc->gfid));
+            /* We need to send zero-xattrop so that stale index entry could be
+             * removed. We need not take lock on this entry to do so as
+             * xattrop on a brick is atomic. */
+            ec_heal_purge_stale_index(frame, ec, loc->inode);
+            goto out;
+        } else if (need_heal == EC_HEAL_NONEED) {
+            gf_msg(ec->xl->name, GF_LOG_DEBUG, 0, EC_MSG_HEAL_FAIL,
+                   "Heal is not required for : %s ", uuid_utoa(loc->gfid));
+            goto out;
+        }
+    }
+
+    sources = alloca0(ec->nodes);
+    healed_sinks = alloca0(ec->nodes);
+    if (IA_ISREG(loc->inode->ia_type)) {
+        ret = ec_heal_data(frame, ec, blocking, loc->inode, sources,
+                           healed_sinks);
+    } else if (IA_ISDIR(loc->inode->ia_type) && !partial) {
+        ret = ec_heal_entry(frame, ec, loc->inode, sources, healed_sinks,
+                            &pending);
+    } else {
+        ret = 0;
+        memcpy(sources, participants, ec->nodes);
+        memcpy(healed_sinks, participants, ec->nodes);
+    }
+
+    if (ret == 0) {
+        good = ec_char_array_to_mask(sources, ec->nodes);
+        bad = ec_char_array_to_mask(healed_sinks, ec->nodes);
+    } else {
+        op_ret = -1;
+        op_errno = -ret;
+    }
+    msources = alloca0(ec->nodes);
+    mhealed_sinks = alloca0(ec->nodes);
+    ret = ec_heal_metadata(frame, ec, loc->inode, msources, mhealed_sinks);
+    if (ret == 0) {
+        mgood = ec_char_array_to_mask(msources, ec->nodes);
+        mbad = ec_char_array_to_mask(mhealed_sinks, ec->nodes);
+    } else {
+        op_ret = -1;
+        op_errno = -ret;
+    }
+
+out:
+    ec_reset_entry_healing(fop);
+    if (fop->cbks.heal) {
+        fop->cbks.heal(fop->req_frame, fop->data, fop->xl, op_ret, op_errno,
+                       ec_char_array_to_mask(participants, ec->nodes),
+                       mgood & good, mbad & bad, pending, NULL);
+    }
+    if (frame)
+        STACK_DESTROY(frame->root);
+
+    return;
+}
+
+int
+ec_synctask_heal_wrap(void *opaque)
+{
+    ec_fop_data_t *fop = opaque;
+    ec_heal_do(fop->xl, fop, &fop->loc[0], fop->int32);
+    return 0;
+}
+
+int
+ec_heal_done(int ret, call_frame_t *heal, void *opaque)
+{
+    if (opaque)
+        ec_fop_data_release(opaque);
+    return 0;
+}
+
+ec_fop_data_t *
+__ec_dequeue_heals(ec_t *ec)
+{
+    ec_fop_data_t *fop = NULL;
+
+    if (list_empty(&ec->heal_waiting))
+        goto none;
+
+    if ((ec->background_heals > 0) && (ec->healers >= ec->background_heals))
+        goto none;
+
+    fop = list_entry(ec->heal_waiting.next, ec_fop_data_t, healer);
+    ec->heal_waiters--;
+    list_del_init(&fop->healer);
+    list_add(&fop->healer, &ec->healing);
+    ec->healers++;
+    return fop;
+none:
+    gf_msg_debug(ec->xl->name, 0, "Num healers: %d, Num Waiters: %d",
+                 ec->healers, ec->heal_waiters);
+    return NULL;
+}
+
+void
+ec_heal_fail(ec_t *ec, ec_fop_data_t *fop)
+{
+    if (fop->cbks.heal) {
+        fop->cbks.heal(fop->req_frame, fop->data, ec->xl, -1, fop->error, 0, 0,
+                       0, 0, NULL);
+    }
+    ec_fop_data_release(fop);
+}
+
+void
+ec_launch_heal(ec_t *ec, ec_fop_data_t *fop)
+{
+    int ret = 0;
+    call_frame_t *frame = NULL;
+
+    frame = create_frame(ec->xl, ec->xl->ctx->pool);
+    if (!frame) {
+        ret = -1;
+        goto out;
+    }
+
+    ec_owner_set(frame, frame->root);
+    /*Do heal as root*/
+    frame->root->uid = 0;
+    frame->root->gid = 0;
+    /*Mark the fops as internal*/
+    frame->root->pid = GF_CLIENT_PID_SELF_HEALD;
+
+    ret = synctask_new(ec->xl->ctx->env, ec_synctask_heal_wrap, ec_heal_done,
+                       frame, fop);
+out:
+    if (ret < 0) {
+        ec_fop_set_error(fop, ENOMEM);
+        ec_heal_fail(ec, fop);
+    }
+
+    if (frame)
+        STACK_DESTROY(frame->root);
+}
+
+void
+ec_handle_healers_done(ec_fop_data_t *fop)
+{
+    ec_t *ec = fop->xl->private;
+    ec_fop_data_t *heal_fop = NULL;
+
+    if (list_empty(&fop->healer))
+        return;
+
+    LOCK(&ec->lock);
+
+    list_del_init(&fop->healer);
+
+    do {
+        ec->healers--;
+        heal_fop = __ec_dequeue_heals(ec);
+
+        if ((heal_fop != NULL) && ec->shutdown) {
+            /* This will prevent ec_handle_healers_done() to be
+             * called recursively. That would be problematic if
+             * the queue is too big. */
+            list_del_init(&heal_fop->healer);
+
+            UNLOCK(&ec->lock);
+
+            ec_fop_set_error(fop, ENOTCONN);
+            ec_heal_fail(ec, heal_fop);
+
+            LOCK(&ec->lock);
+        }
+    } while ((heal_fop != NULL) && ec->shutdown);
+
+    UNLOCK(&ec->lock);
+
+    if (heal_fop)
+        ec_launch_heal(ec, heal_fop);
+}
+
+gf_boolean_t
+ec_is_entry_healing(ec_fop_data_t *fop)
+{
+    ec_inode_t *ctx = NULL;
+    int32_t heal_count = 0;
+    loc_t *loc = NULL;
+
+    loc = &fop->loc[0];
+
+    LOCK(&loc->inode->lock);
+    {
+        ctx = __ec_inode_get(loc->inode, fop->xl);
+        if (ctx) {
+            heal_count = ctx->heal_count;
+        }
+    }
+    UNLOCK(&loc->inode->lock);
+    GF_ASSERT(heal_count >= 0);
+    return heal_count;
+}
+
+void
+ec_heal_throttle(xlator_t *this, ec_fop_data_t *fop)
+{
+    gf_boolean_t can_heal = _gf_true;
+    ec_t *ec = this->private;
+    ec_fop_data_t *fop_rel = NULL;
+
+    if (fop->req_frame == NULL) {
+        LOCK(&ec->lock);
+        {
+            if ((ec->background_heals > 0) &&
+                (ec->heal_wait_qlen + ec->background_heals) >
+                    (ec->heal_waiters + ec->healers)) {
+                if (!ec_is_entry_healing(fop)) {
+                    list_add_tail(&fop->healer, &ec->heal_waiting);
+                    ec->heal_waiters++;
+                    ec_set_entry_healing(fop);
+                } else {
+                    fop_rel = fop;
+                }
+                fop = __ec_dequeue_heals(ec);
+            } else {
+                can_heal = _gf_false;
+            }
+        }
+        UNLOCK(&ec->lock);
+    }
+
+    if (can_heal) {
+        if (fop) {
+            if (fop->req_frame != NULL) {
+                ec_set_entry_healing(fop);
+            }
+            ec_launch_heal(ec, fop);
+        }
+    } else {
+        gf_msg_debug(this->name, 0,
+                     "Max number of heals are "
+                     "pending, background self-heal rejected");
+        ec_fop_set_error(fop, EBUSY);
+        ec_heal_fail(ec, fop);
+    }
+    if (fop_rel) {
+        ec_heal_done(0, NULL, fop_rel);
+    }
+}
+
+void
+ec_heal(call_frame_t *frame, xlator_t *this, uintptr_t target,
+        uint32_t fop_flags, fop_heal_cbk_t func, void *data, loc_t *loc,
+        int32_t partial, dict_t *xdata)
+{
+    ec_cbk_t callback = {.heal = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t err = EINVAL;
+
+    gf_msg_trace("ec", 0, "EC(HEAL) %p", frame);
+
+    VALIDATE_OR_GOTO(this, fail);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, fail);
+
+    if (!loc || !loc->inode || gf_uuid_is_null(loc->inode->gfid))
+        goto fail;
+
+    if (frame && frame->local)
+        goto fail;
+    fop = ec_fop_data_allocate(frame, this, EC_FOP_HEAL, 0, target, fop_flags,
+                               NULL, NULL, callback, data);
+
+    err = ENOMEM;
+
+    if (fop == NULL)
+        goto fail;
+
+    fop->int32 = partial;
+
+    if (loc) {
+        if (loc_copy(&fop->loc[0], loc) != 0)
+            goto fail;
+    }
+
+    if (xdata)
+        fop->xdata = dict_ref(xdata);
+
+    ec_heal_throttle(this, fop);
+
+    return;
+
+fail:
+    if (fop)
+        ec_fop_data_release(fop);
+    if (func)
+        func(frame, data, this, -1, err, 0, 0, 0, 0, NULL);
+}
+
+int
+ec_replace_heal_done(int ret, call_frame_t *heal, void *opaque)
+{
+    ec_t *ec = opaque;
+    gf_boolean_t last_fop = _gf_false;
+
+    if (GF_ATOMIC_DEC(ec->async_fop_count) == 0) {
+        LOCK(&ec->lock);
+        {
+            last_fop = __ec_is_last_fop(ec);
+        }
+        UNLOCK(&ec->lock);
+    }
+    gf_msg_debug(ec->xl->name, 0, "getxattr on bricks is done ret %d", ret);
+
+    if (last_fop)
+        ec_pending_fops_completed(ec);
+
+    return 0;
+}
+
+int32_t
+ec_replace_heal(ec_t *ec, inode_t *inode)
+{
+    loc_t loc = {0};
+    int ret = 0;
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+    ret = syncop_getxattr(ec->xl, &loc, NULL, EC_XATTR_HEAL, NULL, NULL);
+    if (ret < 0)
+        gf_msg_debug(ec->xl->name, 0, "Heal failed for replace brick ret = %d",
+                     ret);
+
+    /* Once the root inode has been checked, it might have triggered a
+     * self-heal on it after a replace brick command or for some other
+     * reason. It can also happen that the volume already had damaged
+     * files in the index, even if the heal on the root directory failed.
+     * In both cases we need to wake all index healers to continue
+     * healing remaining entries that are marked as dirty. */
+    ec_shd_index_healer_wake(ec);
+
+    loc_wipe(&loc);
+    return ret;
+}
+
+int32_t
+ec_replace_brick_heal_wrap(void *opaque)
+{
+    ec_t *ec = opaque;
+    inode_table_t *itable = NULL;
+    int32_t ret = -1;
+
+    if (ec->xl->itable)
+        itable = ec->xl->itable;
+    else
+        goto out;
+
+    if (xlator_is_cleanup_starting(ec->xl))
+        goto out;
+
+    ret = ec_replace_heal(ec, itable->root);
+out:
+    return ret;
+}
+
+int32_t
+ec_launch_replace_heal(ec_t *ec)
+{
+    int ret = -1;
+
+    ret = synctask_new(ec->xl->ctx->env, ec_replace_brick_heal_wrap,
+                       ec_replace_heal_done, NULL, ec);
+
+    if (ret < 0) {
+        gf_msg_debug(ec->xl->name, 0, "Heal failed for replace brick ret = %d",
+                     ret);
+        ec_replace_heal_done(-1, NULL, ec);
+    }
+
+    return ret;
+}
+
+int32_t
+ec_set_heal_info(dict_t **dict_rsp, char *status)
+{
+    dict_t *dict = NULL;
+    int ret = 0;
+
+    dict = dict_new();
+    if (!dict) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    ret = dict_set_str(dict, "heal-info", status);
+    if (ret) {
+        gf_msg(THIS->name, GF_LOG_WARNING, -ret, EC_MSG_HEAL_FAIL,
+               "Failed to set heal-info key to "
+               "%s",
+               status);
+        dict_unref(dict);
+        dict = NULL;
+    }
+    *dict_rsp = dict;
+out:
+    return ret;
+}
+
+static int32_t
+_need_heal_calculate(ec_t *ec, uint64_t *dirty, unsigned char *sources,
+                     gf_boolean_t self_locked, int32_t lock_count,
+                     ec_heal_need_t *need_heal, uint64_t *versions)
+{
+    int i = 0;
+    int source_count = 0;
+
+    source_count = EC_COUNT(sources, ec->nodes);
+    if (source_count == ec->nodes) {
+        *need_heal = EC_HEAL_NONEED;
+        if (self_locked || lock_count == 0) {
+            for (i = 0; i < ec->nodes; i++) {
+                if (dirty[i] || (versions[i] != versions[0])) {
+                    *need_heal = EC_HEAL_MUST;
+                    goto out;
+                }
+            }
+            /* If lock count is 0, all dirty flags are 0 and all the
+             * versions are macthing then why are we here. It looks
+             * like something went wrong while removing the index entries
+             * after completing a successful heal or fop. In this case
+             * we need to remove this index entry to avoid triggering heal
+             * in a loop and causing lookups again and again*/
+            *need_heal = EC_HEAL_PURGE_INDEX;
+        } else {
+            for (i = 0; i < ec->nodes; i++) {
+                /* Since each lock can only increment the dirty
+                 * count once, if dirty is > 1 it means that
+                 * another operation has left the dirty count
+                 * set and this indicates a problem in the
+                 * inode.*/
+                if (dirty[i] > 1) {
+                    *need_heal = EC_HEAL_MUST;
+                    goto out;
+                }
+                if (dirty[i] != dirty[0] || (versions[i] != versions[0])) {
+                    *need_heal = EC_HEAL_MAYBE;
+                }
+            }
+        }
+    } else {
+        *need_heal = EC_HEAL_MUST;
+    }
+
+out:
+    return source_count;
+}
+
+static int32_t
+ec_need_metadata_heal(ec_t *ec, inode_t *inode, default_args_cbk_t *replies,
+                      int32_t lock_count, gf_boolean_t self_locked,
+                      gf_boolean_t thorough, ec_heal_need_t *need_heal)
+{
+    uint64_t *dirty = NULL;
+    unsigned char *sources = NULL;
+    unsigned char *healed_sinks = NULL;
+    uint64_t *meta_versions = NULL;
+    int ret = 0;
+
+    sources = alloca0(ec->nodes);
+    healed_sinks = alloca0(ec->nodes);
+    dirty = alloca0(ec->nodes * sizeof(*dirty));
+    meta_versions = alloca0(ec->nodes * sizeof(*meta_versions));
+    ret = ec_heal_metadata_find_direction(ec, replies, meta_versions, dirty,
+                                          sources, healed_sinks);
+    if (ret < 0 && ret != -EIO) {
+        goto out;
+    }
+
+    ret = _need_heal_calculate(ec, dirty, sources, self_locked, lock_count,
+                               need_heal, meta_versions);
+out:
+    return ret;
+}
+
+static int32_t
+ec_need_data_heal(ec_t *ec, inode_t *inode, default_args_cbk_t *replies,
+                  int32_t lock_count, gf_boolean_t self_locked,
+                  gf_boolean_t thorough, ec_heal_need_t *need_heal)
+{
+    uint64_t *dirty = NULL;
+    unsigned char *sources = NULL;
+    unsigned char *healed_sinks = NULL;
+    uint64_t *data_versions = NULL;
+    uint64_t *size = NULL;
+    int ret = 0;
+
+    sources = alloca0(ec->nodes);
+    healed_sinks = alloca0(ec->nodes);
+    dirty = alloca0(ec->nodes * sizeof(*dirty));
+    data_versions = alloca0(ec->nodes * sizeof(*data_versions));
+    size = alloca0(ec->nodes * sizeof(*size));
+
+    /* When dd is going on and heal info is called there is a very good
+     * chance for on disk sizes to mismatch even though nothing is wrong
+     * we don't need ondisk size check there. But if the file is either
+     * self-locked or the caller wants a thorough check then make sure to
+     * perform on disk check also. */
+    ret = ec_heal_data_find_direction(
+        ec, replies, data_versions, dirty, size, sources, healed_sinks,
+        self_locked || thorough, EC_COMBINE_XDATA);
+    if (ret < 0 && ret != -EIO) {
+        goto out;
+    }
+
+    ret = _need_heal_calculate(ec, dirty, sources, self_locked, lock_count,
+                               need_heal, data_versions);
+out:
+    return ret;
+}
+
+static int32_t
+ec_need_entry_heal(ec_t *ec, inode_t *inode, default_args_cbk_t *replies,
+                   int32_t lock_count, gf_boolean_t self_locked,
+                   gf_boolean_t thorough, ec_heal_need_t *need_heal)
+{
+    uint64_t *dirty = NULL;
+    unsigned char *sources = NULL;
+    unsigned char *healed_sinks = NULL;
+    uint64_t *data_versions = NULL;
+    int ret = 0;
+
+    sources = alloca0(ec->nodes);
+    healed_sinks = alloca0(ec->nodes);
+    dirty = alloca0(ec->nodes * sizeof(*dirty));
+    data_versions = alloca0(ec->nodes * sizeof(*data_versions));
+
+    ret = ec_heal_entry_find_direction(ec, replies, data_versions, dirty,
+                                       sources, healed_sinks);
+    if (ret < 0 && ret != -EIO) {
+        goto out;
+    }
+
+    ret = _need_heal_calculate(ec, dirty, sources, self_locked, lock_count,
+                               need_heal, data_versions);
+out:
+    return ret;
+}
+
+static int32_t
+ec_need_heal(ec_t *ec, inode_t *inode, default_args_cbk_t *replies,
+             int32_t lock_count, gf_boolean_t self_locked,
+             gf_boolean_t thorough, ec_heal_need_t *need_heal)
+{
+    int ret = 0;
+
+    ret = ec_need_metadata_heal(ec, inode, replies, lock_count, self_locked,
+                                thorough, need_heal);
+    if (ret < 0)
+        goto out;
+
+    if (*need_heal == EC_HEAL_MUST)
+        goto out;
+
+    if (inode->ia_type == IA_IFREG) {
+        ret = ec_need_data_heal(ec, inode, replies, lock_count, self_locked,
+                                thorough, need_heal);
+    } else if (inode->ia_type == IA_IFDIR) {
+        ret = ec_need_entry_heal(ec, inode, replies, lock_count, self_locked,
+                                 thorough, need_heal);
+    }
+
+out:
+    return ret;
+}
+
+int32_t
+ec_heal_inspect(call_frame_t *frame, ec_t *ec, inode_t *inode,
+                unsigned char *locked_on, gf_boolean_t self_locked,
+                gf_boolean_t thorough, ec_heal_need_t *need_heal)
+{
+    loc_t loc = {0};
+    int i = 0;
+    int ret = 0;
+    dict_t *xdata = NULL;
+    uint64_t zero_array[2] = {0};
+    uint64_t zero_value = 0;
+    unsigned char *output = NULL;
+    default_args_cbk_t *replies = NULL;
+    int32_t lock_count = 0;
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    output = alloca0(ec->nodes);
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+
+    xdata = dict_new();
+    if (!xdata ||
+        dict_set_static_bin(xdata, EC_XATTR_VERSION, zero_array,
+                            sizeof(zero_array)) ||
+        dict_set_static_bin(xdata, EC_XATTR_DIRTY, zero_array,
+                            sizeof(zero_array)) ||
+        dict_set_static_bin(xdata, EC_XATTR_SIZE, &zero_value,
+                            sizeof(zero_value))) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    if (!self_locked) {
+        ret = dict_set_str(xdata, GLUSTERFS_INODELK_DOM_COUNT, ec->xl->name);
+        if (ret) {
+            ret = -ENOMEM;
+            goto out;
+        }
+    }
+
+    ret = cluster_lookup(ec->xl_list, locked_on, ec->nodes, replies, output,
+                         frame, ec->xl, &loc, xdata);
+
+    if (ret != ec->nodes) {
+        ret = ec->nodes;
+        *need_heal = EC_HEAL_MUST;
+        goto out;
+    }
+
+    if (self_locked)
+        goto need_heal;
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (!output[i] || !replies[i].xdata) {
+            continue;
+        }
+        if ((dict_get_int32(replies[i].xdata, GLUSTERFS_INODELK_COUNT,
+                            &lock_count) == 0) &&
+            lock_count > 0) {
+            break;
+        }
+    }
+need_heal:
+    ret = ec_need_heal(ec, inode, replies, lock_count, self_locked, thorough,
+                       need_heal);
+out:
+    cluster_replies_wipe(replies, ec->nodes);
+    loc_wipe(&loc);
+    if (xdata) {
+        dict_unref(xdata);
+    }
+    return ret;
+}
+
+int32_t
+ec_heal_locked_inspect(call_frame_t *frame, ec_t *ec, inode_t *inode,
+                       ec_heal_need_t *need_heal)
+{
+    unsigned char *locked_on = NULL;
+    unsigned char *up_subvols = NULL;
+    unsigned char *output = NULL;
+    default_args_cbk_t *replies = NULL;
+    int ret = 0;
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    locked_on = alloca0(ec->nodes);
+    output = alloca0(ec->nodes);
+    up_subvols = alloca0(ec->nodes);
+    ec_mask_to_char_array(ec->xl_up, up_subvols, ec->nodes);
+
+    ret = cluster_inodelk(ec->xl_list, up_subvols, ec->nodes, replies,
+                          locked_on, frame, ec->xl, ec->xl->name, inode, 0, 0);
+    if (ret != ec->nodes) {
+        *need_heal = EC_HEAL_MUST;
+        goto unlock;
+    }
+    ret = ec_heal_inspect(frame, ec, inode, locked_on, _gf_true, _gf_true,
+                          need_heal);
+unlock:
+    cluster_uninodelk(ec->xl_list, locked_on, ec->nodes, replies, output, frame,
+                      ec->xl, ec->xl->name, inode, 0, 0);
+    cluster_replies_wipe(replies, ec->nodes);
+    return ret;
+}
+
+int32_t
+ec_get_heal_info(xlator_t *this, loc_t *entry_loc, dict_t **dict_rsp)
+{
+    int ret = -ENOMEM;
+    ec_heal_need_t need_heal = EC_HEAL_NONEED;
+    call_frame_t *frame = NULL;
+    ec_t *ec = NULL;
+    unsigned char *up_subvols = NULL;
+    loc_t loc = {
+        0,
+    };
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, entry_loc, out);
+
+    ec = this->private;
+    up_subvols = alloca0(ec->nodes);
+    ec_mask_to_char_array(ec->xl_up, up_subvols, ec->nodes);
+
+    if (EC_COUNT(up_subvols, ec->nodes) != ec->nodes) {
+        need_heal = EC_HEAL_MUST;
+        goto set_heal;
+    }
+    frame = create_frame(this, this->ctx->pool);
+    if (!frame) {
+        goto out;
+    }
+    ec_owner_set(frame, frame->root);
+    frame->root->uid = 0;
+    frame->root->gid = 0;
+    frame->root->pid = GF_CLIENT_PID_SELF_HEALD;
+
+    if (loc_copy(&loc, entry_loc) != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+               "Failed to copy a location.");
+        goto out;
+    }
+    if (!loc.inode) {
+        ret = syncop_inode_find(this, this, loc.gfid, &loc.inode, NULL, NULL);
+        if (ret < 0)
+            goto out;
+    }
+
+    ret = ec_heal_inspect(frame, ec, loc.inode, up_subvols, _gf_false,
+                          _gf_false, &need_heal);
+    if (ret == ec->nodes && need_heal != EC_HEAL_MAYBE) {
+        goto set_heal;
+    }
+    need_heal = EC_HEAL_NONEED;
+    ret = ec_heal_locked_inspect(frame, ec, loc.inode, &need_heal);
+    if (ret < 0)
+        goto out;
+set_heal:
+    if (need_heal == EC_HEAL_MUST) {
+        ret = ec_set_heal_info(dict_rsp, "heal");
+    } else {
+        ret = ec_set_heal_info(dict_rsp, "no-heal");
+    }
+out:
+    if (frame) {
+        STACK_DESTROY(frame->root);
+    }
+    loc_wipe(&loc);
+    return ret;
+}
diff --git a/xlators/cluster/ec/src/ec-heald.c b/xlators/cluster/ec/src/ec-heald.c
new file mode 100644
index 00000000000..5c1586bc9c5
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-heald.c
@@ -0,0 +1,681 @@
+/*
+  Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/defaults.h>
+#include <glusterfs/compat-errno.h>
+#include "ec.h"
+#include "ec-messages.h"
+#include "ec-heald.h"
+#include "ec-mem-types.h"
+#include <glusterfs/syncop.h>
+#include <glusterfs/syncop-utils.h>
+#include "protocol-common.h"
+
+#define NTH_INDEX_HEALER(this, n)                                              \
+    (&((((ec_t *)this->private))->shd.index_healers[n]))
+#define NTH_FULL_HEALER(this, n)                                               \
+    (&((((ec_t *)this->private))->shd.full_healers[n]))
+
+gf_boolean_t
+ec_shd_is_subvol_local(xlator_t *this, int subvol)
+{
+    ec_t *ec = NULL;
+    gf_boolean_t is_local = _gf_false;
+    loc_t loc = {
+        0,
+    };
+
+    ec = this->private;
+    loc.inode = this->itable->root;
+    syncop_is_subvol_local(ec->xl_list[subvol], &loc, &is_local);
+    return is_local;
+}
+
+char *
+ec_subvol_name(xlator_t *this, int subvol)
+{
+    ec_t *ec = NULL;
+
+    ec = this->private;
+    if (subvol < 0 || subvol > ec->nodes)
+        return NULL;
+
+    return ec->xl_list[subvol]->name;
+}
+
+int
+__ec_shd_healer_wait(struct subvol_healer *healer)
+{
+    ec_t *ec = NULL;
+    struct timespec wait_till = {
+        0,
+    };
+    int ret = 0;
+
+    ec = healer->this->private;
+
+disabled_loop:
+    wait_till.tv_sec = gf_time() + ec->shd.timeout;
+
+    while (!healer->rerun) {
+        ret = pthread_cond_timedwait(&healer->cond, &healer->mutex, &wait_till);
+        if (ret == ETIMEDOUT)
+            break;
+    }
+
+    if (ec->shutdown) {
+        healer->running = _gf_false;
+        return -1;
+    }
+
+    ret = healer->rerun;
+    healer->rerun = 0;
+
+    if (!ec->shd.enabled || !ec->up)
+        goto disabled_loop;
+
+    return ret;
+}
+
+int
+ec_shd_healer_wait(struct subvol_healer *healer)
+{
+    int ret = 0;
+
+    pthread_mutex_lock(&healer->mutex);
+    {
+        ret = __ec_shd_healer_wait(healer);
+    }
+    pthread_mutex_unlock(&healer->mutex);
+
+    return ret;
+}
+
+int
+ec_shd_index_inode(xlator_t *this, xlator_t *subvol, inode_t **inode)
+{
+    loc_t rootloc = {
+        0,
+    };
+    int ret = 0;
+    dict_t *xattr = NULL;
+    void *index_gfid = NULL;
+
+    *inode = NULL;
+    rootloc.inode = inode_ref(this->itable->root);
+    gf_uuid_copy(rootloc.gfid, rootloc.inode->gfid);
+
+    ret = syncop_getxattr(subvol, &rootloc, &xattr, GF_XATTROP_INDEX_GFID, NULL,
+                          NULL);
+    if (ret < 0)
+        goto out;
+    if (!xattr) {
+        ret = -EINVAL;
+        goto out;
+    }
+
+    ret = dict_get_ptr(xattr, GF_XATTROP_INDEX_GFID, &index_gfid);
+    if (ret)
+        goto out;
+
+    gf_msg_debug(this->name, 0, "index-dir gfid for %s: %s", subvol->name,
+                 uuid_utoa(index_gfid));
+
+    ret = syncop_inode_find(this, subvol, index_gfid, inode, NULL, NULL);
+
+out:
+    loc_wipe(&rootloc);
+
+    if (xattr)
+        dict_unref(xattr);
+
+    return ret;
+}
+
+int
+ec_shd_index_purge(xlator_t *subvol, inode_t *inode, char *name)
+{
+    loc_t loc = {
+        0,
+    };
+    int ret = 0;
+
+    loc.parent = inode_ref(inode);
+    loc.name = name;
+
+    ret = syncop_unlink(subvol, &loc, NULL, NULL);
+
+    loc_wipe(&loc);
+    return ret;
+}
+
+static gf_boolean_t
+ec_is_heal_completed(char *status)
+{
+    char *bad_pos = NULL;
+    char *zero_pos = NULL;
+
+    if (!status) {
+        return _gf_false;
+    }
+
+    /*Logic:
+     * Status will be of the form Good: <binary>, Bad: <binary>
+     * If heal completes, if we do strchr for '0' it should be present after
+     * 'Bad:' i.e. strRchr for ':'
+     * */
+
+    zero_pos = strchr(status, '0');
+    bad_pos = strrchr(status, ':');
+    if (!zero_pos || !bad_pos) {
+        /*malformed status*/
+        return _gf_false;
+    }
+
+    if (zero_pos > bad_pos) {
+        return _gf_true;
+    }
+
+    return _gf_false;
+}
+
+int
+ec_shd_selfheal(struct subvol_healer *healer, int child, loc_t *loc,
+                gf_boolean_t full)
+{
+    dict_t *xdata = NULL;
+    dict_t *dict = NULL;
+    uint32_t count;
+    int32_t ret;
+    char *heal_status = NULL;
+    ec_t *ec = healer->this->private;
+
+    GF_ATOMIC_INC(ec->stats.shd.attempted);
+    ret = syncop_getxattr(healer->this, loc, &dict, EC_XATTR_HEAL, NULL,
+                          &xdata);
+    if (ret == 0) {
+        if (dict && (dict_get_str(dict, EC_XATTR_HEAL, &heal_status) == 0)) {
+            if (ec_is_heal_completed(heal_status)) {
+                GF_ATOMIC_INC(ec->stats.shd.completed);
+            }
+        }
+    }
+
+    if (!full && (loc->inode->ia_type == IA_IFDIR)) {
+        /* If we have just healed a directory, it's possible that
+         * other index entries have appeared to be healed. */
+        if ((xdata != NULL) &&
+            (dict_get_uint32(xdata, EC_XATTR_HEAL_NEW, &count) == 0) &&
+            (count > 0)) {
+            /* Force a rerun of the index healer. */
+            gf_msg_debug(healer->this->name, 0, "%d more entries to heal",
+                         count);
+
+            healer->rerun = _gf_true;
+        }
+    }
+
+    if (xdata != NULL) {
+        dict_unref(xdata);
+    }
+
+    if (dict) {
+        dict_unref(dict);
+    }
+
+    return ret;
+}
+
+int
+ec_shd_index_heal(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent,
+                  void *data)
+{
+    struct subvol_healer *healer = data;
+    ec_t *ec = NULL;
+    loc_t loc = {0};
+    int ret = 0;
+
+    ec = healer->this->private;
+    if (ec->xl_up_count <= ec->fragments) {
+        return -ENOTCONN;
+    }
+    if (!ec->shd.enabled)
+        return -EBUSY;
+
+    gf_msg_debug(healer->this->name, 0, "got entry: %s", entry->d_name);
+
+    ret = gf_uuid_parse(entry->d_name, loc.gfid);
+    if (ret)
+        return 0;
+
+    /* If this fails with ENOENT/ESTALE index is stale */
+    ret = syncop_gfid_to_path(healer->this->itable, subvol, loc.gfid,
+                              (char **)&loc.path);
+    if (ret < 0)
+        goto out;
+
+    ret = syncop_inode_find(healer->this, healer->this, loc.gfid, &loc.inode,
+                            NULL, NULL);
+    if (ret < 0)
+        goto out;
+
+    ec_shd_selfheal(healer, healer->subvol, &loc, _gf_false);
+out:
+    if (ret == -ENOENT || ret == -ESTALE) {
+        gf_msg(healer->this->name, GF_LOG_DEBUG, 0, EC_MSG_HEAL_FAIL,
+               "Purging index for gfid %s:", uuid_utoa(loc.gfid));
+        ec_shd_index_purge(subvol, parent->inode, entry->d_name);
+    }
+    loc_wipe(&loc);
+
+    return 0;
+}
+
+int
+ec_shd_index_sweep(struct subvol_healer *healer)
+{
+    loc_t loc = {0};
+    ec_t *ec = NULL;
+    int ret = 0;
+    xlator_t *subvol = NULL;
+    dict_t *xdata = NULL;
+
+    ec = healer->this->private;
+    subvol = ec->xl_list[healer->subvol];
+
+    ret = ec_shd_index_inode(healer->this, subvol, &loc.inode);
+    if (ret < 0) {
+        gf_msg(healer->this->name, GF_LOG_WARNING, errno,
+               EC_MSG_INDEX_DIR_GET_FAIL, "unable to get index-dir on %s",
+               subvol->name);
+        goto out;
+    }
+
+    xdata = dict_new();
+    if (!xdata || dict_set_int32(xdata, "get-gfid-type", 1)) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    _mask_cancellation();
+    ret = syncop_mt_dir_scan(NULL, subvol, &loc, GF_CLIENT_PID_SELF_HEALD,
+                             healer, ec_shd_index_heal, xdata,
+                             ec->shd.max_threads, ec->shd.wait_qlength);
+    _unmask_cancellation();
+out:
+    if (xdata)
+        dict_unref(xdata);
+    loc_wipe(&loc);
+
+    return ret;
+}
+
+int
+ec_shd_full_heal(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent,
+                 void *data)
+{
+    struct subvol_healer *healer = data;
+    xlator_t *this = healer->this;
+    ec_t *ec = NULL;
+    loc_t loc = {0};
+    int ret = 0;
+
+    ec = this->private;
+
+    if (this->cleanup_starting) {
+        return -ENOTCONN;
+    }
+
+    if (ec->xl_up_count <= ec->fragments) {
+        return -ENOTCONN;
+    }
+    if (!ec->shd.enabled)
+        return -EBUSY;
+
+    if (gf_uuid_is_null(entry->d_stat.ia_gfid)) {
+        /* It's possible that an entry has been removed just after
+         * being seen in a directory but before getting its stat info.
+         * In this case we'll receive a NULL gfid here. Since the file
+         * doesn't exist anymore, we can safely ignore it. */
+        return 0;
+    }
+
+    loc.parent = inode_ref(parent->inode);
+    loc.name = entry->d_name;
+    gf_uuid_copy(loc.gfid, entry->d_stat.ia_gfid);
+
+    /* If this fails with ENOENT/ESTALE index is stale */
+    ret = syncop_gfid_to_path(this->itable, subvol, loc.gfid,
+                              (char **)&loc.path);
+    if (ret < 0)
+        goto out;
+
+    ret = syncop_inode_find(this, this, loc.gfid, &loc.inode, NULL, NULL);
+    if (ret < 0)
+        goto out;
+
+    ec_shd_selfheal(healer, healer->subvol, &loc, _gf_true);
+
+    ret = 0;
+
+out:
+    loc_wipe(&loc);
+    return ret;
+}
+
+int
+ec_shd_full_sweep(struct subvol_healer *healer, inode_t *inode)
+{
+    ec_t *ec = NULL;
+    loc_t loc = {0};
+    int ret = -1;
+
+    ec = healer->this->private;
+    loc.inode = inode;
+    _mask_cancellation();
+    ret = syncop_ftw(ec->xl_list[healer->subvol], &loc,
+                     GF_CLIENT_PID_SELF_HEALD, healer, ec_shd_full_heal);
+    _unmask_cancellation();
+    return ret;
+}
+
+void *
+ec_shd_index_healer(void *data)
+{
+    struct subvol_healer *healer = NULL;
+    xlator_t *this = NULL;
+    int run = 0;
+
+    healer = data;
+    THIS = this = healer->this;
+    ec_t *ec = this->private;
+
+    for (;;) {
+        run = ec_shd_healer_wait(healer);
+        if (run == -1)
+            break;
+
+        if (ec->xl_up_count > ec->fragments) {
+            gf_msg_debug(this->name, 0, "starting index sweep on subvol %s",
+                         ec_subvol_name(this, healer->subvol));
+            ec_shd_index_sweep(healer);
+        }
+        gf_msg_debug(this->name, 0, "finished index sweep on subvol %s",
+                     ec_subvol_name(this, healer->subvol));
+    }
+
+    return NULL;
+}
+
+void *
+ec_shd_full_healer(void *data)
+{
+    struct subvol_healer *healer = NULL;
+    xlator_t *this = NULL;
+    loc_t rootloc = {0};
+
+    int run = 0;
+
+    healer = data;
+    THIS = this = healer->this;
+    ec_t *ec = this->private;
+
+    rootloc.inode = this->itable->root;
+    for (;;) {
+        run = ec_shd_healer_wait(healer);
+        if (run < 0) {
+            break;
+        } else if (run == 0) {
+            continue;
+        }
+
+        if (ec->xl_up_count > ec->fragments) {
+            gf_msg(this->name, GF_LOG_INFO, 0, EC_MSG_FULL_SWEEP_START,
+                   "starting full sweep on subvol %s",
+                   ec_subvol_name(this, healer->subvol));
+
+            ec_shd_selfheal(healer, healer->subvol, &rootloc, _gf_true);
+            ec_shd_full_sweep(healer, this->itable->root);
+        }
+
+        gf_msg(this->name, GF_LOG_INFO, 0, EC_MSG_FULL_SWEEP_STOP,
+               "finished full sweep on subvol %s",
+               ec_subvol_name(this, healer->subvol));
+    }
+
+    return NULL;
+}
+
+int
+ec_shd_healer_init(xlator_t *this, struct subvol_healer *healer)
+{
+    int ret = 0;
+
+    ret = pthread_mutex_init(&healer->mutex, NULL);
+    if (ret)
+        goto out;
+
+    ret = pthread_cond_init(&healer->cond, NULL);
+    if (ret)
+        goto out;
+
+    healer->this = this;
+    healer->running = _gf_false;
+    healer->rerun = _gf_false;
+out:
+    return ret;
+}
+
+int
+ec_shd_healer_spawn(xlator_t *this, struct subvol_healer *healer,
+                    void *(threadfn)(void *))
+{
+    int ret = 0;
+
+    pthread_mutex_lock(&healer->mutex);
+    {
+        if (healer->running) {
+            pthread_cond_signal(&healer->cond);
+        } else {
+            ret = gf_thread_create(&healer->thread, NULL, threadfn, healer,
+                                   "ecshd");
+            if (ret)
+                goto unlock;
+            healer->running = 1;
+        }
+
+        healer->rerun = 1;
+    }
+unlock:
+    pthread_mutex_unlock(&healer->mutex);
+
+    return ret;
+}
+
+int
+ec_shd_full_healer_spawn(xlator_t *this, int subvol)
+{
+    if (xlator_is_cleanup_starting(this))
+        return -1;
+
+    return ec_shd_healer_spawn(this, NTH_FULL_HEALER(this, subvol),
+                               ec_shd_full_healer);
+}
+
+int
+ec_shd_index_healer_spawn(xlator_t *this, int subvol)
+{
+    if (xlator_is_cleanup_starting(this))
+        return -1;
+
+    return ec_shd_healer_spawn(this, NTH_INDEX_HEALER(this, subvol),
+                               ec_shd_index_healer);
+}
+
+void
+ec_shd_index_healer_wake(ec_t *ec)
+{
+    int32_t i;
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (((ec->xl_up >> i) & 1) != 0) {
+            ec_shd_index_healer_spawn(ec->xl, i);
+        }
+    }
+}
+
+int
+ec_selfheal_daemon_init(xlator_t *this)
+{
+    ec_t *ec = NULL;
+    ec_self_heald_t *shd = NULL;
+    int ret = -1;
+    int i = 0;
+
+    ec = this->private;
+    shd = &ec->shd;
+
+    shd->index_healers = GF_CALLOC(sizeof(*shd->index_healers), ec->nodes,
+                                   ec_mt_subvol_healer_t);
+    if (!shd->index_healers)
+        goto out;
+
+    for (i = 0; i < ec->nodes; i++) {
+        shd->index_healers[i].subvol = i;
+        ret = ec_shd_healer_init(this, &shd->index_healers[i]);
+        if (ret)
+            goto out;
+    }
+
+    shd->full_healers = GF_CALLOC(sizeof(*shd->full_healers), ec->nodes,
+                                  ec_mt_subvol_healer_t);
+    if (!shd->full_healers)
+        goto out;
+
+    for (i = 0; i < ec->nodes; i++) {
+        shd->full_healers[i].subvol = i;
+        ret = ec_shd_healer_init(this, &shd->full_healers[i]);
+        if (ret)
+            goto out;
+    }
+
+    ret = 0;
+out:
+    return ret;
+}
+
+int
+ec_heal_op(xlator_t *this, dict_t *output, gf_xl_afr_op_t op, int xl_id)
+{
+    char key[64] = {0};
+    int op_ret = 0;
+    ec_t *ec = NULL;
+    int i = 0;
+    GF_UNUSED int ret = 0;
+
+    ec = this->private;
+
+    op_ret = -1;
+    for (i = 0; i < ec->nodes; i++) {
+        snprintf(key, sizeof(key), "%d-%d-status", xl_id, i);
+
+        if (((ec->xl_up >> i) & 1) == 0) {
+            ret = dict_set_str(output, key, "Brick is not connected");
+        } else if (!ec->up) {
+            ret = dict_set_str(output, key, "Disperse subvolume is not up");
+        } else if (!ec_shd_is_subvol_local(this, i)) {
+            ret = dict_set_str(output, key, "Brick is remote");
+        } else {
+            ret = dict_set_str(output, key, "Started self-heal");
+            if (op == GF_SHD_OP_HEAL_FULL) {
+                ec_shd_full_healer_spawn(this, i);
+            } else if (op == GF_SHD_OP_HEAL_INDEX) {
+                ec_shd_index_healer_spawn(this, i);
+            }
+            op_ret = 0;
+        }
+    }
+    return op_ret;
+}
+
+int
+ec_xl_op(xlator_t *this, dict_t *input, dict_t *output)
+{
+    gf_xl_afr_op_t op = GF_SHD_OP_INVALID;
+    int ret = 0;
+    int xl_id = 0;
+
+    ret = dict_get_int32(input, "xl-op", (int32_t *)&op);
+    if (ret)
+        goto out;
+
+    ret = dict_get_int32(input, this->name, &xl_id);
+    if (ret)
+        goto out;
+
+    ret = dict_set_int32(output, this->name, xl_id);
+    if (ret)
+        goto out;
+
+    switch (op) {
+        case GF_SHD_OP_HEAL_FULL:
+            ret = ec_heal_op(this, output, op, xl_id);
+            break;
+
+        case GF_SHD_OP_HEAL_INDEX:
+            ret = ec_heal_op(this, output, op, xl_id);
+            break;
+
+        default:
+            ret = -1;
+            break;
+    }
+out:
+    dict_del(output, this->name);
+    return ret;
+}
+
+void
+ec_destroy_healer_object(xlator_t *this, struct subvol_healer *healer)
+{
+    if (!healer)
+        return;
+
+    pthread_cond_destroy(&healer->cond);
+    pthread_mutex_destroy(&healer->mutex);
+}
+
+void
+ec_selfheal_daemon_fini(xlator_t *this)
+{
+    struct subvol_healer *healer = NULL;
+    ec_self_heald_t *shd = NULL;
+    ec_t *priv = NULL;
+    int i = 0;
+
+    priv = this->private;
+    if (!priv)
+        return;
+
+    shd = &priv->shd;
+    if (!shd->iamshd)
+        return;
+
+    for (i = 0; i < priv->nodes; i++) {
+        healer = &shd->index_healers[i];
+        ec_destroy_healer_object(this, healer);
+
+        healer = &shd->full_healers[i];
+        ec_destroy_healer_object(this, healer);
+    }
+
+    GF_FREE(shd->index_healers);
+    GF_FREE(shd->full_healers);
+}
diff --git a/xlators/cluster/ec/src/ec-heald.h b/xlators/cluster/ec/src/ec-heald.h
new file mode 100644
index 00000000000..6c7da4edc10
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-heald.h
@@ -0,0 +1,30 @@
+/*
+  Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_HEALD_H__
+#define __EC_HEALD_H__
+
+#include "ec-types.h"           // for ec_t
+#include "glusterfs/dict.h"     // for dict_t
+#include "glusterfs/globals.h"  // for xlator_t
+
+int
+ec_xl_op(xlator_t *this, dict_t *input, dict_t *output);
+
+int
+ec_selfheal_daemon_init(xlator_t *this);
+
+void
+ec_shd_index_healer_wake(ec_t *ec);
+
+void
+ec_selfheal_daemon_fini(xlator_t *this);
+
+#endif /* __EC_HEALD_H__ */
diff --git a/xlators/cluster/ec/src/ec-helpers.c b/xlators/cluster/ec/src/ec-helpers.c
new file mode 100644
index 00000000000..48f54475e01
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-helpers.c
@@ -0,0 +1,867 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <libgen.h>
+
+#include <glusterfs/byte-order.h>
+
+#include "ec.h"
+#include "ec-mem-types.h"
+#include "ec-messages.h"
+#include "ec-fops.h"
+#include "ec-method.h"
+#include "ec-helpers.h"
+
+static const char *ec_fop_list[] = {[-EC_FOP_HEAL] = "HEAL"};
+
+const char *
+ec_bin(char *str, size_t size, uint64_t value, int32_t digits)
+{
+    str += size;
+
+    if (size-- < 1) {
+        goto failed;
+    }
+    *--str = 0;
+
+    while ((value != 0) || (digits > 0)) {
+        if (size-- < 1) {
+            goto failed;
+        }
+        *--str = '0' + (value & 1);
+        digits--;
+        value >>= 1;
+    }
+
+    return str;
+
+failed:
+    return "<buffer too small>";
+}
+
+const char *
+ec_fop_name(int32_t id)
+{
+    if (id >= 0) {
+        return gf_fop_list[id];
+    }
+
+    return ec_fop_list[-id];
+}
+
+void
+ec_trace(const char *event, ec_fop_data_t *fop, const char *fmt, ...)
+{
+    char str1[32], str2[32], str3[32];
+    char *msg;
+    ec_t *ec = fop->xl->private;
+    va_list args;
+    int32_t ret;
+
+    va_start(args, fmt);
+    ret = vasprintf(&msg, fmt, args);
+    va_end(args);
+
+    if (ret < 0) {
+        msg = "<memory allocation error>";
+    }
+
+    gf_msg_trace("ec", 0,
+                 "%s(%s) %p(%p) [refs=%d, winds=%d, jobs=%d] "
+                 "frame=%p/%p, min/exp=%d/%d, err=%d state=%d "
+                 "{%s:%s:%s} %s",
+                 event, ec_fop_name(fop->id), fop, fop->parent, fop->refs,
+                 fop->winds, fop->jobs, fop->req_frame, fop->frame,
+                 fop->minimum, fop->expected, fop->error, fop->state,
+                 ec_bin(str1, sizeof(str1), fop->mask, ec->nodes),
+                 ec_bin(str2, sizeof(str2), fop->remaining, ec->nodes),
+                 ec_bin(str3, sizeof(str3), fop->good, ec->nodes), msg);
+
+    if (ret >= 0) {
+        free(msg);
+    }
+}
+
+int32_t
+ec_bits_consume(uint64_t *n)
+{
+    uint64_t tmp;
+
+    tmp = *n;
+    tmp &= -tmp;
+    *n ^= tmp;
+
+    return gf_bits_index(tmp);
+}
+
+size_t
+ec_iov_copy_to(void *dst, struct iovec *vector, int32_t count, off_t offset,
+               size_t size)
+{
+    int32_t i = 0;
+    size_t total = 0, len = 0;
+
+    while (i < count) {
+        if (offset < vector[i].iov_len) {
+            while ((i < count) && (size > 0)) {
+                len = size;
+                if (len > vector[i].iov_len - offset) {
+                    len = vector[i].iov_len - offset;
+                }
+                memcpy(dst, vector[i++].iov_base + offset, len);
+                offset = 0;
+                dst += len;
+                total += len;
+                size -= len;
+            }
+
+            break;
+        }
+
+        offset -= vector[i].iov_len;
+        i++;
+    }
+
+    return total;
+}
+
+int32_t
+ec_buffer_alloc(xlator_t *xl, size_t size, struct iobref **piobref, void **ptr)
+{
+    struct iobref *iobref = NULL;
+    struct iobuf *iobuf = NULL;
+    int32_t ret = -ENOMEM;
+
+    iobuf = iobuf_get_page_aligned(xl->ctx->iobuf_pool, size,
+                                   EC_METHOD_WORD_SIZE);
+    if (iobuf == NULL) {
+        goto out;
+    }
+
+    iobref = *piobref;
+    if (iobref == NULL) {
+        iobref = iobref_new();
+        if (iobref == NULL) {
+            goto out;
+        }
+    }
+
+    ret = iobref_add(iobref, iobuf);
+    if (ret != 0) {
+        if (iobref != *piobref) {
+            iobref_unref(iobref);
+        }
+        iobref = NULL;
+
+        goto out;
+    }
+
+    GF_ASSERT(EC_ALIGN_CHECK(iobuf->ptr, EC_METHOD_WORD_SIZE));
+
+    *ptr = iobuf->ptr;
+
+out:
+    if (iobuf != NULL) {
+        iobuf_unref(iobuf);
+    }
+
+    if (iobref != NULL) {
+        *piobref = iobref;
+    }
+
+    return ret;
+}
+
+int32_t
+ec_dict_set_array(dict_t *dict, char *key, uint64_t value[], int32_t size)
+{
+    int ret = -1;
+    uint64_t *ptr = NULL;
+    int32_t vindex;
+
+    if (value == NULL) {
+        return -EINVAL;
+    }
+
+    ptr = GF_MALLOC(sizeof(uint64_t) * size, gf_common_mt_char);
+    if (ptr == NULL) {
+        return -ENOMEM;
+    }
+    for (vindex = 0; vindex < size; vindex++) {
+        ptr[vindex] = hton64(value[vindex]);
+    }
+    ret = dict_set_bin(dict, key, ptr, sizeof(uint64_t) * size);
+    if (ret)
+        GF_FREE(ptr);
+    return ret;
+}
+
+int32_t
+ec_dict_get_array(dict_t *dict, char *key, uint64_t value[], int32_t size)
+{
+    void *ptr;
+    int32_t len;
+    int32_t vindex;
+    int32_t old_size = 0;
+    int32_t err;
+
+    if (dict == NULL) {
+        return -EINVAL;
+    }
+    err = dict_get_ptr_and_len(dict, key, &ptr, &len);
+    if (err != 0) {
+        return err;
+    }
+
+    if (len > (size * sizeof(uint64_t)) || (len % sizeof(uint64_t))) {
+        return -EINVAL;
+    }
+
+    /* 3.6 version ec would have stored version in 64 bit. In that case treat
+     * metadata versions same as data*/
+    old_size = min(size, len / sizeof(uint64_t));
+    for (vindex = 0; vindex < old_size; vindex++) {
+        value[vindex] = ntoh64(*((uint64_t *)ptr + vindex));
+    }
+
+    if (old_size < size) {
+        for (vindex = old_size; vindex < size; vindex++) {
+            value[vindex] = value[old_size - 1];
+        }
+    }
+
+    return 0;
+}
+
+int32_t
+ec_dict_del_array(dict_t *dict, char *key, uint64_t value[], int32_t size)
+{
+    int ret = 0;
+
+    ret = ec_dict_get_array(dict, key, value, size);
+    if (ret == 0)
+        dict_del(dict, key);
+
+    return ret;
+}
+
+int32_t
+ec_dict_set_number(dict_t *dict, char *key, uint64_t value)
+{
+    int ret = -1;
+    uint64_t *ptr;
+
+    ptr = GF_MALLOC(sizeof(value), gf_common_mt_char);
+    if (ptr == NULL) {
+        return -ENOMEM;
+    }
+
+    *ptr = hton64(value);
+
+    ret = dict_set_bin(dict, key, ptr, sizeof(value));
+    if (ret)
+        GF_FREE(ptr);
+
+    return ret;
+}
+
+int32_t
+ec_dict_del_number(dict_t *dict, char *key, uint64_t *value)
+{
+    void *ptr;
+    int32_t len, err;
+
+    if (dict == NULL) {
+        return -EINVAL;
+    }
+    err = dict_get_ptr_and_len(dict, key, &ptr, &len);
+    if (err != 0) {
+        return err;
+    }
+    if (len != sizeof(uint64_t)) {
+        return -EINVAL;
+    }
+
+    *value = ntoh64(*(uint64_t *)ptr);
+
+    dict_del(dict, key);
+
+    return 0;
+}
+
+int32_t
+ec_dict_set_config(dict_t *dict, char *key, ec_config_t *config)
+{
+    int ret = -1;
+    uint64_t *ptr, data;
+
+    if (config->version > EC_CONFIG_VERSION) {
+        gf_msg("ec", GF_LOG_ERROR, EINVAL, EC_MSG_UNSUPPORTED_VERSION,
+               "Trying to store an unsupported config "
+               "version (%u)",
+               config->version);
+
+        return -EINVAL;
+    }
+
+    ptr = GF_MALLOC(sizeof(uint64_t), gf_common_mt_char);
+    if (ptr == NULL) {
+        return -ENOMEM;
+    }
+
+    data = ((uint64_t)config->version) << 56;
+    data |= ((uint64_t)config->algorithm) << 48;
+    data |= ((uint64_t)config->gf_word_size) << 40;
+    data |= ((uint64_t)config->bricks) << 32;
+    data |= ((uint64_t)config->redundancy) << 24;
+    data |= config->chunk_size;
+
+    *ptr = hton64(data);
+
+    ret = dict_set_bin(dict, key, ptr, sizeof(uint64_t));
+    if (ret)
+        GF_FREE(ptr);
+
+    return ret;
+}
+
+int32_t
+ec_dict_del_config(dict_t *dict, char *key, ec_config_t *config)
+{
+    void *ptr;
+    uint64_t data;
+    int32_t len, err;
+
+    if (dict == NULL) {
+        return -EINVAL;
+    }
+    err = dict_get_ptr_and_len(dict, key, &ptr, &len);
+    if (err != 0) {
+        return err;
+    }
+    if (len != sizeof(uint64_t)) {
+        return -EINVAL;
+    }
+
+    data = ntoh64(*(uint64_t *)ptr);
+    /* Currently we need to get the config xattr for entries of type IA_INVAL.
+     * These entries can later become IA_DIR entries (after inode_link()),
+     * which don't have a config xattr. However, since the xattr is requested
+     * using an xattrop() fop, it will always return a config full of 0's
+     * instead of saying that it doesn't exist.
+     *
+     * We need to filter out this case and consider that a config xattr == 0 is
+     * the same as a non-existent xattr. Otherwise ec_config_check() will fail.
+     */
+    if (data == 0) {
+        return -ENODATA;
+    }
+
+    config->version = (data >> 56) & 0xff;
+    if (config->version > EC_CONFIG_VERSION) {
+        gf_msg("ec", GF_LOG_ERROR, EINVAL, EC_MSG_UNSUPPORTED_VERSION,
+               "Found an unsupported config version (%u)", config->version);
+
+        return -EINVAL;
+    }
+
+    config->algorithm = (data >> 48) & 0xff;
+    config->gf_word_size = (data >> 40) & 0xff;
+    config->bricks = (data >> 32) & 0xff;
+    config->redundancy = (data >> 24) & 0xff;
+    config->chunk_size = data & 0xffffff;
+
+    dict_del(dict, key);
+
+    return 0;
+}
+
+gf_boolean_t
+ec_loc_gfid_check(xlator_t *xl, uuid_t dst, uuid_t src)
+{
+    if (gf_uuid_is_null(src)) {
+        return _gf_true;
+    }
+
+    if (gf_uuid_is_null(dst)) {
+        gf_uuid_copy(dst, src);
+
+        return _gf_true;
+    }
+
+    if (gf_uuid_compare(dst, src) != 0) {
+        gf_msg(xl->name, GF_LOG_WARNING, 0, EC_MSG_GFID_MISMATCH,
+               "Mismatching GFID's in loc");
+
+        return _gf_false;
+    }
+
+    return _gf_true;
+}
+
+int32_t
+ec_loc_setup_inode(xlator_t *xl, inode_table_t *table, loc_t *loc)
+{
+    int32_t ret = -EINVAL;
+
+    if (loc->inode != NULL) {
+        if (!ec_loc_gfid_check(xl, loc->gfid, loc->inode->gfid)) {
+            goto out;
+        }
+    } else if (table != NULL) {
+        if (!gf_uuid_is_null(loc->gfid)) {
+            loc->inode = inode_find(table, loc->gfid);
+        } else if (loc->path && strchr(loc->path, '/')) {
+            loc->inode = inode_resolve(table, (char *)loc->path);
+        }
+    }
+
+    ret = 0;
+
+out:
+    return ret;
+}
+
+int32_t
+ec_loc_setup_parent(xlator_t *xl, inode_table_t *table, loc_t *loc)
+{
+    char *path, *parent;
+    int32_t ret = -EINVAL;
+
+    if (loc->parent != NULL) {
+        if (!ec_loc_gfid_check(xl, loc->pargfid, loc->parent->gfid)) {
+            goto out;
+        }
+    } else if (table != NULL) {
+        if (!gf_uuid_is_null(loc->pargfid)) {
+            loc->parent = inode_find(table, loc->pargfid);
+        } else if (loc->path && strchr(loc->path, '/')) {
+            path = gf_strdup(loc->path);
+            if (path == NULL) {
+                gf_msg(xl->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                       "Unable to duplicate path '%s'", loc->path);
+
+                ret = -ENOMEM;
+
+                goto out;
+            }
+            parent = dirname(path);
+            loc->parent = inode_resolve(table, parent);
+            if (loc->parent != NULL) {
+                gf_uuid_copy(loc->pargfid, loc->parent->gfid);
+            }
+            GF_FREE(path);
+        }
+    }
+
+    /* If 'pargfid' has not been determined, clear 'name' to avoid resolutions
+       based on <gfid:pargfid>/name. */
+    if (gf_uuid_is_null(loc->pargfid)) {
+        loc->name = NULL;
+    }
+
+    ret = 0;
+
+out:
+    return ret;
+}
+
+int32_t
+ec_loc_setup_path(xlator_t *xl, loc_t *loc)
+{
+    static uuid_t root = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1};
+    char *name;
+    int32_t ret = -EINVAL;
+
+    if (loc->path != NULL) {
+        name = strrchr(loc->path, '/');
+        if (name == NULL) {
+            /* Allow gfid paths: <gfid:...> */
+            if (strncmp(loc->path, "<gfid:", 6) == 0) {
+                ret = 0;
+            }
+            goto out;
+        }
+        if (name == loc->path) {
+            if (name[1] == 0) {
+                if (!ec_loc_gfid_check(xl, loc->gfid, root)) {
+                    goto out;
+                }
+            } else {
+                if (!ec_loc_gfid_check(xl, loc->pargfid, root)) {
+                    goto out;
+                }
+            }
+        }
+        name++;
+
+        if (loc->name != NULL) {
+            if (strcmp(loc->name, name) != 0) {
+                gf_msg(xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_INVALID_LOC_NAME,
+                       "Invalid name '%s' in loc", loc->name);
+
+                goto out;
+            }
+        } else {
+            loc->name = name;
+        }
+    }
+
+    ret = 0;
+
+out:
+    return ret;
+}
+
+int32_t
+ec_loc_parent(xlator_t *xl, loc_t *loc, loc_t *parent)
+{
+    inode_table_t *table = NULL;
+    char *str = NULL;
+    int32_t ret = -ENOMEM;
+
+    memset(parent, 0, sizeof(loc_t));
+
+    if (loc->parent != NULL) {
+        table = loc->parent->table;
+        parent->inode = inode_ref(loc->parent);
+    } else if (loc->inode != NULL) {
+        table = loc->inode->table;
+    }
+    if (!gf_uuid_is_null(loc->pargfid)) {
+        gf_uuid_copy(parent->gfid, loc->pargfid);
+    }
+    if (loc->path && strchr(loc->path, '/')) {
+        str = gf_strdup(loc->path);
+        if (str == NULL) {
+            gf_msg(xl->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Unable to duplicate path '%s'", loc->path);
+
+            goto out;
+        }
+        parent->path = gf_strdup(dirname(str));
+        if (parent->path == NULL) {
+            gf_msg(xl->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Unable to duplicate path '%s'", dirname(str));
+
+            goto out;
+        }
+    }
+
+    ret = ec_loc_setup_path(xl, parent);
+    if (ret == 0) {
+        ret = ec_loc_setup_inode(xl, table, parent);
+    }
+    if (ret == 0) {
+        ret = ec_loc_setup_parent(xl, table, parent);
+    }
+    if (ret != 0) {
+        goto out;
+    }
+
+    if ((parent->inode == NULL) && (parent->path == NULL) &&
+        gf_uuid_is_null(parent->gfid)) {
+        gf_msg(xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_LOC_PARENT_INODE_MISSING,
+               "Parent inode missing for loc_t");
+
+        ret = -EINVAL;
+
+        goto out;
+    }
+
+    ret = 0;
+
+out:
+    GF_FREE(str);
+
+    if (ret != 0) {
+        loc_wipe(parent);
+    }
+
+    return ret;
+}
+
+int32_t
+ec_loc_update(xlator_t *xl, loc_t *loc, inode_t *inode, struct iatt *iatt)
+{
+    inode_table_t *table = NULL;
+    int32_t ret = -EINVAL;
+
+    if (inode != NULL) {
+        table = inode->table;
+        if (loc->inode != inode) {
+            if (loc->inode != NULL) {
+                inode_unref(loc->inode);
+            }
+            loc->inode = inode_ref(inode);
+            gf_uuid_copy(loc->gfid, inode->gfid);
+        }
+    } else if (loc->inode != NULL) {
+        table = loc->inode->table;
+    } else if (loc->parent != NULL) {
+        table = loc->parent->table;
+    }
+
+    if (iatt != NULL) {
+        if (!ec_loc_gfid_check(xl, loc->gfid, iatt->ia_gfid)) {
+            goto out;
+        }
+    }
+
+    ret = ec_loc_setup_path(xl, loc);
+    if (ret == 0) {
+        ret = ec_loc_setup_inode(xl, table, loc);
+    }
+    if (ret == 0) {
+        ret = ec_loc_setup_parent(xl, table, loc);
+    }
+    if (ret != 0) {
+        goto out;
+    }
+
+out:
+    return ret;
+}
+
+int32_t
+ec_loc_from_fd(xlator_t *xl, loc_t *loc, fd_t *fd)
+{
+    ec_fd_t *ctx;
+    int32_t ret = -ENOMEM;
+
+    memset(loc, 0, sizeof(*loc));
+
+    ctx = ec_fd_get(fd, xl);
+    if (ctx != NULL) {
+        if (loc_copy(loc, &ctx->loc) != 0) {
+            goto out;
+        }
+    }
+
+    ret = ec_loc_update(xl, loc, fd->inode, NULL);
+    if (ret != 0) {
+        goto out;
+    }
+
+out:
+    if (ret != 0) {
+        loc_wipe(loc);
+    }
+
+    return ret;
+}
+
+int32_t
+ec_loc_from_loc(xlator_t *xl, loc_t *dst, loc_t *src)
+{
+    int32_t ret = -ENOMEM;
+
+    memset(dst, 0, sizeof(*dst));
+
+    if (loc_copy(dst, src) != 0) {
+        goto out;
+    }
+
+    ret = ec_loc_update(xl, dst, NULL, NULL);
+    if (ret != 0) {
+        goto out;
+    }
+
+out:
+    if (ret != 0) {
+        loc_wipe(dst);
+    }
+
+    return ret;
+}
+
+void
+ec_owner_set(call_frame_t *frame, void *owner)
+{
+    set_lk_owner_from_ptr(&frame->root->lk_owner, owner);
+}
+
+void
+ec_owner_copy(call_frame_t *frame, gf_lkowner_t *owner)
+{
+    lk_owner_copy(&frame->root->lk_owner, owner);
+}
+
+static void
+ec_stripe_cache_init(ec_t *ec, ec_inode_t *ctx)
+{
+    ec_stripe_list_t *stripe_cache = NULL;
+
+    stripe_cache = &(ctx->stripe_cache);
+    if (stripe_cache->max == 0) {
+        stripe_cache->max = ec->stripe_cache;
+    }
+}
+
+ec_inode_t *
+__ec_inode_get(inode_t *inode, xlator_t *xl)
+{
+    ec_inode_t *ctx = NULL;
+    uint64_t value = 0;
+
+    if ((__inode_ctx_get(inode, xl, &value) != 0) || (value == 0)) {
+        ctx = GF_MALLOC(sizeof(*ctx), ec_mt_ec_inode_t);
+        if (ctx != NULL) {
+            memset(ctx, 0, sizeof(*ctx));
+            INIT_LIST_HEAD(&ctx->heal);
+            INIT_LIST_HEAD(&ctx->stripe_cache.lru);
+            ctx->heal_count = 0;
+            value = (uint64_t)(uintptr_t)ctx;
+            if (__inode_ctx_set(inode, xl, &value) != 0) {
+                GF_FREE(ctx);
+
+                return NULL;
+            }
+        }
+    } else {
+        ctx = (ec_inode_t *)(uintptr_t)value;
+    }
+    if (ctx)
+        ec_stripe_cache_init(xl->private, ctx);
+
+    return ctx;
+}
+
+ec_inode_t *
+ec_inode_get(inode_t *inode, xlator_t *xl)
+{
+    ec_inode_t *ctx = NULL;
+
+    LOCK(&inode->lock);
+
+    ctx = __ec_inode_get(inode, xl);
+
+    UNLOCK(&inode->lock);
+
+    return ctx;
+}
+
+ec_fd_t *
+__ec_fd_get(fd_t *fd, xlator_t *xl)
+{
+    int i = 0;
+    ec_fd_t *ctx = NULL;
+    ec_inode_t *ictx = NULL;
+    uint64_t value = 0;
+    ec_t *ec = xl->private;
+
+    if ((__fd_ctx_get(fd, xl, &value) != 0) || (value == 0)) {
+        ctx = GF_MALLOC(sizeof(*ctx) + (sizeof(ec_fd_status_t) * ec->nodes),
+                        ec_mt_ec_fd_t);
+        if (ctx != NULL) {
+            memset(ctx, 0, sizeof(*ctx));
+
+            for (i = 0; i < ec->nodes; i++) {
+                if (fd_is_anonymous(fd)) {
+                    ctx->fd_status[i] = EC_FD_OPENED;
+                } else {
+                    ctx->fd_status[i] = EC_FD_NOT_OPENED;
+                }
+            }
+
+            value = (uint64_t)(uintptr_t)ctx;
+            if (__fd_ctx_set(fd, xl, value) != 0) {
+                GF_FREE(ctx);
+                return NULL;
+            }
+            /* Only refering bad-version so no need for lock
+             * */
+            ictx = __ec_inode_get(fd->inode, xl);
+            if (ictx) {
+                ctx->bad_version = ictx->bad_version;
+            }
+        }
+    } else {
+        ctx = (ec_fd_t *)(uintptr_t)value;
+    }
+
+    /* Treat anonymous fd specially */
+    if (fd->anonymous && ctx) {
+        /* Mark the fd open for all subvolumes. */
+        ctx->open = -1;
+        /* Try to populate ctx->loc with fd->inode information. */
+        ec_loc_update(xl, &ctx->loc, fd->inode, NULL);
+    }
+
+    return ctx;
+}
+
+ec_fd_t *
+ec_fd_get(fd_t *fd, xlator_t *xl)
+{
+    ec_fd_t *ctx = NULL;
+
+    LOCK(&fd->lock);
+
+    ctx = __ec_fd_get(fd, xl);
+
+    UNLOCK(&fd->lock);
+
+    return ctx;
+}
+
+gf_boolean_t
+ec_is_internal_xattr(dict_t *dict, char *key, data_t *value, void *data)
+{
+    if (key && (strncmp(key, EC_XATTR_PREFIX, SLEN(EC_XATTR_PREFIX)) == 0))
+        return _gf_true;
+
+    return _gf_false;
+}
+
+void
+ec_filter_internal_xattrs(dict_t *xattr)
+{
+    dict_foreach_match(xattr, ec_is_internal_xattr, NULL,
+                       dict_remove_foreach_fn, NULL);
+}
+
+gf_boolean_t
+ec_is_data_fop(glusterfs_fop_t fop)
+{
+    switch (fop) {
+        case GF_FOP_WRITE:
+        case GF_FOP_TRUNCATE:
+        case GF_FOP_FTRUNCATE:
+        case GF_FOP_FALLOCATE:
+        case GF_FOP_DISCARD:
+        case GF_FOP_ZEROFILL:
+            return _gf_true;
+        default:
+            return _gf_false;
+    }
+    return _gf_false;
+}
+/*
+gf_boolean_t
+ec_is_metadata_fop (int32_t lock_kind, glusterfs_fop_t fop)
+{
+        if (lock_kind == EC_LOCK_ENTRY) {
+                return _gf_false;
+        }
+
+        switch (fop) {
+        case GF_FOP_SETATTR:
+        case GF_FOP_FSETATTR:
+        case GF_FOP_SETXATTR:
+        case GF_FOP_FSETXATTR:
+        case GF_FOP_REMOVEXATTR:
+        case GF_FOP_FREMOVEXATTR:
+                return _gf_true;
+        default:
+                return _gf_false;
+        }
+        return _gf_false;
+}*/
diff --git a/xlators/cluster/ec/src/ec-helpers.h b/xlators/cluster/ec/src/ec-helpers.h
new file mode 100644
index 00000000000..015db514e05
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-helpers.h
@@ -0,0 +1,200 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_HELPERS_H__
+#define __EC_HELPERS_H__
+
+#include "ec-types.h"
+
+#define EC_ERR(_x) ((void *)-(intptr_t)(_x))
+#define EC_IS_ERR(_x) (((uintptr_t)(_x) & ~0xfffULL) == ~0xfffULL)
+#define EC_GET_ERR(_x) ((int32_t)(intptr_t)(_x))
+
+#define EC_ALIGN_CHECK(_ptr, _align) ((((uintptr_t)(_ptr)) & ((_align)-1)) == 0)
+
+const char *
+ec_bin(char *str, size_t size, uint64_t value, int32_t digits);
+const char *
+ec_fop_name(int32_t id);
+void
+ec_trace(const char *event, ec_fop_data_t *fop, const char *fmt, ...);
+int32_t
+ec_bits_consume(uint64_t *n);
+size_t
+ec_iov_copy_to(void *dst, struct iovec *vector, int32_t count, off_t offset,
+               size_t size);
+int32_t
+ec_buffer_alloc(xlator_t *xl, size_t size, struct iobref **piobref, void **ptr);
+int32_t
+ec_dict_set_array(dict_t *dict, char *key, uint64_t *value, int32_t size);
+int32_t
+ec_dict_get_array(dict_t *dict, char *key, uint64_t value[], int32_t size);
+
+int32_t
+ec_dict_del_array(dict_t *dict, char *key, uint64_t *value, int32_t size);
+int32_t
+ec_dict_set_number(dict_t *dict, char *key, uint64_t value);
+int32_t
+ec_dict_del_number(dict_t *dict, char *key, uint64_t *value);
+int32_t
+ec_dict_set_config(dict_t *dict, char *key, ec_config_t *config);
+int32_t
+ec_dict_del_config(dict_t *dict, char *key, ec_config_t *config);
+
+int32_t
+ec_loc_parent(xlator_t *xl, loc_t *loc, loc_t *parent);
+int32_t
+ec_loc_update(xlator_t *xl, loc_t *loc, inode_t *inode, struct iatt *iatt);
+
+int32_t
+ec_loc_from_fd(xlator_t *xl, loc_t *loc, fd_t *fd);
+int32_t
+ec_loc_from_loc(xlator_t *xl, loc_t *dst, loc_t *src);
+
+void
+ec_owner_set(call_frame_t *frame, void *owner);
+void
+ec_owner_copy(call_frame_t *frame, gf_lkowner_t *owner);
+
+ec_inode_t *
+__ec_inode_get(inode_t *inode, xlator_t *xl);
+ec_inode_t *
+ec_inode_get(inode_t *inode, xlator_t *xl);
+ec_fd_t *
+__ec_fd_get(fd_t *fd, xlator_t *xl);
+ec_fd_t *
+ec_fd_get(fd_t *fd, xlator_t *xl);
+
+static inline uint32_t
+ec_adjust_size_down(ec_t *ec, uint64_t *value, gf_boolean_t scale)
+{
+    uint64_t head, tmp;
+
+    tmp = *value;
+    head = tmp % ec->stripe_size;
+    tmp -= head;
+
+    if (scale) {
+        tmp /= ec->fragments;
+    }
+
+    *value = tmp;
+
+    return (uint32_t)head;
+}
+
+/* This function can cause an overflow if the passed value is too near to the
+ * uint64_t limit. If this happens, it returns the tail in negative form and
+ * the value is set to UINT64_MAX. */
+static inline int32_t
+ec_adjust_size_up(ec_t *ec, uint64_t *value, gf_boolean_t scale)
+{
+    uint64_t tmp;
+    int32_t tail;
+
+    tmp = *value;
+    /* We first adjust the value down. This never causes overflow. */
+    tail = ec_adjust_size_down(ec, &tmp, scale);
+
+    /* If the value was already aligned, tail will be 0 and nothing else
+     * needs to be done. */
+    if (tail != 0) {
+        /* Otherwise, we need to compute the real tail and adjust the
+         * returned value to the next stripe. */
+        tail = ec->stripe_size - tail;
+        if (scale) {
+            tmp += ec->fragment_size;
+        } else {
+            tmp += ec->stripe_size;
+            /* If no scaling is requested there's a possibility of
+             * overflow. */
+            if (tmp < ec->stripe_size) {
+                tmp = UINT64_MAX;
+                tail = -tail;
+            }
+        }
+    }
+
+    *value = tmp;
+
+    return tail;
+}
+
+/* This function is equivalent to ec_adjust_size_down() but with a potentially
+ * different parameter size (off_t vs uint64_t). */
+static inline uint32_t
+ec_adjust_offset_down(ec_t *ec, off_t *value, gf_boolean_t scale)
+{
+    off_t head, tmp;
+
+    tmp = *value;
+    head = tmp % ec->stripe_size;
+    tmp -= head;
+
+    if (scale) {
+        tmp /= ec->fragments;
+    }
+
+    *value = tmp;
+
+    return (uint32_t)head;
+}
+
+/* This function is equivalent to ec_adjust_size_up() but with a potentially
+ * different parameter size (off_t vs uint64_t). */
+static inline int32_t
+ec_adjust_offset_up(ec_t *ec, off_t *value, gf_boolean_t scale)
+{
+    uint64_t tail, tmp;
+
+    /* An offset is a signed type that can only have positive values, so
+     * we take advantage of this to avoid overflows. We simply convert it
+     * to an unsigned integer and operate normally. This won't cause an
+     * overflow. Overflow is only checked when converting back to an
+     * off_t. */
+    tmp = *value;
+    tail = ec->stripe_size;
+    tail -= (tmp + tail - 1) % tail + 1;
+    tmp += tail;
+    if (scale) {
+        /* If we are scaling, we'll never get an overflow. */
+        tmp /= ec->fragments;
+    } else {
+        /* Check if there has been an overflow. */
+        if ((off_t)tmp < 0) {
+            tmp = GF_OFF_MAX;
+            tail = -tail;
+        }
+    }
+
+    *value = (off_t)tmp;
+
+    return (int32_t)tail;
+}
+
+static inline int32_t
+ec_is_power_of_2(uint32_t value)
+{
+    return (value != 0) && ((value & (value - 1)) == 0);
+}
+
+gf_boolean_t
+ec_is_internal_xattr(dict_t *dict, char *key, data_t *value, void *data);
+
+void
+ec_filter_internal_xattrs(dict_t *xattr);
+
+gf_boolean_t
+ec_is_data_fop(glusterfs_fop_t fop);
+
+int32_t
+ec_launch_replace_heal(ec_t *ec);
+
+#endif /* __EC_HELPERS_H__ */
diff --git a/xlators/cluster/ec/src/ec-inode-read.c b/xlators/cluster/ec/src/ec-inode-read.c
new file mode 100644
index 00000000000..dad5f4d7018
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-inode-read.c
@@ -0,0 +1,2046 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "ec.h"
+#include "ec-messages.h"
+#include "ec-helpers.h"
+#include "ec-common.h"
+#include "ec-combine.h"
+#include "ec-method.h"
+#include "ec-fops.h"
+
+/* FOP: access */
+
+int32_t
+ec_access_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_ACCESS, idx, op_ret,
+                               op_errno);
+    if (cbk) {
+        if (xdata)
+            cbk->xdata = dict_ref(xdata);
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_access(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_access_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->access,
+                      &fop->loc[0], fop->int32, fop->xdata);
+}
+
+int32_t
+ec_manager_access(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk = NULL;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            ec_lock_prepare_inode(fop, &fop->loc[0], EC_QUERY_INFO, 0,
+                                  EC_RANGE_FULL);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_one(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            if (ec_dispatch_one_retry(fop, NULL)) {
+                return EC_STATE_DISPATCH;
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+            GF_ASSERT(cbk);
+            if (fop->cbks.access != NULL) {
+                if (cbk) {
+                    fop->cbks.access(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                     cbk->op_errno, cbk->xdata);
+                }
+            }
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            if (fop->cbks.access != NULL) {
+                fop->cbks.access(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                 NULL);
+            }
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_access(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_access_cbk_t func, void *data, loc_t *loc,
+          int32_t mask, dict_t *xdata)
+{
+    ec_cbk_t callback = {.access = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(ACCESS) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_ACCESS, EC_FLAG_LOCK_SHARED,
+                               target, fop_flags, ec_wind_access,
+                               ec_manager_access, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->int32 = mask;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
+
+/* FOP: getxattr */
+
+int32_t
+ec_combine_getxattr(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    if (!ec_dict_compare(dst->dict, src->dict)) {
+        gf_msg(fop->xl->name, GF_LOG_NOTICE, 0, EC_MSG_DICT_MISMATCH,
+               "Mismatching dictionary in "
+               "answers of 'GF_FOP_GETXATTR'");
+
+        return 0;
+    }
+
+    return 1;
+}
+
+int32_t
+ec_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_GETXATTR, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            if (dict != NULL) {
+                cbk->dict = dict_ref(dict);
+                if (cbk->dict == NULL) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                           "Failed to reference a "
+                           "dictionary.");
+
+                    goto out;
+                }
+            }
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, ec_combine_getxattr);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_getxattr(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_getxattr_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->getxattr,
+                      &fop->loc[0], fop->str[0], fop->xdata);
+}
+
+void
+ec_handle_special_xattrs(ec_fop_data_t *fop)
+{
+    ec_cbk_data_t *cbk = NULL;
+    /* Stime may not be available on all the bricks, so even if some of the
+     * subvols succeed the operation, treat it as answer.*/
+    if (fop->str[0] && fnmatch(GF_XATTR_STIME_PATTERN, fop->str[0], 0) == 0) {
+        if (!fop->answer || (fop->answer->op_ret < 0)) {
+            list_for_each_entry(cbk, &fop->cbk_list, list)
+            {
+                if (cbk->op_ret >= 0) {
+                    fop->answer = cbk;
+                    break;
+                }
+            }
+        }
+    }
+}
+
+int32_t
+ec_manager_getxattr(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            /* clear-locks commands must be done without any locks acquired
+               to avoid interferences. */
+            if ((fop->str[0] == NULL) ||
+                (strncmp(fop->str[0], GF_XATTR_CLRLK_CMD,
+                         SLEN(GF_XATTR_CLRLK_CMD)) != 0)) {
+                if (fop->fd == NULL) {
+                    ec_lock_prepare_inode(fop, &fop->loc[0], EC_QUERY_INFO, 0,
+                                          EC_RANGE_FULL);
+                } else {
+                    ec_lock_prepare_fd(fop, fop->fd, EC_QUERY_INFO, 0,
+                                       EC_RANGE_FULL);
+                }
+                ec_lock(fop);
+            }
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            if (fop->minimum == EC_MINIMUM_ALL) {
+                ec_dispatch_all(fop);
+            } else {
+                ec_dispatch_one(fop);
+            }
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            ec_handle_special_xattrs(fop);
+            if (fop->minimum == EC_MINIMUM_ALL) {
+                cbk = ec_fop_prepare_answer(fop, _gf_true);
+            } else {
+                if (ec_dispatch_one_retry(fop, &cbk)) {
+                    return EC_STATE_DISPATCH;
+                }
+            }
+            if (cbk != NULL) {
+                int32_t err;
+
+                err = ec_dict_combine(cbk, EC_COMBINE_DICT);
+                if (!ec_cbk_set_error(cbk, -err, _gf_true)) {
+                    if (cbk->xdata != NULL)
+                        ec_filter_internal_xattrs(cbk->xdata);
+
+                    if (cbk->dict != NULL)
+                        ec_filter_internal_xattrs(cbk->dict);
+                }
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.getxattr != NULL) {
+                fop->cbks.getxattr(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                   cbk->op_errno, cbk->dict, cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.getxattr != NULL) {
+                fop->cbks.getxattr(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                   NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+int32_t
+ec_getxattr_heal_cbk(call_frame_t *frame, void *cookie, xlator_t *xl,
+                     int32_t op_ret, int32_t op_errno, uintptr_t mask,
+                     uintptr_t good, uintptr_t bad, uint32_t pending,
+                     dict_t *xdata)
+{
+    fop_getxattr_cbk_t func = cookie;
+    ec_t *ec = xl->private;
+    dict_t *dict = NULL;
+    char *str;
+    char bin1[65], bin2[65];
+
+    /* We try to return the 'pending' information in xdata, but if this cannot
+     * be set, we will ignore it silently. We prefer to report the success or
+     * failure of the heal itself. */
+    if (xdata == NULL) {
+        xdata = dict_new();
+    } else {
+        dict_ref(xdata);
+    }
+    if (xdata != NULL) {
+        if (dict_set_uint32(xdata, EC_XATTR_HEAL_NEW, pending) != 0) {
+            /* dict_set_uint32() is marked as 'warn_unused_result' and gcc
+             * enforces to check the result in this case. However we don't
+             * really care if it succeeded or not. We'll just do the same.
+             *
+             * This empty 'if' avoids the warning, and it will be removed by
+             * the optimizer. */
+        }
+    }
+
+    if (op_ret >= 0) {
+        dict = dict_new();
+        if (dict == NULL) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+        } else {
+            if (gf_asprintf(&str, "Good: %s, Bad: %s",
+                            ec_bin(bin1, sizeof(bin1), good, ec->nodes),
+                            ec_bin(bin2, sizeof(bin2), mask & ~(good | bad),
+                                   ec->nodes)) < 0) {
+                dict_unref(dict);
+                dict = NULL;
+
+                op_ret = -1;
+                op_errno = ENOMEM;
+
+                goto out;
+            }
+
+            if (dict_set_dynstr(dict, EC_XATTR_HEAL, str) != 0) {
+                GF_FREE(str);
+                dict_unref(dict);
+                dict = NULL;
+
+                op_ret = -1;
+                op_errno = ENOMEM;
+
+                goto out;
+            }
+        }
+    }
+
+out:
+    func(frame, NULL, xl, op_ret, op_errno, dict, xdata);
+
+    if (dict != NULL) {
+        dict_unref(dict);
+    }
+    if (xdata != NULL) {
+        dict_unref(xdata);
+    }
+
+    return 0;
+}
+
+void
+ec_getxattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_getxattr_cbk_t func, void *data, loc_t *loc,
+            const char *name, dict_t *xdata)
+{
+    ec_cbk_t callback = {.getxattr = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(GETXATTR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    /* Special handling of an explicit self-heal request */
+    if ((name != NULL) && (strcmp(name, EC_XATTR_HEAL) == 0)) {
+        ec_heal(frame, this, target, EC_MINIMUM_ONE, ec_getxattr_heal_cbk, func,
+                loc, 0, NULL);
+
+        return;
+    }
+
+    fop = ec_fop_data_allocate(
+        frame, this, GF_FOP_GETXATTR, EC_FLAG_LOCK_SHARED, target, fop_flags,
+        ec_wind_getxattr, ec_manager_getxattr, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (name != NULL) {
+        /* In case of list-node-uuids xattr, set flag to indicate
+         * the same and use node-uuid xattr for winding fop */
+        if (XATTR_IS_NODE_UUID_LIST(name)) {
+            fop->int32 = 1;
+            fop->str[0] = gf_strdup(GF_XATTR_NODE_UUID_KEY);
+        } else {
+            fop->str[0] = gf_strdup(name);
+        }
+        if (fop->str[0] == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a string.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
+
+/* FOP: fgetxattr */
+
+int32_t
+ec_fgetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_FGETXATTR, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            if (dict != NULL) {
+                cbk->dict = dict_ref(dict);
+                if (cbk->dict == NULL) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                           "Failed to reference a "
+                           "dictionary.");
+
+                    goto out;
+                }
+            }
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, ec_combine_getxattr);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_fgetxattr(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_fgetxattr_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->fgetxattr,
+                      fop->fd, fop->str[0], fop->xdata);
+}
+
+void
+ec_fgetxattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+             uint32_t fop_flags, fop_fgetxattr_cbk_t func, void *data, fd_t *fd,
+             const char *name, dict_t *xdata)
+{
+    ec_cbk_t callback = {.fgetxattr = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FGETXATTR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(
+        frame, this, GF_FOP_FGETXATTR, EC_FLAG_LOCK_SHARED, target, fop_flags,
+        ec_wind_fgetxattr, ec_manager_getxattr, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (name != NULL) {
+        fop->str[0] = gf_strdup(name);
+        if (fop->str[0] == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a string.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
+
+/* FOP: open */
+
+int32_t
+ec_combine_open(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    if (dst->fd != src->fd) {
+        gf_msg(fop->xl->name, GF_LOG_NOTICE, 0, EC_MSG_FD_MISMATCH,
+               "Mismatching fd in answers "
+               "of 'GF_FOP_OPEN': %p <-> %p",
+               dst->fd, src->fd);
+
+        return 0;
+    }
+
+    return 1;
+}
+
+int32_t
+ec_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, fd_t *fd, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_OPEN, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            if (fd != NULL) {
+                cbk->fd = fd_ref(fd);
+                if (cbk->fd == NULL) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0,
+                           EC_MSG_FILE_DESC_REF_FAIL,
+                           "Failed to reference a "
+                           "file descriptor.");
+
+                    goto out;
+                }
+            }
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, ec_combine_open);
+
+        ec_update_fd_status(fd, this, idx, op_ret);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_open(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_open_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->open,
+                      &fop->loc[0], fop->int32, fop->fd, fop->xdata);
+}
+
+int32_t
+ec_open_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
+{
+    ec_fop_data_t *fop = cookie;
+    int32_t error = 0;
+
+    fop = fop->data;
+    if (op_ret >= 0) {
+        fop->answer->iatt[0] = *postbuf;
+    } else {
+        error = op_errno;
+    }
+
+    ec_resume(fop, error);
+
+    return 0;
+}
+
+int32_t
+ec_manager_open(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+    ec_fd_t *ctx;
+    int32_t err;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            LOCK(&fop->fd->lock);
+
+            ctx = __ec_fd_get(fop->fd, fop->xl);
+            if (ctx == NULL) {
+                UNLOCK(&fop->fd->lock);
+
+                fop->error = ENOMEM;
+
+                return EC_STATE_REPORT;
+            }
+            if (!ctx->loc.inode) {
+                err = ec_loc_from_loc(fop->xl, &ctx->loc, &fop->loc[0]);
+                if (err != 0) {
+                    UNLOCK(&fop->fd->lock);
+
+                    fop->error = -err;
+
+                    return EC_STATE_REPORT;
+                }
+            }
+
+            ctx->flags = fop->int32;
+
+            UNLOCK(&fop->fd->lock);
+
+            /* We need to write to specific offsets on the bricks, so we
+               need to remove O_APPEND from flags (if present).
+               If O_TRUNC is specified, we remove it from open and an
+               ftruncate will be executed later, which will correctly update
+               the file size taking appropriate locks. O_TRUNC flag is saved
+               into fop->uint32 to use it later.*/
+            fop->uint32 = fop->int32 & O_TRUNC;
+            fop->int32 &= ~(O_APPEND | O_TRUNC);
+
+            /* Fall through */
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_true);
+            if (cbk != NULL) {
+                int32_t err;
+
+                err = ec_loc_update(fop->xl, &fop->loc[0], cbk->fd->inode,
+                                    NULL);
+                if (!ec_cbk_set_error(cbk, -err, _gf_true)) {
+                    LOCK(&fop->fd->lock);
+
+                    ctx = __ec_fd_get(fop->fd, fop->xl);
+                    if (ctx != NULL) {
+                        ctx->open |= cbk->mask;
+                    }
+
+                    UNLOCK(&fop->fd->lock);
+
+                    /* If O_TRUNC was specified, call ftruncate to
+                       effectively trunc the file with appropriate locks
+                       acquired. We don't use ctx->flags because self-heal
+                       can use the same fd with different flags. */
+                    if (fop->uint32 != 0) {
+                        ec_sleep(fop);
+                        ec_ftruncate(fop->req_frame, fop->xl, cbk->mask,
+                                     fop->minimum, ec_open_truncate_cbk, fop,
+                                     cbk->fd, 0, NULL);
+                    }
+                }
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.open != NULL) {
+                fop->cbks.open(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                               cbk->op_errno, cbk->fd, cbk->xdata);
+            }
+
+            return EC_STATE_END;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.open != NULL) {
+                fop->cbks.open(fop->req_frame, fop, fop->xl, -1, fop->error,
+                               NULL, NULL);
+            }
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_open(call_frame_t *frame, xlator_t *this, uintptr_t target,
+        uint32_t fop_flags, fop_open_cbk_t func, void *data, loc_t *loc,
+        int32_t flags, fd_t *fd, dict_t *xdata)
+{
+    ec_cbk_t callback = {.open = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(OPEN) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_OPEN, EC_FLAG_LOCK_SHARED,
+                               target, fop_flags, ec_wind_open, ec_manager_open,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->int32 = flags;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
+
+/* FOP: readlink */
+
+int32_t
+ec_combine_readlink(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    if (!ec_iatt_combine(fop, dst->iatt, src->iatt, 1)) {
+        gf_msg(fop->xl->name, GF_LOG_NOTICE, 0, EC_MSG_IATT_MISMATCH,
+               "Mismatching iatt in "
+               "answers of 'GF_FOP_READLINK'");
+
+        return 0;
+    }
+
+    return 1;
+}
+
+int32_t
+ec_readlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, const char *path,
+                struct iatt *buf, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, fop->id, idx, op_ret,
+                               op_errno);
+    if (cbk) {
+        if (xdata)
+            cbk->xdata = dict_ref(xdata);
+
+        if (cbk->op_ret >= 0) {
+            cbk->iatt[0] = *buf;
+            cbk->str = gf_strdup(path);
+            if (!cbk->str) {
+                ec_cbk_set_error(cbk, ENOMEM, _gf_true);
+            }
+        }
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL)
+        ec_complete(fop);
+
+    return 0;
+}
+
+void
+ec_wind_readlink(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_readlink_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->readlink,
+                      &fop->loc[0], fop->size, fop->xdata);
+}
+
+int32_t
+ec_manager_readlink(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk = NULL;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            ec_lock_prepare_inode(fop, &fop->loc[0], EC_QUERY_INFO, 0,
+                                  EC_RANGE_FULL);
+            ec_lock(fop);
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_one(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            if (ec_dispatch_one_retry(fop, &cbk)) {
+                return EC_STATE_DISPATCH;
+            }
+
+            if ((cbk != NULL) && (cbk->op_ret >= 0)) {
+                ec_iatt_rebuild(fop->xl->private, &cbk->iatt[0], 1, 1);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+            GF_ASSERT(cbk);
+            if (fop->cbks.readlink != NULL) {
+                fop->cbks.readlink(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                   cbk->op_errno, cbk->str, &cbk->iatt[0],
+                                   cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            if (fop->cbks.readlink != NULL) {
+                fop->cbks.readlink(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                   NULL, NULL, NULL);
+            }
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_readlink(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_readlink_cbk_t func, void *data, loc_t *loc,
+            size_t size, dict_t *xdata)
+{
+    ec_cbk_t callback = {.readlink = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(READLINK) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(
+        frame, this, GF_FOP_READLINK, EC_FLAG_LOCK_SHARED, target, fop_flags,
+        ec_wind_readlink, ec_manager_readlink, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->size = size;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: readv */
+
+int32_t
+ec_readv_rebuild(ec_t *ec, ec_fop_data_t *fop, ec_cbk_data_t *cbk)
+{
+    struct iovec vector[1];
+    ec_cbk_data_t *ans = NULL;
+    struct iobref *iobref = NULL;
+    void *ptr;
+    uint64_t fsize = 0, size = 0, max = 0;
+    int32_t pos, err = -ENOMEM;
+
+    if (cbk->op_ret < 0) {
+        err = -cbk->op_errno;
+
+        goto out;
+    }
+
+    /* This shouldn't fail because we have the inode locked. */
+    GF_ASSERT(ec_get_inode_size(fop, fop->fd->inode, &cbk->iatt[0].ia_size));
+
+    if (cbk->op_ret > 0) {
+        void *blocks[cbk->count];
+        uint32_t values[cbk->count];
+
+        fsize = cbk->op_ret;
+        size = fsize * ec->fragments;
+        for (ans = cbk; ans != NULL; ans = ans->next) {
+            pos = gf_bits_count(cbk->mask & ((1 << ans->idx) - 1));
+            values[pos] = ans->idx + 1;
+            blocks[pos] = ans->vector[0].iov_base;
+            if ((ans->int32 != 1) ||
+                !EC_ALIGN_CHECK(blocks[pos], EC_METHOD_WORD_SIZE)) {
+                if (iobref == NULL) {
+                    err = ec_buffer_alloc(ec->xl, size, &iobref, &ptr);
+                    if (err != 0) {
+                        goto out;
+                    }
+                }
+                ec_iov_copy_to(ptr, ans->vector, ans->int32, 0, fsize);
+                blocks[pos] = ptr;
+                ptr += fsize;
+            }
+        }
+
+        err = ec_buffer_alloc(ec->xl, size, &iobref, &ptr);
+        if (err != 0) {
+            goto out;
+        }
+
+        err = ec_method_decode(&ec->matrix, fsize, cbk->mask, values, blocks,
+                               ptr);
+        if (err != 0) {
+            goto out;
+        }
+
+        vector[0].iov_base = ptr + fop->head;
+        vector[0].iov_len = size - fop->head;
+
+        max = fop->offset * ec->fragments + size;
+        if (max > cbk->iatt[0].ia_size) {
+            max = cbk->iatt[0].ia_size;
+        }
+        max -= fop->offset * ec->fragments + fop->head;
+        if (max > fop->user_size) {
+            max = fop->user_size;
+        }
+        size -= fop->head;
+        if (size > max) {
+            vector[0].iov_len -= size - max;
+            size = max;
+        }
+
+        cbk->op_ret = size;
+        cbk->int32 = 1;
+
+        iobref_unref(cbk->buffers);
+        cbk->buffers = iobref;
+
+        GF_FREE(cbk->vector);
+        cbk->vector = iov_dup(vector, 1);
+        if (cbk->vector == NULL) {
+            return -ENOMEM;
+        }
+    }
+
+    return 0;
+
+out:
+    if (iobref != NULL) {
+        iobref_unref(iobref);
+    }
+
+    return err;
+}
+
+int32_t
+ec_combine_readv(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    if (!ec_vector_compare(dst->vector, dst->int32, src->vector, src->int32)) {
+        gf_msg(fop->xl->name, GF_LOG_NOTICE, 0, EC_MSG_VECTOR_MISMATCH,
+               "Mismatching vector in "
+               "answers of 'GF_FOP_READ'");
+
+        return 0;
+    }
+
+    if (!ec_iatt_combine(fop, dst->iatt, src->iatt, 1)) {
+        gf_msg(fop->xl->name, GF_LOG_NOTICE, 0, EC_MSG_IATT_MISMATCH,
+               "Mismatching iatt in "
+               "answers of 'GF_FOP_READ'");
+
+        return 0;
+    }
+
+    return 1;
+}
+
+int32_t
+ec_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, struct iovec *vector, int32_t count,
+             struct iatt *stbuf, struct iobref *iobref, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    ec_t *ec = this->private;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_READ, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            cbk->int32 = count;
+
+            if (count > 0) {
+                cbk->vector = iov_dup(vector, count);
+                if (cbk->vector == NULL) {
+                    gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                           "Failed to duplicate a "
+                           "vector list.");
+
+                    goto out;
+                }
+                cbk->int32 = count;
+            }
+            if (stbuf != NULL) {
+                cbk->iatt[0] = *stbuf;
+            }
+            if (iobref != NULL) {
+                cbk->buffers = iobref_ref(iobref);
+                if (cbk->buffers == NULL) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_BUF_REF_FAIL,
+                           "Failed to reference a "
+                           "buffer.");
+
+                    goto out;
+                }
+            }
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        if ((op_ret > 0) && ((op_ret % ec->fragment_size) != 0)) {
+            ec_cbk_set_error(cbk, EIO, _gf_true);
+        }
+
+        ec_combine(cbk, ec_combine_readv);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_readv(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_readv_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->readv, fop->fd,
+                      fop->size, fop->offset, fop->uint32, fop->xdata);
+}
+
+int32_t
+ec_manager_readv(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+    ec_t *ec = fop->xl->private;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            fop->user_size = fop->size;
+            fop->head = ec_adjust_offset_down(fop->xl->private, &fop->offset,
+                                              _gf_true);
+            fop->size += fop->head;
+            ec_adjust_size_up(fop->xl->private, &fop->size, _gf_true);
+
+            /* Fall through */
+
+        case EC_STATE_LOCK:
+            ec_lock_prepare_fd(fop, fop->fd, EC_QUERY_INFO, fop->offset,
+                               fop->size);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            if (ec->read_mask) {
+                fop->mask &= ec->read_mask;
+            }
+            ec_dispatch_min(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_true);
+            if (cbk != NULL) {
+                int32_t err;
+
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 1, cbk->count);
+
+                err = ec_readv_rebuild(fop->xl->private, fop, cbk);
+                if (err != 0) {
+                    ec_cbk_set_error(cbk, -err, _gf_true);
+                }
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.readv != NULL) {
+                fop->cbks.readv(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                cbk->op_errno, cbk->vector, cbk->int32,
+                                &cbk->iatt[0], cbk->buffers, cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.readv != NULL) {
+                fop->cbks.readv(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                NULL, 0, NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_readv(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_readv_cbk_t func, void *data, fd_t *fd,
+         size_t size, off_t offset, uint32_t flags, dict_t *xdata)
+{
+    ec_cbk_t callback = {.readv = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(READ) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_READ, EC_FLAG_LOCK_SHARED,
+                               target, fop_flags, ec_wind_readv,
+                               ec_manager_readv, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->size = size;
+    fop->offset = offset;
+    fop->uint32 = flags;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, 0, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: seek */
+
+int32_t
+ec_seek_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, off_t offset, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    ec_t *ec = this->private;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_SEEK, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            cbk->offset = offset;
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+        }
+
+        if ((op_ret > 0) && ((cbk->offset % ec->fragment_size) != 0)) {
+            cbk->op_ret = -1;
+            cbk->op_errno = EIO;
+        }
+
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_seek(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_seek_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->seek, fop->fd,
+                      fop->offset, fop->seek, fop->xdata);
+}
+
+int32_t
+ec_manager_seek(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+    uint64_t size;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            fop->user_size = fop->offset;
+            fop->head = ec_adjust_offset_down(fop->xl->private, &fop->offset,
+                                              _gf_true);
+
+            /* Fall through */
+
+        case EC_STATE_LOCK:
+            ec_lock_prepare_fd(fop, fop->fd, EC_QUERY_INFO, fop->offset,
+                               EC_RANGE_FULL);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            /* This shouldn't fail because we have the inode locked. */
+            GF_ASSERT(
+                ec_get_inode_size(fop, fop->locks[0].lock->loc.inode, &size));
+
+            if (fop->user_size >= size) {
+                ec_fop_set_error(fop, ENXIO);
+
+                return EC_STATE_REPORT;
+            }
+
+            ec_dispatch_one(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            if (ec_dispatch_one_retry(fop, &cbk)) {
+                return EC_STATE_DISPATCH;
+            }
+            if ((cbk != NULL) && (cbk->op_ret >= 0)) {
+                ec_t *ec = fop->xl->private;
+
+                /* This shouldn't fail because we have the inode locked. */
+                GF_ASSERT(ec_get_inode_size(fop, fop->locks[0].lock->loc.inode,
+                                            &size));
+
+                cbk->offset *= ec->fragments;
+                if (cbk->offset < fop->user_size) {
+                    cbk->offset = fop->user_size;
+                }
+                if (cbk->offset > size) {
+                    cbk->offset = size;
+                }
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.seek != NULL) {
+                fop->cbks.seek(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                               cbk->op_errno, cbk->offset, cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.seek != NULL) {
+                fop->cbks.seek(fop->req_frame, fop, fop->xl, -1, fop->error, 0,
+                               NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, 0, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_seek(call_frame_t *frame, xlator_t *this, uintptr_t target,
+        uint32_t fop_flags, fop_seek_cbk_t func, void *data, fd_t *fd,
+        off_t offset, gf_seek_what_t what, dict_t *xdata)
+{
+    ec_cbk_t callback = {.seek = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = EIO;
+
+    gf_msg_trace("ec", 0, "EC(SEEK) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_SEEK, EC_FLAG_LOCK_SHARED,
+                               target, fop_flags, ec_wind_seek, ec_manager_seek,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->offset = offset;
+    fop->seek = what;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, EIO, 0, NULL);
+    }
+}
+
+/* FOP: stat */
+
+int32_t
+ec_combine_stat(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    if (!ec_iatt_combine(fop, dst->iatt, src->iatt, 1)) {
+        gf_msg(fop->xl->name, GF_LOG_NOTICE, 0, EC_MSG_IATT_MISMATCH,
+               "Mismatching iatt in "
+               "answers of 'GF_FOP_STAT'");
+
+        return 0;
+    }
+
+    return 1;
+}
+
+int32_t
+ec_stat_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, struct iatt *buf, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_STAT, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            if (buf != NULL) {
+                cbk->iatt[0] = *buf;
+            }
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, ec_combine_stat);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_stat(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_stat_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->stat,
+                      &fop->loc[0], fop->xdata);
+}
+
+int32_t
+ec_manager_stat(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            if (fop->fd == NULL) {
+                ec_lock_prepare_inode(fop, &fop->loc[0], EC_QUERY_INFO, 0,
+                                      EC_RANGE_FULL);
+            } else {
+                ec_lock_prepare_fd(fop, fop->fd, EC_QUERY_INFO, 0,
+                                   EC_RANGE_FULL);
+            }
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_true);
+
+            if (cbk != NULL) {
+                if (cbk->iatt[0].ia_type == IA_IFREG) {
+                    ec_iatt_rebuild(fop->xl->private, cbk->iatt, 1, cbk->count);
+
+                    /* This shouldn't fail because we have the inode locked. */
+                    GF_ASSERT(ec_get_inode_size(fop,
+                                                fop->locks[0].lock->loc.inode,
+                                                &cbk->iatt[0].ia_size));
+                }
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->id == GF_FOP_STAT) {
+                if (fop->cbks.stat != NULL) {
+                    fop->cbks.stat(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                   cbk->op_errno, &cbk->iatt[0], cbk->xdata);
+                }
+            } else {
+                if (fop->cbks.fstat != NULL) {
+                    fop->cbks.fstat(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                    cbk->op_errno, &cbk->iatt[0], cbk->xdata);
+                }
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->id == GF_FOP_STAT) {
+                if (fop->cbks.stat != NULL) {
+                    fop->cbks.stat(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                   NULL, NULL);
+                }
+            } else {
+                if (fop->cbks.fstat != NULL) {
+                    fop->cbks.fstat(fop->req_frame, fop, fop->xl, -1,
+                                    fop->error, NULL, NULL);
+                }
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_stat(call_frame_t *frame, xlator_t *this, uintptr_t target,
+        uint32_t fop_flags, fop_stat_cbk_t func, void *data, loc_t *loc,
+        dict_t *xdata)
+{
+    ec_cbk_t callback = {.stat = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(STAT) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_STAT, EC_FLAG_LOCK_SHARED,
+                               target, fop_flags, ec_wind_stat, ec_manager_stat,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
+
+/* FOP: fstat */
+
+int32_t
+ec_fstat_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, struct iatt *buf, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_FSTAT, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            if (buf != NULL) {
+                cbk->iatt[0] = *buf;
+            }
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, ec_combine_stat);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_fstat(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_fstat_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->fstat, fop->fd,
+                      fop->xdata);
+}
+
+void
+ec_fstat(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_fstat_cbk_t func, void *data, fd_t *fd,
+         dict_t *xdata)
+{
+    ec_cbk_t callback = {.fstat = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FSTAT) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FSTAT, EC_FLAG_LOCK_SHARED,
+                               target, fop_flags, ec_wind_fstat,
+                               ec_manager_stat, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
diff --git a/xlators/cluster/ec/src/ec-inode-write.c b/xlators/cluster/ec/src/ec-inode-write.c
new file mode 100644
index 00000000000..9b5fe2a7fdc
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-inode-write.c
@@ -0,0 +1,2369 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "ec-messages.h"
+#include "ec-helpers.h"
+#include "ec-common.h"
+#include "ec-combine.h"
+#include "ec-method.h"
+#include "ec-fops.h"
+#include "ec-mem-types.h"
+
+int32_t
+ec_update_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
+{
+    ec_fop_data_t *fop = cookie;
+    ec_cbk_data_t *cbk = NULL;
+    ec_fop_data_t *parent = fop->parent;
+    int i = 0;
+
+    ec_trace("UPDATE_WRITEV_CBK", cookie, "ret=%d, errno=%d, parent-fop=%s",
+             op_ret, op_errno, ec_fop_name(parent->id));
+
+    if (op_ret < 0) {
+        ec_fop_set_error(parent, op_errno);
+        goto out;
+    }
+    cbk = ec_cbk_data_allocate(parent->frame, this, parent, parent->id, 0,
+                               op_ret, op_errno);
+    if (!cbk) {
+        ec_fop_set_error(parent, ENOMEM);
+        goto out;
+    }
+
+    if (xdata)
+        cbk->xdata = dict_ref(xdata);
+
+    if (prebuf)
+        cbk->iatt[i++] = *prebuf;
+
+    if (postbuf)
+        cbk->iatt[i++] = *postbuf;
+
+    LOCK(&parent->lock);
+    {
+        parent->good &= fop->good;
+
+        if (gf_bits_count(parent->good) < parent->minimum) {
+            __ec_fop_set_error(parent, EIO);
+        } else if (fop->error == 0 && parent->answer == NULL) {
+            parent->answer = cbk;
+        }
+    }
+    UNLOCK(&parent->lock);
+out:
+    return 0;
+}
+
+static int32_t
+ec_update_write(ec_fop_data_t *fop, uintptr_t mask, off_t offset, uint64_t size)
+{
+    struct iobref *iobref = NULL;
+    struct iobuf *iobuf = NULL;
+    struct iovec vector;
+    int32_t err = -ENOMEM;
+
+    iobref = iobref_new();
+    if (iobref == NULL) {
+        goto out;
+    }
+    iobuf = iobuf_get(fop->xl->ctx->iobuf_pool);
+    if (iobuf == NULL) {
+        goto out;
+    }
+    err = iobref_add(iobref, iobuf);
+    if (err != 0) {
+        goto out;
+    }
+
+    if (fop->locks[0].lock)
+        ec_lock_update_good(fop->locks[0].lock, fop);
+    vector.iov_base = iobuf->ptr;
+    vector.iov_len = size;
+    memset(vector.iov_base, 0, vector.iov_len);
+
+    ec_writev(fop->frame, fop->xl, mask, fop->minimum, ec_update_writev_cbk,
+              NULL, fop->fd, &vector, 1, offset, 0, iobref, NULL);
+
+    err = 0;
+
+out:
+    if (iobuf != NULL) {
+        iobuf_unref(iobuf);
+    }
+    if (iobref != NULL) {
+        iobref_unref(iobref);
+    }
+
+    return err;
+}
+
+int
+ec_inode_write_cbk(call_frame_t *frame, xlator_t *this, void *cookie,
+                   int op_ret, int op_errno, struct iatt *prestat,
+                   struct iatt *poststat, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int i = 0;
+    int idx = 0;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+    idx = (int32_t)(uintptr_t)cookie;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, fop->id, idx, op_ret,
+                               op_errno);
+    if (!cbk)
+        goto out;
+
+    if (op_ret < 0)
+        goto out;
+
+    if (xdata)
+        cbk->xdata = dict_ref(xdata);
+
+    if (prestat)
+        cbk->iatt[i++] = *prestat;
+
+    if (poststat)
+        cbk->iatt[i++] = *poststat;
+
+out:
+    if (cbk)
+        ec_combine(cbk, ec_combine_write);
+
+    if (fop)
+        ec_complete(fop);
+    return 0;
+}
+/* FOP: removexattr */
+
+int32_t
+ec_removexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    return ec_inode_write_cbk(frame, this, cookie, op_ret, op_errno, NULL, NULL,
+                              xdata);
+}
+
+void
+ec_wind_removexattr(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_removexattr_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->removexattr,
+                      &fop->loc[0], fop->str[0], fop->xdata);
+}
+
+void
+ec_xattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = cookie;
+    switch (fop->id) {
+        case GF_FOP_SETXATTR:
+            if (fop->cbks.setxattr) {
+                QUORUM_CBK(fop->cbks.setxattr, fop, frame, cookie, this, op_ret,
+                           op_errno, xdata);
+            }
+            break;
+        case GF_FOP_REMOVEXATTR:
+            if (fop->cbks.removexattr) {
+                QUORUM_CBK(fop->cbks.removexattr, fop, frame, cookie, this,
+                           op_ret, op_errno, xdata);
+            }
+            break;
+        case GF_FOP_FSETXATTR:
+            if (fop->cbks.fsetxattr) {
+                QUORUM_CBK(fop->cbks.fsetxattr, fop, frame, cookie, this,
+                           op_ret, op_errno, xdata);
+            }
+            break;
+        case GF_FOP_FREMOVEXATTR:
+            if (fop->cbks.fremovexattr) {
+                QUORUM_CBK(fop->cbks.fremovexattr, fop, frame, cookie, this,
+                           op_ret, op_errno, xdata);
+            }
+            break;
+    }
+}
+
+int32_t
+ec_manager_xattr(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            if (fop->fd == NULL) {
+                ec_lock_prepare_inode(fop, &fop->loc[0],
+                                      EC_UPDATE_META | EC_QUERY_INFO, 0,
+                                      EC_RANGE_FULL);
+            } else {
+                ec_lock_prepare_fd(fop, fop->fd, EC_UPDATE_META | EC_QUERY_INFO,
+                                   0, EC_RANGE_FULL);
+            }
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            ec_fop_prepare_answer(fop, _gf_false);
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            ec_xattr_cbk(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                         cbk->op_errno, cbk->xdata);
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            ec_xattr_cbk(fop->req_frame, fop, fop->xl, -1, fop->error, NULL);
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_removexattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+               uint32_t fop_flags, fop_removexattr_cbk_t func, void *data,
+               loc_t *loc, const char *name, dict_t *xdata)
+{
+    ec_cbk_t callback = {.removexattr = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(REMOVEXATTR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_REMOVEXATTR, 0, target,
+                               fop_flags, ec_wind_removexattr, ec_manager_xattr,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (name != NULL) {
+        fop->str[0] = gf_strdup(name);
+        if (fop->str[0] == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a string.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
+
+/* FOP: fremovexattr */
+
+int32_t
+ec_fremovexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    return ec_inode_write_cbk(frame, this, cookie, op_ret, op_errno, NULL, NULL,
+                              xdata);
+}
+
+void
+ec_wind_fremovexattr(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_fremovexattr_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->fremovexattr,
+                      fop->fd, fop->str[0], fop->xdata);
+}
+
+void
+ec_fremovexattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+                uint32_t fop_flags, fop_fremovexattr_cbk_t func, void *data,
+                fd_t *fd, const char *name, dict_t *xdata)
+{
+    ec_cbk_t callback = {.fremovexattr = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FREMOVEXATTR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FREMOVEXATTR, 0, target,
+                               fop_flags, ec_wind_fremovexattr,
+                               ec_manager_xattr, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (name != NULL) {
+        fop->str[0] = gf_strdup(name);
+        if (fop->str[0] == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a string.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
+
+/* FOP: setattr */
+
+int32_t
+ec_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, struct iatt *prestat,
+               struct iatt *poststat, dict_t *xdata)
+{
+    return ec_inode_write_cbk(frame, this, cookie, op_ret, op_errno, prestat,
+                              poststat, xdata);
+}
+
+void
+ec_wind_setattr(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_setattr_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->setattr,
+                      &fop->loc[0], &fop->iatt, fop->int32, fop->xdata);
+}
+
+int32_t
+ec_manager_setattr(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            if (fop->fd == NULL) {
+                ec_lock_prepare_inode(fop, &fop->loc[0],
+                                      EC_UPDATE_META | EC_QUERY_INFO, 0,
+                                      EC_RANGE_FULL);
+            } else {
+                ec_lock_prepare_fd(fop, fop->fd, EC_UPDATE_META | EC_QUERY_INFO,
+                                   0, EC_RANGE_FULL);
+            }
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                if (cbk->iatt[0].ia_type == IA_IFREG) {
+                    ec_iatt_rebuild(fop->xl->private, cbk->iatt, 2, cbk->count);
+
+                    /* This shouldn't fail because we have the inode locked. */
+                    GF_ASSERT(ec_get_inode_size(fop,
+                                                fop->locks[0].lock->loc.inode,
+                                                &cbk->iatt[0].ia_size));
+                    cbk->iatt[1].ia_size = cbk->iatt[0].ia_size;
+                }
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->id == GF_FOP_SETATTR) {
+                if (fop->cbks.setattr != NULL) {
+                    QUORUM_CBK(fop->cbks.setattr, fop, fop->req_frame, fop,
+                               fop->xl, cbk->op_ret, cbk->op_errno,
+                               &cbk->iatt[0], &cbk->iatt[1], cbk->xdata);
+                }
+            } else {
+                if (fop->cbks.fsetattr != NULL) {
+                    QUORUM_CBK(fop->cbks.fsetattr, fop, fop->req_frame, fop,
+                               fop->xl, cbk->op_ret, cbk->op_errno,
+                               &cbk->iatt[0], &cbk->iatt[1], cbk->xdata);
+                }
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->id == GF_FOP_SETATTR) {
+                if (fop->cbks.setattr != NULL) {
+                    fop->cbks.setattr(fop->req_frame, fop, fop->xl, -1,
+                                      fop->error, NULL, NULL, NULL);
+                }
+            } else {
+                if (fop->cbks.fsetattr != NULL) {
+                    fop->cbks.fsetattr(fop->req_frame, fop, fop->xl, -1,
+                                       fop->error, NULL, NULL, NULL);
+                }
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_setattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_setattr_cbk_t func, void *data, loc_t *loc,
+           struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    ec_cbk_t callback = {.setattr = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(SETATTR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_SETATTR, 0, target,
+                               fop_flags, ec_wind_setattr, ec_manager_setattr,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->int32 = valid;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (stbuf != NULL) {
+        fop->iatt = *stbuf;
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: fsetattr */
+
+int32_t
+ec_fsetattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *prestat,
+                struct iatt *poststat, dict_t *xdata)
+{
+    return ec_inode_write_cbk(frame, this, cookie, op_ret, op_errno, prestat,
+                              poststat, xdata);
+}
+
+void
+ec_wind_fsetattr(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_fsetattr_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->fsetattr,
+                      fop->fd, &fop->iatt, fop->int32, fop->xdata);
+}
+
+void
+ec_fsetattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_fsetattr_cbk_t func, void *data, fd_t *fd,
+            struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    ec_cbk_t callback = {.fsetattr = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FSETATTR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FSETATTR, 0, target,
+                               fop_flags, ec_wind_fsetattr, ec_manager_setattr,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->int32 = valid;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (stbuf != NULL) {
+        fop->iatt = *stbuf;
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: setxattr */
+
+int32_t
+ec_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    return ec_inode_write_cbk(frame, this, cookie, op_ret, op_errno, NULL, NULL,
+                              xdata);
+}
+
+void
+ec_wind_setxattr(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_setxattr_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->setxattr,
+                      &fop->loc[0], fop->dict, fop->int32, fop->xdata);
+}
+
+void
+ec_setxattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_setxattr_cbk_t func, void *data, loc_t *loc,
+            dict_t *dict, int32_t flags, dict_t *xdata)
+{
+    ec_cbk_t callback = {.setxattr = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(SETXATTR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_SETXATTR, 0, target,
+                               fop_flags, ec_wind_setxattr, ec_manager_xattr,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->int32 = flags;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (dict != NULL) {
+        fop->dict = dict_copy_with_ref(dict, NULL);
+        if (fop->dict == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
+
+/* FOP: fsetxattr */
+
+int32_t
+ec_fsetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_FSETXATTR, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_fsetxattr(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_fsetxattr_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->fsetxattr,
+                      fop->fd, fop->dict, fop->int32, fop->xdata);
+}
+
+void
+ec_fsetxattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+             uint32_t fop_flags, fop_fsetxattr_cbk_t func, void *data, fd_t *fd,
+             dict_t *dict, int32_t flags, dict_t *xdata)
+{
+    ec_cbk_t callback = {.fsetxattr = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FSETXATTR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FSETXATTR, 0, target,
+                               fop_flags, ec_wind_fsetxattr, ec_manager_xattr,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->int32 = flags;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (dict != NULL) {
+        fop->dict = dict_copy_with_ref(dict, NULL);
+        if (fop->dict == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
+
+/*********************************************************************
+ *
+ * File Operation : fallocate
+ *
+ *********************************************************************/
+
+int32_t
+ec_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                 struct iatt *postbuf, dict_t *xdata)
+{
+    return ec_inode_write_cbk(frame, this, cookie, op_ret, op_errno, prebuf,
+                              postbuf, xdata);
+}
+
+void
+ec_wind_fallocate(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_fallocate_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->fallocate,
+                      fop->fd, fop->int32, fop->offset, fop->size, fop->xdata);
+}
+
+int32_t
+ec_manager_fallocate(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk = NULL;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            if (fop->size == 0) {
+                ec_fop_set_error(fop, EINVAL);
+                return EC_STATE_REPORT;
+            }
+            if (fop->int32 &
+                (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
+                 FALLOC_FL_ZERO_RANGE | FALLOC_FL_PUNCH_HOLE)) {
+                ec_fop_set_error(fop, ENOTSUP);
+                return EC_STATE_REPORT;
+            }
+            fop->user_size = fop->offset + fop->size;
+            fop->head = ec_adjust_offset_down(fop->xl->private, &fop->offset,
+                                              _gf_true);
+            fop->size += fop->head;
+            ec_adjust_size_up(fop->xl->private, &fop->size, _gf_true);
+
+            /* Fall through */
+
+        case EC_STATE_LOCK:
+            ec_lock_prepare_fd(fop, fop->fd,
+                               EC_UPDATE_DATA | EC_UPDATE_META | EC_QUERY_INFO,
+                               fop->offset, fop->size);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 2, cbk->count);
+
+                /* This shouldn't fail because we have the inode locked. */
+                LOCK(&fop->locks[0].lock->loc.inode->lock);
+                {
+                    GF_ASSERT(__ec_get_inode_size(fop,
+                                                  fop->locks[0].lock->loc.inode,
+                                                  &cbk->iatt[0].ia_size));
+
+                    /*If mode has FALLOC_FL_KEEP_SIZE keep the size */
+                    if (fop->int32 & FALLOC_FL_KEEP_SIZE) {
+                        cbk->iatt[1].ia_size = cbk->iatt[0].ia_size;
+                    } else if (fop->user_size > cbk->iatt[0].ia_size) {
+                        cbk->iatt[1].ia_size = fop->user_size;
+
+                        /* This shouldn't fail because we have the inode
+                         * locked. */
+                        GF_ASSERT(__ec_set_inode_size(
+                            fop, fop->locks[0].lock->loc.inode,
+                            cbk->iatt[1].ia_size));
+                    } else {
+                        cbk->iatt[1].ia_size = cbk->iatt[0].ia_size;
+                    }
+                }
+                UNLOCK(&fop->locks[0].lock->loc.inode->lock);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.fallocate != NULL) {
+                QUORUM_CBK(fop->cbks.fallocate, fop, fop->req_frame, fop,
+                           fop->xl, cbk->op_ret, cbk->op_errno, &cbk->iatt[0],
+                           &cbk->iatt[1], cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.fallocate != NULL) {
+                fop->cbks.fallocate(fop->req_frame, fop, fop->xl, -1,
+                                    fop->error, NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_fallocate(call_frame_t *frame, xlator_t *this, uintptr_t target,
+             uint32_t fop_flags, fop_fallocate_cbk_t func, void *data, fd_t *fd,
+             int32_t mode, off_t offset, size_t len, dict_t *xdata)
+{
+    ec_cbk_t callback = {.fallocate = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FALLOCATE) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FALLOCATE, 0, target,
+                               fop_flags, ec_wind_fallocate,
+                               ec_manager_fallocate, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+    fop->int32 = mode;
+    fop->offset = offset;
+    fop->size = len;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+            goto out;
+        }
+    }
+
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL);
+    }
+}
+
+/*********************************************************************
+ *
+ * File Operation : Discard
+ *
+ *********************************************************************/
+void
+ec_update_discard_write(ec_fop_data_t *fop, uintptr_t mask)
+{
+    ec_t *ec = fop->xl->private;
+    off_t off_head = 0;
+    off_t off_tail = 0;
+    uint64_t size_head = 0;
+    uint64_t size_tail = 0;
+    int error = 0;
+
+    off_head = fop->offset * ec->fragments - fop->int32;
+    if (fop->size == 0) {
+        error = ec_update_write(fop, mask, off_head, fop->user_size);
+    } else {
+        size_head = fop->int32;
+        size_tail = (off_head + fop->user_size) % ec->stripe_size;
+        off_tail = off_head + fop->user_size - size_tail;
+        if (size_head) {
+            error = ec_update_write(fop, mask, off_head, size_head);
+            if (error) {
+                goto out;
+            }
+        }
+        if (size_tail) {
+            error = ec_update_write(fop, mask, off_tail, size_tail);
+        }
+    }
+out:
+    if (error)
+        ec_fop_set_error(fop, -error);
+}
+
+void
+ec_discard_adjust_offset_size(ec_fop_data_t *fop)
+{
+    ec_t *ec = fop->xl->private;
+
+    fop->user_size = fop->size;
+    /* If discard length covers at least a fragment on brick, we will
+     * perform discard operation(when fop->size is non-zero) else we just
+     * write zeros.
+     */
+    fop->int32 = ec_adjust_offset_up(ec, &fop->offset, _gf_true);
+    fop->frag_range.first = fop->offset;
+    if (fop->size < fop->int32) {
+        fop->size = 0;
+    } else {
+        fop->size -= fop->int32;
+        ec_adjust_size_down(ec, &fop->size, _gf_true);
+    }
+    fop->frag_range.last = fop->offset + fop->size;
+}
+
+int32_t
+ec_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+               struct iatt *postbuf, dict_t *xdata)
+{
+    return ec_inode_write_cbk(frame, this, cookie, op_ret, op_errno, prebuf,
+                              postbuf, xdata);
+}
+
+void
+ec_wind_discard(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_discard_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->discard,
+                      fop->fd, fop->offset, fop->size, fop->xdata);
+}
+
+int32_t
+ec_manager_discard(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk = NULL;
+    off_t fl_start = 0;
+    uint64_t fl_size = 0;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            if ((fop->size <= 0) || (fop->offset < 0)) {
+                ec_fop_set_error(fop, EINVAL);
+                return EC_STATE_REPORT;
+            }
+            /* Because of the head/tail writes, "discard" happens on the
+             * remaining regions, but we need to compute region including
+             * head/tail writes so compute them separately*/
+            fl_start = fop->offset;
+            fl_size = fop->size;
+            fl_size += ec_adjust_offset_down(fop->xl->private, &fl_start,
+                                             _gf_true);
+            ec_adjust_size_up(fop->xl->private, &fl_size, _gf_true);
+
+            ec_discard_adjust_offset_size(fop);
+
+            /* Fall through */
+
+        case EC_STATE_LOCK:
+            ec_lock_prepare_fd(fop, fop->fd,
+                               EC_UPDATE_DATA | EC_UPDATE_META | EC_QUERY_INFO,
+                               fl_start, fl_size);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+
+            /* Dispatch discard fop only if we have whole fragment
+             * to deallocate */
+            if (fop->size) {
+                ec_dispatch_all(fop);
+                return EC_STATE_DELAYED_START;
+            } else {
+                /* Assume discard to have succeeded on all bricks */
+                ec_succeed_all(fop);
+            }
+
+            /* Fall through */
+
+        case EC_STATE_DELAYED_START:
+
+            if (fop->size) {
+                if (fop->answer && fop->answer->op_ret == 0)
+                    ec_update_discard_write(fop, fop->answer->mask);
+            } else {
+                ec_update_discard_write(fop, fop->mask);
+            }
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 2, cbk->count);
+
+                /* This shouldn't fail because we have the inode locked. */
+                GF_ASSERT(ec_get_inode_size(fop, fop->locks[0].lock->loc.inode,
+                                            &cbk->iatt[0].ia_size));
+
+                cbk->iatt[1].ia_size = cbk->iatt[0].ia_size;
+            }
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.discard != NULL) {
+                QUORUM_CBK(fop->cbks.discard, fop, fop->req_frame, fop, fop->xl,
+                           cbk->op_ret, cbk->op_errno, &cbk->iatt[0],
+                           &cbk->iatt[1], cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_DELAYED_START:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.discard != NULL) {
+                fop->cbks.discard(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                  NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_discard(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_discard_cbk_t func, void *data, fd_t *fd,
+           off_t offset, size_t len, dict_t *xdata)
+{
+    ec_cbk_t callback = {.discard = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(DISCARD) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_DISCARD, 0, target,
+                               fop_flags, ec_wind_discard, ec_manager_discard,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+    fop->offset = offset;
+    fop->size = len;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+    }
+
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL);
+    }
+}
+
+/*********************************************************************
+ *
+ * File Operation : truncate
+ *
+ *********************************************************************/
+
+int32_t
+ec_update_truncate_write(ec_fop_data_t *fop, uintptr_t mask)
+{
+    ec_t *ec = fop->xl->private;
+    uint64_t size = fop->offset * ec->fragments - fop->user_size;
+    return ec_update_write(fop, mask, fop->user_size, size);
+}
+
+int32_t
+ec_truncate_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
+{
+    ec_fop_data_t *fop = cookie;
+    int32_t err;
+
+    fop->parent->good &= fop->good;
+    if (op_ret >= 0) {
+        fd_bind(fd);
+        err = ec_update_truncate_write(fop->parent, fop->answer->mask);
+        if (err != 0) {
+            ec_fop_set_error(fop->parent, -err);
+        }
+    }
+
+    return 0;
+}
+
+int32_t
+ec_truncate_clean(ec_fop_data_t *fop)
+{
+    if (fop->fd == NULL) {
+        fop->fd = fd_create(fop->loc[0].inode, fop->frame->root->pid);
+        if (fop->fd == NULL) {
+            return -ENOMEM;
+        }
+
+        ec_open(fop->frame, fop->xl, fop->answer->mask, fop->minimum,
+                ec_truncate_open_cbk, fop, &fop->loc[0], O_RDWR, fop->fd, NULL);
+
+        return 0;
+    } else {
+        return ec_update_truncate_write(fop, fop->answer->mask);
+    }
+}
+
+int32_t
+ec_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *prestat,
+                struct iatt *poststat, dict_t *xdata)
+{
+    return ec_inode_write_cbk(frame, this, cookie, op_ret, op_errno, prestat,
+                              poststat, xdata);
+}
+
+void
+ec_wind_truncate(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_truncate_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->truncate,
+                      &fop->loc[0], fop->offset, fop->xdata);
+}
+
+int32_t
+ec_manager_truncate(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+    off_t offset_down;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            fop->user_size = fop->offset;
+            ec_adjust_offset_up(fop->xl->private, &fop->offset, _gf_true);
+            fop->frag_range.first = fop->offset;
+            fop->frag_range.last = UINT64_MAX;
+
+            /* Fall through */
+
+        case EC_STATE_LOCK:
+            offset_down = fop->user_size;
+            ec_adjust_offset_down(fop->xl->private, &offset_down, _gf_true);
+
+            if (fop->id == GF_FOP_TRUNCATE) {
+                ec_lock_prepare_inode(
+                    fop, &fop->loc[0],
+                    EC_UPDATE_DATA | EC_UPDATE_META | EC_QUERY_INFO,
+                    offset_down, EC_RANGE_FULL);
+            } else {
+                ec_lock_prepare_fd(
+                    fop, fop->fd,
+                    EC_UPDATE_DATA | EC_UPDATE_META | EC_QUERY_INFO,
+                    offset_down, EC_RANGE_FULL);
+            }
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                int32_t err;
+
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 2, cbk->count);
+
+                /* This shouldn't fail because we have the inode locked. */
+                /* Inode size doesn't need to be updated under locks, because
+                 * conflicting operations won't be in-flight
+                 */
+                GF_ASSERT(ec_get_inode_size(fop, fop->locks[0].lock->loc.inode,
+                                            &cbk->iatt[0].ia_size));
+                cbk->iatt[1].ia_size = fop->user_size;
+                /* This shouldn't fail because we have the inode locked. */
+                GF_ASSERT(ec_set_inode_size(fop, fop->locks[0].lock->loc.inode,
+                                            fop->user_size));
+                if ((cbk->iatt[0].ia_size > cbk->iatt[1].ia_size) &&
+                    (fop->user_size != fop->offset)) {
+                    err = ec_truncate_clean(fop);
+                    if (err != 0) {
+                        ec_cbk_set_error(cbk, -err, _gf_false);
+                    }
+                }
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->id == GF_FOP_TRUNCATE) {
+                if (fop->cbks.truncate != NULL) {
+                    QUORUM_CBK(fop->cbks.truncate, fop, fop->req_frame, fop,
+                               fop->xl, cbk->op_ret, cbk->op_errno,
+                               &cbk->iatt[0], &cbk->iatt[1], cbk->xdata);
+                }
+            } else {
+                if (fop->cbks.ftruncate != NULL) {
+                    QUORUM_CBK(fop->cbks.ftruncate, fop, fop->req_frame, fop,
+                               fop->xl, cbk->op_ret, cbk->op_errno,
+                               &cbk->iatt[0], &cbk->iatt[1], cbk->xdata);
+                }
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->id == GF_FOP_TRUNCATE) {
+                if (fop->cbks.truncate != NULL) {
+                    fop->cbks.truncate(fop->req_frame, fop, fop->xl, -1,
+                                       fop->error, NULL, NULL, NULL);
+                }
+            } else {
+                if (fop->cbks.ftruncate != NULL) {
+                    fop->cbks.ftruncate(fop->req_frame, fop, fop->xl, -1,
+                                        fop->error, NULL, NULL, NULL);
+                }
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_truncate(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_truncate_cbk_t func, void *data, loc_t *loc,
+            off_t offset, dict_t *xdata)
+{
+    ec_cbk_t callback = {.truncate = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(TRUNCATE) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_TRUNCATE, 0, target,
+                               fop_flags, ec_wind_truncate, ec_manager_truncate,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->offset = offset;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: ftruncate */
+
+int32_t
+ec_ftruncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct iatt *prestat,
+                 struct iatt *poststat, dict_t *xdata)
+{
+    return ec_inode_write_cbk(frame, this, cookie, op_ret, op_errno, prestat,
+                              poststat, xdata);
+}
+
+void
+ec_wind_ftruncate(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_ftruncate_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->ftruncate,
+                      fop->fd, fop->offset, fop->xdata);
+}
+
+void
+ec_ftruncate(call_frame_t *frame, xlator_t *this, uintptr_t target,
+             uint32_t fop_flags, fop_ftruncate_cbk_t func, void *data, fd_t *fd,
+             off_t offset, dict_t *xdata)
+{
+    ec_cbk_t callback = {.ftruncate = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FTRUNCATE) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FTRUNCATE, 0, target,
+                               fop_flags, ec_wind_ftruncate,
+                               ec_manager_truncate, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->offset = offset;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: writev */
+static ec_stripe_t *
+ec_allocate_stripe(ec_t *ec, ec_stripe_list_t *stripe_cache)
+{
+    ec_stripe_t *stripe = NULL;
+
+    if (stripe_cache->count >= stripe_cache->max) {
+        GF_ASSERT(!list_empty(&stripe_cache->lru));
+        stripe = list_first_entry(&stripe_cache->lru, ec_stripe_t, lru);
+        list_move_tail(&stripe->lru, &stripe_cache->lru);
+        GF_ATOMIC_INC(ec->stats.stripe_cache.evicts);
+    } else {
+        stripe = GF_MALLOC(sizeof(ec_stripe_t) + ec->stripe_size,
+                           ec_mt_ec_stripe_t);
+        if (stripe != NULL) {
+            stripe_cache->count++;
+            list_add_tail(&stripe->lru, &stripe_cache->lru);
+            GF_ATOMIC_INC(ec->stats.stripe_cache.allocs);
+        } else {
+            GF_ATOMIC_INC(ec->stats.stripe_cache.errors);
+        }
+    }
+
+    return stripe;
+}
+
+static void
+ec_write_stripe_data(ec_t *ec, ec_fop_data_t *fop, ec_stripe_t *stripe)
+{
+    off_t base;
+
+    base = fop->size - ec->stripe_size;
+    memcpy(stripe->data, fop->vector[0].iov_base + base, ec->stripe_size);
+    stripe->frag_offset = fop->frag_range.last - ec->fragment_size;
+}
+
+static void
+ec_add_stripe_in_cache(ec_t *ec, ec_fop_data_t *fop)
+{
+    ec_inode_t *ctx = NULL;
+    ec_stripe_t *stripe = NULL;
+    ec_stripe_list_t *stripe_cache = NULL;
+    gf_boolean_t failed = _gf_true;
+
+    LOCK(&fop->fd->inode->lock);
+
+    ctx = __ec_inode_get(fop->fd->inode, fop->xl);
+    if (ctx == NULL) {
+        goto out;
+    }
+
+    stripe_cache = &ctx->stripe_cache;
+    if (stripe_cache->max > 0) {
+        stripe = ec_allocate_stripe(ec, stripe_cache);
+        if (stripe == NULL) {
+            goto out;
+        }
+
+        ec_write_stripe_data(ec, fop, stripe);
+    }
+
+    failed = _gf_false;
+
+out:
+    UNLOCK(&fop->fd->inode->lock);
+
+    if (failed) {
+        gf_msg(ec->xl->name, GF_LOG_DEBUG, ENOMEM, EC_MSG_FILE_DESC_REF_FAIL,
+               "Failed to create and add stripe in cache");
+    }
+}
+
+int32_t
+ec_writev_merge_tail(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iovec *vector,
+                     int32_t count, struct iatt *stbuf, struct iobref *iobref,
+                     dict_t *xdata)
+{
+    ec_t *ec = this->private;
+    ec_fop_data_t *fop = frame->local;
+    uint64_t size, base, tmp;
+
+    if (op_ret >= 0) {
+        tmp = 0;
+        size = fop->size - fop->user_size - fop->head;
+        base = ec->stripe_size - size;
+        if (op_ret > base) {
+            tmp = min(op_ret - base, size);
+            ec_iov_copy_to(fop->vector[0].iov_base + fop->size - size, vector,
+                           count, base, tmp);
+
+            size -= tmp;
+        }
+
+        if (size > 0) {
+            memset(fop->vector[0].iov_base + fop->size - size, 0, size);
+        }
+
+        if (ec->stripe_cache) {
+            ec_add_stripe_in_cache(ec, fop);
+        }
+    }
+    return 0;
+}
+
+int32_t
+ec_writev_merge_head(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iovec *vector,
+                     int32_t count, struct iatt *stbuf, struct iobref *iobref,
+                     dict_t *xdata)
+{
+    ec_t *ec = this->private;
+    ec_fop_data_t *fop = frame->local;
+    uint64_t size, base;
+
+    if (op_ret >= 0) {
+        size = fop->head;
+        base = 0;
+
+        if (op_ret > 0) {
+            base = min(op_ret, size);
+            ec_iov_copy_to(fop->vector[0].iov_base, vector, count, 0, base);
+
+            size -= base;
+        }
+
+        if (size > 0) {
+            memset(fop->vector[0].iov_base + base, 0, size);
+        }
+
+        size = fop->size - fop->user_size - fop->head;
+        if ((size > 0) && (fop->size == ec->stripe_size)) {
+            ec_writev_merge_tail(frame, cookie, this, op_ret, op_errno, vector,
+                                 count, stbuf, iobref, xdata);
+        }
+    }
+
+    return 0;
+}
+
+static int
+ec_make_internal_fop_xdata(dict_t **xdata)
+{
+    dict_t *dict = NULL;
+
+    if (*xdata)
+        return 0;
+
+    dict = dict_new();
+    if (!dict)
+        goto out;
+
+    if (dict_set_str(dict, GLUSTERFS_INTERNAL_FOP_KEY, "yes"))
+        goto out;
+
+    *xdata = dict;
+    return 0;
+out:
+    if (dict)
+        dict_unref(dict);
+    return -1;
+}
+
+static int32_t
+ec_writev_prepare_buffers(ec_t *ec, ec_fop_data_t *fop)
+{
+    struct iobref *iobref = NULL;
+    struct iovec *iov;
+    void *ptr;
+    int32_t err;
+
+    fop->user_size = iov_length(fop->vector, fop->int32);
+    fop->head = ec_adjust_offset_down(ec, &fop->offset, _gf_false);
+    fop->frag_range.first = fop->offset / ec->fragments;
+    fop->size = fop->user_size + fop->head;
+    ec_adjust_size_up(ec, &fop->size, _gf_false);
+    fop->frag_range.last = fop->frag_range.first + fop->size / ec->fragments;
+
+    if ((fop->int32 != 1) || (fop->head != 0) || (fop->size > fop->user_size) ||
+        !EC_ALIGN_CHECK(fop->vector[0].iov_base, EC_METHOD_WORD_SIZE)) {
+        err = ec_buffer_alloc(ec->xl, fop->size, &iobref, &ptr);
+        if (err != 0) {
+            goto out;
+        }
+
+        ec_iov_copy_to(ptr + fop->head, fop->vector, fop->int32, 0,
+                       fop->user_size);
+
+        fop->vector[0].iov_base = ptr;
+        fop->vector[0].iov_len = fop->size;
+
+        iobref_unref(fop->buffers);
+        fop->buffers = iobref;
+    }
+
+    if (fop->int32 != 2) {
+        iov = GF_MALLOC(VECTORSIZE(2), gf_common_mt_iovec);
+        if (iov == NULL) {
+            err = -ENOMEM;
+
+            goto out;
+        }
+        iov[0].iov_base = fop->vector[0].iov_base;
+        iov[0].iov_len = fop->vector[0].iov_len;
+
+        GF_FREE(fop->vector);
+        fop->vector = iov;
+    }
+
+    fop->vector[1].iov_len = fop->size / ec->fragments;
+    err = ec_buffer_alloc(ec->xl, fop->vector[1].iov_len * ec->nodes,
+                          &fop->buffers, &fop->vector[1].iov_base);
+    if (err != 0) {
+        goto out;
+    }
+
+    err = 0;
+
+out:
+    return err;
+}
+
+static void
+ec_merge_stripe_head_locked(ec_t *ec, ec_fop_data_t *fop, ec_stripe_t *stripe)
+{
+    uint32_t head, size;
+
+    head = fop->head;
+    memcpy(fop->vector[0].iov_base, stripe->data, head);
+
+    size = ec->stripe_size - head;
+    if (size > fop->user_size) {
+        head += fop->user_size;
+        size = ec->stripe_size - head;
+        memcpy(fop->vector[0].iov_base + head, stripe->data + head, size);
+    }
+}
+
+static void
+ec_merge_stripe_tail_locked(ec_t *ec, ec_fop_data_t *fop, ec_stripe_t *stripe)
+{
+    uint32_t head, tail;
+    off_t offset;
+
+    offset = fop->user_size + fop->head;
+    tail = fop->size - offset;
+    head = ec->stripe_size - tail;
+
+    memcpy(fop->vector[0].iov_base + offset, stripe->data + head, tail);
+}
+
+static ec_stripe_t *
+ec_get_stripe_from_cache_locked(ec_t *ec, ec_fop_data_t *fop,
+                                uint64_t frag_offset)
+{
+    ec_inode_t *ctx = NULL;
+    ec_stripe_t *stripe = NULL;
+    ec_stripe_list_t *stripe_cache = NULL;
+
+    ctx = __ec_inode_get(fop->fd->inode, fop->xl);
+    if (ctx == NULL) {
+        GF_ATOMIC_INC(ec->stats.stripe_cache.errors);
+        return NULL;
+    }
+
+    stripe_cache = &ctx->stripe_cache;
+    list_for_each_entry(stripe, &stripe_cache->lru, lru)
+    {
+        if (stripe->frag_offset == frag_offset) {
+            list_move_tail(&stripe->lru, &stripe_cache->lru);
+            GF_ATOMIC_INC(ec->stats.stripe_cache.hits);
+            return stripe;
+        }
+    }
+
+    GF_ATOMIC_INC(ec->stats.stripe_cache.misses);
+
+    return NULL;
+}
+
+static gf_boolean_t
+ec_get_and_merge_stripe(ec_t *ec, ec_fop_data_t *fop, ec_stripe_part_t which)
+{
+    uint64_t frag_offset;
+    ec_stripe_t *stripe = NULL;
+    gf_boolean_t found = _gf_false;
+
+    if (!ec->stripe_cache) {
+        return found;
+    }
+
+    LOCK(&fop->fd->inode->lock);
+    if (which == EC_STRIPE_HEAD) {
+        frag_offset = fop->frag_range.first;
+        stripe = ec_get_stripe_from_cache_locked(ec, fop, frag_offset);
+        if (stripe) {
+            ec_merge_stripe_head_locked(ec, fop, stripe);
+            found = _gf_true;
+        }
+    }
+
+    if (which == EC_STRIPE_TAIL) {
+        frag_offset = fop->frag_range.last - ec->fragment_size;
+        stripe = ec_get_stripe_from_cache_locked(ec, fop, frag_offset);
+        if (stripe) {
+            ec_merge_stripe_tail_locked(ec, fop, stripe);
+            found = _gf_true;
+        }
+    }
+    UNLOCK(&fop->fd->inode->lock);
+
+    return found;
+}
+
+static uintptr_t
+ec_get_lock_good_mask(inode_t *inode, xlator_t *xl)
+{
+    ec_lock_t *lock = NULL;
+    ec_inode_t *ictx = NULL;
+    LOCK(&inode->lock);
+    {
+        ictx = __ec_inode_get(inode, xl);
+        if (ictx)
+            lock = ictx->inode_lock;
+    }
+    UNLOCK(&inode->lock);
+    if (lock)
+        return lock->good_mask;
+    return 0;
+}
+
+void
+ec_writev_start(ec_fop_data_t *fop)
+{
+    ec_t *ec = fop->xl->private;
+    ec_fd_t *ctx;
+    fd_t *fd;
+    dict_t *xdata = NULL;
+    uint64_t tail, current;
+    int32_t err = -ENOMEM;
+    gf_boolean_t found_stripe = _gf_false;
+
+    /* This shouldn't fail because we have the inode locked. */
+    GF_ASSERT(ec_get_inode_size(fop, fop->fd->inode, &current));
+
+    fd = fd_anonymous(fop->fd->inode);
+    if (fd == NULL) {
+        goto failed;
+    }
+
+    fop->frame->root->uid = 0;
+    fop->frame->root->gid = 0;
+
+    ctx = ec_fd_get(fop->fd, fop->xl);
+    if (ctx != NULL) {
+        if ((ctx->flags & O_APPEND) != 0) {
+            /* Appending writes take full locks so size won't change because
+             * of any parallel operations
+             */
+            fop->offset = current;
+        }
+    }
+
+    err = ec_writev_prepare_buffers(ec, fop);
+    if (err != 0) {
+        goto failed_fd;
+    }
+    tail = fop->size - fop->user_size - fop->head;
+    if (fop->head > 0) {
+        if (current > fop->offset) {
+            found_stripe = ec_get_and_merge_stripe(ec, fop, EC_STRIPE_HEAD);
+            if (!found_stripe) {
+                if (ec_make_internal_fop_xdata(&xdata)) {
+                    err = -ENOMEM;
+                    goto failed_xdata;
+                }
+                ec_readv(fop->frame, fop->xl,
+                         ec_get_lock_good_mask(fop->fd->inode, fop->xl),
+                         EC_MINIMUM_MIN, ec_writev_merge_head, NULL, fd,
+                         ec->stripe_size, fop->offset, 0, xdata);
+            }
+        } else {
+            memset(fop->vector[0].iov_base, 0, fop->head);
+            memset(fop->vector[0].iov_base + fop->size - tail, 0, tail);
+            if (ec->stripe_cache && (fop->size <= ec->stripe_size)) {
+                ec_add_stripe_in_cache(ec, fop);
+            }
+        }
+    }
+
+    if ((tail > 0) && ((fop->head == 0) || (fop->size > ec->stripe_size))) {
+        /* Current locking scheme will make sure the 'current' below will
+         * never decrease while the fop is in progress, so the checks will
+         * work as expected
+         */
+        if (current > fop->offset + fop->head + fop->user_size) {
+            found_stripe = ec_get_and_merge_stripe(ec, fop, EC_STRIPE_TAIL);
+            if (!found_stripe) {
+                if (ec_make_internal_fop_xdata(&xdata)) {
+                    err = -ENOMEM;
+                    goto failed_xdata;
+                }
+                ec_readv(fop->frame, fop->xl,
+                         ec_get_lock_good_mask(fop->fd->inode, fop->xl),
+                         EC_MINIMUM_MIN, ec_writev_merge_tail, NULL, fd,
+                         ec->stripe_size,
+                         fop->offset + fop->size - ec->stripe_size, 0, xdata);
+            }
+        } else {
+            memset(fop->vector[0].iov_base + fop->size - tail, 0, tail);
+            if (ec->stripe_cache) {
+                ec_add_stripe_in_cache(ec, fop);
+            }
+        }
+    }
+
+    err = 0;
+
+failed_xdata:
+    if (xdata) {
+        dict_unref(xdata);
+    }
+failed_fd:
+    fd_unref(fd);
+failed:
+    ec_fop_set_error(fop, -err);
+}
+
+int32_t
+ec_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct iatt *prestat, struct iatt *poststat,
+              dict_t *xdata)
+{
+    ec_t *ec = NULL;
+    if (this && this->private) {
+        ec = this->private;
+        if ((op_ret > 0) && ((op_ret % ec->fragment_size) != 0)) {
+            op_ret = -1;
+            op_errno = EIO;
+        }
+    }
+    return ec_inode_write_cbk(frame, this, cookie, op_ret, op_errno, prestat,
+                              poststat, xdata);
+}
+
+void
+ec_wind_writev(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    struct iovec vector[1];
+    size_t size;
+
+    size = fop->vector[1].iov_len;
+
+    vector[0].iov_base = fop->vector[1].iov_base + idx * size;
+    vector[0].iov_len = size;
+
+    STACK_WIND_COOKIE(fop->frame, ec_writev_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->writev, fop->fd,
+                      vector, 1, fop->offset / ec->fragments, fop->uint32,
+                      fop->buffers, fop->xdata);
+}
+
+static void
+ec_writev_encode(ec_fop_data_t *fop)
+{
+    ec_t *ec = fop->xl->private;
+    void *blocks[ec->nodes];
+    uint32_t i;
+
+    blocks[0] = fop->vector[1].iov_base;
+    for (i = 1; i < ec->nodes; i++) {
+        blocks[i] = blocks[i - 1] + fop->vector[1].iov_len;
+    }
+    ec_method_encode(&ec->matrix, fop->vector[0].iov_len,
+                     fop->vector[0].iov_base, blocks);
+}
+
+int32_t
+ec_manager_writev(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+    ec_fd_t *ctx = NULL;
+    ec_t *ec = fop->xl->private;
+    off_t fl_start = 0;
+    uint64_t fl_size = LONG_MAX;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            ctx = ec_fd_get(fop->fd, fop->xl);
+            if (ctx != NULL) {
+                if ((ctx->flags & O_APPEND) == 0) {
+                    off_t user_size = 0;
+                    off_t head = 0;
+
+                    fl_start = fop->offset;
+                    user_size = iov_length(fop->vector, fop->int32);
+                    head = ec_adjust_offset_down(ec, &fl_start, _gf_true);
+                    fl_size = user_size + head;
+                    ec_adjust_size_up(ec, &fl_size, _gf_true);
+                }
+            }
+            ec_lock_prepare_fd(fop, fop->fd,
+                               EC_UPDATE_DATA | EC_UPDATE_META | EC_QUERY_INFO,
+                               fl_start, fl_size);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_writev_start(fop);
+
+            return EC_STATE_DELAYED_START;
+
+        case EC_STATE_DELAYED_START:
+            /* Restore uid, gid if they were changed to do some partial
+             * reads. */
+            fop->frame->root->uid = fop->uid;
+            fop->frame->root->gid = fop->gid;
+
+            ec_writev_encode(fop);
+
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                ec_t *ec = fop->xl->private;
+                uint64_t size;
+
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 2, cbk->count);
+
+                /* This shouldn't fail because we have the inode locked. */
+                LOCK(&fop->fd->inode->lock);
+                {
+                    GF_ASSERT(__ec_get_inode_size(fop, fop->fd->inode,
+                                                  &cbk->iatt[0].ia_size));
+                    cbk->iatt[1].ia_size = cbk->iatt[0].ia_size;
+                    size = fop->offset + fop->head + fop->user_size;
+                    if (size > cbk->iatt[0].ia_size) {
+                        /* Only update inode size if this is a top level fop.
+                         * Otherwise this is an internal write and the top
+                         * level fop should take care of the real inode size.
+                         */
+                        if (fop->parent == NULL) {
+                            /* This shouldn't fail because we have the inode
+                             * locked. */
+                            GF_ASSERT(
+                                __ec_set_inode_size(fop, fop->fd->inode, size));
+                        }
+                        cbk->iatt[1].ia_size = size;
+                    }
+                }
+                UNLOCK(&fop->fd->inode->lock);
+
+                if (fop->error == 0) {
+                    cbk->op_ret *= ec->fragments;
+                    if (cbk->op_ret < fop->head) {
+                        cbk->op_ret = 0;
+                    } else {
+                        cbk->op_ret -= fop->head;
+                    }
+                    if (cbk->op_ret > fop->user_size) {
+                        cbk->op_ret = fop->user_size;
+                    }
+                }
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.writev != NULL) {
+                QUORUM_CBK(fop->cbks.writev, fop, fop->req_frame, fop, fop->xl,
+                           cbk->op_ret, cbk->op_errno, &cbk->iatt[0],
+                           &cbk->iatt[1], cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_DELAYED_START:
+            /* We have failed while doing partial reads. We need to restore
+             * original uid, gid. */
+            fop->frame->root->uid = fop->uid;
+            fop->frame->root->gid = fop->gid;
+
+            /* Fall through */
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.writev != NULL) {
+                fop->cbks.writev(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                 NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_writev(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_writev_cbk_t func, void *data, fd_t *fd,
+          struct iovec *vector, int32_t count, off_t offset, uint32_t flags,
+          struct iobref *iobref, dict_t *xdata)
+{
+    ec_cbk_t callback = {.writev = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(WRITE) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_WRITE, 0, target, fop_flags,
+                               ec_wind_writev, ec_manager_writev, callback,
+                               data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->int32 = count;
+    fop->offset = offset;
+    fop->uint32 = flags;
+
+    fop->use_fd = 1;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (count > 0) {
+        fop->vector = iov_dup(vector, count);
+        if (fop->vector == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a "
+                   "vector list.");
+
+            goto out;
+        }
+        fop->int32 = count;
+    }
+    if (iobref != NULL) {
+        fop->buffers = iobref_ref(iobref);
+        if (fop->buffers == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_BUF_REF_FAIL,
+                   "Failed to reference a "
+                   "buffer.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL);
+    }
+}
diff --git a/xlators/cluster/ec/src/ec-locks.c b/xlators/cluster/ec/src/ec-locks.c
new file mode 100644
index 00000000000..601960d6154
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-locks.c
@@ -0,0 +1,1128 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "ec-helpers.h"
+#include "ec-common.h"
+#include "ec-combine.h"
+#include "ec-fops.h"
+#include "ec-messages.h"
+
+#define EC_LOCK_MODE_NONE 0
+#define EC_LOCK_MODE_INC 1
+#define EC_LOCK_MODE_ALL 2
+
+int32_t
+ec_lock_check(ec_fop_data_t *fop, uintptr_t *mask)
+{
+    ec_t *ec = fop->xl->private;
+    ec_cbk_data_t *ans = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    uintptr_t locked = 0;
+    int32_t good = 0;
+    int32_t eagain = 0;
+    int32_t estale = 0;
+    int32_t error = -1;
+
+    /* There are some errors that we'll handle in an special way while trying
+     * to acquire a lock.
+     *
+     *   EAGAIN:  If it's found during a parallel non-blocking lock request, we
+     *            consider that there's contention on the inode, so we consider
+     *            the acquisition a failure and try again with a sequential
+     *            blocking lock request. This will ensure that we get a lock on
+     *            as many bricks as possible (ignoring EAGAIN here would cause
+     *            unnecessary triggers of self-healing).
+     *
+     *            If it's found during a sequential blocking lock request, it's
+     *            considered an error. Lock will only succeed if there are
+     *            enough other bricks locked.
+     *
+     *   ESTALE:  This can appear during parallel or sequential lock request if
+     *            the inode has just been unlinked. We consider this error is
+     *            not recoverable, but we also don't consider it as fatal. So,
+     *            if it happens during parallel lock, we won't attempt a
+     *            sequential one unless there are EAGAIN errors on other
+     *            bricks (and are enough to form a quorum), but if we reach
+     *            quorum counting the ESTALE bricks, we consider the whole
+     *            result of the operation is ESTALE instead of EIO.
+     */
+
+    list_for_each_entry(ans, &fop->cbk_list, list)
+    {
+        if (ans->op_ret >= 0) {
+            if (locked != 0) {
+                error = EIO;
+            }
+            locked |= ans->mask;
+            good = ans->count;
+            cbk = ans;
+        } else if (ans->op_errno == ESTALE) {
+            estale += ans->count;
+        } else if ((ans->op_errno == EAGAIN) &&
+                   (fop->uint32 != EC_LOCK_MODE_INC)) {
+            eagain += ans->count;
+        }
+    }
+
+    if (error == -1) {
+        /* If we have enough quorum with succeeded and EAGAIN answers, we
+         * ignore for now any ESTALE answer. If there are EAGAIN answers,
+         * we retry with a sequential blocking lock request if needed.
+         * Otherwise we succeed. */
+        if ((good + eagain) >= ec->fragments) {
+            if (eagain == 0) {
+                if (fop->answer == NULL) {
+                    fop->answer = cbk;
+                }
+
+                ec_update_good(fop, locked);
+
+                error = 0;
+            } else {
+                switch (fop->uint32) {
+                    case EC_LOCK_MODE_NONE:
+                        error = EAGAIN;
+                        break;
+                    case EC_LOCK_MODE_ALL:
+                        fop->uint32 = EC_LOCK_MODE_INC;
+                        break;
+                    default:
+                        /* This shouldn't happen because eagain cannot be > 0
+                         * when fop->uint32 is EC_LOCK_MODE_INC. */
+                        error = EIO;
+                        break;
+                }
+            }
+        } else {
+            /* We have been unable to find enough candidates that will be able
+             * to take the lock. If we have quorum on some answer, we return
+             * it. Otherwise we check if ESTALE answers allow us to reach
+             * quorum. If so, we return ESTALE. */
+            if (fop->answer && fop->answer->op_ret < 0) {
+                error = fop->answer->op_errno;
+            } else if ((good + eagain + estale) >= ec->fragments) {
+                error = ESTALE;
+            } else {
+                error = EIO;
+            }
+        }
+    }
+
+    *mask = locked;
+
+    return error;
+}
+
+int32_t
+ec_lock_unlocked(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    if (op_ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, EC_MSG_UNLOCK_FAILED,
+               "Failed to unlock an entry/inode");
+    }
+
+    return 0;
+}
+
+int32_t
+ec_lock_lk_unlocked(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct gf_flock *flock,
+                    dict_t *xdata)
+{
+    if (op_ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, EC_MSG_LK_UNLOCK_FAILED,
+               "Failed to unlock an lk");
+    }
+
+    return 0;
+}
+
+/* FOP: entrylk */
+
+int32_t
+ec_entrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_ENTRYLK, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_entrylk(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_entrylk_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->entrylk,
+                      fop->str[0], &fop->loc[0], fop->str[1], fop->entrylk_cmd,
+                      fop->entrylk_type, fop->xdata);
+}
+
+int32_t
+ec_manager_entrylk(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            if (fop->entrylk_cmd == ENTRYLK_LOCK) {
+                fop->uint32 = EC_LOCK_MODE_ALL;
+                fop->entrylk_cmd = ENTRYLK_LOCK_NB;
+            }
+
+            /* Fall through */
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_PREPARE_ANSWER:
+            if (fop->entrylk_cmd != ENTRYLK_UNLOCK) {
+                uintptr_t mask;
+
+                ec_fop_set_error(fop, ec_lock_check(fop, &mask));
+                if (fop->error != 0) {
+                    if (mask != 0) {
+                        if (fop->id == GF_FOP_ENTRYLK) {
+                            ec_entrylk(
+                                fop->frame, fop->xl, mask, 1, ec_lock_unlocked,
+                                NULL, fop->str[0], &fop->loc[0], fop->str[1],
+                                ENTRYLK_UNLOCK, fop->entrylk_type, fop->xdata);
+                        } else {
+                            ec_fentrylk(fop->frame, fop->xl, mask, 1,
+                                        ec_lock_unlocked, NULL, fop->str[0],
+                                        fop->fd, fop->str[1], ENTRYLK_UNLOCK,
+                                        fop->entrylk_type, fop->xdata);
+                        }
+                    }
+                    if (fop->error < 0) {
+                        fop->error = 0;
+
+                        fop->entrylk_cmd = ENTRYLK_LOCK;
+
+                        ec_dispatch_inc(fop);
+
+                        return EC_STATE_PREPARE_ANSWER;
+                    }
+                }
+            } else {
+                ec_fop_prepare_answer(fop, _gf_true);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->id == GF_FOP_ENTRYLK) {
+                if (fop->cbks.entrylk != NULL) {
+                    fop->cbks.entrylk(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                      cbk->op_errno, cbk->xdata);
+                }
+            } else {
+                if (fop->cbks.fentrylk != NULL) {
+                    fop->cbks.fentrylk(fop->req_frame, fop, fop->xl,
+                                       cbk->op_ret, cbk->op_errno, cbk->xdata);
+                }
+            }
+
+            return EC_STATE_END;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->id == GF_FOP_ENTRYLK) {
+                if (fop->cbks.entrylk != NULL) {
+                    fop->cbks.entrylk(fop->req_frame, fop, fop->xl, -1,
+                                      fop->error, NULL);
+                }
+            } else {
+                if (fop->cbks.fentrylk != NULL) {
+                    fop->cbks.fentrylk(fop->req_frame, fop, fop->xl, -1,
+                                       fop->error, NULL);
+                }
+            }
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_entrylk(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_entrylk_cbk_t func, void *data,
+           const char *volume, loc_t *loc, const char *basename,
+           entrylk_cmd cmd, entrylk_type type, dict_t *xdata)
+{
+    ec_cbk_t callback = {.entrylk = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(ENTRYLK) %p", frame);
+
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_ENTRYLK, 0, target,
+                               fop_flags, ec_wind_entrylk, ec_manager_entrylk,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->entrylk_cmd = cmd;
+    fop->entrylk_type = type;
+
+    if (volume != NULL) {
+        fop->str[0] = gf_strdup(volume);
+        if (fop->str[0] == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a string.");
+
+            goto out;
+        }
+    }
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (basename != NULL) {
+        fop->str[1] = gf_strdup(basename);
+        if (fop->str[1] == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a string.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
+
+/* FOP: fentrylk */
+
+int32_t
+ec_fentrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_FENTRYLK, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_fentrylk(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_fentrylk_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->fentrylk,
+                      fop->str[0], fop->fd, fop->str[1], fop->entrylk_cmd,
+                      fop->entrylk_type, fop->xdata);
+}
+
+void
+ec_fentrylk(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_fentrylk_cbk_t func, void *data,
+            const char *volume, fd_t *fd, const char *basename, entrylk_cmd cmd,
+            entrylk_type type, dict_t *xdata)
+{
+    ec_cbk_t callback = {.fentrylk = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FENTRYLK) %p", frame);
+
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FENTRYLK, 0, target,
+                               fop_flags, ec_wind_fentrylk, ec_manager_entrylk,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->entrylk_cmd = cmd;
+    fop->entrylk_type = type;
+
+    if (volume != NULL) {
+        fop->str[0] = gf_strdup(volume);
+        if (fop->str[0] == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a string.");
+
+            goto out;
+        }
+    }
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (basename != NULL) {
+        fop->str[1] = gf_strdup(basename);
+        if (fop->str[1] == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a string.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
+
+/* FOP: inodelk */
+
+int32_t
+ec_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_INODELK, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_inodelk(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_inodelk_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->inodelk,
+                      fop->str[0], &fop->loc[0], fop->int32, &fop->flock,
+                      fop->xdata);
+}
+
+int32_t
+ec_manager_inodelk(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            fop->flock.l_len += ec_adjust_offset_down(
+                fop->xl->private, &fop->flock.l_start, _gf_true);
+            ec_adjust_offset_up(fop->xl->private, &fop->flock.l_len, _gf_true);
+            if ((fop->int32 == F_SETLKW) && (fop->flock.l_type != F_UNLCK)) {
+                fop->uint32 = EC_LOCK_MODE_ALL;
+                fop->int32 = F_SETLK;
+            }
+
+            /* Fall through */
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_PREPARE_ANSWER:
+            if (fop->flock.l_type != F_UNLCK) {
+                uintptr_t mask;
+
+                ec_fop_set_error(fop, ec_lock_check(fop, &mask));
+                if (fop->error != 0) {
+                    if (mask != 0) {
+                        ec_t *ec = fop->xl->private;
+                        struct gf_flock flock;
+
+                        flock.l_type = F_UNLCK;
+                        flock.l_whence = fop->flock.l_whence;
+                        flock.l_start = fop->flock.l_start * ec->fragments;
+                        flock.l_len = fop->flock.l_len * ec->fragments;
+                        flock.l_pid = 0;
+                        flock.l_owner.len = 0;
+
+                        if (fop->id == GF_FOP_INODELK) {
+                            ec_inodelk(fop->frame, fop->xl,
+                                       &fop->frame->root->lk_owner, mask, 1,
+                                       ec_lock_unlocked, NULL, fop->str[0],
+                                       &fop->loc[0], F_SETLK, &flock,
+                                       fop->xdata);
+                        } else {
+                            ec_finodelk(fop->frame, fop->xl,
+                                        &fop->frame->root->lk_owner, mask, 1,
+                                        ec_lock_unlocked, NULL, fop->str[0],
+                                        fop->fd, F_SETLK, &flock, fop->xdata);
+                        }
+                    }
+                    if (fop->error < 0) {
+                        fop->error = 0;
+
+                        fop->int32 = F_SETLKW;
+
+                        ec_dispatch_inc(fop);
+
+                        return EC_STATE_PREPARE_ANSWER;
+                    }
+                }
+            } else {
+                ec_fop_prepare_answer(fop, _gf_true);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->id == GF_FOP_INODELK) {
+                if (fop->cbks.inodelk != NULL) {
+                    fop->cbks.inodelk(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                      cbk->op_errno, cbk->xdata);
+                }
+            } else {
+                if (fop->cbks.finodelk != NULL) {
+                    fop->cbks.finodelk(fop->req_frame, fop, fop->xl,
+                                       cbk->op_ret, cbk->op_errno, cbk->xdata);
+                }
+            }
+
+            return EC_STATE_END;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->id == GF_FOP_INODELK) {
+                if (fop->cbks.inodelk != NULL) {
+                    fop->cbks.inodelk(fop->req_frame, fop, fop->xl, -1,
+                                      fop->error, NULL);
+                }
+            } else {
+                if (fop->cbks.finodelk != NULL) {
+                    fop->cbks.finodelk(fop->req_frame, fop, fop->xl, -1,
+                                       fop->error, NULL);
+                }
+            }
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_inodelk(call_frame_t *frame, xlator_t *this, gf_lkowner_t *owner,
+           uintptr_t target, uint32_t fop_flags, fop_inodelk_cbk_t func,
+           void *data, const char *volume, loc_t *loc, int32_t cmd,
+           struct gf_flock *flock, dict_t *xdata)
+{
+    ec_cbk_t callback = {.inodelk = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(INODELK) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_INODELK, 0, target,
+                               fop_flags, ec_wind_inodelk, ec_manager_inodelk,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->int32 = cmd;
+    ec_owner_copy(fop->frame, owner);
+
+    if (volume != NULL) {
+        fop->str[0] = gf_strdup(volume);
+        if (fop->str[0] == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a string.");
+
+            goto out;
+        }
+    }
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (flock != NULL) {
+        fop->flock.l_type = flock->l_type;
+        fop->flock.l_whence = flock->l_whence;
+        fop->flock.l_start = flock->l_start;
+        fop->flock.l_len = flock->l_len;
+        fop->flock.l_pid = flock->l_pid;
+        fop->flock.l_owner.len = flock->l_owner.len;
+        if (flock->l_owner.len > 0) {
+            memcpy(fop->flock.l_owner.data, flock->l_owner.data,
+                   flock->l_owner.len);
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
+
+/* FOP: finodelk */
+
+int32_t
+ec_finodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_FINODELK, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_finodelk(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_finodelk_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->finodelk,
+                      fop->str[0], fop->fd, fop->int32, &fop->flock,
+                      fop->xdata);
+}
+
+void
+ec_finodelk(call_frame_t *frame, xlator_t *this, gf_lkowner_t *owner,
+            uintptr_t target, uint32_t fop_flags, fop_finodelk_cbk_t func,
+            void *data, const char *volume, fd_t *fd, int32_t cmd,
+            struct gf_flock *flock, dict_t *xdata)
+{
+    ec_cbk_t callback = {.finodelk = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FINODELK) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FINODELK, 0, target,
+                               fop_flags, ec_wind_finodelk, ec_manager_inodelk,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->int32 = cmd;
+    ec_owner_copy(fop->frame, owner);
+
+    if (volume != NULL) {
+        fop->str[0] = gf_strdup(volume);
+        if (fop->str[0] == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a string.");
+
+            goto out;
+        }
+    }
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (flock != NULL) {
+        fop->flock.l_type = flock->l_type;
+        fop->flock.l_whence = flock->l_whence;
+        fop->flock.l_start = flock->l_start;
+        fop->flock.l_len = flock->l_len;
+        fop->flock.l_pid = flock->l_pid;
+        fop->flock.l_owner.len = flock->l_owner.len;
+        if (flock->l_owner.len > 0) {
+            memcpy(fop->flock.l_owner.data, flock->l_owner.data,
+                   flock->l_owner.len);
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
+
+/* FOP: lk */
+
+int32_t
+ec_combine_lk(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    if (!ec_flock_compare(&dst->flock, &src->flock)) {
+        gf_msg(fop->xl->name, GF_LOG_NOTICE, 0, EC_MSG_LOCK_MISMATCH,
+               "Mismatching lock in "
+               "answers of 'GF_FOP_LK'");
+
+        return 0;
+    }
+
+    return 1;
+}
+
+int32_t
+ec_lk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+          int32_t op_errno, struct gf_flock *flock, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_LK, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            if (flock != NULL) {
+                cbk->flock.l_type = flock->l_type;
+                cbk->flock.l_whence = flock->l_whence;
+                cbk->flock.l_start = flock->l_start;
+                cbk->flock.l_len = flock->l_len;
+                cbk->flock.l_pid = flock->l_pid;
+                cbk->flock.l_owner.len = flock->l_owner.len;
+                if (flock->l_owner.len > 0) {
+                    memcpy(cbk->flock.l_owner.data, flock->l_owner.data,
+                           flock->l_owner.len);
+                }
+            }
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, ec_combine_lk);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_lk(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_lk_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->lk, fop->fd,
+                      fop->int32, &fop->flock, fop->xdata);
+}
+
+int32_t
+ec_manager_lk(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            if ((fop->int32 == F_SETLKW) && (fop->flock.l_type != F_UNLCK)) {
+                fop->uint32 = EC_LOCK_MODE_ALL;
+                fop->int32 = F_SETLK;
+            }
+
+            /* Fall through */
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_PREPARE_ANSWER:
+            if (fop->flock.l_type != F_UNLCK) {
+                uintptr_t mask;
+
+                ec_fop_set_error(fop, ec_lock_check(fop, &mask));
+                if (fop->error != 0) {
+                    if (mask != 0) {
+                        struct gf_flock flock = {0};
+
+                        flock.l_type = F_UNLCK;
+                        flock.l_whence = fop->flock.l_whence;
+                        flock.l_start = fop->flock.l_start;
+                        flock.l_len = fop->flock.l_len;
+                        flock.l_pid = fop->flock.l_pid;
+                        lk_owner_copy(&flock.l_owner, &fop->flock.l_owner);
+
+                        ec_lk(fop->frame, fop->xl, mask, 1, ec_lock_lk_unlocked,
+                              NULL, fop->fd, F_SETLK, &flock, fop->xdata);
+                    }
+
+                    if (fop->error < 0) {
+                        fop->error = 0;
+
+                        fop->int32 = F_SETLKW;
+
+                        ec_dispatch_inc(fop);
+
+                        return EC_STATE_PREPARE_ANSWER;
+                    }
+                }
+            } else {
+                ec_fop_prepare_answer(fop, _gf_true);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.lk != NULL) {
+                fop->cbks.lk(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                             cbk->op_errno, &cbk->flock, cbk->xdata);
+            }
+
+            return EC_STATE_END;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.lk != NULL) {
+                fop->cbks.lk(fop->req_frame, fop, fop->xl, -1, fop->error, NULL,
+                             NULL);
+            }
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_lk(call_frame_t *frame, xlator_t *this, uintptr_t target, uint32_t fop_flags,
+      fop_lk_cbk_t func, void *data, fd_t *fd, int32_t cmd,
+      struct gf_flock *flock, dict_t *xdata)
+{
+    ec_cbk_t callback = {.lk = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(LK) %p", frame);
+
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_LK, 0, target, fop_flags,
+                               ec_wind_lk, ec_manager_lk, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->int32 = cmd;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (flock != NULL) {
+        fop->flock.l_type = flock->l_type;
+        fop->flock.l_whence = flock->l_whence;
+        fop->flock.l_start = flock->l_start;
+        fop->flock.l_len = flock->l_len;
+        fop->flock.l_pid = flock->l_pid;
+        fop->flock.l_owner.len = flock->l_owner.len;
+        if (flock->l_owner.len > 0) {
+            memcpy(fop->flock.l_owner.data, flock->l_owner.data,
+                   flock->l_owner.len);
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
diff --git a/xlators/cluster/ec/src/ec-mem-types.h b/xlators/cluster/ec/src/ec-mem-types.h
new file mode 100644
index 00000000000..3252c4c1c58
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-mem-types.h
@@ -0,0 +1,30 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_MEM_TYPES_H__
+#define __EC_MEM_TYPES_H__
+
+#include <glusterfs/mem-types.h>
+
+enum gf_ec_mem_types_ {
+    ec_mt_ec_t = gf_common_mt_end + 1,
+    ec_mt_xlator_t,
+    ec_mt_ec_inode_t,
+    ec_mt_ec_fd_t,
+    ec_mt_subvol_healer_t,
+    ec_mt_ec_gf_t,
+    ec_mt_ec_code_t,
+    ec_mt_ec_code_builder_t,
+    ec_mt_ec_matrix_t,
+    ec_mt_ec_stripe_t,
+    ec_mt_end
+};
+
+#endif /* __EC_MEM_TYPES_H__ */
diff --git a/xlators/cluster/ec/src/ec-messages.h b/xlators/cluster/ec/src/ec-messages.h
new file mode 100644
index 00000000000..72e98f11286
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-messages.h
@@ -0,0 +1,61 @@
+/*
+  Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _EC_MESSAGES_H_
+#define _EC_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(EC, EC_MSG_INVALID_CONFIG, EC_MSG_HEAL_FAIL,
+           EC_MSG_DICT_COMBINE_FAIL, EC_MSG_STIME_COMBINE_FAIL,
+           EC_MSG_INVALID_DICT_NUMS, EC_MSG_IATT_COMBINE_FAIL,
+           EC_MSG_INVALID_FORMAT, EC_MSG_DICT_GET_FAILED,
+           EC_MSG_UNHANDLED_STATE, EC_MSG_FILE_DESC_REF_FAIL,
+           EC_MSG_LOC_COPY_FAIL, EC_MSG_BUF_REF_FAIL, EC_MSG_DICT_REF_FAIL,
+           EC_MSG_LK_UNLOCK_FAILED, EC_MSG_UNLOCK_FAILED,
+           EC_MSG_LOC_PARENT_INODE_MISSING, EC_MSG_INVALID_LOC_NAME,
+           EC_MSG_NO_MEMORY, EC_MSG_GFID_MISMATCH, EC_MSG_UNSUPPORTED_VERSION,
+           EC_MSG_FD_CREATE_FAIL, EC_MSG_READDIRP_REQ_PREP_FAIL,
+           EC_MSG_LOOKUP_REQ_PREP_FAIL, EC_MSG_INODE_REF_FAIL,
+           EC_MSG_LOOKUP_READAHEAD_FAIL, EC_MSG_FRAME_MISMATCH,
+           EC_MSG_XLATOR_MISMATCH, EC_MSG_VECTOR_MISMATCH, EC_MSG_IATT_MISMATCH,
+           EC_MSG_FD_MISMATCH, EC_MSG_DICT_MISMATCH, EC_MSG_INDEX_DIR_GET_FAIL,
+           EC_MSG_PREOP_LOCK_FAILED, EC_MSG_CHILDS_INSUFFICIENT,
+           EC_MSG_OP_EXEC_UNAVAIL, EC_MSG_UNLOCK_DELAY_FAILED,
+           EC_MSG_SIZE_VERS_UPDATE_FAIL, EC_MSG_INVALID_REQUEST,
+           EC_MSG_INVALID_LOCK_TYPE, EC_MSG_SIZE_VERS_GET_FAIL,
+           EC_MSG_FILE_SIZE_GET_FAIL, EC_MSG_FOP_MISMATCH,
+           EC_MSG_SUBVOL_ID_DICT_SET_FAIL, EC_MSG_SUBVOL_BUILD_FAIL,
+           EC_MSG_XLATOR_INIT_FAIL, EC_MSG_NO_PARENTS, EC_MSG_TIMER_CREATE_FAIL,
+           EC_MSG_TOO_MANY_SUBVOLS, EC_MSG_DATA_UNAVAILABLE,
+           EC_MSG_INODE_REMOVE_FAIL, EC_MSG_INVALID_REDUNDANCY,
+           EC_MSG_XLATOR_PARSE_OPT_FAIL, EC_MSG_OP_FAIL_ON_SUBVOLS,
+           EC_MSG_INVALID_INODE, EC_MSG_LOCK_MISMATCH, EC_MSG_XDATA_MISMATCH,
+           EC_MSG_HEALING_INFO, EC_MSG_HEAL_SUCCESS, EC_MSG_FULL_SWEEP_START,
+           EC_MSG_FULL_SWEEP_STOP, EC_MSG_INVALID_FOP, EC_MSG_EC_UP,
+           EC_MSG_EC_DOWN, EC_MSG_SIZE_XATTR_GET_FAIL,
+           EC_MSG_VER_XATTR_GET_FAIL, EC_MSG_CONFIG_XATTR_GET_FAIL,
+           EC_MSG_CONFIG_XATTR_INVALID, EC_MSG_EXTENSION, EC_MSG_EXTENSION_NONE,
+           EC_MSG_EXTENSION_UNKNOWN, EC_MSG_EXTENSION_UNSUPPORTED,
+           EC_MSG_EXTENSION_FAILED, EC_MSG_NO_GF, EC_MSG_MATRIX_FAILED,
+           EC_MSG_DYN_CREATE_FAILED, EC_MSG_DYN_CODEGEN_FAILED,
+           EC_MSG_THREAD_CLEANUP_FAILED, EC_MSG_FD_BAD);
+
+#endif /* !_EC_MESSAGES_H_ */
diff --git a/xlators/cluster/ec/src/ec-method.c b/xlators/cluster/ec/src/ec-method.c
new file mode 100644
index 00000000000..55faed0b193
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-method.c
@@ -0,0 +1,433 @@
+/*
+  Copyright (c) 2012-2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <string.h>
+#include <inttypes.h>
+
+#include "ec-types.h"
+#include "ec-mem-types.h"
+#include "ec-galois.h"
+#include "ec-code.h"
+#include "ec-method.h"
+#include "ec-helpers.h"
+
+static void
+ec_method_matrix_normal(ec_gf_t *gf, uint32_t *matrix, uint32_t columns,
+                        uint32_t *values, uint32_t count)
+{
+    uint32_t i, j, v, tmp;
+
+    columns--;
+    for (i = 0; i < count; i++) {
+        v = *values++;
+        *matrix++ = tmp = ec_gf_exp(gf, v, columns);
+        for (j = 0; j < columns; j++) {
+            *matrix++ = tmp = ec_gf_div(gf, tmp, v);
+        }
+    }
+}
+
+static void
+ec_method_matrix_inverse(ec_gf_t *gf, uint32_t *matrix, uint32_t *values,
+                         uint32_t count)
+{
+    uint32_t a[count];
+    uint32_t i, j, p, last, tmp;
+
+    last = count - 1;
+    for (i = 0; i < last; i++) {
+        a[i] = 1;
+    }
+    a[i] = values[0];
+    for (i = last; i > 0; i--) {
+        for (j = i - 1; j < last; j++) {
+            a[j] = a[j + 1] ^ ec_gf_mul(gf, values[i], a[j]);
+        }
+        a[j] = ec_gf_mul(gf, values[i], a[j]);
+    }
+    for (i = 0; i < count; i++) {
+        p = a[0];
+        matrix += count;
+        *matrix = tmp = p ^ values[i];
+        for (j = 1; j < last; j++) {
+            matrix += count;
+            *matrix = tmp = a[j] ^ ec_gf_mul(gf, values[i], tmp);
+            p = tmp ^ ec_gf_mul(gf, values[i], p);
+        }
+        for (j = 0; j < last; j++) {
+            *matrix = ec_gf_div(gf, *matrix, p);
+            matrix -= count;
+        }
+        *matrix = ec_gf_div(gf, 1, p);
+        matrix++;
+    }
+}
+
+static void
+ec_method_matrix_init(ec_matrix_list_t *list, ec_matrix_t *matrix,
+                      uintptr_t mask, uint32_t *rows, gf_boolean_t inverse)
+{
+    uint32_t i;
+
+    matrix->refs = 1;
+    matrix->mask = mask;
+    matrix->code = list->code;
+    matrix->columns = list->columns;
+    INIT_LIST_HEAD(&matrix->lru);
+
+    if (inverse) {
+        matrix->rows = list->columns;
+        ec_method_matrix_inverse(matrix->code->gf, matrix->values, rows,
+                                 matrix->rows);
+        for (i = 0; i < matrix->rows; i++) {
+            matrix->row_data[i].values = matrix->values + i * matrix->columns;
+            matrix->row_data[i].func.interleaved = ec_code_build_interleaved(
+                matrix->code, EC_METHOD_WORD_SIZE, matrix->row_data[i].values,
+                matrix->columns);
+        }
+    } else {
+        matrix->rows = list->rows;
+        ec_method_matrix_normal(matrix->code->gf, matrix->values,
+                                matrix->columns, rows, matrix->rows);
+        for (i = 0; i < matrix->rows; i++) {
+            matrix->row_data[i].values = matrix->values + i * matrix->columns;
+            matrix->row_data[i].func.linear = ec_code_build_linear(
+                matrix->code, EC_METHOD_WORD_SIZE, matrix->row_data[i].values,
+                matrix->columns);
+        }
+    }
+}
+
+static void
+ec_method_matrix_release(ec_matrix_t *matrix)
+{
+    uint32_t i;
+
+    for (i = 0; i < matrix->rows; i++) {
+        if (matrix->row_data[i].func.linear != NULL) {
+            ec_code_release(matrix->code, &matrix->row_data[i].func);
+            matrix->row_data[i].func.linear = NULL;
+        }
+    }
+}
+
+static void
+ec_method_matrix_destroy(ec_matrix_list_t *list, ec_matrix_t *matrix)
+{
+    list_del_init(&matrix->lru);
+
+    ec_method_matrix_release(matrix);
+
+    mem_put(matrix);
+
+    list->count--;
+}
+
+static void
+ec_method_matrix_unref(ec_matrix_list_t *list, ec_matrix_t *matrix)
+{
+    if (--matrix->refs == 0) {
+        list_add_tail(&matrix->lru, &list->lru);
+        if (list->count > list->max) {
+            matrix = list_first_entry(&list->lru, ec_matrix_t, lru);
+            ec_method_matrix_destroy(list, matrix);
+        }
+    }
+}
+
+static ec_matrix_t *
+ec_method_matrix_lookup(ec_matrix_list_t *list, uintptr_t mask, uint32_t *pos)
+{
+    ec_matrix_t *matrix;
+    uint32_t i, j, k;
+
+    i = 0;
+    j = list->count;
+    while (i < j) {
+        k = (i + j) >> 1;
+        matrix = list->objects[k];
+        if (matrix->mask == mask) {
+            *pos = k;
+            return matrix;
+        }
+        if (matrix->mask < mask) {
+            i = k + 1;
+        } else {
+            j = k;
+        }
+    }
+    *pos = i;
+
+    return NULL;
+}
+
+static void
+ec_method_matrix_remove(ec_matrix_list_t *list, uintptr_t mask)
+{
+    uint32_t pos;
+
+    if (ec_method_matrix_lookup(list, mask, &pos) != NULL) {
+        list->count--;
+        if (pos < list->count) {
+            memmove(list->objects + pos, list->objects + pos + 1,
+                    sizeof(ec_matrix_t *) * (list->count - pos));
+        }
+    }
+}
+
+static void
+ec_method_matrix_insert(ec_matrix_list_t *list, ec_matrix_t *matrix)
+{
+    uint32_t pos;
+
+    GF_ASSERT(ec_method_matrix_lookup(list, matrix->mask, &pos) == NULL);
+
+    if (pos < list->count) {
+        memmove(list->objects + pos + 1, list->objects + pos,
+                sizeof(ec_matrix_t *) * (list->count - pos));
+    }
+    list->objects[pos] = matrix;
+    list->count++;
+}
+
+static ec_matrix_t *
+ec_method_matrix_get(ec_matrix_list_t *list, uintptr_t mask, uint32_t *rows)
+{
+    ec_matrix_t *matrix;
+    uint32_t pos;
+
+    LOCK(&list->lock);
+
+    matrix = ec_method_matrix_lookup(list, mask, &pos);
+    if (matrix != NULL) {
+        list_del_init(&matrix->lru);
+        matrix->refs++;
+
+        goto out;
+    }
+
+    if ((list->count >= list->max) && !list_empty(&list->lru)) {
+        matrix = list_first_entry(&list->lru, ec_matrix_t, lru);
+        list_del_init(&matrix->lru);
+
+        ec_method_matrix_remove(list, matrix->mask);
+
+        ec_method_matrix_release(matrix);
+    } else {
+        matrix = mem_get0(list->pool);
+        if (matrix == NULL) {
+            matrix = EC_ERR(ENOMEM);
+            goto out;
+        }
+        matrix->values = (uint32_t *)((uintptr_t)matrix + sizeof(ec_matrix_t) +
+                                      sizeof(ec_matrix_row_t) * list->columns);
+    }
+
+    ec_method_matrix_init(list, matrix, mask, rows, _gf_true);
+
+    if (list->count < list->max) {
+        ec_method_matrix_insert(list, matrix);
+    } else {
+        matrix->mask = 0;
+    }
+
+out:
+    UNLOCK(&list->lock);
+
+    return matrix;
+}
+
+static void
+ec_method_matrix_put(ec_matrix_list_t *list, ec_matrix_t *matrix)
+{
+    LOCK(&list->lock);
+
+    ec_method_matrix_unref(list, matrix);
+
+    UNLOCK(&list->lock);
+}
+
+static int32_t
+ec_method_setup(xlator_t *xl, ec_matrix_list_t *list, const char *gen)
+{
+    ec_matrix_t *matrix;
+    uint32_t values[list->rows];
+    uint32_t i;
+    int32_t err;
+
+    matrix = GF_MALLOC(sizeof(ec_matrix_t) +
+                           sizeof(ec_matrix_row_t) * list->rows +
+                           sizeof(uint32_t) * list->columns * list->rows,
+                       ec_mt_ec_matrix_t);
+    if (matrix == NULL) {
+        err = -ENOMEM;
+        goto failed;
+    }
+    memset(matrix, 0, sizeof(ec_matrix_t));
+    matrix->values = (uint32_t *)((uintptr_t)matrix + sizeof(ec_matrix_t) +
+                                  sizeof(ec_matrix_row_t) * list->rows);
+
+    list->code = ec_code_create(list->gf, ec_code_detect(xl, gen));
+    if (EC_IS_ERR(list->code)) {
+        err = EC_GET_ERR(list->code);
+        list->code = NULL;
+        goto failed_matrix;
+    }
+
+    for (i = 0; i < list->rows; i++) {
+        values[i] = i + 1;
+    }
+    ec_method_matrix_init(list, matrix, 0, values, _gf_false);
+
+    list->encode = matrix;
+
+    return 0;
+
+failed_matrix:
+    GF_FREE(matrix);
+failed:
+    return err;
+}
+
+int32_t
+ec_method_init(xlator_t *xl, ec_matrix_list_t *list, uint32_t columns,
+               uint32_t rows, uint32_t max, const char *gen)
+{
+    list->columns = columns;
+    list->rows = rows;
+    list->max = max;
+    list->stripe = EC_METHOD_CHUNK_SIZE * list->columns;
+    INIT_LIST_HEAD(&list->lru);
+    int32_t err;
+
+    list->pool = mem_pool_new_fn(xl->ctx,
+                                 sizeof(ec_matrix_t) +
+                                     sizeof(ec_matrix_row_t) * columns +
+                                     sizeof(uint32_t) * columns * columns,
+                                 128, "ec_matrix_t");
+    if (list->pool == NULL) {
+        err = -ENOMEM;
+        goto failed;
+    }
+
+    list->objects = GF_MALLOC(sizeof(ec_matrix_t *) * max, ec_mt_ec_matrix_t);
+    if (list->objects == NULL) {
+        err = -ENOMEM;
+        goto failed_pool;
+    }
+
+    list->gf = ec_gf_prepare(EC_GF_BITS, EC_GF_MOD);
+    if (EC_IS_ERR(list->gf)) {
+        err = EC_GET_ERR(list->gf);
+        goto failed_objects;
+    }
+
+    err = ec_method_setup(xl, list, gen);
+    if (err != 0) {
+        goto failed_gf;
+    }
+
+    LOCK_INIT(&list->lock);
+
+    return 0;
+
+failed_gf:
+    ec_gf_destroy(list->gf);
+failed_objects:
+    GF_FREE(list->objects);
+failed_pool:
+    mem_pool_destroy(list->pool);
+failed:
+    list->pool = NULL;
+    list->objects = NULL;
+    list->gf = NULL;
+
+    return err;
+}
+
+void
+ec_method_fini(ec_matrix_list_t *list)
+{
+    ec_matrix_t *matrix;
+
+    if (list->encode == NULL) {
+        return;
+    }
+
+    while (!list_empty(&list->lru)) {
+        matrix = list_first_entry(&list->lru, ec_matrix_t, lru);
+        ec_method_matrix_destroy(list, matrix);
+    }
+
+    GF_ASSERT(list->count == 0);
+
+    if (list->pool) /*Init was successful*/
+        LOCK_DESTROY(&list->lock);
+
+    ec_method_matrix_release(list->encode);
+    GF_FREE(list->encode);
+
+    ec_code_destroy(list->code);
+    ec_gf_destroy(list->gf);
+    GF_FREE(list->objects);
+
+    if (list->pool)
+        mem_pool_destroy(list->pool);
+}
+
+int32_t
+ec_method_update(xlator_t *xl, ec_matrix_list_t *list, const char *gen)
+{
+    /* TODO: Allow changing code generator */
+
+    return 0;
+}
+
+void
+ec_method_encode(ec_matrix_list_t *list, uint64_t size, void *in, void **out)
+{
+    ec_matrix_t *matrix;
+    uint64_t pos;
+    uint32_t i;
+
+    matrix = list->encode;
+    for (pos = 0; pos < size; pos += list->stripe) {
+        for (i = 0; i < matrix->rows; i++) {
+            matrix->row_data[i].func.linear(
+                out[i], in, pos, matrix->row_data[i].values, list->columns);
+            out[i] += EC_METHOD_CHUNK_SIZE;
+        }
+    }
+}
+
+int32_t
+ec_method_decode(ec_matrix_list_t *list, uint64_t size, uintptr_t mask,
+                 uint32_t *rows, void **in, void *out)
+{
+    ec_matrix_t *matrix;
+    uint64_t pos;
+    uint32_t i;
+
+    matrix = ec_method_matrix_get(list, mask, rows);
+    if (EC_IS_ERR(matrix)) {
+        return EC_GET_ERR(matrix);
+    }
+    for (pos = 0; pos < size; pos += EC_METHOD_CHUNK_SIZE) {
+        for (i = 0; i < matrix->rows; i++) {
+            matrix->row_data[i].func.interleaved(
+                out, in, pos, matrix->row_data[i].values, list->columns);
+            out += EC_METHOD_CHUNK_SIZE;
+        }
+    }
+
+    ec_method_matrix_put(list, matrix);
+
+    return 0;
+}
diff --git a/xlators/cluster/ec/src/ec-method.h b/xlators/cluster/ec/src/ec-method.h
new file mode 100644
index 00000000000..f91233b2f88
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-method.h
@@ -0,0 +1,48 @@
+/*
+  Copyright (c) 2012-2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_METHOD_H__
+#define __EC_METHOD_H__
+
+#include "ec-types.h"
+#include "ec-galois.h"
+
+#define EC_GF_BITS 8
+#define EC_GF_MOD 0x11D
+
+#define EC_GF_SIZE (1 << EC_GF_BITS)
+
+/* Determines the maximum size of the matrix used to encode/decode data */
+#define EC_METHOD_MAX_FRAGMENTS 16
+/* Determines the maximum number of usable elements in the Galois Field */
+#define EC_METHOD_MAX_NODES (EC_GF_SIZE - 1)
+
+#define EC_METHOD_WORD_SIZE 64
+
+#define EC_METHOD_CHUNK_SIZE (EC_METHOD_WORD_SIZE * EC_GF_BITS)
+
+int32_t
+ec_method_init(xlator_t *xl, ec_matrix_list_t *list, uint32_t columns,
+               uint32_t rows, uint32_t max, const char *gen);
+
+void
+ec_method_fini(ec_matrix_list_t *list);
+
+int32_t
+ec_method_update(xlator_t *xl, ec_matrix_list_t *list, const char *gen);
+
+void
+ec_method_encode(ec_matrix_list_t *list, uint64_t size, void *in, void **out);
+
+int32_t
+ec_method_decode(ec_matrix_list_t *list, uint64_t size, uintptr_t mask,
+                 uint32_t *rows, void **in, void *out);
+
+#endif /* __EC_METHOD_H__ */
diff --git a/xlators/cluster/ec/src/ec-types.h b/xlators/cluster/ec/src/ec-types.h
new file mode 100644
index 00000000000..de9b89bb2c9
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-types.h
@@ -0,0 +1,690 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_TYPES_H__
+#define __EC_TYPES_H__
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/timer.h>
+#include "libxlator.h"
+#include <glusterfs/atomic.h>
+
+#define EC_GF_MAX_REGS 16
+
+enum _ec_heal_need;
+typedef enum _ec_heal_need ec_heal_need_t;
+
+enum _ec_stripe_part;
+typedef enum _ec_stripe_part ec_stripe_part_t;
+
+enum _ec_read_policy;
+typedef enum _ec_read_policy ec_read_policy_t;
+
+struct _ec_config;
+typedef struct _ec_config ec_config_t;
+
+struct _ec_fd;
+typedef struct _ec_fd ec_fd_t;
+
+struct _ec_fragment_range;
+typedef struct _ec_fragment_range ec_fragment_range_t;
+
+struct _ec_inode;
+typedef struct _ec_inode ec_inode_t;
+
+union _ec_cbk;
+typedef union _ec_cbk ec_cbk_t;
+
+struct _ec_lock;
+typedef struct _ec_lock ec_lock_t;
+
+struct _ec_lock_link;
+typedef struct _ec_lock_link ec_lock_link_t;
+
+struct _ec_fop_data;
+typedef struct _ec_fop_data ec_fop_data_t;
+
+struct _ec_cbk_data;
+typedef struct _ec_cbk_data ec_cbk_data_t;
+
+enum _ec_gf_opcode;
+typedef enum _ec_gf_opcode ec_gf_opcode_t;
+
+struct _ec_gf_op;
+typedef struct _ec_gf_op ec_gf_op_t;
+
+struct _ec_gf_mul;
+typedef struct _ec_gf_mul ec_gf_mul_t;
+
+struct _ec_gf;
+typedef struct _ec_gf ec_gf_t;
+
+struct _ec_code_gen;
+typedef struct _ec_code_gen ec_code_gen_t;
+
+struct _ec_code;
+typedef struct _ec_code ec_code_t;
+
+struct _ec_code_arg;
+typedef struct _ec_code_arg ec_code_arg_t;
+
+struct _ec_code_op;
+typedef struct _ec_code_op ec_code_op_t;
+
+struct _ec_code_builder;
+typedef struct _ec_code_builder ec_code_builder_t;
+
+struct _ec_code_chunk;
+typedef struct _ec_code_chunk ec_code_chunk_t;
+
+struct _ec_stripe;
+typedef struct _ec_stripe ec_stripe_t;
+
+struct _ec_stripe_list;
+typedef struct _ec_stripe_list ec_stripe_list_t;
+
+struct _ec_code_space;
+typedef struct _ec_code_space ec_code_space_t;
+
+typedef void (*ec_code_func_linear_t)(void *dst, void *src, uint64_t offset,
+                                      uint32_t *values, uint32_t count);
+
+typedef void (*ec_code_func_interleaved_t)(void *dst, void **src,
+                                           uint64_t offset, uint32_t *values,
+                                           uint32_t count);
+
+union _ec_code_func;
+typedef union _ec_code_func ec_code_func_t;
+
+struct _ec_matrix_row;
+typedef struct _ec_matrix_row ec_matrix_row_t;
+
+struct _ec_matrix;
+typedef struct _ec_matrix ec_matrix_t;
+
+struct _ec_matrix_list;
+typedef struct _ec_matrix_list ec_matrix_list_t;
+
+struct _ec_heal;
+typedef struct _ec_heal ec_heal_t;
+
+struct _ec_self_heald;
+typedef struct _ec_self_heald ec_self_heald_t;
+
+struct _ec_statistics;
+typedef struct _ec_statistics ec_statistics_t;
+
+struct _ec;
+typedef struct _ec ec_t;
+
+typedef void (*ec_wind_f)(ec_t *, ec_fop_data_t *, int32_t);
+typedef int32_t (*ec_handler_f)(ec_fop_data_t *, int32_t);
+typedef void (*ec_resume_f)(ec_fop_data_t *, int32_t);
+
+enum _ec_read_policy { EC_ROUND_ROBIN, EC_GFID_HASH, EC_READ_POLICY_MAX };
+
+enum _ec_heal_need {
+    EC_HEAL_NONEED,
+    EC_HEAL_MAYBE,
+    EC_HEAL_MUST,
+    EC_HEAL_PURGE_INDEX
+};
+
+enum _ec_stripe_part { EC_STRIPE_HEAD, EC_STRIPE_TAIL };
+
+/* Enumartions to indicate FD status. */
+typedef enum { EC_FD_NOT_OPENED, EC_FD_OPENED, EC_FD_OPENING } ec_fd_status_t;
+
+struct _ec_config {
+    uint32_t version;
+    uint8_t algorithm;
+    uint8_t gf_word_size;
+    uint8_t bricks;
+    uint8_t redundancy;
+    uint32_t chunk_size;
+};
+
+struct _ec_fd {
+    loc_t loc;
+    uintptr_t open;
+    int32_t flags;
+    uint64_t bad_version;
+    ec_fd_status_t fd_status[0];
+};
+
+struct _ec_stripe {
+    struct list_head lru; /* LRU list member */
+    uint64_t frag_offset; /* Fragment offset of this stripe */
+    char data[];          /* Contents of the stripe */
+};
+
+struct _ec_stripe_list {
+    struct list_head lru;
+    uint32_t count;
+    uint32_t max;
+};
+
+struct _ec_inode {
+    ec_lock_t *inode_lock;
+    gf_boolean_t have_info;
+    gf_boolean_t have_config;
+    gf_boolean_t have_version;
+    gf_boolean_t have_size;
+    int32_t heal_count;
+    ec_config_t config;
+    uint64_t pre_version[2];
+    uint64_t post_version[2];
+    uint64_t pre_size;
+    uint64_t post_size;
+    uint64_t dirty[2];
+    struct list_head heal;
+    ec_stripe_list_t stripe_cache;
+    uint64_t bad_version;
+};
+
+typedef int32_t (*fop_heal_cbk_t)(call_frame_t *, void *, xlator_t *, int32_t,
+                                  int32_t, uintptr_t, uintptr_t, uintptr_t,
+                                  uint32_t, dict_t *);
+typedef int32_t (*fop_fheal_cbk_t)(call_frame_t *, void *, xlator_t *, int32_t,
+                                   int32_t, uintptr_t, uintptr_t, uintptr_t,
+                                   uint32_t, dict_t *);
+
+union _ec_cbk {
+    fop_access_cbk_t access;
+    fop_create_cbk_t create;
+    fop_discard_cbk_t discard;
+    fop_entrylk_cbk_t entrylk;
+    fop_fentrylk_cbk_t fentrylk;
+    fop_fallocate_cbk_t fallocate;
+    fop_flush_cbk_t flush;
+    fop_fsync_cbk_t fsync;
+    fop_fsyncdir_cbk_t fsyncdir;
+    fop_getxattr_cbk_t getxattr;
+    fop_fgetxattr_cbk_t fgetxattr;
+    fop_heal_cbk_t heal;
+    fop_fheal_cbk_t fheal;
+    fop_inodelk_cbk_t inodelk;
+    fop_finodelk_cbk_t finodelk;
+    fop_link_cbk_t link;
+    fop_lk_cbk_t lk;
+    fop_lookup_cbk_t lookup;
+    fop_mkdir_cbk_t mkdir;
+    fop_mknod_cbk_t mknod;
+    fop_open_cbk_t open;
+    fop_opendir_cbk_t opendir;
+    fop_readdir_cbk_t readdir;
+    fop_readdirp_cbk_t readdirp;
+    fop_readlink_cbk_t readlink;
+    fop_readv_cbk_t readv;
+    fop_removexattr_cbk_t removexattr;
+    fop_fremovexattr_cbk_t fremovexattr;
+    fop_rename_cbk_t rename;
+    fop_rmdir_cbk_t rmdir;
+    fop_setattr_cbk_t setattr;
+    fop_fsetattr_cbk_t fsetattr;
+    fop_setxattr_cbk_t setxattr;
+    fop_fsetxattr_cbk_t fsetxattr;
+    fop_stat_cbk_t stat;
+    fop_fstat_cbk_t fstat;
+    fop_statfs_cbk_t statfs;
+    fop_symlink_cbk_t symlink;
+    fop_truncate_cbk_t truncate;
+    fop_ftruncate_cbk_t ftruncate;
+    fop_unlink_cbk_t unlink;
+    fop_writev_cbk_t writev;
+    fop_xattrop_cbk_t xattrop;
+    fop_fxattrop_cbk_t fxattrop;
+    fop_zerofill_cbk_t zerofill;
+    fop_seek_cbk_t seek;
+    fop_ipc_cbk_t ipc;
+};
+
+struct _ec_lock {
+    ec_inode_t *ctx;
+    gf_timer_t *timer;
+
+    /* List of owners of this lock. All fops added to this list are running
+     * concurrently. */
+    struct list_head owners;
+
+    /* List of fops waiting to be an owner of the lock. Fops are added to this
+     * list when the current owner has an incompatible access (conflicting lock)
+     * or the lock is not acquired yet. */
+    struct list_head waiting;
+
+    /* List of fops that will wait until the next unlock/lock cycle. This
+     * happens when the currently acquired lock is decided to be released as
+     * soon as possible. In this case, all frozen fops will be continued only
+     * after the lock is reacquired. */
+    struct list_head frozen;
+
+    uintptr_t mask;
+    uintptr_t good_mask;
+    uintptr_t healing;
+    uint32_t refs_owners;   /* Refs for fops owning the lock */
+    uint32_t refs_pending;  /* Refs assigned to fops being prepared */
+    uint32_t waiting_flags; /*Track xattrop/dirty marking*/
+    gf_boolean_t acquired;
+    gf_boolean_t contention;
+    gf_boolean_t unlock_now;
+    gf_boolean_t release;
+    gf_boolean_t query;
+    fd_t *fd;
+    loc_t loc;
+    union {
+        entrylk_type type;
+        struct gf_flock flock;
+    };
+};
+
+struct _ec_lock_link {
+    ec_lock_t *lock;
+    ec_fop_data_t *fop;
+    struct list_head owner_list;
+    struct list_head wait_list;
+    gf_boolean_t update[2];
+    gf_boolean_t dirty[2];
+    gf_boolean_t optimistic_changelog;
+    loc_t *base;
+    uint64_t size;
+    uint32_t waiting_flags;
+    off_t fl_start;
+    off_t fl_end;
+};
+
+/* This structure keeps a range of fragment offsets affected by a fop. Since
+ * real file offsets can be difficult to handle correctly because of overflows,
+ * we use the 'scaled' offset, which corresponds to the offset of the fragment
+ * seen by the bricks, which is always smaller and cannot overflow. */
+struct _ec_fragment_range {
+    uint64_t first; /* Address of the first affected fragment as seen by the
+                       bricks (offset on brick) */
+    uint64_t last;  /* Address of the first non affected fragment as seen by
+                       the bricks (offset on brick) */
+};
+
+/* EC xlator data structure to collect all the data required to perform
+ * the file operation.*/
+struct _ec_fop_data {
+    int32_t id; /* ID of the file operation */
+    int32_t refs;
+    int32_t state;
+    uint32_t minimum; /* Minimum number of successful
+                         operation required to conclude a
+                         fop as successful */
+    int32_t expected;
+    int32_t winds;
+    int32_t jobs;
+    int32_t error;
+    ec_fop_data_t *parent;
+    xlator_t *xl;                  /* points to EC xlator */
+    call_frame_t *req_frame;       /* frame of the calling xlator */
+    call_frame_t *frame;           /* frame used by this fop */
+    struct list_head cbk_list;     /* sorted list of groups of answers */
+    struct list_head answer_list;  /* list of answers */
+    struct list_head pending_list; /* member of ec_t.pending_fops */
+    ec_cbk_data_t *answer;         /* accepted answer */
+    int32_t lock_count;
+    int32_t locked;
+    gf_lock_t lock;
+    ec_lock_link_t locks[2];
+    int32_t first_lock;
+
+    uint32_t fop_flags; /* Flags passed by the caller. */
+    uint32_t flags;     /* Internal flags. */
+    uint32_t first;
+    uintptr_t mask;
+    uintptr_t healing; /*Dispatch is done but call is successful only
+                         if fop->minimum number of subvolumes succeed
+                         which are not healing*/
+    uintptr_t remaining;
+    uintptr_t received; /* Mask of responses */
+    uintptr_t good;
+
+    uid_t uid;
+    gid_t gid;
+
+    ec_wind_f wind;       /* Function to wind to */
+    ec_handler_f handler; /* FOP manager function */
+    ec_resume_f resume;
+    ec_cbk_t cbks; /* Callback function for this FOP */
+    void *data;
+    ec_heal_t *heal;
+    struct list_head healer;
+
+    uint64_t user_size;
+    uint32_t head;
+
+    int32_t use_fd; /* Indicates whether this FOP uses FD or
+                       not */
+
+    dict_t *xdata;
+    dict_t *dict;
+    int32_t int32;
+    uint32_t uint32;
+    uint64_t size;
+    off_t offset;
+    mode_t mode[2];
+    entrylk_cmd entrylk_cmd;
+    entrylk_type entrylk_type;
+    gf_xattrop_flags_t xattrop_flags;
+    dev_t dev;
+    inode_t *inode;
+    fd_t *fd; /* FD of the file on which FOP is
+                 being carried upon */
+    struct iatt iatt;
+    char *str[2];
+    loc_t loc[2]; /* Holds the location details for
+                     the file */
+    struct gf_flock flock;
+    struct iovec *vector;
+    struct iobref *buffers;
+    gf_seek_what_t seek;
+    ec_fragment_range_t frag_range; /* This will hold the range of stripes
+                                        affected by the fop. */
+    char *errstr;                   /*String of fop name, path and gfid
+                                     to be used in gf_msg. */
+};
+
+struct _ec_cbk_data {
+    struct list_head list;        /* item in the sorted list of groups */
+    struct list_head answer_list; /* item in the list of answers */
+    ec_fop_data_t *fop;
+    ec_cbk_data_t *next; /* next answer in the same group */
+    uint32_t idx;
+    int32_t op_ret;
+    int32_t op_errno;
+    int32_t count;
+    uintptr_t mask;
+
+    dict_t *xdata;
+    dict_t *dict;
+    int32_t int32;
+    uintptr_t uintptr[3];
+    uint64_t size;
+    uint64_t version[2];
+    inode_t *inode;
+    fd_t *fd;
+    struct statvfs statvfs;
+    struct iatt iatt[5];
+    struct gf_flock flock;
+    struct iovec *vector;
+    struct iobref *buffers;
+    char *str;
+    gf_dirent_t entries;
+    off_t offset;
+    gf_seek_what_t what;
+};
+
+enum _ec_gf_opcode {
+    EC_GF_OP_LOAD,
+    EC_GF_OP_STORE,
+    EC_GF_OP_COPY,
+    EC_GF_OP_XOR2,
+    EC_GF_OP_XOR3,
+    EC_GF_OP_XORM,
+    EC_GF_OP_END
+};
+
+struct _ec_gf_op {
+    ec_gf_opcode_t op;
+    uint32_t arg1;
+    uint32_t arg2;
+    uint32_t arg3;
+};
+
+struct _ec_gf_mul {
+    uint32_t regs;
+    uint32_t map[EC_GF_MAX_REGS];
+    ec_gf_op_t *ops;
+};
+
+struct _ec_gf {
+    uint32_t bits;
+    uint32_t size;
+    uint32_t mod;
+    uint32_t min_ops;
+    uint32_t max_ops;
+    uint32_t avg_ops;
+    uint32_t *log;
+    uint32_t *pow;
+    ec_gf_mul_t **table;
+};
+
+struct _ec_code_gen {
+    char *name;
+    char **flags;
+    uint32_t width;
+
+    void (*prolog)(ec_code_builder_t *builder);
+    void (*epilog)(ec_code_builder_t *builder);
+    void (*load)(ec_code_builder_t *builder, uint32_t reg, uint32_t offset,
+                 uint32_t bit);
+    void (*store)(ec_code_builder_t *builder, uint32_t reg, uint32_t bit);
+    void (*copy)(ec_code_builder_t *builder, uint32_t dst, uint32_t src);
+    void (*xor2)(ec_code_builder_t *builder, uint32_t dst, uint32_t src);
+    void (*xor3)(ec_code_builder_t *builder, uint32_t dst, uint32_t src1,
+                 uint32_t src2);
+    void (*xorm)(ec_code_builder_t *builder, uint32_t dst, uint32_t offset,
+                 uint32_t bit);
+};
+
+struct _ec_code {
+    gf_lock_t lock;
+    struct list_head spaces;
+    ec_gf_t *gf;
+    ec_code_gen_t *gen;
+};
+
+struct _ec_code_arg {
+    uint32_t value;
+};
+
+struct _ec_code_op {
+    ec_gf_opcode_t op;
+    ec_code_arg_t arg1;
+    ec_code_arg_t arg2;
+    ec_code_arg_t arg3;
+};
+
+struct _ec_code_builder {
+    ec_code_t *code;
+    uint64_t address;
+    uint8_t *data;
+    uint32_t size;
+    int32_t error;
+    uint32_t regs;
+    uint32_t bits;
+    uint32_t width;
+    uint32_t count;
+    uint32_t base;
+    uint32_t map[EC_GF_MAX_REGS];
+    gf_boolean_t linear;
+    uint64_t loop;
+    ec_code_op_t ops[0];
+};
+
+struct _ec_code_chunk {
+    struct list_head list;
+    size_t size;
+    ec_code_space_t *space;
+};
+
+struct _ec_code_space {
+    struct list_head list;
+    struct list_head chunks;
+    ec_code_t *code;
+    void *exec;
+    size_t size;
+};
+
+union _ec_code_func {
+    ec_code_func_linear_t linear;
+    ec_code_func_interleaved_t interleaved;
+};
+
+struct _ec_matrix_row {
+    ec_code_func_t func;
+    uint32_t *values;
+};
+
+struct _ec_matrix {
+    struct list_head lru;
+    uint32_t refs;
+    uint32_t columns;
+    uint32_t rows;
+    uintptr_t mask;
+    ec_code_t *code;
+    uint32_t *values;
+    ec_matrix_row_t row_data[0];
+};
+
+struct _ec_matrix_list {
+    struct list_head lru;
+    gf_lock_t lock;
+    uint32_t columns;
+    uint32_t rows;
+    uint32_t max;
+    uint32_t count;
+    uint32_t stripe;
+    struct mem_pool *pool;
+    ec_gf_t *gf;
+    ec_code_t *code;
+    ec_matrix_t *encode;
+    ec_matrix_t **objects;
+};
+
+struct _ec_heal {
+    struct list_head list;
+    gf_lock_t lock;
+    xlator_t *xl;
+    ec_fop_data_t *fop;
+    void *data;
+    ec_fop_data_t *lookup;
+    loc_t loc;
+    struct iatt iatt;
+    char *symlink;
+    fd_t *fd;
+    int32_t partial;
+    int32_t done;
+    int32_t error;
+    gf_boolean_t nameheal;
+    uintptr_t available;
+    uintptr_t good;
+    uintptr_t bad;
+    uintptr_t open;
+    uintptr_t fixed;
+    uint64_t offset;
+    uint64_t size;
+    uint64_t total_size;
+    uint64_t version[2];
+    uint64_t raw_size;
+};
+
+struct subvol_healer {
+    xlator_t *this;
+    int subvol;
+    gf_boolean_t running;
+    gf_boolean_t rerun;
+    pthread_mutex_t mutex;
+    pthread_cond_t cond;
+    pthread_t thread;
+};
+
+struct _ec_self_heald {
+    gf_boolean_t iamshd;
+    gf_boolean_t enabled;
+    int timeout;
+    uint32_t max_threads;
+    uint32_t wait_qlength;
+    struct subvol_healer *index_healers;
+    struct subvol_healer *full_healers;
+};
+
+struct _ec_statistics {
+    struct {
+        gf_atomic_t hits;    /* Cache hits. */
+        gf_atomic_t misses;  /* Cache misses. */
+        gf_atomic_t updates; /* Number of times an existing stripe has
+                                been updated with new content. */
+        gf_atomic_t invals;  /* Number of times an existing stripe has
+                                been invalidated because of truncates
+                                or discards. */
+        gf_atomic_t evicts;  /* Number of times that an existing entry
+                                has been evicted to make room for newer
+                                entries. */
+        gf_atomic_t allocs;  /* Number of memory allocations made to
+                                store stripes. */
+        gf_atomic_t errors;  /* Number of errors that have caused extra
+                                requests. (Basically memory allocation
+                                errors). */
+    } stripe_cache;
+    struct {
+        gf_atomic_t attempted; /*Number of heals attempted on
+                                files/directories*/
+        gf_atomic_t completed; /*Number of heals complted on files/directories*/
+    } shd;
+};
+
+struct _ec {
+    xlator_t *xl;
+    int32_t healers;
+    int32_t heal_waiters;
+    int32_t nodes; /* Total number of bricks(n) */
+    int32_t bits_for_nodes;
+    int32_t fragments;      /* Data bricks(k) */
+    int32_t redundancy;     /* Redundant bricks(m) */
+    uint32_t fragment_size; /* Size of fragment/chunk on a
+                               brick. */
+    uint32_t stripe_size;   /* (fragment_size * fragments)
+                               maximum size of user data
+                               stored in one stripe. */
+    int32_t up;             /* Represents whether EC volume is
+                               up or not. */
+    uint32_t idx;
+    uint32_t xl_up_count;     /* Number of UP bricks. */
+    uintptr_t xl_up;          /* Bit flag representing UP
+                                 bricks */
+    uint32_t xl_notify_count; /* Number of notifications. */
+    uintptr_t xl_notify;      /* Bit flag representing
+                                 notification for bricks. */
+    uintptr_t node_mask;
+    uintptr_t read_mask;         /*Stores user defined read-mask*/
+    gf_atomic_t async_fop_count; /* Number of on going asynchronous fops. */
+    xlator_t **xl_list;
+    gf_lock_t lock;
+    gf_timer_t *timer;
+    gf_boolean_t shutdown;
+    gf_boolean_t eager_lock;
+    gf_boolean_t other_eager_lock;
+    gf_boolean_t optimistic_changelog;
+    gf_boolean_t parallel_writes;
+    uint32_t stripe_cache;
+    uint32_t quorum_count;
+    uint32_t background_heals;
+    uint32_t heal_wait_qlen;
+    uint32_t self_heal_window_size; /* max size of read/writes */
+    uint32_t eager_lock_timeout;
+    uint32_t other_eager_lock_timeout;
+    struct list_head pending_fops;
+    struct list_head heal_waiting;
+    struct list_head healing;
+    struct mem_pool *fop_pool;
+    struct mem_pool *cbk_pool;
+    struct mem_pool *lock_pool;
+    ec_self_heald_t shd;
+    char vol_uuid[UUID_SIZE + 1];
+    dict_t *leaf_to_subvolid;
+    ec_read_policy_t read_policy;
+    ec_matrix_list_t matrix;
+    ec_statistics_t stats;
+};
+
+#endif /* __EC_TYPES_H__ */
diff --git a/xlators/cluster/ec/src/ec.c b/xlators/cluster/ec/src/ec.c
new file mode 100644
index 00000000000..7344be4968d
--- /dev/null
+++ b/xlators/cluster/ec/src/ec.c
@@ -0,0 +1,1873 @@
+/*
+  Copyright (c) 2012-2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/defaults.h>
+#include <glusterfs/statedump.h>
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/upcall-utils.h>
+
+#include "ec.h"
+#include "ec-messages.h"
+#include "ec-mem-types.h"
+#include "ec-types.h"
+#include "ec-helpers.h"
+#include "ec-common.h"
+#include "ec-fops.h"
+#include "ec-method.h"
+#include "ec-code.h"
+#include "ec-heald.h"
+#include <glusterfs/events.h>
+
+static char *ec_read_policies[EC_READ_POLICY_MAX + 1] = {
+    [EC_ROUND_ROBIN] = "round-robin",
+    [EC_GFID_HASH] = "gfid-hash",
+    [EC_READ_POLICY_MAX] = NULL};
+
+#define EC_INTERNAL_XATTR_OR_GOTO(name, xattr, op_errno, label)                \
+    do {                                                                       \
+        if (ec_is_internal_xattr(NULL, (char *)name, NULL, NULL)) {            \
+            op_errno = EPERM;                                                  \
+            goto label;                                                        \
+        }                                                                      \
+        if (name && (strlen(name) == 0) && xattr) {                            \
+            /* Bulk [f]removexattr/[f]setxattr */                              \
+            GF_IF_INTERNAL_XATTR_GOTO(EC_XATTR_PREFIX "*", xattr, op_errno,    \
+                                      label);                                  \
+        }                                                                      \
+    } while (0)
+
+int32_t
+ec_parse_options(xlator_t *this)
+{
+    ec_t *ec = this->private;
+    int32_t error = EINVAL;
+    uintptr_t mask;
+
+    GF_OPTION_INIT("redundancy", ec->redundancy, int32, out);
+    ec->fragments = ec->nodes - ec->redundancy;
+    if ((ec->redundancy < 1) || (ec->redundancy >= ec->fragments) ||
+        (ec->fragments > EC_MAX_FRAGMENTS)) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, EC_MSG_INVALID_REDUNDANCY,
+               "Invalid redundancy (must be between "
+               "1 and %d)",
+               (ec->nodes - 1) / 2);
+
+        goto out;
+    }
+
+    ec->bits_for_nodes = 1;
+    mask = 2;
+    while (ec->nodes > mask) {
+        ec->bits_for_nodes++;
+        mask <<= 1;
+    }
+    ec->node_mask = (1ULL << ec->nodes) - 1ULL;
+    ec->fragment_size = EC_METHOD_CHUNK_SIZE;
+    ec->stripe_size = ec->fragment_size * ec->fragments;
+
+    gf_msg_debug("ec", 0,
+                 "Initialized with: nodes=%u, fragments=%u, "
+                 "stripe_size=%u, node_mask=%" PRIxFAST32,
+                 ec->nodes, ec->fragments, ec->stripe_size, ec->node_mask);
+
+    error = 0;
+
+out:
+    return error;
+}
+
+int32_t
+ec_prepare_childs(xlator_t *this)
+{
+    ec_t *ec = this->private;
+    xlator_list_t *child = NULL;
+    int32_t count = 0;
+
+    for (child = this->children; child != NULL; child = child->next) {
+        count++;
+    }
+    if (count > EC_MAX_NODES) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, EC_MSG_TOO_MANY_SUBVOLS,
+               "Too many subvolumes");
+
+        return EINVAL;
+    }
+    ec->nodes = count;
+
+    ec->xl_list = GF_CALLOC(count, sizeof(ec->xl_list[0]), ec_mt_xlator_t);
+    if (ec->xl_list == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+               "Allocation of xlator list failed");
+
+        return ENOMEM;
+    }
+    ec->xl_up = 0;
+    ec->xl_up_count = 0;
+
+    count = 0;
+    for (child = this->children; child != NULL; child = child->next) {
+        ec->xl_list[count++] = child->xlator;
+    }
+
+    return 0;
+}
+
+/* This function transforms the subvol to subvol-id*/
+static int
+_subvol_to_subvolid(dict_t *this, char *key, data_t *value, void *data)
+{
+    ec_t *ec = data;
+    xlator_t *subvol = NULL;
+    int i = 0;
+    int ret = -1;
+
+    subvol = data_to_ptr(value);
+    for (i = 0; i < ec->nodes; i++) {
+        if (ec->xl_list[i] == subvol) {
+            ret = dict_set_int32(this, key, i);
+            /* -1 stops dict_foreach and returns -1*/
+            if (ret < 0)
+                ret = -1;
+            goto out;
+        }
+    }
+out:
+    return ret;
+}
+
+int
+ec_subvol_to_subvol_id_transform(ec_t *ec, dict_t *leaf_to_subvolid)
+{
+    return dict_foreach(leaf_to_subvolid, _subvol_to_subvolid, ec);
+}
+
+void
+__ec_destroy_private(xlator_t *this)
+{
+    ec_t *ec = this->private;
+
+    if (ec != NULL) {
+        LOCK(&ec->lock);
+
+        if (ec->timer != NULL) {
+            gf_timer_call_cancel(this->ctx, ec->timer);
+            ec->timer = NULL;
+        }
+
+        UNLOCK(&ec->lock);
+
+        /* There is a race with timer because there is no way to know if
+         * timer callback has really been cancelled or it has been scheduled
+         * for execution. If it has been scheduled, it will crash if we
+         * destroy ec too fast.
+         *
+         * Not sure how this can be solved without using global variables or
+         * having support from gf_timer_call_cancel()
+         */
+        sleep(2);
+
+        this->private = NULL;
+        if (ec->xl_list != NULL) {
+            GF_FREE(ec->xl_list);
+            ec->xl_list = NULL;
+        }
+
+        if (ec->fop_pool != NULL) {
+            mem_pool_destroy(ec->fop_pool);
+        }
+
+        if (ec->cbk_pool != NULL) {
+            mem_pool_destroy(ec->cbk_pool);
+        }
+
+        if (ec->lock_pool != NULL) {
+            mem_pool_destroy(ec->lock_pool);
+        }
+
+        LOCK_DESTROY(&ec->lock);
+
+        if (ec->leaf_to_subvolid)
+            dict_unref(ec->leaf_to_subvolid);
+
+        ec_method_fini(&ec->matrix);
+
+        GF_FREE(ec);
+    }
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    if (xlator_mem_acct_init(this, ec_mt_end + 1) != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+               "Memory accounting initialization "
+               "failed.");
+
+        return -1;
+    }
+
+    return 0;
+}
+
+void
+ec_configure_background_heal_opts(ec_t *ec, int background_heals,
+                                  int heal_wait_qlen)
+{
+    if (background_heals == 0) {
+        ec->heal_wait_qlen = 0;
+    } else {
+        ec->heal_wait_qlen = heal_wait_qlen;
+    }
+    ec->background_heals = background_heals;
+}
+
+int
+ec_assign_read_policy(ec_t *ec, char *read_policy)
+{
+    int read_policy_idx = -1;
+
+    read_policy_idx = gf_get_index_by_elem(ec_read_policies, read_policy);
+    if (read_policy_idx < 0 || read_policy_idx >= EC_READ_POLICY_MAX)
+        return -1;
+
+    ec->read_policy = read_policy_idx;
+    return 0;
+}
+
+int32_t
+reconfigure(xlator_t *this, dict_t *options)
+{
+    ec_t *ec = this->private;
+    char *read_policy = NULL;
+    char *extensions = NULL;
+    uint32_t heal_wait_qlen = 0;
+    uint32_t background_heals = 0;
+    int32_t ret = -1;
+    int32_t err;
+
+    GF_OPTION_RECONF("cpu-extensions", extensions, options, str, failed);
+
+    GF_OPTION_RECONF("self-heal-daemon", ec->shd.enabled, options, bool,
+                     failed);
+    GF_OPTION_RECONF("iam-self-heal-daemon", ec->shd.iamshd, options, bool,
+                     failed);
+    GF_OPTION_RECONF("eager-lock", ec->eager_lock, options, bool, failed);
+    GF_OPTION_RECONF("other-eager-lock", ec->other_eager_lock, options, bool,
+                     failed);
+    GF_OPTION_RECONF("eager-lock-timeout", ec->eager_lock_timeout, options,
+                     uint32, failed);
+    GF_OPTION_RECONF("other-eager-lock-timeout", ec->other_eager_lock_timeout,
+                     options, uint32, failed);
+    GF_OPTION_RECONF("background-heals", background_heals, options, uint32,
+                     failed);
+    GF_OPTION_RECONF("heal-wait-qlength", heal_wait_qlen, options, uint32,
+                     failed);
+    GF_OPTION_RECONF("self-heal-window-size", ec->self_heal_window_size,
+                     options, uint32, failed);
+    GF_OPTION_RECONF("heal-timeout", ec->shd.timeout, options, int32, failed);
+    ec_configure_background_heal_opts(ec, background_heals, heal_wait_qlen);
+    GF_OPTION_RECONF("shd-max-threads", ec->shd.max_threads, options, uint32,
+                     failed);
+    GF_OPTION_RECONF("shd-wait-qlength", ec->shd.wait_qlength, options, uint32,
+                     failed);
+
+    GF_OPTION_RECONF("read-policy", read_policy, options, str, failed);
+
+    GF_OPTION_RECONF("optimistic-change-log", ec->optimistic_changelog, options,
+                     bool, failed);
+    GF_OPTION_RECONF("parallel-writes", ec->parallel_writes, options, bool,
+                     failed);
+    GF_OPTION_RECONF("stripe-cache", ec->stripe_cache, options, uint32, failed);
+    GF_OPTION_RECONF("quorum-count", ec->quorum_count, options, uint32, failed);
+    ret = 0;
+    if (ec_assign_read_policy(ec, read_policy)) {
+        ret = -1;
+    }
+
+    err = ec_method_update(this, &ec->matrix, extensions);
+    if (err != 0) {
+        ret = -1;
+    }
+
+failed:
+    return ret;
+}
+
+glusterfs_event_t
+ec_get_event_from_state(ec_t *ec)
+{
+    int down_count = 0;
+
+    if (ec->xl_up_count >= ec->fragments) {
+        /* If ec is up but some subvolumes are yet to notify, give
+         * grace time for other subvols to notify to prevent start of
+         * I/O which may result in self-heals */
+        if (ec->xl_notify_count < ec->nodes)
+            return GF_EVENT_MAXVAL;
+
+        return GF_EVENT_CHILD_UP;
+    } else {
+        down_count = ec->xl_notify_count - ec->xl_up_count;
+        if (down_count > ec->redundancy)
+            return GF_EVENT_CHILD_DOWN;
+    }
+
+    return GF_EVENT_MAXVAL;
+}
+
+void
+ec_up(xlator_t *this, ec_t *ec)
+{
+    char str1[32], str2[32];
+
+    if (ec->timer != NULL) {
+        gf_timer_call_cancel(this->ctx, ec->timer);
+        ec->timer = NULL;
+    }
+
+    ec->up = 1;
+    gf_msg(this->name, GF_LOG_INFO, 0, EC_MSG_EC_UP,
+           "Going UP : Child UP = %s Child Notify = %s",
+           ec_bin(str1, sizeof(str1), ec->xl_up, ec->nodes),
+           ec_bin(str2, sizeof(str2), ec->xl_notify, ec->nodes));
+
+    gf_event(EVENT_EC_MIN_BRICKS_UP, "subvol=%s", this->name);
+}
+
+void
+ec_down(xlator_t *this, ec_t *ec)
+{
+    char str1[32], str2[32];
+
+    if (ec->timer != NULL) {
+        gf_timer_call_cancel(this->ctx, ec->timer);
+        ec->timer = NULL;
+    }
+
+    ec->up = 0;
+    gf_msg(this->name, GF_LOG_INFO, 0, EC_MSG_EC_DOWN,
+           "Going DOWN : Child UP = %s Child Notify = %s",
+           ec_bin(str1, sizeof(str1), ec->xl_up, ec->nodes),
+           ec_bin(str2, sizeof(str2), ec->xl_notify, ec->nodes));
+
+    gf_event(EVENT_EC_MIN_BRICKS_NOT_UP, "subvol=%s", this->name);
+}
+
+void
+ec_notify_cbk(void *data)
+{
+    ec_t *ec = data;
+    glusterfs_event_t event = GF_EVENT_MAXVAL;
+    gf_boolean_t propagate = _gf_false;
+    gf_boolean_t launch_heal = _gf_false;
+
+    LOCK(&ec->lock);
+    {
+        if (!ec->timer) {
+            /*
+             * Either child_up/child_down is already sent to parent
+             * This is a spurious wake up.
+             */
+            goto unlock;
+        }
+
+        gf_timer_call_cancel(ec->xl->ctx, ec->timer);
+        ec->timer = NULL;
+
+        /* The timeout has expired, so any subvolume that has not
+         * already reported its state, will be considered to be down.
+         * We mark as if all bricks had reported. */
+        ec->xl_notify = (1ULL << ec->nodes) - 1ULL;
+        ec->xl_notify_count = ec->nodes;
+
+        /* Since we have marked all subvolumes as notified, it's
+         * guaranteed that ec_get_event_from_state() will return
+         * CHILD_UP or CHILD_DOWN, but not MAXVAL. */
+        event = ec_get_event_from_state(ec);
+        if (event == GF_EVENT_CHILD_UP) {
+            /* We are ready to bring the volume up. If there are
+             * still bricks DOWN, they will be healed when they
+             * come up. */
+            ec_up(ec->xl, ec);
+
+            if (ec->shd.iamshd && !ec->shutdown) {
+                launch_heal = _gf_true;
+                GF_ATOMIC_INC(ec->async_fop_count);
+            }
+        }
+
+        propagate = _gf_true;
+    }
+unlock:
+    UNLOCK(&ec->lock);
+
+    if (launch_heal) {
+        /* We have just brought the volume UP, so we trigger
+         * a self-heal check on the root directory. */
+        ec_launch_replace_heal(ec);
+    }
+    if (propagate) {
+        default_notify(ec->xl, event, NULL);
+    }
+}
+
+void
+ec_launch_notify_timer(xlator_t *this, ec_t *ec)
+{
+    struct timespec delay = {
+        0,
+    };
+
+    gf_msg_debug(this->name, 0, "Initiating child-down timer");
+    delay.tv_sec = 10;
+    delay.tv_nsec = 0;
+    ec->timer = gf_timer_call_after(this->ctx, delay, ec_notify_cbk, ec);
+    if (ec->timer == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_TIMER_CREATE_FAIL,
+               "Cannot create timer "
+               "for delayed initialization");
+    }
+}
+
+gf_boolean_t
+ec_disable_delays(ec_t *ec)
+{
+    ec->shutdown = _gf_true;
+
+    return __ec_is_last_fop(ec);
+}
+
+void
+ec_cleanup_healer_object(ec_t *ec)
+{
+    struct subvol_healer *healer = NULL;
+    ec_self_heald_t *shd = NULL;
+    void *res = NULL;
+    int i = 0;
+    gf_boolean_t is_join = _gf_false;
+
+    shd = &ec->shd;
+    if (!shd->iamshd)
+        return;
+
+    for (i = 0; i < ec->nodes; i++) {
+        healer = &shd->index_healers[i];
+        pthread_mutex_lock(&healer->mutex);
+        {
+            healer->rerun = 1;
+            if (healer->running) {
+                pthread_cond_signal(&healer->cond);
+                is_join = _gf_true;
+            }
+        }
+        pthread_mutex_unlock(&healer->mutex);
+        if (is_join) {
+            pthread_join(healer->thread, &res);
+            is_join = _gf_false;
+        }
+
+        healer = &shd->full_healers[i];
+        pthread_mutex_lock(&healer->mutex);
+        {
+            healer->rerun = 1;
+            if (healer->running) {
+                pthread_cond_signal(&healer->cond);
+                is_join = _gf_true;
+            }
+        }
+        pthread_mutex_unlock(&healer->mutex);
+        if (is_join) {
+            pthread_join(healer->thread, &res);
+            is_join = _gf_false;
+        }
+    }
+}
+void
+ec_pending_fops_completed(ec_t *ec)
+{
+    if (ec->shutdown) {
+        default_notify(ec->xl, GF_EVENT_PARENT_DOWN, NULL);
+    }
+}
+
+static gf_boolean_t
+ec_set_up_state(ec_t *ec, uintptr_t index_mask, uintptr_t new_state)
+{
+    uintptr_t current_state = 0;
+
+    if (xlator_is_cleanup_starting(ec->xl))
+        return _gf_false;
+
+    if ((ec->xl_notify & index_mask) == 0) {
+        ec->xl_notify |= index_mask;
+        ec->xl_notify_count++;
+    }
+    current_state = ec->xl_up & index_mask;
+    if (current_state != new_state) {
+        ec->xl_up ^= index_mask;
+        ec->xl_up_count += (current_state ? -1 : 1);
+
+        return _gf_true;
+    }
+
+    return _gf_false;
+}
+
+static gf_boolean_t
+ec_upcall(ec_t *ec, struct gf_upcall *upcall)
+{
+    struct gf_upcall_cache_invalidation *ci = NULL;
+    struct gf_upcall_inodelk_contention *lc = NULL;
+    inode_t *inode;
+    inode_table_t *table;
+
+    switch (upcall->event_type) {
+        case GF_UPCALL_CACHE_INVALIDATION:
+            ci = upcall->data;
+            ci->flags |= UP_INVAL_ATTR;
+            return _gf_true;
+
+        case GF_UPCALL_INODELK_CONTENTION:
+            lc = upcall->data;
+            if (strcmp(lc->domain, ec->xl->name) != 0) {
+                /* The lock is not owned by EC, ignore it. */
+                return _gf_true;
+            }
+            table = ((xlator_t *)ec->xl->graph->top)->itable;
+            if (table == NULL) {
+                /* Self-heal daemon doesn't have an inode table on the top
+                 * xlator because it doesn't need it. In this case we should
+                 * use the inode table managed by EC itself where all inodes
+                 * being healed should be present. However self-heal doesn't
+                 * use eager-locking and inodelk's are already released as
+                 * soon as possible. In this case we can safely ignore these
+                 * notifications. */
+                return _gf_false;
+            }
+            inode = inode_find(table, upcall->gfid);
+            /* If inode is not found, it means that it's already released,
+             * so we can ignore it. Probably it has been released and
+             * destroyed while the contention notification was being sent.
+             */
+            if (inode != NULL) {
+                ec_lock_release(ec, inode);
+                inode_unref(inode);
+            }
+
+            return _gf_false;
+
+        default:
+            return _gf_true;
+    }
+}
+
+int32_t
+ec_notify(xlator_t *this, int32_t event, void *data, void *data2)
+{
+    ec_t *ec = this->private;
+    int32_t idx = 0;
+    int32_t error = 0;
+    glusterfs_event_t old_event = GF_EVENT_MAXVAL;
+    dict_t *input = NULL;
+    dict_t *output = NULL;
+    gf_boolean_t propagate = _gf_true;
+    gf_boolean_t needs_shd_check = _gf_false;
+    int32_t orig_event = event;
+    uintptr_t mask = 0;
+
+    gf_msg_trace(this->name, 0, "NOTIFY(%d): %p, %p", event, data, data2);
+
+    if (event == GF_EVENT_UPCALL) {
+        propagate = ec_upcall(ec, data);
+        goto done;
+    }
+
+    if (event == GF_EVENT_TRANSLATOR_OP) {
+        if (!ec->up) {
+            error = -1;
+        } else {
+            input = data;
+            output = data2;
+            error = ec_xl_op(this, input, output);
+        }
+        goto out;
+    }
+
+    for (idx = 0; idx < ec->nodes; idx++) {
+        if (ec->xl_list[idx] == data) {
+            break;
+        }
+    }
+
+    LOCK(&ec->lock);
+
+    if (event == GF_EVENT_PARENT_UP) {
+        /*
+         * Start a timer which sends appropriate event to parent
+         * xlator to prevent the 'mount' syscall from hanging.
+         */
+        ec_launch_notify_timer(this, ec);
+        goto unlock;
+    } else if (event == GF_EVENT_PARENT_DOWN) {
+        /* If there aren't pending fops running after we have waken up
+         * them, we immediately propagate the notification. */
+        propagate = ec_disable_delays(ec);
+        ec_cleanup_healer_object(ec);
+        goto unlock;
+    }
+
+    if (idx < ec->nodes) { /* CHILD_* events */
+        old_event = ec_get_event_from_state(ec);
+
+        mask = 1ULL << idx;
+        if (event == GF_EVENT_CHILD_UP) {
+            /* We need to trigger a selfheal if a brick changes
+             * to UP state. */
+            if (ec_set_up_state(ec, mask, mask) && ec->shd.iamshd &&
+                !ec->shutdown) {
+                needs_shd_check = _gf_true;
+            }
+        } else if (event == GF_EVENT_CHILD_DOWN) {
+            ec_set_up_state(ec, mask, 0);
+        }
+
+        event = ec_get_event_from_state(ec);
+
+        if (event == GF_EVENT_CHILD_UP) {
+            if (!ec->up) {
+                ec_up(this, ec);
+            }
+        } else {
+            /* If the volume is not UP, it's irrelevant if one
+             * brick has come up. We cannot heal anything. */
+            needs_shd_check = _gf_false;
+
+            if ((event == GF_EVENT_CHILD_DOWN) && ec->up) {
+                ec_down(this, ec);
+            }
+        }
+
+        if (event != GF_EVENT_MAXVAL) {
+            if (event == old_event) {
+                if (orig_event == GF_EVENT_CHILD_UP)
+                    event = GF_EVENT_SOME_DESCENDENT_UP;
+                else /* orig_event has to be GF_EVENT_CHILD_DOWN */
+                    event = GF_EVENT_SOME_DESCENDENT_DOWN;
+            }
+        } else {
+            propagate = _gf_false;
+            needs_shd_check = _gf_false;
+        }
+
+        if (needs_shd_check) {
+            GF_ATOMIC_INC(ec->async_fop_count);
+        }
+    }
+unlock:
+    UNLOCK(&ec->lock);
+
+done:
+    if (needs_shd_check) {
+        ec_launch_replace_heal(ec);
+    }
+    if (propagate) {
+        error = default_notify(this, event, data);
+    }
+
+out:
+    return error;
+}
+
+int32_t
+notify(xlator_t *this, int32_t event, void *data, ...)
+{
+    int ret = -1;
+    va_list ap;
+    void *data2 = NULL;
+
+    va_start(ap, data);
+    data2 = va_arg(ap, dict_t *);
+    va_end(ap);
+    ret = ec_notify(this, event, data, data2);
+
+    return ret;
+}
+
+static void
+ec_statistics_init(ec_t *ec)
+{
+    GF_ATOMIC_INIT(ec->stats.stripe_cache.hits, 0);
+    GF_ATOMIC_INIT(ec->stats.stripe_cache.misses, 0);
+    GF_ATOMIC_INIT(ec->stats.stripe_cache.updates, 0);
+    GF_ATOMIC_INIT(ec->stats.stripe_cache.invals, 0);
+    GF_ATOMIC_INIT(ec->stats.stripe_cache.evicts, 0);
+    GF_ATOMIC_INIT(ec->stats.stripe_cache.allocs, 0);
+    GF_ATOMIC_INIT(ec->stats.stripe_cache.errors, 0);
+    GF_ATOMIC_INIT(ec->stats.shd.attempted, 0);
+    GF_ATOMIC_INIT(ec->stats.shd.completed, 0);
+}
+
+static int
+ec_assign_read_mask(ec_t *ec, char *read_mask_str)
+{
+    char *mask = NULL;
+    char *maskptr = NULL;
+    char *saveptr = NULL;
+    char *id_str = NULL;
+    int id = 0;
+    int ret = 0;
+    uintptr_t read_mask = 0;
+
+    if (!read_mask_str) {
+        ec->read_mask = 0;
+        ret = 0;
+        goto out;
+    }
+
+    mask = gf_strdup(read_mask_str);
+    if (!mask) {
+        ret = -1;
+        goto out;
+    }
+    maskptr = mask;
+
+    for (;;) {
+        id_str = strtok_r(maskptr, ":", &saveptr);
+        if (id_str == NULL)
+            break;
+        if (gf_string2int(id_str, &id)) {
+            gf_msg(ec->xl->name, GF_LOG_ERROR, 0, EC_MSG_XLATOR_INIT_FAIL,
+                   "In read-mask \"%s\" id %s is not a valid integer",
+                   read_mask_str, id_str);
+            ret = -1;
+            goto out;
+        }
+
+        if ((id < 0) || (id >= ec->nodes)) {
+            gf_msg(ec->xl->name, GF_LOG_ERROR, 0, EC_MSG_XLATOR_INIT_FAIL,
+                   "In read-mask \"%s\" id %d is not in range [0 - %d]",
+                   read_mask_str, id, ec->nodes - 1);
+            ret = -1;
+            goto out;
+        }
+        read_mask |= (1UL << id);
+        maskptr = NULL;
+    }
+
+    if (gf_bits_count(read_mask) < ec->fragments) {
+        gf_msg(ec->xl->name, GF_LOG_ERROR, 0, EC_MSG_XLATOR_INIT_FAIL,
+               "read-mask \"%s\" should contain at least %d ids", read_mask_str,
+               ec->fragments);
+        ret = -1;
+        goto out;
+    }
+    ec->read_mask = read_mask;
+    ret = 0;
+out:
+    GF_FREE(mask);
+    return ret;
+}
+
+int32_t
+init(xlator_t *this)
+{
+    ec_t *ec = NULL;
+    char *read_policy = NULL;
+    char *extensions = NULL;
+    int32_t err;
+    char *read_mask_str = NULL;
+
+    if (this->parents == NULL) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, EC_MSG_NO_PARENTS,
+               "Volume does not have parents.");
+    }
+
+    ec = GF_MALLOC(sizeof(*ec), ec_mt_ec_t);
+    if (ec == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+               "Failed to allocate private memory.");
+
+        return -1;
+    }
+    memset(ec, 0, sizeof(*ec));
+
+    this->private = ec;
+
+    ec->xl = this;
+    LOCK_INIT(&ec->lock);
+
+    GF_ATOMIC_INIT(ec->async_fop_count, 0);
+    INIT_LIST_HEAD(&ec->pending_fops);
+    INIT_LIST_HEAD(&ec->heal_waiting);
+    INIT_LIST_HEAD(&ec->healing);
+
+    ec->fop_pool = mem_pool_new(ec_fop_data_t, 1024);
+    ec->cbk_pool = mem_pool_new(ec_cbk_data_t, 4096);
+    ec->lock_pool = mem_pool_new(ec_lock_t, 1024);
+    if ((ec->fop_pool == NULL) || (ec->cbk_pool == NULL) ||
+        (ec->lock_pool == NULL)) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+               "Failed to create memory pools.");
+
+        goto failed;
+    }
+
+    if (ec_prepare_childs(this) != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_XLATOR_INIT_FAIL,
+               "Failed to initialize xlator");
+
+        goto failed;
+    }
+
+    if (ec_parse_options(this) != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, EC_MSG_XLATOR_PARSE_OPT_FAIL,
+               "Failed to parse xlator options");
+
+        goto failed;
+    }
+
+    GF_OPTION_INIT("cpu-extensions", extensions, str, failed);
+
+    err = ec_method_init(this, &ec->matrix, ec->fragments, ec->nodes,
+                         ec->nodes * 2, extensions);
+    if (err != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, -err, EC_MSG_MATRIX_FAILED,
+               "Failed to initialize matrix management");
+
+        goto failed;
+    }
+
+    GF_OPTION_INIT("self-heal-daemon", ec->shd.enabled, bool, failed);
+    GF_OPTION_INIT("iam-self-heal-daemon", ec->shd.iamshd, bool, failed);
+    GF_OPTION_INIT("eager-lock", ec->eager_lock, bool, failed);
+    GF_OPTION_INIT("other-eager-lock", ec->other_eager_lock, bool, failed);
+    GF_OPTION_INIT("eager-lock-timeout", ec->eager_lock_timeout, uint32,
+                   failed);
+    GF_OPTION_INIT("other-eager-lock-timeout", ec->other_eager_lock_timeout,
+                   uint32, failed);
+    GF_OPTION_INIT("background-heals", ec->background_heals, uint32, failed);
+    GF_OPTION_INIT("heal-wait-qlength", ec->heal_wait_qlen, uint32, failed);
+    GF_OPTION_INIT("self-heal-window-size", ec->self_heal_window_size, uint32,
+                   failed);
+    ec_configure_background_heal_opts(ec, ec->background_heals,
+                                      ec->heal_wait_qlen);
+    GF_OPTION_INIT("read-policy", read_policy, str, failed);
+    if (ec_assign_read_policy(ec, read_policy))
+        goto failed;
+
+    GF_OPTION_INIT("heal-timeout", ec->shd.timeout, int32, failed);
+    GF_OPTION_INIT("shd-max-threads", ec->shd.max_threads, uint32, failed);
+    GF_OPTION_INIT("shd-wait-qlength", ec->shd.wait_qlength, uint32, failed);
+    GF_OPTION_INIT("optimistic-change-log", ec->optimistic_changelog, bool,
+                   failed);
+    GF_OPTION_INIT("parallel-writes", ec->parallel_writes, bool, failed);
+    GF_OPTION_INIT("stripe-cache", ec->stripe_cache, uint32, failed);
+    GF_OPTION_INIT("quorum-count", ec->quorum_count, uint32, failed);
+    GF_OPTION_INIT("ec-read-mask", read_mask_str, str, failed);
+
+    if (ec_assign_read_mask(ec, read_mask_str))
+        goto failed;
+
+    this->itable = inode_table_new(EC_SHD_INODE_LRU_LIMIT, this);
+    if (!this->itable)
+        goto failed;
+
+    if (ec->shd.iamshd)
+        ec_selfheal_daemon_init(this);
+    gf_msg_debug(this->name, 0, "Disperse translator initialized.");
+
+    ec->leaf_to_subvolid = dict_new();
+    if (!ec->leaf_to_subvolid)
+        goto failed;
+    if (glusterfs_reachable_leaves(this, ec->leaf_to_subvolid)) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_SUBVOL_BUILD_FAIL,
+               "Failed to build subvol "
+               "dictionary");
+        goto failed;
+    }
+
+    if (ec_subvol_to_subvol_id_transform(ec, ec->leaf_to_subvolid) < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_SUBVOL_ID_DICT_SET_FAIL,
+               "Failed to build subvol-id "
+               "dictionary");
+        goto failed;
+    }
+
+    ec_statistics_init(ec);
+
+    return 0;
+
+failed:
+    __ec_destroy_private(this);
+
+    return -1;
+}
+
+void
+fini(xlator_t *this)
+{
+    ec_selfheal_daemon_fini(this);
+    __ec_destroy_private(this);
+}
+
+int32_t
+ec_gf_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+             dict_t *xdata)
+{
+    ec_access(frame, this, -1, EC_MINIMUM_ONE, default_access_cbk, NULL, loc,
+              mask, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+             mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
+{
+    ec_create(frame, this, -1, EC_MINIMUM_MIN, default_create_cbk, NULL, loc,
+              flags, mode, umask, fd, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+              size_t len, dict_t *xdata)
+{
+    ec_discard(frame, this, -1, EC_MINIMUM_MIN, default_discard_cbk, NULL, fd,
+               offset, len, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_entrylk(call_frame_t *frame, xlator_t *this, const char *volume,
+              loc_t *loc, const char *basename, entrylk_cmd cmd,
+              entrylk_type type, dict_t *xdata)
+{
+    uint32_t fop_flags = EC_MINIMUM_ALL;
+
+    if (cmd == ENTRYLK_UNLOCK)
+        fop_flags = EC_MINIMUM_ONE;
+    ec_entrylk(frame, this, -1, fop_flags, default_entrylk_cbk, NULL, volume,
+               loc, basename, cmd, type, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume,
+               fd_t *fd, const char *basename, entrylk_cmd cmd,
+               entrylk_type type, dict_t *xdata)
+{
+    uint32_t fop_flags = EC_MINIMUM_ALL;
+
+    if (cmd == ENTRYLK_UNLOCK)
+        fop_flags = EC_MINIMUM_ONE;
+    ec_fentrylk(frame, this, -1, fop_flags, default_fentrylk_cbk, NULL, volume,
+                fd, basename, cmd, type, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+                off_t offset, size_t len, dict_t *xdata)
+{
+    ec_fallocate(frame, this, -1, EC_MINIMUM_MIN, default_fallocate_cbk, NULL,
+                 fd, mode, offset, len, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    ec_flush(frame, this, -1, EC_MINIMUM_MIN, default_flush_cbk, NULL, fd,
+             xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+            dict_t *xdata)
+{
+    ec_fsync(frame, this, -1, EC_MINIMUM_MIN, default_fsync_cbk, NULL, fd,
+             datasync, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+               dict_t *xdata)
+{
+    ec_fsyncdir(frame, this, -1, EC_MINIMUM_MIN, default_fsyncdir_cbk, NULL, fd,
+                datasync, xdata);
+
+    return 0;
+}
+
+int
+ec_marker_populate_args(call_frame_t *frame, int type, int *gauge,
+                        xlator_t **subvols)
+{
+    xlator_t *this = frame->this;
+    ec_t *ec = this->private;
+
+    memcpy(subvols, ec->xl_list, sizeof(*subvols) * ec->nodes);
+
+    if (type == MARKER_XTIME_TYPE) {
+        /*Don't error out on ENOENT/ENOTCONN */
+        gauge[MCNT_NOTFOUND] = 0;
+        gauge[MCNT_ENOTCONN] = 0;
+    }
+
+    return ec->nodes;
+}
+
+int32_t
+ec_handle_heal_commands(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                        const char *name, dict_t *xdata)
+{
+    dict_t *dict_rsp = NULL;
+    int op_ret = -1;
+    int op_errno = ENOMEM;
+
+    if (!name || strcmp(name, GF_HEAL_INFO))
+        return -1;
+
+    op_errno = -ec_get_heal_info(this, loc, &dict_rsp);
+    if (op_errno <= 0) {
+        op_errno = op_ret = 0;
+    }
+
+    STACK_UNWIND_STRICT(getxattr, frame, op_ret, op_errno, dict_rsp, NULL);
+    if (dict_rsp)
+        dict_unref(dict_rsp);
+    return 0;
+}
+
+int32_t
+ec_gf_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               const char *name, dict_t *xdata)
+{
+    int error = 0;
+    ec_t *ec = this->private;
+    int32_t fop_flags = EC_MINIMUM_ONE;
+
+    if (name && strcmp(name, EC_XATTR_HEAL) != 0) {
+        EC_INTERNAL_XATTR_OR_GOTO(name, NULL, error, out);
+    }
+
+    if (ec_handle_heal_commands(frame, this, loc, name, xdata) == 0)
+        return 0;
+
+    if (cluster_handle_marker_getxattr(frame, loc, name, ec->vol_uuid, NULL,
+                                       ec_marker_populate_args) == 0)
+        return 0;
+
+    if (name && ((fnmatch(GF_XATTR_STIME_PATTERN, name, 0) == 0) ||
+                 XATTR_IS_NODE_UUID(name) || XATTR_IS_NODE_UUID_LIST(name))) {
+        fop_flags = EC_MINIMUM_ALL;
+    }
+
+    ec_getxattr(frame, this, -1, fop_flags, default_getxattr_cbk, NULL, loc,
+                name, xdata);
+
+    return 0;
+out:
+    error = ENODATA;
+    STACK_UNWIND_STRICT(getxattr, frame, -1, error, NULL, NULL);
+    return 0;
+}
+
+int32_t
+ec_gf_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+                dict_t *xdata)
+{
+    int error = 0;
+
+    EC_INTERNAL_XATTR_OR_GOTO(name, NULL, error, out);
+
+    ec_fgetxattr(frame, this, -1, EC_MINIMUM_ONE, default_fgetxattr_cbk, NULL,
+                 fd, name, xdata);
+    return 0;
+out:
+    error = ENODATA;
+    STACK_UNWIND_STRICT(fgetxattr, frame, -1, error, NULL, NULL);
+    return 0;
+}
+
+int32_t
+ec_gf_inodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+              loc_t *loc, int32_t cmd, struct gf_flock *flock, dict_t *xdata)
+{
+    int32_t fop_flags = EC_MINIMUM_ALL;
+
+    if (flock->l_type == F_UNLCK)
+        fop_flags = EC_MINIMUM_ONE;
+
+    ec_inodelk(frame, this, &frame->root->lk_owner, -1, fop_flags,
+               default_inodelk_cbk, NULL, volume, loc, cmd, flock, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_finodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+               fd_t *fd, int32_t cmd, struct gf_flock *flock, dict_t *xdata)
+{
+    int32_t fop_flags = EC_MINIMUM_ALL;
+
+    if (flock->l_type == F_UNLCK)
+        fop_flags = EC_MINIMUM_ONE;
+    ec_finodelk(frame, this, &frame->root->lk_owner, -1, fop_flags,
+                default_finodelk_cbk, NULL, volume, fd, cmd, flock, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+           dict_t *xdata)
+{
+    ec_link(frame, this, -1, EC_MINIMUM_MIN, default_link_cbk, NULL, oldloc,
+            newloc, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
+         struct gf_flock *flock, dict_t *xdata)
+{
+    int32_t fop_flags = EC_MINIMUM_ALL;
+
+    if (flock->l_type == F_UNLCK)
+        fop_flags = EC_MINIMUM_ONE;
+    ec_lk(frame, this, -1, fop_flags, default_lk_cbk, NULL, fd, cmd, flock,
+          xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    ec_lookup(frame, this, -1, EC_MINIMUM_MIN, default_lookup_cbk, NULL, loc,
+              xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+            mode_t umask, dict_t *xdata)
+{
+    ec_mkdir(frame, this, -1, EC_MINIMUM_MIN, default_mkdir_cbk, NULL, loc,
+             mode, umask, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+            dev_t rdev, mode_t umask, dict_t *xdata)
+{
+    ec_mknod(frame, this, -1, EC_MINIMUM_MIN, default_mknod_cbk, NULL, loc,
+             mode, rdev, umask, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+           fd_t *fd, dict_t *xdata)
+{
+    ec_open(frame, this, -1, EC_MINIMUM_MIN, default_open_cbk, NULL, loc, flags,
+            fd, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+              dict_t *xdata)
+{
+    ec_opendir(frame, this, -1, EC_MINIMUM_MIN, default_opendir_cbk, NULL, loc,
+               fd, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+              off_t offset, dict_t *xdata)
+{
+    ec_readdir(frame, this, -1, EC_MINIMUM_ONE, default_readdir_cbk, NULL, fd,
+               size, offset, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+               off_t offset, dict_t *xdata)
+{
+    ec_readdirp(frame, this, -1, EC_MINIMUM_ONE, default_readdirp_cbk, NULL, fd,
+                size, offset, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+               dict_t *xdata)
+{
+    ec_readlink(frame, this, -1, EC_MINIMUM_ONE, default_readlink_cbk, NULL,
+                loc, size, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t offset, uint32_t flags, dict_t *xdata)
+{
+    ec_readv(frame, this, -1, EC_MINIMUM_MIN, default_readv_cbk, NULL, fd, size,
+             offset, flags, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                  const char *name, dict_t *xdata)
+{
+    int error = 0;
+
+    EC_INTERNAL_XATTR_OR_GOTO(name, xdata, error, out);
+
+    ec_removexattr(frame, this, -1, EC_MINIMUM_MIN, default_removexattr_cbk,
+                   NULL, loc, name, xdata);
+
+    return 0;
+out:
+    STACK_UNWIND_STRICT(removexattr, frame, -1, error, NULL);
+    return 0;
+}
+
+int32_t
+ec_gf_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                   const char *name, dict_t *xdata)
+{
+    int error = 0;
+
+    EC_INTERNAL_XATTR_OR_GOTO(name, xdata, error, out);
+
+    ec_fremovexattr(frame, this, -1, EC_MINIMUM_MIN, default_fremovexattr_cbk,
+                    NULL, fd, name, xdata);
+
+    return 0;
+out:
+    STACK_UNWIND_STRICT(fremovexattr, frame, -1, error, NULL);
+    return 0;
+}
+
+int32_t
+ec_gf_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+             dict_t *xdata)
+{
+    ec_rename(frame, this, -1, EC_MINIMUM_MIN, default_rename_cbk, NULL, oldloc,
+              newloc, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflags,
+            dict_t *xdata)
+{
+    ec_rmdir(frame, this, -1, EC_MINIMUM_MIN, default_rmdir_cbk, NULL, loc,
+             xflags, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+              struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    ec_setattr(frame, this, -1, EC_MINIMUM_MIN, default_setattr_cbk, NULL, loc,
+               stbuf, valid, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+               struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    ec_fsetattr(frame, this, -1, EC_MINIMUM_MIN, default_fsetattr_cbk, NULL, fd,
+                stbuf, valid, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+               int32_t flags, dict_t *xdata)
+{
+    int error = 0;
+
+    EC_INTERNAL_XATTR_OR_GOTO("", dict, error, out);
+
+    ec_setxattr(frame, this, -1, EC_MINIMUM_MIN, default_setxattr_cbk, NULL,
+                loc, dict, flags, xdata);
+
+    return 0;
+out:
+    STACK_UNWIND_STRICT(setxattr, frame, -1, error, NULL);
+    return 0;
+}
+
+int32_t
+ec_gf_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+                int32_t flags, dict_t *xdata)
+{
+    int error = 0;
+
+    EC_INTERNAL_XATTR_OR_GOTO("", dict, error, out);
+
+    ec_fsetxattr(frame, this, -1, EC_MINIMUM_MIN, default_fsetxattr_cbk, NULL,
+                 fd, dict, flags, xdata);
+
+    return 0;
+out:
+    STACK_UNWIND_STRICT(fsetxattr, frame, -1, error, NULL);
+    return 0;
+}
+
+int32_t
+ec_gf_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    ec_stat(frame, this, -1, EC_MINIMUM_MIN, default_stat_cbk, NULL, loc,
+            xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    ec_fstat(frame, this, -1, EC_MINIMUM_MIN, default_fstat_cbk, NULL, fd,
+             xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    ec_statfs(frame, this, -1, EC_MINIMUM_MIN, default_statfs_cbk, NULL, loc,
+              xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_symlink(call_frame_t *frame, xlator_t *this, const char *linkname,
+              loc_t *loc, mode_t umask, dict_t *xdata)
+{
+    ec_symlink(frame, this, -1, EC_MINIMUM_MIN, default_symlink_cbk, NULL,
+               linkname, loc, umask, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+               dict_t *xdata)
+{
+    ec_truncate(frame, this, -1, EC_MINIMUM_MIN, default_truncate_cbk, NULL,
+                loc, offset, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                dict_t *xdata)
+{
+    ec_ftruncate(frame, this, -1, EC_MINIMUM_MIN, default_ftruncate_cbk, NULL,
+                 fd, offset, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflags,
+             dict_t *xdata)
+{
+    ec_unlink(frame, this, -1, EC_MINIMUM_MIN, default_unlink_cbk, NULL, loc,
+              xflags, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_writev(call_frame_t *frame, xlator_t *this, fd_t *fd,
+             struct iovec *vector, int32_t count, off_t offset, uint32_t flags,
+             struct iobref *iobref, dict_t *xdata)
+{
+    ec_writev(frame, this, -1, EC_MINIMUM_MIN, default_writev_cbk, NULL, fd,
+              vector, count, offset, flags, iobref, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+              gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+    ec_xattrop(frame, this, -1, EC_MINIMUM_MIN, default_xattrop_cbk, NULL, loc,
+               optype, xattr, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+               gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+    ec_fxattrop(frame, this, -1, EC_MINIMUM_MIN, default_fxattrop_cbk, NULL, fd,
+                optype, xattr, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+               off_t len, dict_t *xdata)
+{
+    default_zerofill_failure_cbk(frame, ENOTSUP);
+
+    return 0;
+}
+
+int32_t
+ec_gf_seek(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+           gf_seek_what_t what, dict_t *xdata)
+{
+    ec_seek(frame, this, -1, EC_MINIMUM_ONE, default_seek_cbk, NULL, fd, offset,
+            what, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_ipc(call_frame_t *frame, xlator_t *this, int32_t op, dict_t *xdata)
+{
+    ec_ipc(frame, this, -1, EC_MINIMUM_MIN, default_ipc_cbk, NULL, op, xdata);
+    return 0;
+}
+
+int32_t
+ec_gf_forget(xlator_t *this, inode_t *inode)
+{
+    uint64_t value = 0;
+    ec_inode_t *ctx = NULL;
+
+    if ((inode_ctx_del(inode, this, &value) == 0) && (value != 0)) {
+        ctx = (ec_inode_t *)(uintptr_t)value;
+        /* We can only forget an inode if it has been unlocked, so the stripe
+         * cache should also be empty. */
+        GF_ASSERT(list_empty(&ctx->stripe_cache.lru));
+        GF_FREE(ctx);
+    }
+
+    return 0;
+}
+
+void
+ec_gf_release_fd(xlator_t *this, fd_t *fd)
+{
+    uint64_t value = 0;
+    ec_fd_t *ctx = NULL;
+
+    if ((fd_ctx_del(fd, this, &value) == 0) && (value != 0)) {
+        ctx = (ec_fd_t *)(uintptr_t)value;
+        loc_wipe(&ctx->loc);
+        GF_FREE(ctx);
+    }
+}
+
+int32_t
+ec_gf_release(xlator_t *this, fd_t *fd)
+{
+    ec_gf_release_fd(this, fd);
+
+    return 0;
+}
+
+int32_t
+ec_gf_releasedir(xlator_t *this, fd_t *fd)
+{
+    ec_gf_release_fd(this, fd);
+
+    return 0;
+}
+
+int32_t
+ec_dump_private(xlator_t *this)
+{
+    ec_t *ec = NULL;
+    char key_prefix[GF_DUMP_MAX_BUF_LEN];
+    char tmp[65];
+
+    GF_ASSERT(this);
+
+    ec = this->private;
+    GF_ASSERT(ec);
+
+    snprintf(key_prefix, GF_DUMP_MAX_BUF_LEN, "%s.%s", this->type, this->name);
+    gf_proc_dump_add_section("%s", key_prefix);
+    gf_proc_dump_write("up", "%u", ec->up);
+    gf_proc_dump_write("nodes", "%u", ec->nodes);
+    gf_proc_dump_write("redundancy", "%u", ec->redundancy);
+    gf_proc_dump_write("fragment_size", "%u", ec->fragment_size);
+    gf_proc_dump_write("stripe_size", "%u", ec->stripe_size);
+    gf_proc_dump_write("childs_up", "%u", ec->xl_up_count);
+    gf_proc_dump_write("childs_up_mask", "%s",
+                       ec_bin(tmp, sizeof(tmp), ec->xl_up, ec->nodes));
+    if (ec->read_mask) {
+        gf_proc_dump_write("read-mask", "%s",
+                           ec_bin(tmp, sizeof(tmp), ec->read_mask, ec->nodes));
+    }
+    gf_proc_dump_write("background-heals", "%d", ec->background_heals);
+    gf_proc_dump_write("heal-wait-qlength", "%d", ec->heal_wait_qlen);
+    gf_proc_dump_write("self-heal-window-size", "%" PRIu32,
+                       ec->self_heal_window_size);
+    gf_proc_dump_write("healers", "%d", ec->healers);
+    gf_proc_dump_write("heal-waiters", "%d", ec->heal_waiters);
+    gf_proc_dump_write("read-policy", "%s", ec_read_policies[ec->read_policy]);
+    gf_proc_dump_write("parallel-writes", "%d", ec->parallel_writes);
+    gf_proc_dump_write("quorum-count", "%u", ec->quorum_count);
+
+    snprintf(key_prefix, GF_DUMP_MAX_BUF_LEN, "%s.%s.stats.stripe_cache",
+             this->type, this->name);
+    gf_proc_dump_add_section("%s", key_prefix);
+
+    gf_proc_dump_write("hits", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(ec->stats.stripe_cache.hits));
+    gf_proc_dump_write("misses", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(ec->stats.stripe_cache.misses));
+    gf_proc_dump_write("updates", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(ec->stats.stripe_cache.updates));
+    gf_proc_dump_write("invalidations", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(ec->stats.stripe_cache.invals));
+    gf_proc_dump_write("evicts", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(ec->stats.stripe_cache.evicts));
+    gf_proc_dump_write("allocations", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(ec->stats.stripe_cache.allocs));
+    gf_proc_dump_write("errors", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(ec->stats.stripe_cache.errors));
+    gf_proc_dump_write("heals-attempted", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(ec->stats.shd.attempted));
+    gf_proc_dump_write("heals-completed", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(ec->stats.shd.completed));
+
+    return 0;
+}
+
+struct xlator_fops fops = {.lookup = ec_gf_lookup,
+                           .stat = ec_gf_stat,
+                           .fstat = ec_gf_fstat,
+                           .truncate = ec_gf_truncate,
+                           .ftruncate = ec_gf_ftruncate,
+                           .access = ec_gf_access,
+                           .readlink = ec_gf_readlink,
+                           .mknod = ec_gf_mknod,
+                           .mkdir = ec_gf_mkdir,
+                           .unlink = ec_gf_unlink,
+                           .rmdir = ec_gf_rmdir,
+                           .symlink = ec_gf_symlink,
+                           .rename = ec_gf_rename,
+                           .link = ec_gf_link,
+                           .create = ec_gf_create,
+                           .open = ec_gf_open,
+                           .readv = ec_gf_readv,
+                           .writev = ec_gf_writev,
+                           .flush = ec_gf_flush,
+                           .fsync = ec_gf_fsync,
+                           .opendir = ec_gf_opendir,
+                           .readdir = ec_gf_readdir,
+                           .readdirp = ec_gf_readdirp,
+                           .fsyncdir = ec_gf_fsyncdir,
+                           .statfs = ec_gf_statfs,
+                           .setxattr = ec_gf_setxattr,
+                           .getxattr = ec_gf_getxattr,
+                           .fsetxattr = ec_gf_fsetxattr,
+                           .fgetxattr = ec_gf_fgetxattr,
+                           .removexattr = ec_gf_removexattr,
+                           .fremovexattr = ec_gf_fremovexattr,
+                           .lk = ec_gf_lk,
+                           .inodelk = ec_gf_inodelk,
+                           .finodelk = ec_gf_finodelk,
+                           .entrylk = ec_gf_entrylk,
+                           .fentrylk = ec_gf_fentrylk,
+                           .xattrop = ec_gf_xattrop,
+                           .fxattrop = ec_gf_fxattrop,
+                           .setattr = ec_gf_setattr,
+                           .fsetattr = ec_gf_fsetattr,
+                           .fallocate = ec_gf_fallocate,
+                           .discard = ec_gf_discard,
+                           .zerofill = ec_gf_zerofill,
+                           .seek = ec_gf_seek,
+                           .ipc = ec_gf_ipc};
+
+struct xlator_cbks cbks = {.forget = ec_gf_forget,
+                           .release = ec_gf_release,
+                           .releasedir = ec_gf_releasedir};
+
+struct xlator_dumpops dumpops = {.priv = ec_dump_private};
+
+struct volume_options options[] = {
+    {.key = {"redundancy"},
+     .type = GF_OPTION_TYPE_INT,
+     .default_value = "{{ volume.redundancy }}",
+     .description = "Maximum number of bricks that can fail "
+                    "simultaneously without losing data."},
+    {
+        .key = {"self-heal-daemon"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .description = "self-heal daemon enable/disable",
+        .default_value = "enable",
+        .op_version = {GD_OP_VERSION_3_7_0},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"disperse"},
+    },
+    {.key = {"iam-self-heal-daemon"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description = "This option differentiates if the disperse "
+                    "translator is running as part of self-heal-daemon "
+                    "or not."},
+    {.key = {"eager-lock"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .op_version = {GD_OP_VERSION_3_7_10},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"disperse"},
+     .description = "Enable/Disable eager lock for regular files on a "
+                    "disperse volume. If a fop takes a lock and completes "
+                    "its operation, it waits for next 1 second before "
+                    "releasing the lock, to see if the lock can be reused "
+                    "for next fop from the same client. If ec finds any lock "
+                    "contention within 1 second it releases the lock "
+                    "immediately before time expires. This improves the "
+                    "performance of file operations. However, as it takes "
+                    "lock on first brick, for few operations like read, "
+                    "discovery of lock contention might take long time and "
+                    "can actually degrade the performance. If eager lock is "
+                    "disabled, lock will be released as soon as fop "
+                    "completes."},
+    {.key = {"other-eager-lock"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .op_version = {GD_OP_VERSION_3_13_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"disperse"},
+     .description = "It's equivalent to the eager-lock option but for non "
+                    "regular files."},
+    {.key = {"eager-lock-timeout"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = 60,
+     .default_value = "1",
+     .op_version = {GD_OP_VERSION_4_0_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"disperse", "locks", "timeout"},
+     .description = "Maximum time (in seconds) that a lock on an inode is "
+                    "kept held if no new operations on the inode are "
+                    "received."},
+    {.key = {"other-eager-lock-timeout"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = 60,
+     .default_value = "1",
+     .op_version = {GD_OP_VERSION_4_0_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"disperse", "locks", "timeout"},
+     .description = "It's equivalent to eager-lock-timeout option but for "
+                    "non regular files."},
+    {
+        .key = {"background-heals"},
+        .type = GF_OPTION_TYPE_INT,
+        .min = 0, /*Disabling background heals*/
+        .max = 256,
+        .default_value = "8",
+        .op_version = {GD_OP_VERSION_3_7_3},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+        .tags = {"disperse"},
+        .description = "This option can be used to control number of parallel"
+                       " heals",
+    },
+    {
+        .key = {"heal-wait-qlength"},
+        .type = GF_OPTION_TYPE_INT,
+        .min = 0,
+        .max =
+            65536, /*Around 100MB as of now with sizeof(ec_fop_data_t) at 1800*/
+        .default_value = "128",
+        .op_version = {GD_OP_VERSION_3_7_3},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+        .tags = {"disperse"},
+        .description = "This option can be used to control number of heals"
+                       " that can wait",
+    },
+    {.key = {"heal-timeout"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 60,
+     .max = INT_MAX,
+     .default_value = "600",
+     .op_version = {GD_OP_VERSION_3_7_3},
+     .flags = OPT_FLAG_SETTABLE,
+     .tags = {"disperse"},
+     .description = "time interval for checking the need to self-heal "
+                    "in self-heal-daemon"},
+    {
+        .key = {"read-policy"},
+        .type = GF_OPTION_TYPE_STR,
+        .value = {"round-robin", "gfid-hash"},
+        .default_value = "gfid-hash",
+        .op_version = {GD_OP_VERSION_3_7_6},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+        .tags = {"disperse"},
+        .description =
+            "inode-read fops happen only on 'k' number of bricks in"
+            " n=k+m disperse subvolume. 'round-robin' selects the read"
+            " subvolume using round-robin algo. 'gfid-hash' selects read"
+            " subvolume based on hash of the gfid of that file/directory.",
+    },
+    {.key = {"shd-max-threads"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = 64,
+     .default_value = "1",
+     .op_version = {GD_OP_VERSION_3_9_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"disperse"},
+     .description = "Maximum number of parallel heals SHD can do per local "
+                    "brick.  This can substantially lower heal times, "
+                    "but can also crush your bricks if you don't have "
+                    "the storage hardware to support this."},
+    {.key = {"shd-wait-qlength"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = 65536,
+     .default_value = "1024",
+     .op_version = {GD_OP_VERSION_3_9_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"disperse"},
+     .description = "This option can be used to control number of heals"
+                    " that can wait in SHD per subvolume"},
+    {.key = {"cpu-extensions"},
+     .type = GF_OPTION_TYPE_STR,
+     .value = {"none", "auto", "x64", "sse", "avx"},
+     .default_value = "auto",
+     .op_version = {GD_OP_VERSION_3_9_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"disperse"},
+     .description = "force the cpu extensions to be used to accelerate the "
+                    "galois field computations."},
+    {.key = {"self-heal-window-size"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = 1024,
+     .default_value = "1",
+     .op_version = {GD_OP_VERSION_3_11_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"disperse"},
+     .description = "Maximum number blocks(128KB) per file for which "
+                    "self-heal process would be applied simultaneously."},
+    {.key = {"optimistic-change-log"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .op_version = {GD_OP_VERSION_3_10_1},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT,
+     .tags = {"disperse"},
+     .description = "Set/Unset dirty flag for every update fop at the start"
+                    "of the fop. If OFF, this option impacts performance of"
+                    "entry  operations or metadata operations as it will"
+                    "set dirty flag at the start and unset it at the end of"
+                    "ALL update fop. If ON and all the bricks are good,"
+                    "dirty flag will be set at the start only for file fops"
+                    "For metadata and entry fops dirty flag will not be set"
+                    "at the start, if all the bricks are good. This does"
+                    "not impact performance for metadata operations and"
+                    "entry operation but has a very small window to miss"
+                    "marking entry as dirty in case it is required to be"
+                    "healed"},
+    {.key = {"parallel-writes"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .description = "This controls if writes can be wound in parallel as long"
+                    "as it doesn't modify same stripes"},
+    {.key = {"stripe-cache"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 0, /*Disabling stripe_cache*/
+     .max = EC_STRIPE_CACHE_MAX_SIZE,
+     .default_value = "4",
+     .description = "This option will keep the last stripe of write fop"
+                    "in memory. If next write falls in this stripe, we need"
+                    "not to read it again from backend and we can save READ"
+                    "fop going over the network. This will improve performance,"
+                    "specially for sequential writes. However, this will also"
+                    "lead to extra memory consumption, maximum "
+                    "(cache size * stripe size) Bytes per open file."},
+    {
+        .key = {"quorum-count"},
+        .type = GF_OPTION_TYPE_INT,
+        .default_value = "0",
+        .description =
+            "This option can be used to define how many successes on"
+            "the bricks constitute a success to the application. This"
+            " count should be in the range"
+            "[disperse-data-count,  disperse-count] (inclusive)",
+    },
+    {
+        .key = {"ec-read-mask"},
+        .type = GF_OPTION_TYPE_STR,
+        .default_value = NULL,
+        .description = "This option can be used to choose which bricks can be"
+                       " used for reading data/metadata of a file/directory",
+    },
+    {
+        .key = {NULL},
+    },
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1},
+    .dumpops = &dumpops,
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "disperse",
+    .category = GF_MAINTAINED,
+};
diff --git a/xlators/cluster/ec/src/ec.h b/xlators/cluster/ec/src/ec.h
new file mode 100644
index 00000000000..6f6de6d5981
--- /dev/null
+++ b/xlators/cluster/ec/src/ec.h
@@ -0,0 +1,34 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_H__
+#define __EC_H__
+
+#include "ec-method.h"
+
+#define EC_XATTR_PREFIX "trusted.ec."
+#define EC_XATTR_CONFIG EC_XATTR_PREFIX "config"
+#define EC_XATTR_SIZE EC_XATTR_PREFIX "size"
+#define EC_XATTR_VERSION EC_XATTR_PREFIX "version"
+#define EC_XATTR_HEAL EC_XATTR_PREFIX "heal"
+#define EC_XATTR_HEAL_NEW EC_XATTR_PREFIX "heal-new"
+#define EC_XATTR_DIRTY EC_XATTR_PREFIX "dirty"
+#define EC_STRIPE_CACHE_MAX_SIZE 10
+#define EC_VERSION_SIZE 2
+#define EC_SHD_INODE_LRU_LIMIT 10
+
+#define EC_MAX_FRAGMENTS EC_METHOD_MAX_FRAGMENTS
+/* The maximum number of nodes is derived from the maximum allowed fragments
+ * using the rule that redundancy cannot be equal or greater than the number
+ * of fragments.
+ */
+#define EC_MAX_NODES min(EC_MAX_FRAGMENTS * 2 - 1, EC_METHOD_MAX_NODES)
+
+#endif /* __EC_H__ */
diff --git a/xlators/cluster/ha/src/Makefile.am b/xlators/cluster/ha/src/Makefile.am
deleted file mode 100644
index 5c1364b7f9d..00000000000
--- a/xlators/cluster/ha/src/Makefile.am
+++ /dev/null
@@ -1,16 +0,0 @@
-xlator_LTLIBRARIES = ha.la
-xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/testing/cluster
-
-ha_la_LDFLAGS = -module -avoid-version 
-
-ha_la_SOURCES = ha-helpers.c ha.c
-ha_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
-
-noinst_HEADERS = ha.h
-
-AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src
-
-AM_CFLAGS = -Wall $(GF_CFLAGS)
-
-CLEANFILES = 
-
diff --git a/xlators/cluster/ha/src/ha-helpers.c b/xlators/cluster/ha/src/ha-helpers.c
deleted file mode 100644
index 19be1ed2773..00000000000
--- a/xlators/cluster/ha/src/ha-helpers.c
+++ /dev/null
@@ -1,194 +0,0 @@
-/*
-   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
-   This file is part of GlusterFS.
-
-   This file is licensed to you under your choice of the GNU Lesser
-   General Public License, version 3 or any later version (LGPLv3 or
-   later), or the GNU General Public License, version 2 (GPLv2), in all
-   cases as published by the Free Software Foundation.
-*/
-#include "xlator.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "dict.h"
-#include "compat-errno.h"
-#include "ha.h"
-
-#define HA_TRANSPORT_NOTCONN(_ret, _errno, _fd) \
-	((_ret == -1) && (_fd ? (_errno == EBADFD):(_errno == ENOTCONN)))
-
-int ha_alloc_init_fd (call_frame_t *frame, fd_t *fd)
-{
-	ha_local_t *local = NULL;
-	int i = -1;
-	ha_private_t *pvt = NULL;
-	int child_count = 0;
-	int ret = -1;
-	hafd_t *hafdp = NULL;
-	xlator_t *this = NULL;
-	uint64_t tmp_hafdp = 0;
-
-	this = frame->this;
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-
-	if (local == NULL) {
-		ret = fd_ctx_get (fd, this, &tmp_hafdp);
-		if (ret < 0) {
-			goto out;
-		}
-		hafdp = (hafd_t *)(long)tmp_hafdp;
-                local = frame->local = GF_CALLOC (1, sizeof (*local), 
-                                                  gf_ha_mt_ha_local_t);
-		if (local == NULL) {
-			ret = -ENOMEM;
-			goto out;
-		}
-                local->state = GF_CALLOC (1, child_count, 
-                                          gf_ha_mt_child_count);
-		if (local->state == NULL) {
-			ret = -ENOMEM;
-			goto out;
-		}
-
-		/* take care of the preferred subvolume */
-		if (pvt->pref_subvol == -1)
-			local->active = hafdp->active;
-		else
-			local->active = pvt->pref_subvol;
-
-		LOCK (&hafdp->lock);
-		memcpy (local->state, hafdp->fdstate, child_count);
-		UNLOCK (&hafdp->lock);
-
-		/* in case the preferred subvolume is down */
-		if ((local->active != -1) && (local->state[local->active] == 0))
-			local->active = -1;
-
-		for (i = 0; i < child_count; i++) {
-			if (local->state[i]) {
-				if (local->active == -1)
-					local->active = i;
-				local->tries++;
-			}
-		}
-		if (local->active == -1) {
-			ret = -ENOTCONN;
-			goto out;
-		}
-		local->fd = fd_ref (fd);
-	}
-	ret = 0;
-out:
-	return ret;
-}
-
-int ha_handle_cbk (call_frame_t *frame, void *cookie, int op_ret, int op_errno) 
-{
-	xlator_t *xl = NULL;
-	ha_private_t *pvt = NULL;
-	xlator_t **children = NULL;
-	int prev_child = -1;
-	hafd_t *hafdp = NULL;
-	int ret = -1;
-	call_stub_t *stub = NULL;
-	ha_local_t *local = NULL;
-	uint64_t tmp_hafdp = 0;
-
-	xl = frame->this;
-	pvt = xl->private;
-	children = pvt->children;
-	prev_child = (long) cookie;
-	local = frame->local;
-
-	if (op_ret == -1) {
-		gf_log (xl->name, GF_LOG_ERROR ,"(child=%s) (op_ret=%d op_errno=%s)",
-			children[prev_child]->name, op_ret, strerror (op_errno));
-	}
-
-	if (HA_TRANSPORT_NOTCONN (op_ret, op_errno, (local->fd))) {
-		ret = 0;
-		if (local->fd) {
-			ret = fd_ctx_get (local->fd, xl, &tmp_hafdp);
-		}
-		hafdp = (hafd_t *)(long)tmp_hafdp;		
-		if (ret == 0) {
-			if (local->fd) {
-				LOCK(&hafdp->lock);
-				hafdp->fdstate[prev_child] = 0;
-				UNLOCK(&hafdp->lock);
-			}
-			local->tries--;
-			if (local->tries != 0) {
-				while (1) {
-					local->active = (local->active + 1) % pvt->child_count;
-					if (local->state[local->active])
-						break;
-				}
-				stub = local->stub;
-				local->stub = NULL;
-				call_resume (stub);
-				return -1;
-			}
-		}
-	}
-	if (local->stub) {
-		call_stub_destroy (local->stub);
-                local->stub = NULL;
-        }
-
-	if (local->fd) {
-		GF_FREE (local->state);
-                local->state = NULL;
-
-		fd_unref (local->fd);
-                local->fd = NULL;
-	}
-	return 0;
-}
-
-int ha_alloc_init_inode (call_frame_t *frame, inode_t *inode)
-{
-	int i = -1;
-	ha_private_t *pvt = NULL;
-	xlator_t *xl = NULL;
-	int ret = -1;
-	ha_local_t *local = NULL;
-	uint64_t tmp_state = 0;
-
-	xl = frame->this;
-	pvt = xl->private;
-	local = frame->local;
-
-	if (local == NULL) {
-                local = frame->local = GF_CALLOC (1, sizeof (*local), 
-                                                  gf_ha_mt_ha_local_t);
-		if (local == NULL) {
-			ret = -ENOMEM;
-			goto out;
-		}
-		local->active = pvt->pref_subvol;
-		ret = inode_ctx_get (inode, xl, &tmp_state);
-		if (ret < 0) {
-			goto out;
-		}
-		local->state = (char *)(long)tmp_state;
-		if (local->active != -1 && local->state[local->active] == 0)
-			local->active = -1;
-		for (i = 0; i < pvt->child_count; i++) {
-			if (local->state[i]) {
-				if (local->active == -1)
-					local->active = i;
-				local->tries++;
-			}
-		}
-		if (local->active == -1) {
-			ret = -ENOTCONN;
-			goto out;
-		}
-	}
-	ret = 0;
-out:
-	return ret;
-}
diff --git a/xlators/cluster/ha/src/ha-mem-types.h b/xlators/cluster/ha/src/ha-mem-types.h
deleted file mode 100644
index e5e97d237dc..00000000000
--- a/xlators/cluster/ha/src/ha-mem-types.h
+++ /dev/null
@@ -1,26 +0,0 @@
-/*
-   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
-   This file is part of GlusterFS.
-
-   This file is licensed to you under your choice of the GNU Lesser
-   General Public License, version 3 or any later version (LGPLv3 or
-   later), or the GNU General Public License, version 2 (GPLv2), in all
-   cases as published by the Free Software Foundation.
-*/
-
-#ifndef __HA_MEM_TYPES_H__
-#define __HA_MEM_TYPES_H__
-
-#include "mem-types.h"
-
-enum gf_ha_mem_types_ {
-        gf_ha_mt_ha_local_t = gf_common_mt_end + 1,
-        gf_ha_mt_hafd_t,
-        gf_ha_mt_char,
-        gf_ha_mt_child_count,
-        gf_ha_mt_xlator_t,
-        gf_ha_mt_ha_private_t,
-        gf_ha_mt_end
-};
-#endif
-
diff --git a/xlators/cluster/ha/src/ha.c b/xlators/cluster/ha/src/ha.c
deleted file mode 100644
index 3eccb516b65..00000000000
--- a/xlators/cluster/ha/src/ha.c
+++ /dev/null
@@ -1,4009 +0,0 @@
-/*
-   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
-   This file is part of GlusterFS.
-
-   This file is licensed to you under your choice of the GNU Lesser
-   General Public License, version 3 or any later version (LGPLv3 or
-   later), or the GNU General Public License, version 2 (GPLv2), in all
-   cases as published by the Free Software Foundation.
-*/
-/* generate errors randomly, code is simple now, better alogorithm
- * can be written to decide what error to be returned and when
- */
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "xlator.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "dict.h"
-#include "compat-errno.h"
-#include "ha.h"
-
-/*
- * TODO:
- * - dbench fails if ha over server side afr
- * - lock calls - lock on all subvols.
- * - support preferred-subvolume option. code already there.
- * - do not alloc the call-stub in case only one subvol is up.
- */
-
-void
-ha_local_wipe (ha_local_t *local)
-{
-        if (local->stub) {
-                call_stub_destroy (local->stub);
-                local->stub = NULL;
-        }
-
-        if (local->state) {
-                GF_FREE (local->state);
-                local->state = NULL;
-        }
-
-        if (local->dict) {
-                dict_unref (local->dict);
-                local->dict = NULL;
-        }
-
-        loc_wipe (&local->loc);
-
-        if (local->fd) {
-                fd_unref (local->fd);
-                local->fd = NULL;
-        }
-
-        if (local->inode) {
-                inode_unref (local->inode);
-                local->inode = NULL;
-        }
-
-        GF_FREE (local);
-        return;
-}
-
-
-int
-ha_forget (xlator_t *this,
-	   inode_t *inode)
-{
-	uint64_t stateino = 0;
-	char *state = NULL;
-	if (!inode_ctx_del (inode, this, &stateino)) {
-		state =  ((char *)(long)stateino);
-		GF_FREE (state);
-	}
-
-	return 0;
-
-}
-
-int32_t 
-ha_lookup_cbk (call_frame_t *frame,
-	       void *cookie,
-	       xlator_t *this,
-	       int32_t op_ret,
-	       int32_t op_errno,
-	       inode_t *inode,
-	       struct iatt *buf,
-               dict_t *dict,
-               struct iatt *postparent)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	int child_count = 0, i = 0, callcnt = 0;
-	char *state = NULL;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	uint64_t tmp_state = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-	children = pvt->children;
-
-	for (i = 0; i < child_count; i++) {
-		if (pvt->children[i] == prev_frame->this)
-			break;
-	}
-	if ((op_ret == -1) && (op_errno != ENOENT)) {
-		gf_log (this->name, GF_LOG_ERROR, "(child=%s) (op_ret=%d op_errno=%s)", 
-			  children[i]->name, op_ret, strerror (op_errno));
-	}
-	inode_ctx_get (local->inode, this, &tmp_state);
-	state = (char *)(long)tmp_state;
-
-	LOCK (&frame->lock);
-	if (local->revalidate == 1) {
-		if ((!op_ret) != state[i]) {
-			local->revalidate_error = 1;
-			gf_log (this->name, GF_LOG_DEBUG, "revalidate error on %s", 
-				pvt->children[i]->name);
-		}
-	} else {
-		if (op_ret == 0) {
-			state[i] = 1;
-		}
-	}
-	if (local->op_ret == -1 && op_ret == 0) {
-		local->op_ret = 0;
-		local->buf = *buf;
-                local->postparent = *postparent;
-		if (dict)
-			local->dict = dict_ref (dict);
-	}
-	if (op_ret == -1 && op_ret != ENOTCONN)
-		local->op_errno = op_errno;
-	callcnt = --local->call_count;
-	UNLOCK (&frame->lock);
-
-	if (callcnt == 0) {
-		dict_t *ctx = local->dict;
-		inode_t *inode = local->inode;
-		if (local->revalidate_error == 1) {
-			local->op_ret = -1;
-			local->op_errno = EIO;
-			gf_log (this->name, GF_LOG_DEBUG, "revalidate error, returning EIO");
-		}
-		STACK_UNWIND (frame,
-			      local->op_ret,
-			      local->op_errno,
-			      inode,
-			      &local->buf,
-			      ctx,
-                              &local->postparent);
-		if (inode)
-			inode_unref (inode);
-		if (ctx)
-			dict_unref (ctx);
-	}
-	return 0;
-}
-
-int32_t
-ha_lookup (call_frame_t *frame,
-	   xlator_t *this,
-	   loc_t *loc,
-	   dict_t *xattr_req)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	int child_count = 0, i = 0;
-	char *state = NULL;
-	xlator_t **children = NULL;
-	int ret = -1;
-        int32_t op_errno = EINVAL;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	children = pvt->children;
-
-	frame->local = local = GF_CALLOC (1, sizeof (*local), 
-                                          gf_ha_mt_ha_local_t);
-        if (!local) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto unwind;
-        }
-
-	child_count = pvt->child_count;
-	local->inode = inode_ref (loc->inode);
-
-	ret = inode_ctx_get (loc->inode, this, NULL);
-	if (ret) {
-		state = GF_CALLOC (1, child_count, gf_ha_mt_child_count);
-                if (state == NULL) {
-                        gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                        op_errno = ENOMEM;
-                        goto unwind;
-                }
-
-		inode_ctx_put (loc->inode, this, (uint64_t)(long)state);
-	} else
-		local->revalidate = 1;
-
-	local->op_ret = -1;
-	local->op_errno = ENOTCONN;
-	local->call_count = child_count;
-
-	for (i = 0; i < child_count; i++) {
-		STACK_WIND (frame,
-			    ha_lookup_cbk,
-			    children[i],
-			    children[i]->fops->lookup,
-			    loc,
-			    xattr_req);
-	}
-	return 0;
-
-unwind:
-        local = frame->local;
-        frame->local = NULL;
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL, NULL, NULL);
-
-        ha_local_wipe (local);
-        return 0;
-}
-
- int32_t
-ha_stat_cbk (call_frame_t *frame,
-	     void *cookie,
-	     xlator_t *this,
-	     int32_t op_ret,
-	     int32_t op_errno,
-	     struct iatt *buf)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno,
-			      buf);
-	}
-	return 0;
-}
-
-int32_t
-ha_stat (call_frame_t *frame,
-	 xlator_t *this,
-	 loc_t *loc)
-{
-	ha_local_t *local = NULL;
-	int op_errno = ENOTCONN;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_stat_stub (frame, ha_stat, loc);
-
-	STACK_WIND_COOKIE (frame,
-			   ha_stat_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->stat,
-			   loc);
-	return 0;
-err:
-	STACK_UNWIND (frame, -1, op_errno, NULL);
-	return 0;	
-}
-
-int32_t
-ha_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                int32_t op_ret, int32_t op_errno, struct iatt *statpre,
-                struct iatt *statpost)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame, op_ret, op_errno, statpre, statpost);
-	}
-	return 0;
-}
-
-
-int32_t
-ha_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *stbuf,
-            int32_t valid)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_setattr_stub (frame, ha_setattr, loc, stbuf, valid);
-
-	STACK_WIND_COOKIE (frame,
-			   ha_setattr_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->setattr,
-			   loc, stbuf, valid);
-	return 0;
-err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-	return 0;
-}
-
-
-int32_t
-ha_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf,
-             int32_t valid)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_fsetattr_stub (frame, ha_fsetattr, fd, stbuf, valid);
-
-	STACK_WIND_COOKIE (frame,	      
-			   ha_setattr_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->fsetattr,
-			   fd, stbuf, valid);
-	return 0;
-err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-	return 0;
-}
-
-
-int32_t
-ha_truncate_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 struct iatt *prebuf,
-                 struct iatt *postbuf)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno,
-			      prebuf,
-                              postbuf);
-	}
-	return 0;
-}
-
-int32_t
-ha_truncate (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc,
-	     off_t offset)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_truncate_stub (frame, ha_truncate, loc, offset);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_truncate_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->truncate,
-			   loc,
-			   offset);
-	return 0;
-err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-	return 0;
-}
-
- int32_t
-ha_ftruncate_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  struct iatt *prebuf,
-                  struct iatt *postbuf)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno,
-			      prebuf, postbuf);
-	}
-	return 0;
-}
-
-int32_t
-ha_ftruncate (call_frame_t *frame,
-	      xlator_t *this,
-	      fd_t *fd,
-	      off_t offset)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_ftruncate_stub (frame, ha_ftruncate, fd, offset);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_ftruncate_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->ftruncate,
-			   fd,
-			   offset);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
-int32_t
-ha_access_cbk (call_frame_t *frame,
-	       void *cookie,
-	       xlator_t *this,
-	       int32_t op_ret,
-	       int32_t op_errno)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno);
-	}
-	return 0;
-}
-
-int32_t
-ha_access (call_frame_t *frame,
-	   xlator_t *this,
-	   loc_t *loc,
-	   int32_t mask)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_access_stub (frame, ha_access, loc, mask);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_access_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->access,
-			   loc,
-			   mask);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
-
- int32_t
-ha_readlink_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 const char *path,
-                 struct iatt *sbuf)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno,
-			      path,
-                              sbuf);
-	}
-	return 0;
-}
-
-int32_t
-ha_readlink (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc,
-	     size_t size)
-{
-	ha_local_t *local = frame->local;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_readlink_stub (frame, ha_readlink, loc, size);
-
-	STACK_WIND_COOKIE (frame,
-			   ha_readlink_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->readlink,
-			   loc,
-			   size);
-	return 0;
-err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-	return 0;
-}
-
-int
-ha_mknod_lookup_cbk (call_frame_t *frame,
-		     void *cookie,
-		     xlator_t *this,
-		     int32_t op_ret,
-		     int32_t op_errno,
-		     inode_t *inode,
-		     struct iatt *buf,
-                     dict_t *dict,
-                     struct iatt *postparent)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	char *stateino = NULL;
-	int child_count = 0, i = 0, cnt = 0, ret = 0;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	uint64_t tmp_stateino = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-	children = pvt->children;
-
-	for (i = 0; i < child_count; i++)
-		if (prev_frame->this == children[i])
-			break;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"(path=%s) (op_ret=%d op_errno=%d)", 
-			local->stub->args.mknod.loc.path, op_ret, op_errno);
-	}
-	ret = inode_ctx_get (local->stub->args.mknod.loc.inode, 
-			     this, &tmp_stateino);
-	stateino = (char *)(long)tmp_stateino;
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"unwind(-1), inode_ctx_get() error");
-		/* It is difficult to handle this error at this stage
-		 * as we still expect more cbks, we can't return as
-		 * of now
-		 */
-	} else if (op_ret == 0) {
-		stateino[i] = 1;
-	}
-	LOCK (&frame->lock);
-	cnt = --local->call_count;
-	UNLOCK (&frame->lock);
-
-	if (cnt == 0) {
-		call_stub_t *stub = local->stub;
-		GF_FREE (local->state);
-		STACK_UNWIND (frame,
-			      local->op_ret,
-			      local->op_errno,
-			      local->stub->args.mknod.loc.inode,
-			      &local->buf, &local->preparent,
-                              &local->postparent);
-		call_stub_destroy (stub);
-	}
-	return 0;
-}
-
-int32_t
-ha_mknod_cbk (call_frame_t *frame,
-	      void *cookie,
-	      xlator_t *this,
-	      int32_t op_ret,
-	      int32_t op_errno,
-	      inode_t *inode,
-              struct iatt *buf,
-              struct iatt *preparent,
-              struct iatt *postparent)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	char *stateino = NULL;
-	int child_count = 0, i = 0, cnt = 0, ret = 0;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	uint64_t tmp_stateino = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-	children = pvt->children;
-
-	for (i = 0; i < child_count; i++)
-		if (prev_frame->this == children[i])
-			break;
-
-	if (op_ret == -1) {
-		local->op_errno = op_errno;
-		gf_log (this->name, GF_LOG_ERROR, "(path=%s) (op_ret=%d op_errno=%d)", local->stub->args.mknod.loc.path, op_ret, op_errno);
-	}
-
-	ret = inode_ctx_get (local->stub->args.mknod.loc.inode, 
-			     this, &tmp_stateino);
-	stateino = (char *)(long)tmp_stateino;
-
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR, "inode_ctx_get() error");
-		/* FIXME: handle the case */
-	}
-	if (op_ret == 0) {
-		stateino[i] = 1;
-		local->op_ret = 0;
-		local->first_success = 1;
-		local->buf = *buf;
-                local->preparent = *preparent;
-                local->postparent = *postparent;
-	}
-	cnt = --local->call_count;
-	for (i = local->active + 1; i < child_count; i++) {
-		if (local->state[i])
-			break;
-	}
-
-	if (cnt == 0 || i == child_count) {
-		call_stub_t *stub = local->stub;
-		GF_FREE (local->state);
-		stub = local->stub;
-		STACK_UNWIND (frame, local->op_ret, local->op_errno,
-                              local->stub->args.mknod.loc.inode, &local->buf,
-                              &local->preparent, &local->postparent);
-		call_stub_destroy (stub);
-		return 0;
-	}
-
-	local->active = i;
-
-	if (local->first_success == 0) {
-		STACK_WIND (frame,
-			    ha_mknod_cbk,
-			    children[i],
-			    children[i]->fops->mknod,
-			    &local->stub->args.mknod.loc,
-			    local->stub->args.mknod.mode,
-			    local->stub->args.mknod.rdev);
-		return 0;
-	}
-	cnt = local->call_count;
-
-	for (; i < child_count; i++) {
-		if (local->state[i]) {
-			STACK_WIND (frame,
-				    ha_mknod_lookup_cbk,
-				    children[i],
-				    children[i]->fops->lookup,
-				    &local->stub->args.mknod.loc,
-				    0);
-			if (--cnt == 0)
-				break;
-		}
-	}
-	return 0;
-}
-
-int32_t
-ha_mknod (call_frame_t *frame,
-	  xlator_t *this,
-	  loc_t *loc,
-	  mode_t mode,
-	  dev_t rdev)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	int child_count = 0, i = 0;
-	char *stateino = NULL;
-        int32_t op_errno = EINVAL;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-
-	frame->local = local = GF_CALLOC (1, sizeof (*local), 
-                                          gf_ha_mt_ha_local_t);
-        if (!local) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	local->stub = fop_mknod_stub (frame, ha_mknod, loc, mode, rdev);
-        if (!local->stub) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	local->op_ret = -1;
-	local->op_errno = ENOTCONN;
-	local->state = GF_CALLOC (1, child_count, gf_ha_mt_char);
-        if (!local->state) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	memcpy (local->state, pvt->state, child_count);
-	local->active = -1;
-
-	stateino = GF_CALLOC (1, child_count, gf_ha_mt_char);
-        if (!stateino) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	inode_ctx_put (loc->inode, this, (uint64_t)(long)stateino);
-
-	for (i = 0; i < child_count; i++) {
-		if (local->state[i]) {
-			local->call_count++;
-			if (local->active == -1) 
-				local->active = i;
-		}
-	}
-
-	STACK_WIND (frame,
-		    ha_mknod_cbk,
-		    HA_ACTIVE_CHILD(this, local),
-		    HA_ACTIVE_CHILD(this, local)->fops->mknod,
-		    loc, mode, rdev);
-	return 0;
-
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL, NULL, NULL);
-        ha_local_wipe (local);
-        return 0;
-}
-
-
-int
-ha_mkdir_lookup_cbk (call_frame_t *frame,
-		     void *cookie,
-		     xlator_t *this,
-		     int32_t op_ret,
-		     int32_t op_errno,
-		     inode_t *inode,
-		     struct iatt *buf,
-                     dict_t *dict,
-                     struct iatt *postparent)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	char *stateino = NULL;
-	int child_count = 0, i = 0, cnt = 0;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	uint64_t tmp_stateino = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-	children = pvt->children;
-
-	for (i = 0; i < child_count; i++)
-		if (prev_frame->this == children[i])
-			break;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR, "(path=%s) (op_ret=%d op_errno=%d)", local->stub->args.mkdir.loc.path, op_ret, op_errno);
-	}
-	inode_ctx_get (local->stub->args.mkdir.loc.inode, 
-		       this, &tmp_stateino);  
-	stateino = (char *)(long)tmp_stateino;
-
-	if (op_ret == 0)
-		stateino[i] = 1;
-
-	LOCK (&frame->lock);
-	cnt = --local->call_count;
-	UNLOCK (&frame->lock);
-
-	if (cnt == 0) {
-		call_stub_t *stub = local->stub;
-		GF_FREE (local->state);
-		STACK_UNWIND (frame,
-			      local->op_ret,
-			      local->op_errno,
-			      local->stub->args.mkdir.loc.inode, &local->buf,
-			      &local->preparent, &local->postparent);
-		call_stub_destroy (stub);
-	}
-	return 0;
-}
-
-int32_t
-ha_mkdir_cbk (call_frame_t *frame,
-	      void *cookie,
-	      xlator_t *this,
-	      int32_t op_ret,
-	      int32_t op_errno,
-	      inode_t *inode,
-              struct iatt *buf,
-              struct iatt *preparent,
-              struct iatt *postparent)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	char *stateino = NULL;
-	int child_count = 0, i = 0, cnt = 0;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	uint64_t tmp_stateino = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-	children = pvt->children;
-	
-	for (i = 0; i < child_count; i++)
-		if (prev_frame->this == children[i])
-			break;
-
-	if (op_ret == -1) {
-		local->op_errno = op_errno;
-		gf_log (this->name, GF_LOG_ERROR, "(path=%s) (op_ret=%d op_errno=%d)", local->stub->args.mkdir.loc.path, op_ret, op_errno);
-	}
-
-	inode_ctx_get (local->stub->args.mkdir.loc.inode, 
-		       this, &tmp_stateino);
-	stateino = (char *)(long)tmp_stateino;
-
-	if (op_ret == 0) {
-		stateino[i] = 1;
-		local->op_ret = 0;
-		local->first_success = 1;
-		local->buf = *buf;
-                local->preparent = *preparent;
-                local->postparent = *postparent;
-	}
-	cnt = --local->call_count;
-	for (i = local->active + 1; i < child_count; i++) {
-		if (local->state[i])
-			break;
-	}
-
-	if (cnt == 0 || i == child_count) {
-		call_stub_t *stub = local->stub;
-		GF_FREE (local->state);
-		stub = local->stub;
-		STACK_UNWIND (frame, local->op_ret, local->op_errno,
-                              local->stub->args.mkdir.loc.inode, &local->buf,
-                              &local->preparent, &local->postparent);
-		call_stub_destroy (stub);
-		return 0;
-	}
-
-	local->active = i;
-
-	if (local->first_success == 0) {
-		STACK_WIND (frame,
-			    ha_mkdir_cbk,
-			    children[i],
-			    children[i]->fops->mkdir,
-			    &local->stub->args.mkdir.loc,
-			    local->stub->args.mkdir.mode);
-		return 0;
-	}
-	cnt = local->call_count;
-
-	for (; i < child_count; i++) {
-		if (local->state[i]) {
-			STACK_WIND (frame,
-				    ha_mkdir_lookup_cbk,
-				    children[i],
-				    children[i]->fops->lookup,
-				    &local->stub->args.mkdir.loc,
-				    0);
-			if (--cnt == 0)
-				break;
-		}
-	}
-	return 0;
-}
-
-int32_t
-ha_mkdir (call_frame_t *frame,
-	  xlator_t *this,
-	  loc_t *loc,
-	  mode_t mode)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	int child_count = 0, i = 0;
-	char *stateino = NULL;
-        int32_t op_errno = EINVAL;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-
-	frame->local = local = GF_CALLOC (1, sizeof (*local), 
-                                          gf_ha_mt_ha_local_t);
-        if (!frame->local) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	local->stub = fop_mkdir_stub (frame, ha_mkdir, loc, mode);
-        if (!local->stub) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	local->op_ret = -1;
-	local->op_errno = ENOTCONN;
-	local->state = GF_CALLOC (1, child_count, gf_ha_mt_char);
-        if (!local->state) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	memcpy (local->state, pvt->state, child_count);
-	local->active = -1;
-
-	stateino = GF_CALLOC (1, child_count, gf_ha_mt_char);
-        if (!stateino) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	inode_ctx_put (loc->inode, this, (uint64_t)(long)stateino);
-	for (i = 0; i < child_count; i++) {
-		if (local->state[i]) {
-			local->call_count++;
-			if (local->active == -1)
-				local->active = i;
-		}
-	}
-
-	STACK_WIND (frame,
-		    ha_mkdir_cbk,
-		    HA_ACTIVE_CHILD(this, local),
-		    HA_ACTIVE_CHILD(this, local)->fops->mkdir,
-		    loc, mode);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL, NULL, NULL);
-        ha_local_wipe (local);
-        return 0;
-}
-
- int32_t
-ha_unlink_cbk (call_frame_t *frame,
-	       void *cookie,
-	       xlator_t *this,
-	       int32_t op_ret,
-	       int32_t op_errno,
-               struct iatt *preparent,
-               struct iatt *postparent)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-	if (ret == 0) {
-		STACK_UNWIND (frame, op_ret, op_errno, preparent, postparent);
-	}
-	return 0;
-}
-
-int32_t
-ha_unlink (call_frame_t *frame,
-	   xlator_t *this,
-	   loc_t *loc)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_unlink_stub (frame, ha_unlink, loc);
-        if (!local->stub) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_unlink_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->unlink,
-			   loc);
-	return 0;
-err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-	return 0;
-}
-
- int32_t
-ha_rmdir_cbk (call_frame_t *frame,
-	      void *cookie,
-	      xlator_t *this,
-	      int32_t op_ret,
-	      int32_t op_errno,
-              struct iatt *preparent,
-              struct iatt *postparent)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno,
-                              preparent,
-                              postparent);
-	}
-	return 0;
-}
-
-int32_t
-ha_rmdir (call_frame_t *frame,
-	  xlator_t *this,
-	  loc_t *loc)
-{
-	ha_local_t *local = frame->local;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_rmdir_stub (frame, ha_rmdir, loc);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_rmdir_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->rmdir,
-			   loc);
-	return 0;
-err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-	return 0;
-}
-
-
-int
-ha_symlink_lookup_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno,
-		       inode_t *inode,
-		       struct iatt *buf,
-                       dict_t *dict,
-                       struct iatt *postparent)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	char *stateino = NULL;
-	int child_count = 0, i = 0, cnt = 0;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	uint64_t tmp_stateino = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-	children = pvt->children;
-
-	for (i = 0; i < child_count; i++)
-		if (prev_frame->this == children[i])
-			break;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR, "(path=%s) (op_ret=%d op_errno=%d)", local->stub->args.symlink.loc.path, op_ret, op_errno);
-	}
-	inode_ctx_get (local->stub->args.symlink.loc.inode,
-		       this, &tmp_stateino);  
-	stateino = (char *)(long)tmp_stateino;
-
-	if (op_ret == 0)
-		stateino[i] = 1;
-
-	LOCK (&frame->lock);
-	cnt = --local->call_count;
-	UNLOCK (&frame->lock);
-
-	if (cnt == 0) {
-		call_stub_t *stub = local->stub;
-		GF_FREE (local->state);
-		STACK_UNWIND (frame,
-			      local->op_ret,
-			      local->op_errno,
-			      local->stub->args.symlink.loc.inode, &local->buf,
-                              &local->preparent, &local->postparent);
-		call_stub_destroy (stub);
-	}
-	return 0;
-}
-
-int32_t
-ha_symlink_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		inode_t *inode,
-                struct iatt *buf,
-                struct iatt *preparent,
-                struct iatt *postparent)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	char *stateino = NULL;
-	int child_count = 0, i = 0, cnt = 0;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	uint64_t tmp_stateino = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-	children = pvt->children;
-
-	for (i = 0; i < child_count; i++)
-		if (prev_frame->this == children[i])
-			break;
-
-	if (op_ret == -1) {
-		local->op_errno = op_errno;
-		gf_log (this->name, GF_LOG_ERROR, "(path=%s) (op_ret=%d op_errno=%d)", local->stub->args.symlink.loc.path, op_ret, op_errno);
-	}
-	inode_ctx_get (local->stub->args.symlink.loc.inode, 
-		       this, &tmp_stateino);
-	stateino = (char *)(long)tmp_stateino;
-
-	if (op_ret == 0) {
-		stateino[i] = 1;
-		local->op_ret = 0;
-		local->first_success = 1;
-		local->buf = *buf;
-                local->preparent = *preparent;
-                local->postparent = *postparent;
-	}
-	cnt = --local->call_count;
-	for (i = local->active + 1; i < child_count; i++) {
-		if (local->state[i])
-			break;
-	}
-
-	if (cnt == 0 || i == child_count) {
-		call_stub_t *stub = local->stub;
-		GF_FREE (local->state);
-		stub = local->stub;
-		STACK_UNWIND (frame, local->op_ret, local->op_errno, 
-			      local->stub->args.symlink.loc.inode, &local->buf,
-                              &local->preparent, &local->postparent);
-		call_stub_destroy (stub);
-		return 0;
-	}
-
-	local->active = i;
-
-	if (local->first_success == 0) {
-		STACK_WIND (frame,
-			    ha_symlink_cbk,
-			    children[i],
-			    children[i]->fops->symlink,
-			    local->stub->args.symlink.linkname,
-			    &local->stub->args.symlink.loc);
-		return 0;
-	}
-	cnt = local->call_count;
-
-	for (; i < child_count; i++) {
-		if (local->state[i]) {
-			STACK_WIND (frame,
-				    ha_symlink_lookup_cbk,
-				    children[i],
-				    children[i]->fops->lookup,
-				    &local->stub->args.symlink.loc,
-				    0);
-			if (--cnt == 0)
-				break;
-		}
-	}
-	return 0;
-}
-
-int32_t
-ha_symlink (call_frame_t *frame,
-	    xlator_t *this,
-	    const char *linkname,
-	    loc_t *loc)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	int child_count = 0, i = 0;
-	char *stateino = NULL;
-        int32_t op_errno = EINVAL;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-
-	frame->local = local = GF_CALLOC (1, sizeof (*local), 
-                                          gf_ha_mt_ha_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	local->stub = fop_symlink_stub (frame, ha_symlink, linkname, loc);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	local->op_ret = -1;
-	local->op_errno = ENOTCONN;
-	local->state = GF_CALLOC (1, child_count, gf_ha_mt_char);
-        if (!local->state) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	memcpy (local->state, pvt->state, child_count);
-	local->active = -1;
-
-	stateino = GF_CALLOC (1, child_count, gf_ha_mt_char);
-        if (!stateino) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	inode_ctx_put (loc->inode, this, (uint64_t)(long)stateino);
-
-	for (i = 0; i < child_count; i++) {
-		if (local->state[i]) {
-			local->call_count++;
-			if (local->active == -1) {
-				local->active = i;
-			}
-		}
-	}
-
-	STACK_WIND (frame,
-		    ha_symlink_cbk,
-		    HA_ACTIVE_CHILD(this, local),
-		    HA_ACTIVE_CHILD(this, local)->fops->symlink,
-		    linkname, loc);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-        STACK_UNWIND (frame, -1, op_errno, NULL, NULL, NULL, NULL);
-        ha_local_wipe (local);
-        return 0;
-}
-
- int32_t
-ha_rename_cbk (call_frame_t *frame,
-	       void *cookie,
-	       xlator_t *this,
-	       int32_t op_ret,
-	       int32_t op_errno,
-	       struct iatt *buf,
-               struct iatt *preoldparent,
-               struct iatt *postoldparent,
-               struct iatt *prenewparent,
-               struct iatt *postnewparent)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame, op_ret, op_errno, buf, preoldparent,
-                              postoldparent, prenewparent, postnewparent);
-	}
-	return 0;
-}
-
-int32_t
-ha_rename (call_frame_t *frame,
-	   xlator_t *this,
-	   loc_t *oldloc,
-	   loc_t *newloc)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, oldloc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_rename_stub (frame, ha_rename, oldloc, newloc);
-        if (!local->stub) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_rename_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->rename,
-			   oldloc, newloc);
-	return 0;
-err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
-	return 0;
-}
-
-int
-ha_link_lookup_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    inode_t *inode,
-		    struct iatt *buf,
-                    dict_t *dict,
-                    struct iatt *postparent)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	char *stateino = NULL;
-	int child_count = 0, i = 0, cnt = 0;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	uint64_t tmp_stateino = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-	children = pvt->children;
-
-	for (i = 0; i < child_count; i++)
-		if (prev_frame->this == children[i])
-			break;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR, "(path=%s) (op_ret=%d op_errno=%d)", local->stub->args.link.newloc.path, op_ret, op_errno);
-	}
-	inode_ctx_get (local->stub->args.link.newloc.inode, 
-		       this, &tmp_stateino);  
-	stateino = (char *)(long)tmp_stateino;
-
-	if (op_ret == 0)
-		stateino[i] = 1;
-
-	LOCK (&frame->lock);
-	cnt = --local->call_count;
-	UNLOCK (&frame->lock);
-
-	if (cnt == 0) {
-		call_stub_t *stub = local->stub;
-		GF_FREE (local->state);
-		STACK_UNWIND (frame,
-			      local->op_ret,
-			      local->op_errno,
-			      local->stub->args.link.oldloc.inode, &local->buf,
-                              &local->preparent, &local->postparent);
-		call_stub_destroy (stub);
-	}
-	return 0;
-}
-
-int32_t
-ha_link_cbk (call_frame_t *frame,
-	     void *cookie,
-	     xlator_t *this,
-	     int32_t op_ret,
-	     int32_t op_errno,
-	     inode_t *inode,
-             struct iatt *buf,
-             struct iatt *preparent,
-             struct iatt *postparent)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	char *stateino = NULL;
-	int child_count = 0, i = 0, cnt = 0;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	uint64_t tmp_stateino = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-	children = pvt->children;
-
-	for (i = 0; i < child_count; i++)
-		if (prev_frame->this == children[i])
-			break;
-
-	if (op_ret == -1) {
-		local->op_errno = op_errno;
-		gf_log (this->name, GF_LOG_ERROR, "(path=%s) (op_ret=%d op_errno=%d)", local->stub->args.link.newloc.path, op_ret, op_errno);
-	}
-	inode_ctx_get (local->stub->args.link.newloc.inode, 
-		       this, &tmp_stateino);
-	stateino = (char *)(long)tmp_stateino;
-
-	if (op_ret == 0) {
-		stateino[i] = 1;
-		local->op_ret = 0;
-		local->first_success = 1;
-		local->buf = *buf;
-                local->preparent = *preparent;
-                local->postparent = *postparent;
-	}
-	cnt = --local->call_count;
-	for (i = local->active + 1; i < child_count; i++) {
-		if (local->state[i])
-			break;
-	}
-
-	if (cnt == 0 || i == child_count) {
-		call_stub_t *stub = local->stub;
-		GF_FREE (local->state);
-		stub = local->stub;
-		STACK_UNWIND (frame, local->op_ret, local->op_errno,
-                              local->stub->args.link.oldloc.inode, &local->buf,
-                              &local->preparent, &local->postparent);
-		call_stub_destroy (stub);
-		return 0;
-	}
-
-	local->active = i;
-
-	if (local->first_success == 0) {
-		STACK_WIND (frame,
-			    ha_link_cbk,
-			    children[i],
-			    children[i]->fops->link,
-			    &local->stub->args.link.oldloc,
-			    &local->stub->args.link.newloc);
-		return 0;
-	}
-	cnt = local->call_count;
-
-	for (; i < child_count; i++) {
-		if (local->state[i]) {
-			STACK_WIND (frame,
-				    ha_link_lookup_cbk,
-				    children[i],
-				    children[i]->fops->lookup,
-				    &local->stub->args.link.newloc,
-				    0);
-			if (--cnt == 0)
-				break;
-		}
-	}
-	return 0;
-}
-
-int32_t
-ha_link (call_frame_t *frame,
-	 xlator_t *this,
-	 loc_t *oldloc,
-	 loc_t *newloc)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	int child_count = 0, i = 0;
-	char *stateino = NULL;
-	int32_t ret = 0, op_errno = 0;
-	uint64_t tmp_stateino = 0;
-
-	ret = inode_ctx_get (newloc->inode, this, &tmp_stateino);
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR, "dict_ptr_error()");
-	}
-	stateino = (char *)(long)tmp_stateino;
-
-	if (stateino == NULL) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"newloc->inode's ctx is NULL, returning EINVAL");
-		STACK_UNWIND (frame, -1, EINVAL, oldloc->inode, NULL, NULL,
-                              NULL);
-		return 0;
-	}
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-
-	frame->local = local = GF_CALLOC (1, sizeof (*local), 
-                                          gf_ha_mt_ha_local_t);
-        if (!frame->local) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	local->stub = fop_link_stub (frame, ha_link, oldloc, newloc);
-        if (!local->stub) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	local->op_ret = -1;
-	local->op_errno = ENOTCONN;
-	local->state = GF_CALLOC (1, child_count, gf_ha_mt_char);
-        if (!local->state) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	memcpy (local->state, pvt->state, child_count);
-	local->active = -1;
-
-	for (i = 0; i < child_count; i++) {
-		if (local->state[i]) {
-			local->call_count++;
-			if (local->active == -1)
-				local->active = i;
-		}
-	}
-
-	STACK_WIND (frame,
-		    ha_link_cbk,
-		    HA_ACTIVE_CHILD(this, local),
-		    HA_ACTIVE_CHILD(this, local)->fops->link,
-		    oldloc,
-		    newloc);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-        STACK_UNWIND (frame, -1, op_errno, NULL, NULL, NULL, NULL);
-        return 0;
-}
-
-int32_t
-ha_create_cbk (call_frame_t *frame,
-	       void *cookie,
-	       xlator_t *this,
-	       int32_t op_ret,
-	       int32_t op_errno,
-	       fd_t *fd,
-	       inode_t *inode,
-	       struct iatt *buf,
-               struct iatt *preparent,
-               struct iatt *postparent)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	int i, child_count = 0, cnt = 0, ret = 0;
-	char *stateino = NULL;
-	hafd_t *hafdp = NULL;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	uint64_t tmp_stateino = 0;
-	uint64_t tmp_hafdp = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-	children = pvt->children;
-
-	ret = inode_ctx_get (local->stub->args.create.loc.inode, 
-			     this, &tmp_stateino);
-	stateino = (char *)(long)tmp_stateino;
-
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR, "dict_to_ptr() error");
-		/* FIXME: handle */
-	}
-	ret = fd_ctx_get (local->stub->args.create.fd, this, &tmp_hafdp);
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR, "dict_to_ptr() error");
-		/* FIXME: handle */
-	}
-	hafdp = (hafd_t *)(long)tmp_hafdp;
-
-	for (i = 0; i < child_count; i++) {
-		if (prev_frame->this == children[i])
-			break;
-	}
-
-	if (op_ret == -1) {
-		local->op_errno = op_errno;
-		gf_log (this->name, GF_LOG_ERROR, "(path=%s) (op_ret=%d op_errno=%d)", local->stub->args.create.loc.path, op_ret, op_errno);
-	}
-	if (op_ret != -1) {
-		stateino[i] = 1;
-		hafdp->fdstate[i] = 1;
-		if (local->op_ret == -1) {
-			local->op_ret = 0;
-			local->buf = *buf;
-                        local->preparent = *preparent;
-                        local->postparent = *postparent;
-			local->first_success = 1;
-		}
-		local->stub->args.create.flags &= (~O_EXCL);
-	}
-	LOCK (&frame->lock);
-	cnt = --local->call_count;
-	UNLOCK (&frame->lock);
-
-	for (i = local->active + 1; i < child_count; i++) {
-		if (local->state[i])
-			break;
-	}
-
-	if (cnt == 0 || i == child_count) {
-		char *state = local->state;
-		call_stub_t *stub = local->stub;
-		STACK_UNWIND (frame, local->op_ret, local->op_errno,
-			      stub->args.create.fd,
-			      stub->args.create.loc.inode, &local->buf,
-                              &local->preparent, &local->postparent);
-		GF_FREE (state);
-		call_stub_destroy (stub);
-		return 0;
-	}
-	local->active = i;
-	cnt = local->call_count;
-	for (; i < child_count; i++) {
-		if (local->state[i]) {
-			STACK_WIND (frame,
-				    ha_create_cbk,
-				    children[i],
-				    children[i]->fops->create,
-				    &local->stub->args.create.loc,
-				    local->stub->args.create.flags,
-				    local->stub->args.create.mode,
-				    local->stub->args.create.fd);
-			if ((local->first_success == 0) || (cnt == 0))
-				break;
-		}
-	}
-	return 0;
-}
-
-int32_t
-ha_create (call_frame_t *frame,
-	   xlator_t *this,
-	   loc_t *loc,
-	   int32_t flags,
-	   mode_t mode, fd_t *fd)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	int i, child_count = 0;
-	char *stateino = NULL;
-	xlator_t **children = NULL;
-	hafd_t *hafdp = NULL;
-        int32_t op_errno = EINVAL;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	children = pvt->children;
-
-	if (local == NULL) {
-                frame->local = local = GF_CALLOC (1, sizeof (*local), 
-                                                  gf_ha_mt_ha_local_t);
-                if (!local) {
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                        goto err;
-                }
-
-		local->stub = fop_create_stub (frame, ha_create, loc, flags, mode, fd);
-                if (!local->stub) {
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                        goto err;
-                }
-
-                local->state = GF_CALLOC (1, child_count, gf_ha_mt_char);
-                if (!local->state) {
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                        goto err;
-                }
-
-		local->active = -1;
-		local->op_ret = -1;
-		local->op_errno = ENOTCONN;
-		memcpy (local->state, pvt->state, child_count);
-
-		for (i = 0; i < pvt->child_count; i++) {
-			if (local->state[i]) {
-				local->call_count++;
-				if (local->active == -1)
-					local->active = i;
-			}
-		}
-		/* FIXME handle active -1 */
-		stateino = GF_CALLOC (1, child_count, gf_ha_mt_char);
-                if (!stateino) {
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                        goto err;
-                }
-
-		hafdp = GF_CALLOC (1, sizeof (*hafdp), gf_ha_mt_hafd_t);
-                if (!hafdp) {
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                        goto err;
-                }
-
-		hafdp->fdstate = GF_CALLOC (1, child_count, gf_ha_mt_char);
-                if (!hafdp->fdstate) {
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                        goto err;
-                }
-
-		hafdp->path = gf_strdup(loc->path);
-                if (!hafdp->path) {
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                        goto err;
-                }
-
-		LOCK_INIT (&hafdp->lock);
-		fd_ctx_set (fd, this, (uint64_t)(long)hafdp);
-		inode_ctx_put (loc->inode, this, (uint64_t)(long)stateino);
-	}
-
-	STACK_WIND (frame,
-		    ha_create_cbk,
-		    children[local->active],
-		    children[local->active]->fops->create,
-		    loc, flags, mode, fd);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-        STACK_UNWIND (frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
-        ha_local_wipe (local);
-
-        if (stateino) {
-                GF_FREE (stateino);
-                stateino = NULL;
-        }
-
-        if (hafdp) {
-                GF_FREE (hafdp->fdstate);
-
-                GF_FREE (hafdp->path);
-
-                GF_FREE (hafdp);
-        }
-
-        return 0;
-}
-
- int32_t
-ha_open_cbk (call_frame_t *frame,
-	     void *cookie,
-	     xlator_t *this,
-	     int32_t op_ret,
-	     int32_t op_errno,
-	     fd_t *fd)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	xlator_t **children = NULL;
-	int i = 0, child_count = 0, callcnt = 0, ret = 0;
-	call_frame_t *prev_frame = NULL;
-	hafd_t *hafdp = NULL;
-	uint64_t tmp_hafdp = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	children = pvt->children;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-
-	ret = fd_ctx_get (local->fd, this, &tmp_hafdp);
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR, "dict_ptr_error()");
-	}
-	hafdp = (hafd_t *)(long)tmp_hafdp;
-
-	for (i = 0; i < child_count; i++)
-		if (children[i] == prev_frame->this)
-			break;
-	LOCK (&frame->lock);
-	if (op_ret != -1) {
-		hafdp->fdstate[i] = 1;
-		local->op_ret = 0;
-	}
-	if (op_ret == -1 && op_errno != ENOTCONN)
-		local->op_errno = op_errno;
-	callcnt = --local->call_count;
-	UNLOCK (&frame->lock);
-
-	if (callcnt == 0) {
-		STACK_UNWIND (frame,
-			      local->op_ret,
-			      local->op_errno,
-			      local->fd);
-	}
-	return 0;
-}
-
-int32_t
-ha_open (call_frame_t *frame,
-	 xlator_t *this,
-	 loc_t *loc,
-	 int32_t flags, fd_t *fd, int wbflags)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	char *stateino = NULL;
-	xlator_t **children = NULL;
-	int cnt = 0, i, child_count = 0, ret = 0;
-	hafd_t *hafdp = NULL;
-	uint64_t tmp_stateino = 0;
-        int32_t op_errno = ENOMEM;
-
-	local = frame->local;
-	pvt = this->private;
-	children = pvt->children;
-	child_count = pvt->child_count;
-
-
-	frame->local = local = GF_CALLOC (1, sizeof (*local), 
-                                          gf_ha_mt_ha_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-                
-	local->op_ret = -1;
-	local->op_errno = ENOTCONN;
-	local->fd = fd;
-
-	hafdp = GF_CALLOC (1, sizeof (*hafdp), gf_ha_mt_hafd_t);
-        if (!hafdp) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	hafdp->fdstate = GF_CALLOC (1, child_count, gf_ha_mt_char);
-        if (!hafdp->fdstate) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	hafdp->path = gf_strdup (loc->path);
-        if (!hafdp->path) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	hafdp->active = -1;
-	if (pvt->pref_subvol == -1) {
-		hafdp->active = fd->inode->ino % child_count;
-	}
-
-	LOCK_INIT (&hafdp->lock);
-	fd_ctx_set (fd, this, (uint64_t)(long)hafdp);
-	ret = inode_ctx_get (loc->inode, this, &tmp_stateino);
-	stateino = (char *)(long)tmp_stateino;
-
-	for (i = 0; i < child_count; i++)
-		if (stateino[i])
-			cnt++;
-	local->call_count = cnt;
-	for (i = 0; i < child_count; i++) {
-		if (stateino[i]) {
-			STACK_WIND (frame,
-				    ha_open_cbk,
-				    children[i],
-				    children[i]->fops->open,
-				    loc, flags, fd, wbflags);
-			if (--cnt == 0)
-				break;
-		}
-	}
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, fd);
-        if (hafdp) {
-                if (hafdp->fdstate) {
-                        GF_FREE (hafdp->fdstate);
-                        hafdp->fdstate = NULL;
-                }
-
-                if (hafdp->path) {
-                        GF_FREE (hafdp->path);
-                        hafdp->path = NULL;
-                }
-
-                GF_FREE (hafdp);
-        }
-
-        ha_local_wipe (local);
-        return 0;
-}
-
- int32_t
-ha_readv_cbk (call_frame_t *frame,
-	      void *cookie,
-	      xlator_t *this,
-	      int32_t op_ret,
-	      int32_t op_errno,
-	      struct iovec *vector,
-	      int32_t count,
-	      struct iatt *stbuf,
-              struct iobref *iobref)
-{
-	int ret = 0;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno,
-			      vector,
-			      count,
-			      stbuf,
-                              iobref);
-	}
-	return 0;
-}
-
-int32_t
-ha_readv (call_frame_t *frame,
-	  xlator_t *this,
-	  fd_t *fd,
-	  size_t size,
-	  off_t offset)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_readv_stub (frame, ha_readv, fd, size, offset);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_readv_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->readv,
-			   fd,
-			   size,
-			   offset);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, NULL, 0, NULL, NULL);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
- int32_t
-ha_writev_cbk (call_frame_t *frame,
-	       void *cookie,
-	       xlator_t *this,
-	       int32_t op_ret,
-	       int32_t op_errno,
-               struct iatt *prebuf,
-	       struct iatt *postbuf)
-{
-	int ret = 0;
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno,
-                              prebuf,
-			      postbuf);
-	}
-	return 0;
-}
-
-int32_t
-ha_writev (call_frame_t *frame,
-	   xlator_t *this,
-	   fd_t *fd,
-	   struct iovec *vector,
-	   int32_t count,
-	   off_t off,
-           struct iobref *iobref)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_writev_stub (frame, ha_writev, fd, vector, count, off,
-                                       iobref);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_writev_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->writev,
-			   fd,
-			   vector,
-			   count,
-			   off,
-                           iobref);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-        ha_local_wipe (local);
-	return 0;	
-}
-
- int32_t
-ha_flush_cbk (call_frame_t *frame,
-	      void *cookie,
-	      xlator_t *this,
-	      int32_t op_ret,
-	      int32_t op_errno)
-{
-	int ret = 0;
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno);
-	}
-	return 0;
-}
-
-int32_t
-ha_flush (call_frame_t *frame,
-	  xlator_t *this,
-	  fd_t *fd)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_flush_stub (frame, ha_flush, fd);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-                
-	STACK_WIND_COOKIE (frame,
-			   ha_flush_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->flush,
-			   fd);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
-
- int32_t
-ha_fsync_cbk (call_frame_t *frame,
-	      void *cookie,
-	      xlator_t *this,
-	      int32_t op_ret,
-	      int32_t op_errno,
-              struct iatt *prebuf,
-              struct iatt *postbuf)
-{
-	int ret = 0;
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno);
-	}
-	return 0;
-}
-
-int32_t
-ha_fsync (call_frame_t *frame,
-	  xlator_t *this,
-	  fd_t *fd,
-	  int32_t flags)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_fsync_stub (frame, ha_fsync, fd, flags);
-        if (!local->stub) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_fsync_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->fsync,
-			   fd,
-			   flags);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
- int32_t
-ha_fstat_cbk (call_frame_t *frame,
-	      void *cookie,
-	      xlator_t *this,
-	      int32_t op_ret,
-	      int32_t op_errno,
-	      struct iatt *buf)
-{
-	int ret = 0;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno,
-			      buf);
-	}
-	return 0;
-}
-
-int32_t
-ha_fstat (call_frame_t *frame,
-	  xlator_t *this,
-	  fd_t *fd)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_fstat_stub (frame, ha_fstat, fd);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_fstat_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->fstat,
-			   fd);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, NULL);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
-int32_t
-ha_opendir_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		fd_t *fd)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	xlator_t **children = NULL;
-	int i = 0, child_count = 0, callcnt = 0, ret = 0;
-	call_frame_t *prev_frame = NULL;
-	hafd_t *hafdp = NULL;
-	uint64_t tmp_hafdp = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	children = pvt->children;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-
-	ret = fd_ctx_get (local->fd, this, &tmp_hafdp);
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR, "dict_ptr_error()");
-	}
-	hafdp = (hafd_t *)(long)tmp_hafdp;
-
-	for (i = 0; i < child_count; i++)
-		if (children[i] == prev_frame->this)
-			break;
-	LOCK (&frame->lock);
-	if (op_ret != -1) {
-		hafdp->fdstate[i] = 1;
-		local->op_ret = 0;
-	}
-	if (op_ret == -1 && op_errno != ENOTCONN)
-		local->op_errno = op_errno;
-	callcnt = --local->call_count;
-	UNLOCK (&frame->lock);
-
-	if (callcnt == 0) {
-		STACK_UNWIND (frame,
-			      local->op_ret,
-			      local->op_errno,
-			      local->fd);
-	}
-	return 0;
-}
-
-int32_t
-ha_opendir (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *loc, fd_t *fd)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	char *stateino = NULL;
-	xlator_t **children = NULL;
-	int cnt = 0, i, child_count = 0, ret = 0;
-	hafd_t *hafdp = NULL;
-	uint64_t tmp_stateino = 0;
-        int32_t op_errno = EINVAL;
-
-	local = frame->local;
-	pvt = this->private;
-	children = pvt->children;
-	child_count = pvt->child_count;
-
-	frame->local = local = GF_CALLOC (1, sizeof (*local), 
-                                          gf_ha_mt_ha_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	local->op_ret = -1;
-	local->op_errno = ENOTCONN;
-	local->fd = fd;
-
-	hafdp = GF_CALLOC (1, sizeof (*hafdp), gf_ha_mt_hafd_t);
-        if (!hafdp) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	hafdp->fdstate = GF_CALLOC (1, child_count, gf_ha_mt_char);
-        if (!hafdp->fdstate) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	hafdp->path = gf_strdup (loc->path);
-        if (!hafdp->path) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	LOCK_INIT (&hafdp->lock);
-	fd_ctx_set (fd, this, (uint64_t)(long)hafdp);
-	ret = inode_ctx_get (loc->inode, this, &tmp_stateino);
-	stateino = (char *)(long)tmp_stateino;
-	
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR, "inode_ctx_get() error");
-	}
-	for (i = 0; i < child_count; i++)
-		if (stateino[i])
-			cnt++;
-	local->call_count = cnt;
-	for (i = 0; i < child_count; i++) {
-		if (stateino[i]) {
-			STACK_WIND (frame,
-				    ha_opendir_cbk,
-				    children[i],
-				    children[i]->fops->opendir,
-				    loc, fd);
-			if (--cnt == 0)
-				break;
-		}
-	}
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, NULL);
-        ha_local_wipe (local);
-        if (hafdp) {
-                if (hafdp->fdstate) {
-                        GF_FREE (hafdp->fdstate);
-                        hafdp->fdstate = NULL;
-                }
-
-                if (hafdp->path) {
-                        GF_FREE (hafdp->path);
-                        hafdp->path = NULL;
-                }
-
-                GF_FREE (hafdp);
-        }
-        return 0;
-}
-
- int32_t
-ha_getdents_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 dir_entry_t *entries,
-		 int32_t count)
-{
-	int ret = 0;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno,
-			      entries,
-			      count);
-	}
-	return 0;
-}
-
-int32_t
-ha_getdents (call_frame_t *frame,
-	     xlator_t *this,
-	     fd_t *fd,
-	     size_t size,
-	     off_t offset,
-	     int32_t flag)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_getdents_stub (frame, ha_getdents, fd, size, offset,
-                                         flag);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_getdents_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->getdents,
-			   fd,
-			   size,
-			   offset,
-			   flag);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, NULL, 0);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
- int32_t
-ha_setdents_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno)
-{
-	int ret = 0;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno);
-	}
-	return 0;
-}
-
-int32_t
-ha_setdents (call_frame_t *frame,
-	     xlator_t *this,
-	     fd_t *fd,
-	     int32_t flags,
-	     dir_entry_t *entries,
-	     int32_t count)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-
-	local->stub = fop_setdents_stub (frame, ha_setdents, fd, flags, entries,
-                                         count);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_setdents_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->setdents,
-			   fd,
-			   flags,
-			   entries,
-			   count);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
- int32_t
-ha_fsyncdir_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno)
-{
-	int ret = 0;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno);
-	}
-	return 0;
-}
-
-int32_t
-ha_fsyncdir (call_frame_t *frame,
-	     xlator_t *this,
-	     fd_t *fd,
-	     int32_t flags)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_fsyncdir_stub (frame, ha_fsyncdir, fd, flags);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_fsyncdir_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->fsyncdir,
-			   fd,
-			   flags);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
- 
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
-
- int32_t
-ha_statfs_cbk (call_frame_t *frame,
-	       void *cookie,
-	       xlator_t *this,
-	       int32_t op_ret,
-	       int32_t op_errno,
-	       struct statvfs *buf)
-{
-        ha_local_t   *local = NULL;
-        ha_private_t *priv  = NULL;
-
-        local = frame->local;
-        if (-1 == op_ret) {
-                local->active = (local->active + 1) % priv->child_count;
-                local->tries--;
-                if (!local->tries)
-                        goto out;
-                
-                STACK_WIND (frame, ha_statfs_cbk,
-                            HA_ACTIVE_CHILD(this, local),
-                            HA_ACTIVE_CHILD(this, local)->fops->statfs, 
-                            &local->loc);
-                return 0;
-        }
-
- out:
-        loc_wipe (&local->loc);
-        STACK_UNWIND (frame, op_ret, op_errno, buf);
-        
-	return 0;
-}
-
-int32_t
-ha_statfs (call_frame_t *frame,
-	   xlator_t *this,
-	   loc_t *loc)
-{
-        ha_private_t *priv = NULL;
-	ha_local_t   *local = NULL;
-	int           op_errno = 0;
-
-        /* The normal way of handling failover doesn't work here
-         * as loc->inode may be null in this case.
-         */
-        local = GF_CALLOC (1, sizeof (*local), 
-                           gf_ha_mt_ha_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        priv = this->private;
-        frame->local  = local;
-        local->active = priv->pref_subvol;
-        if (-1 == local->active)
-                local->active = 0;
-        local->tries  = priv->child_count;
-        loc_copy (&local->loc, loc);
-
-	STACK_WIND (frame, ha_statfs_cbk, HA_ACTIVE_CHILD(this, local),
-                    HA_ACTIVE_CHILD(this, local)->fops->statfs, loc);
-	return 0;
-err:
-	STACK_UNWIND (frame, -1, op_errno, NULL);
-	return 0;
-}
-
- int32_t
-ha_setxattr_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno);
-	}
-	return 0;
-}
-
-int32_t
-ha_setxattr (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc,
-	     dict_t *dict,
-	     int32_t flags)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_setxattr_stub (frame, ha_setxattr, loc, dict, flags);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_setxattr_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->setxattr,
-			   loc,
-			   dict,
-			   flags);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
- int32_t
-ha_getxattr_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 dict_t *dict)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno,
-			      dict);
-	}
-	return 0;
-}
-
-int32_t
-ha_getxattr (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc,
-	     const char *name)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_getxattr_stub (frame, ha_getxattr, loc, name);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_getxattr_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->getxattr,
-			   loc,
-			   name);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, NULL);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
-int32_t
-ha_xattrop_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		dict_t *dict)
-{
-	int ret = -1;
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-	if (ret == 0) {
-		STACK_UNWIND (frame, op_ret, op_errno, dict);
-	}
-	return 0;
-}
-
-
-int32_t
-ha_xattrop (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *loc,
-	    gf_xattrop_flags_t flags,
-	    dict_t *dict)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-
-	local->stub = fop_xattrop_stub (frame, ha_xattrop, loc, flags, dict);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_xattrop_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->xattrop,
-			   loc,
-			   flags,
-			   dict);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, dict);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
-int32_t
-ha_fxattrop_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 dict_t *dict)
-{
-	int ret = -1;
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-	if (ret == 0)
-		STACK_UNWIND (frame, op_ret, op_errno, dict);
-	return 0;
-}
-
-int32_t
-ha_fxattrop (call_frame_t *frame,
-	     xlator_t *this,
-	     fd_t *fd,
-	     gf_xattrop_flags_t flags,
-	     dict_t *dict)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_fxattrop_stub (frame, ha_fxattrop, fd, flags, dict);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_fxattrop_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->fxattrop,
-			   fd,
-			   flags,
-			   dict);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, dict);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
- int32_t
-ha_removexattr_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno)
-{
-	int ret = -1;
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno);
-	}
-	return 0;
-}
-
-int32_t
-ha_removexattr (call_frame_t *frame,
-		xlator_t *this,
-		loc_t *loc,
-		const char *name)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	
-	local->stub = fop_removexattr_stub (frame, ha_removexattr, loc, name);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_removexattr_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->removexattr,
-			   loc,
-			   name);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
-int32_t
-ha_lk_setlk_unlck_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno,
-		       struct gf_flock *lock)
-{
-	ha_local_t *local = NULL;
-	int cnt = 0;
-	call_stub_t *stub = NULL;
-
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	cnt = --local->call_count;
-	if (op_ret == 0)
-		local->op_ret = 0;
-	UNLOCK (&frame->lock);
-
-	if (cnt == 0) {
-		stub = local->stub;
-		GF_FREE (local->state);
-		if (stub->args.lk.lock.l_type == F_UNLCK) {
-			STACK_UNWIND (frame, local->op_ret, local->op_errno, &stub->args.lk.lock);
-		} else {
-			STACK_UNWIND (frame, -1, EIO, NULL);
-		}
-		call_stub_destroy (stub);
-	}
-	return 0;
-}
-
-int32_t
-ha_lk_setlk_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 struct gf_flock *lock)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	xlator_t **children = NULL;
-	int i = 0, cnt = 0, j = 0;
-	int child_count = 0;
-	call_frame_t *prev_frame = NULL;
-	char *state = NULL;
-
-	local = frame->local;
-	pvt = this->private;
-	children = pvt->children;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-	state = local->state;
-
-	if (op_ret == 0)
-		local->op_ret = 0;
-
-	if ((op_ret == 0) || (op_ret == -1 && op_errno == ENOTCONN)) {
-		for (i = 0; i < child_count; i++) {
-			if (prev_frame->this == cookie)
-				break;
-		}
-		i++;
-		for (; i < child_count; i++) {
-			if (local->state[i])
-				break;
-		}
-		if (i == child_count) {
-			call_stub_t *stub = local->stub;
-			GF_FREE (local->state);
-			STACK_UNWIND (frame, 0, op_errno, &stub->args.lk.lock);
-			call_stub_destroy (stub);
-			return 0;
-		}
-		STACK_WIND (frame,
-			    ha_lk_setlk_cbk,
-			    children[i],
-			    children[i]->fops->lk,
-			    local->stub->args.lk.fd,
-			    local->stub->args.lk.cmd,
-			    &local->stub->args.lk.lock);
-		return 0;
-	} else {
-		for (i = 0; i < child_count; i++) {
-			if (prev_frame->this == cookie)
-				break;
-		}
-		cnt = 0;
-		for (j = 0; j < i; j++) {
-			if (state[i])
-				cnt++;
-		}
-		if (cnt) {
-			struct gf_flock lock;
-			lock = local->stub->args.lk.lock;
-			for (i = 0; i < child_count; i++) {
-				if (state[i]) {
-					STACK_WIND (frame,
-						    ha_lk_setlk_unlck_cbk,
-						    children[i],
-						    children[i]->fops->lk,
-						    local->stub->args.lk.fd,
-						    local->stub->args.lk.cmd,
-						    &lock);
-					if (--cnt == 0)
-						break;
-				}
-			}
-			return 0;
-		} else {
-			GF_FREE (local->state);
-			call_stub_destroy (local->stub);
-			STACK_UNWIND (frame,
-				      op_ret,
-				      op_errno,
-				      lock);
-			return 0;
-		}
-	}
-}
-
-int32_t
-ha_lk_getlk_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 struct gf_flock *lock)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	fd_t *fd = NULL;
-	int child_count = 0, i = 0;
-	xlator_t **children = NULL;
-	call_frame_t *prev_frame = NULL;
-
-	local = frame->local;
-	pvt = this->private;
-	fd = local->stub->args.lk.fd;
-	child_count = pvt->child_count;
-	children = pvt->children;
-	prev_frame = cookie;
-
-	if (op_ret == 0) {
-		GF_FREE (local->state);
-		call_stub_destroy (local->stub);
-		STACK_UNWIND (frame, 0, 0, lock);
-		return 0;
-	}
-
-	for (i = 0; i < child_count; i++) {
-		if (prev_frame->this == children[i])
-			break;
-	}
-
-	for (; i < child_count; i++) {
-		if (local->state[i])
-			break;
-	}
-
-	if (i == child_count) {
-		GF_FREE (local->state);
-		call_stub_destroy (local->stub);
-		STACK_UNWIND (frame, op_ret, op_errno, lock);
-		return 0;
-	}
-
-	STACK_WIND (frame,
-		    ha_lk_getlk_cbk,
-		    children[i],
-		    children[i]->fops->lk,
-		    fd,
-		    local->stub->args.lk.cmd,
-		    &local->stub->args.lk.lock);
-	return 0;
-}
-
-int32_t
-ha_lk (call_frame_t *frame,
-       xlator_t *this,
-       fd_t *fd,
-       int32_t cmd,
-       struct gf_flock *lock)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	hafd_t *hafdp = NULL;
-	char *state = NULL;
-	int child_count = 0, i = 0, cnt = 0, ret = 0;
-	xlator_t **children = NULL;
-	uint64_t tmp_hafdp = 0;
-        int32_t op_errno = EINVAL;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	children = pvt->children;
-	ret = fd_ctx_get (fd, this, &tmp_hafdp);
-	if (ret < 0)
-		gf_log (this->name, GF_LOG_ERROR, "fd_ctx_get failed");
-
-	if (local == NULL) {
-                local = frame->local = GF_CALLOC (1, sizeof (*local),
-                                                  gf_ha_mt_ha_local_t);
-                if (!local) {
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                        goto err;
-                }
-                        
-		local->active = -1;
-		local->op_ret = -1;
-		local->op_errno = ENOTCONN;
-	}
-	hafdp = (hafd_t *)(long)tmp_hafdp;
-
-	if (local->active == -1) {
-                op_errno = ENOTCONN;
-                goto err;
-	}
-
-	local->stub = fop_lk_stub (frame, ha_lk, fd, cmd, lock);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	local->state = GF_CALLOC (1, child_count, gf_ha_mt_char);
-        if (!local->state) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	state = hafdp->fdstate;
-	LOCK (&hafdp->lock);
-	memcpy (local->state, state, child_count);
-	UNLOCK (&hafdp->lock);
-	if (cmd == F_GETLK) {
-		for (i = 0; i < child_count; i++) {
-			if (local->state[i])
-				break;
-		}
-		STACK_WIND (frame,
-			    ha_lk_getlk_cbk,
-			    children[i],
-			    children[i]->fops->lk,
-			    fd,
-			    cmd,
-			    lock);
-	} else if (cmd == F_SETLK && lock->l_type == F_UNLCK) {
-		for (i = 0; i < child_count; i++) {
-			if (local->state[i])
-				local->call_count++;
-		}
-		cnt = local->call_count;
-		for (i = 0; i < child_count; i++) {
-			if (local->state[i]) {
-				STACK_WIND (frame,
-					    ha_lk_setlk_unlck_cbk,
-					    children[i],
-					    children[i]->fops->lk,
-					    fd, cmd, lock);
-				if (--cnt == 0)
-					break;
-			}
-		}
-	} else {
-		for (i = 0; i < child_count; i++) {
-			if (local->state[i])
-				break;
-		}
-		STACK_WIND (frame,
-			    ha_lk_setlk_cbk,
-			    children[i],
-			    children[i]->fops->lk,
-			    fd,
-			    cmd,
-			    lock);
-	}
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-        STACK_UNWIND (frame, -1, op_errno, NULL);
-
-        ha_local_wipe (local);
-        return 0;
-}
-
- int32_t
-ha_inode_entry_lk_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno);
-	}
-	return 0;
-}
-
-int32_t
-ha_inodelk (call_frame_t *frame,
-	    xlator_t *this,
-            const char *volume,
-	    loc_t *loc,
-	    int32_t cmd,
-	    struct gf_flock *lock)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_inodelk_stub (frame, ha_inodelk, volume,
-                                        loc, cmd, lock);
-	STACK_WIND_COOKIE (frame,
-			   ha_inode_entry_lk_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->inodelk,
-			   volume,
-                           loc,
-			   cmd,
-			   lock);
-	return 0;
-err:
-	STACK_UNWIND (frame, -1, op_errno);
-	return 0;
-}
-
-int32_t
-ha_entrylk (call_frame_t *frame,
-	    xlator_t *this,
-            const char *volume,
-	    loc_t *loc,
-	    const char *basename,
-	    entrylk_cmd cmd,
-	    entrylk_type type)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_entrylk_stub (frame, ha_entrylk, volume,
-                                        loc, basename, cmd, type);
-	STACK_WIND_COOKIE (frame,
-			   ha_inode_entry_lk_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->entrylk,
-			   volume, loc, basename, cmd, type);
-	return 0;
-err:
-	STACK_UNWIND (frame, -1, op_errno);
-	return 0;
-}
-
- int32_t
-ha_checksum_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 uint8_t *file_checksum,
-		 uint8_t *dir_checksum)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno,
-			      file_checksum,
-			      dir_checksum);
-	}
-	return 0;
-}
-
-int32_t
-ha_checksum (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc,
-	     int32_t flag)
-{
-	int op_errno = 0;
-	ha_local_t *local = NULL;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_checksum_stub (frame, ha_checksum, loc, flag);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_checksum_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->checksum,
-			   loc,
-			   flag);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
-int32_t
-ha_common_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, gf_dirent_t *entries)
-{
-	int ret = 0;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-	if (ret == 0)
-		STACK_UNWIND (frame, op_ret, op_errno, entries);
-	return 0;
-}
-
-int32_t
-ha_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-            off_t off);
-
-int32_t
-ha_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-             off_t off);
-int32_t
-ha_do_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-               off_t off, int whichop)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-        if (whichop == GF_FOP_READDIR)
-                local->stub = fop_readdir_stub (frame, ha_readdir, fd, size,
-                                                off);
-        else
-                local->stub = fop_readdirp_stub (frame, ha_readdirp, fd, size,
-                                                 off);
-        if (!local->stub) {
-                op_errno = ENOMEM; 
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-        if (whichop == GF_FOP_READDIR)
-                STACK_WIND_COOKIE (frame, ha_common_readdir_cbk,
-                                   (void *)(long)local->active,
-                                   HA_ACTIVE_CHILD(this, local),
-                                   HA_ACTIVE_CHILD(this, local)->fops->readdir,
-                                   fd, size, off);
-        else
-                STACK_WIND_COOKIE (frame, ha_common_readdir_cbk,
-                                   (void *)(long)local->active,
-                                   HA_ACTIVE_CHILD(this, local),
-                                   HA_ACTIVE_CHILD(this, local)->fops->readdirp,
-                                   fd, size, off);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, ENOTCONN, NULL);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
-
-int32_t
-ha_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-            off_t off)
-{
-        ha_do_readdir (frame, this, fd, size, off, GF_FOP_READDIR);
-        return 0;
-}
-
-int32_t
-ha_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-             off_t off)
-{
-        ha_do_readdir (frame, this, fd, size, off, GF_FOP_READDIRP);
-        return 0;
-}
-
-/* Management operations */
-
- int32_t
-ha_stats_cbk (call_frame_t *frame,
-	      void *cookie,
-	      xlator_t *this,
-	      int32_t op_ret,
-	      int32_t op_errno,
-	      struct xlator_stats *stats)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	int i = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	prev_frame = cookie;
-	children = pvt->children;
-
-	if (op_ret == -1 && op_errno == ENOTCONN) {
-		for (i = 0; i < pvt->child_count; i++) {
-			if (prev_frame->this == children[i])
-				break;
-		}
-		i++;
-		for (; i < pvt->child_count; i++) {
-			if (pvt->state[i])
-				break;
-		}
-
-		if (i == pvt->child_count) {
-			STACK_UNWIND (frame, -1, ENOTCONN, NULL);
-			return 0;
-		}
-		STACK_WIND (frame,
-			    ha_stats_cbk,
-			    children[i],
-			    children[i]->mops->stats,
-			    local->flags);
-		return 0;
-	}
-
-	STACK_UNWIND (frame,
-		      op_ret,
-		      op_errno,
-		      stats);
-	return 0;
-}
-
-int32_t
-ha_stats (call_frame_t *frame,
-	  xlator_t *this,
-	  int32_t flags)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	xlator_t **children = NULL;
-	int i = 0;
-        int32_t op_errno = EINVAL;
-
-	local = frame->local = GF_CALLOC (1, sizeof (*local),
-                                          gf_ha_mt_ha_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	pvt = this->private;
-	children = pvt->children;
-	for (i = 0; i < pvt->child_count; i++) {
-		if (pvt->state[i])
-			break;
-	}
-
-	if (i == pvt->child_count) {
-                op_errno = ENOTCONN;
-                goto err;
-	}
-	local->flags = flags;
-
-	STACK_WIND (frame,
-		    ha_stats_cbk,
-		    children[i],
-		    children[i]->mops->stats,
-		    flags);
-	return 0;
-
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-        STACK_UNWIND (frame, -1, ENOTCONN, NULL);
-
-        ha_local_wipe (local);
-        return 0;
-
-}
-
-
-int32_t
-ha_getspec_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		char *spec_data)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	int i = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	prev_frame = cookie;
-	children = pvt->children;
-
-	if (op_ret == -1 && op_errno == ENOTCONN) {
-		for (i = 0; i < pvt->child_count; i++) {
-			if (prev_frame->this == children[i])
-				break;
-		}
-		i++;
-		for (; i < pvt->child_count; i++) {
-			if (pvt->state[i])
-				break;
-		}
-
-		if (i == pvt->child_count) {
-			STACK_UNWIND (frame, -1, ENOTCONN, NULL);
-			return 0;
-		}
-		STACK_WIND (frame,
-			    ha_getspec_cbk,
-			    children[i],
-			    children[i]->mops->getspec,
-			    local->pattern,
-			    local->flags);
-		return 0;
-	}
-
-	STACK_UNWIND (frame,
-		      op_ret,
-		      op_errno,
-		      spec_data);
-	return 0;
-}
-
-int32_t
-ha_getspec (call_frame_t *frame,
-	    xlator_t *this,
-	    const char *key,
-	    int32_t flags)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	xlator_t **children = NULL;
-	int i = 0;
-        int32_t op_errno = EINVAL;
-
-	local = frame->local = GF_CALLOC (1, sizeof (*local), 
-                                          gf_ha_mt_ha_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	pvt = this->private;
-	children = pvt->children;
-
-	for (i = 0; i < pvt->child_count; i++) {
-		if (pvt->state[i])
-			break;
-	}
-
-	if (i == pvt->child_count) {
-                op_errno = ENOTCONN;
-                goto err;
-	}
-	local->flags = flags;
-	local->pattern = (char *)key;
-
-	STACK_WIND (frame,
-		    ha_getspec_cbk,
-		    children[i],
-		    children[i]->mops->getspec,
-		    key, flags);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-        STACK_UNWIND (frame, -1, ENOTCONN, NULL);
-
-        ha_local_wipe (local);
-        return 0;
-
-}
-
-int32_t
-ha_closedir (xlator_t *this,
-	     fd_t *fd)
-{
-	hafd_t *hafdp = NULL;
-	int op_errno = 0;
-	uint64_t tmp_hafdp = 0;
-
-	op_errno = fd_ctx_del (fd, this, &tmp_hafdp);
-	if (op_errno != 0) {
-		gf_log (this->name, GF_LOG_ERROR, "fd_ctx_del() error");
-		return 0;
-	}
-	hafdp = (hafd_t *)(long)tmp_hafdp;
-
-	GF_FREE (hafdp->fdstate);
-	GF_FREE (hafdp->path);
-	LOCK_DESTROY (&hafdp->lock);
-	return 0;
-}
-
-int32_t
-ha_close (xlator_t *this,
-	  fd_t *fd)
-{
-	hafd_t *hafdp = NULL;
-	int op_errno = 0;
-	uint64_t tmp_hafdp = 0;
-
-	op_errno = fd_ctx_del (fd, this, &tmp_hafdp);
-	if (op_errno != 0) {
-		gf_log (this->name, GF_LOG_ERROR, "fd_ctx_del() error");
-		return 0;
-	}
-	hafdp = (hafd_t *)(long)tmp_hafdp;
-
-	GF_FREE (hafdp->fdstate);
-	GF_FREE (hafdp->path);
-	LOCK_DESTROY (&hafdp->lock);
-	return 0;
-}
-
-/* notify */
-int32_t
-notify (xlator_t *this,
-	int32_t event,
-	void *data,
-	...)
-{
-	ha_private_t *pvt = NULL;
-	int32_t i = 0, upcnt = 0;
-
-	pvt = this->private;
-	if (pvt == NULL) {
-		gf_log (this->name, GF_LOG_DEBUG, "got notify before init()");
-		return 0;
-	}
-
-	switch (event)
-	{
-	case GF_EVENT_CHILD_DOWN:
-	{
-		for (i = 0; i < pvt->child_count; i++) {
-			if (data == pvt->children[i])
-				break;
-		}
-	        gf_log (this->name, GF_LOG_DEBUG, "GF_EVENT_CHILD_DOWN from %s", pvt->children[i]->name);
-		pvt->state[i] = 0;
-		for (i = 0; i < pvt->child_count; i++) {
-			if (pvt->state[i])
-				break;
-		}
-		if (i == pvt->child_count) {
-			default_notify (this, event, data);
-		}
-	}
-	break;
-	case GF_EVENT_CHILD_UP:
-	{
-		for (i = 0; i < pvt->child_count; i++) {
-			if (data == pvt->children[i])
-				break;
-		}
-
-		gf_log (this->name, GF_LOG_DEBUG, "GF_EVENT_CHILD_UP from %s", pvt->children[i]->name);
-
-		pvt->state[i] = 1;
-
-		for (i = 0; i < pvt->child_count; i++) {
-			if (pvt->state[i])
-				upcnt++;
-		}
-
-		if (upcnt == 1) {
-			default_notify (this, event, data);
-		}
-	}
-	break;
-
-	default:
-	{
-		default_notify (this, event, data);
-	}
-	}
-
-	return 0;
-}
-
-int32_t
-mem_acct_init (xlator_t *this)
-{
-        int     ret = -1;
-
-        if (!this)
-                return ret;
-
-        ret = xlator_mem_acct_init (this, gf_ha_mt_end + 1);
-        
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
-                                "failed");
-                return ret;
-        }
-
-        return ret;
-}
-
-int
-init (xlator_t *this)
-{
-	ha_private_t *pvt = NULL;
-	xlator_list_t *trav = NULL;
-	int count = 0, ret = 0;
-
-
-	if (!this->children) {
-		gf_log (this->name,GF_LOG_ERROR, 
-			"FATAL: ha should have one or more child defined");
-		return -1;
-	}
-
-	if (!this->parents) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"dangling volume. check volfile ");
-	}
-  
-	trav = this->children;
-	pvt = GF_CALLOC (1, sizeof (ha_private_t), gf_ha_mt_ha_private_t);
-
-	ret = dict_get_int32 (this->options, "preferred-subvolume", 
-			      &pvt->pref_subvol);
-	if (ret < 0) {
-		pvt->pref_subvol = -1;
-	}
-
-	trav = this->children;
-	while (trav) {
-		count++;
-		trav = trav->next;
-	}
-
-	pvt->child_count = count;
-	pvt->children = GF_CALLOC (count, sizeof (xlator_t*), 
-                                   gf_ha_mt_xlator_t);
-
-	trav = this->children;
-	count = 0;
-	while (trav) {
-		pvt->children[count] = trav->xlator;
-		count++;
-		trav = trav->next;
-	}
-
-	pvt->state = GF_CALLOC (1, count, gf_ha_mt_char);
-	this->private = pvt;
-	return 0;
-}
-
-void
-fini (xlator_t *this)
-{
-	ha_private_t *priv = NULL;
-	priv = this->private;
-	GF_FREE (priv);
-	return;
-}
-
-
-struct xlator_fops fops = {
-	.lookup      = ha_lookup,
-	.stat        = ha_stat,
-	.readlink    = ha_readlink,
-	.mknod       = ha_mknod,
-	.mkdir       = ha_mkdir,
-	.unlink      = ha_unlink,
-	.rmdir       = ha_rmdir,
-	.symlink     = ha_symlink,
-	.rename      = ha_rename,
-	.link        = ha_link,
-	.truncate    = ha_truncate,
-	.create      = ha_create,
-	.open        = ha_open,
-	.readv       = ha_readv,
-	.writev      = ha_writev,
-	.statfs      = ha_statfs,
-	.flush       = ha_flush,
-	.fsync       = ha_fsync,
-	.setxattr    = ha_setxattr,
-	.getxattr    = ha_getxattr,
-	.removexattr = ha_removexattr,
-	.opendir     = ha_opendir,
-	.readdir     = ha_readdir,
-	.readdirp    = ha_readdirp,
-	.getdents    = ha_getdents,
-	.fsyncdir    = ha_fsyncdir,
-	.access      = ha_access,
-	.ftruncate   = ha_ftruncate,
-	.fstat       = ha_fstat,
-	.lk          = ha_lk,
-	.setdents    = ha_setdents,
-	.lookup_cbk  = ha_lookup_cbk,
-	.checksum    = ha_checksum,
-	.xattrop     = ha_xattrop,
-	.fxattrop    = ha_fxattrop,
-        .setattr     = ha_setattr,
-        .fsetattr    = ha_fsetattr,
-};
-
-struct xlator_cbks cbks = {
-	.release    = ha_close,
-	.releasedir = ha_closedir,
-	.forget     = ha_forget,
-};
diff --git a/xlators/cluster/ha/src/ha.h b/xlators/cluster/ha/src/ha.h
deleted file mode 100644
index e2ed7eaa68a..00000000000
--- a/xlators/cluster/ha/src/ha.h
+++ /dev/null
@@ -1,53 +0,0 @@
-/*
-   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
-   This file is part of GlusterFS.
-
-   This file is licensed to you under your choice of the GNU Lesser
-   General Public License, version 3 or any later version (LGPLv3 or
-   later), or the GNU General Public License, version 2 (GPLv2), in all
-   cases as published by the Free Software Foundation.
-*/
-#ifndef __HA_H_
-#define __HA_H_
-
-#include "ha-mem-types.h"
-
-typedef struct {
-	call_stub_t *stub;
-	int32_t op_ret, op_errno;
-	int32_t active, tries, revalidate, revalidate_error;
-	int32_t call_count;
-	char *state, *pattern;
-	dict_t *dict;
-	loc_t loc;
-	struct iatt buf;
-        struct iatt postparent;
-        struct iatt preparent;
-	fd_t *fd;
-	inode_t *inode;
-	int32_t flags;
-	int32_t first_success;
-} ha_local_t;
-
-typedef struct {
-	char *state;
-	xlator_t **children;
-	int child_count, pref_subvol;
-} ha_private_t;
-
-typedef struct {
-	char *fdstate;
-	char *path;
-	gf_lock_t lock;
-	int active;
-} hafd_t;
-
-#define HA_ACTIVE_CHILD(this, local) (((ha_private_t *)this->private)->children[local->active])
-
-extern int ha_alloc_init_fd (call_frame_t *frame, fd_t *fd);
-
-extern int ha_handle_cbk (call_frame_t *frame, void *cookie, int op_ret, int op_errno) ;
-
-extern int ha_alloc_init_inode (call_frame_t *frame, inode_t *inode);
-
-#endif
diff --git a/xlators/cluster/map/src/Makefile.am b/xlators/cluster/map/src/Makefile.am
deleted file mode 100644
index a278b05e2d1..00000000000
--- a/xlators/cluster/map/src/Makefile.am
+++ /dev/null
@@ -1,16 +0,0 @@
-xlator_LTLIBRARIES = map.la
-xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/testing/cluster
-
-map_la_LDFLAGS = -module -avoid-version 
-
-map_la_SOURCES = map.c map-helper.c
-map_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
-
-noinst_HEADERS = map.h
-
-AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src
-
-AM_CFLAGS = -Wall $(GF_CFLAGS)
-
-CLEANFILES = 
-
diff --git a/xlators/cluster/map/src/map-helper.c b/xlators/cluster/map/src/map-helper.c
deleted file mode 100644
index 851397b68d3..00000000000
--- a/xlators/cluster/map/src/map-helper.c
+++ /dev/null
@@ -1,348 +0,0 @@
-/*
-   Copyright (c) 2009-2012 Red Hat, Inc. <http://www.redhat.com>
-   This file is part of GlusterFS.
-
-   This file is licensed to you under your choice of the GNU Lesser
-   General Public License, version 3 or any later version (LGPLv3 or
-   later), or the GNU General Public License, version 2 (GPLv2), in all
-   cases as published by the Free Software Foundation.
-*/
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "xlator.h"
-#include "map.h"
-
-
-xlator_t *
-map_subvol_next (xlator_t *this, xlator_t *prev)
-{
-	map_private_t *priv = NULL;
-	xlator_t      *next = NULL;
-	int            i = 0;
-
-	priv = this->private;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (priv->xlarray[i].xl == prev) {
-			if ((i + 1) < priv->child_count)
-				next = priv->xlarray[i + 1].xl;
-			break;
-		}
-	}
-
-	return next;
-}
-
-int
-map_subvol_cnt (xlator_t *this, xlator_t *subvol)
-{
-	int i = 0;
-	int ret = -1;
-	map_private_t *priv = NULL;
-
-	priv = this->private;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (subvol == priv->xlarray[i].xl) {
-			ret = i;
-			break;
-		}
-	}
-
-	return ret;
-}
-
-int
-map_itransform (xlator_t *this, xlator_t *subvol, uint64_t x, uint64_t *y_p)
-{
-	map_private_t *priv = NULL;
-	int         cnt = 0;
-	int         max = 0;
-	uint64_t    y = 0;
-
-	if (x == ((uint64_t) -1)) {
-		y = (uint64_t) -1;
-		goto out;
-	}
-
-	priv = this->private;
-
-	max = priv->child_count;
-	cnt = map_subvol_cnt (this, subvol);
-
-	y = ((x * max) + cnt);
-
-out:
-	if (y_p)
-		*y_p = y;
-
-	return 0;
-}
-
-
-int
-map_deitransform (xlator_t *this, uint64_t y, xlator_t **subvol_p,
-		  uint64_t *x_p)
-{
-	int         cnt = 0;
-	int         max = 0;
-	uint64_t    x = 0;
-	xlator_t   *subvol = 0;
-	map_private_t *priv = NULL;
-
-	priv = this->private;
-	max = priv->child_count;
-
-	cnt = y % max;
-	x   = y / max;
-
-	subvol = priv->xlarray[cnt].xl;
-
-	if (subvol_p)
-		*subvol_p = subvol;
-
-	if (x_p)
-		*x_p = x;
-
-	return 0;
-}
-
-
-xlator_t *
-get_mapping_subvol_from_path (xlator_t *this, const char *path) 
-{
-	map_private_t      *priv = NULL;
-	struct map_pattern *map = NULL;
-
-	/* To make sure we handle '/' properly */
-	if (!strcmp (path, "/"))
-		return NULL;
-
-	priv = this->private;
-
-	map = priv->map;
-	while (map) {
-		if (!strncmp (map->directory, path, map->dir_len)) {
-			if ((path[map->dir_len] == '/') ||
-			    (path[map->dir_len] == '\0')) {
-				return map->xl;
-			}
-		}
-		
-		map = map->next;
-	}
-
-	return priv->default_xl;
-}
-
-xlator_t *
-get_mapping_subvol_from_ctx (xlator_t *this, inode_t *inode)
-{
-	uint64_t subvol = 0;
-	int      ret    = -1;
-
-	ret = inode_ctx_get (inode, this, &subvol);
-	if (ret != 0) 
-		return NULL;
-
-	return (xlator_t *)(long)subvol;
-}
-
-int
-check_multiple_volume_entry (xlator_t *this, 
-			     xlator_t *subvol)
-{
-	int ret = -1;
-	int idx = 0;
-	map_private_t *priv = NULL;
-
-	priv = this->private;
-	
-	for (idx = 0; idx < priv->child_count; idx++) {
-		if (priv->xlarray[idx].xl == subvol) {
-			if (priv->xlarray[idx].mapped) {
-				gf_log (this->name, GF_LOG_ERROR,
-					"subvolume '%s' is already mapped",
-					subvol->name);
-				goto out;
-			}
-			priv->xlarray[idx].mapped = 1;
-			ret = 0;
-			goto out;
-		}
-	}
-
-	gf_log (this->name, GF_LOG_ERROR,
-		"subvolume '%s' is not found",
-		subvol->name);
-	
- out:
-	return ret;
-}
-
-int
-verify_dir_and_assign_subvol (xlator_t *this, 
-			      const char *directory, 
-			      const char *subvol)
-{
-	int            default_flag = 0;
-	int            ret  = -1;
-	int            idx  = 0;
-	map_private_t *priv = NULL;
-	xlator_list_t *trav = NULL;
-	struct map_pattern *tmp_map = NULL;
-
-	priv = this->private;
-
-	/* check if directory is valid, ie, its a top level dir, and 
-	 * not includes a '*' in it.
-	 */
-	if (!strcmp ("*", directory)) {
-		default_flag = 1;
-	} else {
-		if (directory[0] != '/') {
-			gf_log (this->name, GF_LOG_ERROR,
-				"map takes absolute path, starting with '/'. "
-				"not '%s'", directory);
-			goto out;
-		}
-		for (idx = 1; idx < (strlen (directory) - 1); idx++) {
-			if (directory[idx] == '/') {
-				gf_log (this->name, GF_LOG_ERROR,
-					"map takes only top level directory, "
-					"not '%s'", directory);
-				goto out;
-			}
-		}
-	}
-
-	/* Assign proper subvolume */
-	trav = this->children;
-	while (trav) {
-		if (!strcmp (trav->xlator->name, subvol)) {
-			
-			/* Check if there is another directory for 
-			 * same volume, if yes, return error.
-			 */
-			ret = check_multiple_volume_entry (this, 
-							   trav->xlator);
-			if (ret != 0) {
-				goto out;
-			}
-
-			ret = 0;
-			if (default_flag) {
-				if (priv->default_xl) {
-					ret = -1;
-					gf_log (this->name, GF_LOG_ERROR,
-						"'*' specified more than "
-						"once. don't confuse me!!!");
-				}
-
-				priv->default_xl = trav->xlator;
-				goto out;
-			}
-
-			tmp_map = GF_CALLOC (1, sizeof (struct map_pattern),
-                                             gf_map_mt_map_pattern);
-			tmp_map->xl = trav->xlator;
-			tmp_map->dir_len = strlen (directory);
-
-			/* make sure that the top level directory starts 
-			 * with '/' and ends without '/'
-			 */
-			tmp_map->directory = gf_strdup (directory);
-			if (directory[tmp_map->dir_len - 1] == '/') {
-				tmp_map->dir_len--;
-			}
-
-			if (!priv->map) 
-				priv->map = tmp_map;
-			else {
-				struct map_pattern *trav_map = NULL;
-				trav_map = priv->map;
-				while (trav_map->next)
-					trav_map = trav_map->next;
-				trav_map->next = tmp_map;
-			}
-			
-			goto out;
-		}
-
-		trav = trav->next;
-	}
-
-	gf_log (this->name, GF_LOG_ERROR, 
-		"map volume '%s' is not proper subvolume", subvol);
-
- out:
-	return ret;
-}
-
-int 
-assign_default_subvol (xlator_t *this, const char *default_xl)
-{
-	int ret = -1;
-	map_private_t *priv = NULL;
-	xlator_list_t *trav = NULL;
-
-	priv = this->private;
-	trav = this->children;
-
-	while (trav) {
-		if (!strcmp (trav->xlator->name, default_xl)) {
-			ret = check_multiple_volume_entry (this, 
-							   trav->xlator);
-			if (ret != 0) {
-				goto out;
-			}
-			if (priv->default_xl)
-				gf_log (this->name, GF_LOG_WARNING,
-					"default-volume option provided, "
-					"overriding earlier '*' option");
-			priv->default_xl = trav->xlator;
-			return 0;
-		}
-		trav = trav->next;
-	}
-
-	gf_log (this->name, GF_LOG_ERROR,
-		"default-volume value is not an valid subvolume. check again");
- out:
-	return -1;
-}
-
-void
-verify_if_all_subvolumes_got_used (xlator_t *this)
-{
-	int idx = 0;
-	map_private_t *priv = NULL;
-
-	priv = this->private;
-	
-	for (idx = 0; idx < priv->child_count; idx++) {
-		if (!priv->xlarray[idx].mapped) {
-			if (!priv->default_xl) {
-				priv->default_xl = priv->xlarray[idx].xl;
-				priv->xlarray[idx].mapped = 1;
-			} else {
-				gf_log (this->name, GF_LOG_WARNING,
-					"subvolume '%s' is not mapped to "
-					"any directory",
-					priv->xlarray[idx].xl->name);
-			}
-		}
-	}
-
-	if (!priv->default_xl) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"default subvolume not specified, filesystem "
-			"may not work properly. Check 'map' translator "
-			"documentation for more info");
-	}
-
-	return ;
-}
diff --git a/xlators/cluster/map/src/map-mem-types.h b/xlators/cluster/map/src/map-mem-types.h
deleted file mode 100644
index 3e89f4736e4..00000000000
--- a/xlators/cluster/map/src/map-mem-types.h
+++ /dev/null
@@ -1,24 +0,0 @@
-/*
-   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
-   This file is part of GlusterFS.
-
-   This file is licensed to you under your choice of the GNU Lesser
-   General Public License, version 3 or any later version (LGPLv3 or
-   later), or the GNU General Public License, version 2 (GPLv2), in all
-   cases as published by the Free Software Foundation.
-*/
-
-#ifndef __MAP_MEM_TYPES_H__
-#define __MAP_MEM_TYPES_H__
-
-#include "mem-types.h"
-
-enum gf_map_mem_types_ {
-        gf_map_mt_map_private_t = gf_common_mt_end + 1,
-        gf_map_mt_map_local_t,
-        gf_map_mt_map_xlator_array,
-        gf_map_mt_map_pattern,
-        gf_map_mt_end
-};
-#endif
-
diff --git a/xlators/cluster/map/src/map.c b/xlators/cluster/map/src/map.c
deleted file mode 100644
index 6150a33cec8..00000000000
--- a/xlators/cluster/map/src/map.c
+++ /dev/null
@@ -1,2566 +0,0 @@
-/*
-   Copyright (c) 2010-2012 Red Hat, Inc. <http://www.redhat.com>
-   This file is part of GlusterFS.
-
-   This file is licensed to you under your choice of the GNU Lesser
-   General Public License, version 3 or any later version (LGPLv3 or
-   later), or the GNU General Public License, version 2 (GPLv2), in all
-   cases as published by the Free Software Foundation.
-*/
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "xlator.h"
-#include "map.h"
-
-/* TODO : 
- *     -> support for 'get' 'put' API in through xattrs.
- *     -> define the behavior of notify()
- */
-
-static int32_t
-map_stat_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  struct iatt *buf)
-
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-	
-	map_itransform (this, prev->this, buf->ia_ino, &buf->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, buf);
-	return 0;
-}
-
-static int32_t
-map_setattr_cbk (call_frame_t *frame,
-                 void *cookie,
-                 xlator_t *this,
-                 int32_t op_ret,
-                 int32_t op_errno,
-                 struct iatt *statpre,
-                 struct iatt *statpost)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-
-	map_itransform (this, prev->this, statpre->ia_ino, &statpre->ia_ino);
-	map_itransform (this, prev->this, statpost->ia_ino, &statpost->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, statpre, statpost);
-	return 0;
-}
-
-static int32_t
-map_fsetattr_cbk (call_frame_t *frame,
-                  void *cookie,
-                  xlator_t *this,
-                  int32_t op_ret,
-                  int32_t op_errno,
-                  struct iatt *statpre,
-                  struct iatt *statpost)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-
-	map_itransform (this, prev->this, statpre->ia_ino, &statpre->ia_ino);
-	map_itransform (this, prev->this, statpost->ia_ino, &statpost->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, statpre, statpost);
-	return 0;
-}
-
-static int32_t
-map_truncate_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno,
-		      struct iatt *prebuf,
-                      struct iatt *postbuf)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-	
-	map_itransform (this, prev->this, postbuf->ia_ino, &postbuf->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, prebuf, postbuf);
-	return 0;
-}
-
-static int32_t
-map_ftruncate_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno,
-		       struct iatt *prebuf,
-                       struct iatt *postbuf)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-	
-	map_itransform (this, prev->this, postbuf->ia_ino, &postbuf->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, prebuf, postbuf);
-	return 0;
-}
-
-
-static int32_t
-map_access_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-static int32_t
-map_readlink_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno,
-		      const char *path,
-                      struct iatt *sbuf)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, path, sbuf);
-	return 0;
-}
-
-static int32_t
-map_unlink_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-                    struct iatt *preparent,
-                    struct iatt *postparent)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, preparent, postparent);
-	return 0;
-}
-
-static int32_t
-map_rmdir_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-                   struct iatt *preparent,
-                   struct iatt *postparent)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, preparent, postparent);
-	return 0;
-}
-
-
-static int32_t
-map_rename_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    struct iatt *buf,
-                    struct iatt *preoldparent,
-                    struct iatt *postoldparent,
-                    struct iatt *prenewparent,
-                    struct iatt *postnewparent)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-	
-	map_itransform (this, prev->this, buf->ia_ino, &buf->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, buf);
-	return 0;
-}
-
-static int32_t
-map_link_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  inode_t *inode,
-                  struct iatt *buf,
-                  struct iatt *preparent,
-                  struct iatt *postparent)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-	
-	map_itransform (this, prev->this, buf->ia_ino, &buf->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, inode,	buf);
-	return 0;
-}
-
-static int32_t
-map_open_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  fd_t *fd)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, fd);
-	return 0;
-}
-
-static int32_t
-map_readv_cbk (call_frame_t *frame,
-               void *cookie,
-               xlator_t *this,
-               int32_t op_ret,
-               int32_t op_errno,
-               struct iovec *vector,
-               int32_t count,
-               struct iatt *stbuf,
-               struct iobref *iobref)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-	
-	map_itransform (this, prev->this, stbuf->ia_ino, &stbuf->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, vector, count, stbuf, iobref);
-	return 0;
-}
-
-static int32_t
-map_writev_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-                    struct iatt *prebuf,
-		    struct iatt *postbuf)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-	
-	map_itransform (this, prev->this, postbuf->ia_ino, &postbuf->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, prebuf, postbuf);
-	return 0;
-}
-
-static int32_t
-map_flush_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-
-static int32_t
-map_fsync_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-                   struct iatt *prebuf,
-                   struct iatt *postbuf)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, prebuf, postbuf);
-	return 0;
-}
-
-
-static int32_t
-map_fstat_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-		   struct iatt *buf)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-	
-	map_itransform (this, prev->this, buf->ia_ino, &buf->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, buf);
-	return 0;
-}
-
-
-static int32_t
-map_getdents_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno,
-		      dir_entry_t *entries,
-		      int32_t count)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, entries, count);
-	return 0;
-}
-
-
-static int32_t
-map_setdents_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-static int32_t
-map_fsyncdir_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-
-static int32_t
-map_setxattr_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-
-static int32_t
-map_fsetxattr_cbk (call_frame_t *frame,
-                       void *cookie,
-                       xlator_t *this,
-                       int32_t op_ret,
-                       int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-
-static int32_t
-map_fgetxattr_cbk (call_frame_t *frame,
-                       void *cookie,
-                       xlator_t *this,
-                       int32_t op_ret,
-                       int32_t op_errno,
-                       dict_t *dict)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, dict);
-	return 0;
-}
-
-
-
-static int32_t
-map_getxattr_cbk (call_frame_t *frame,
-                      void *cookie,
-                      xlator_t *this,
-                      int32_t op_ret,
-                      int32_t op_errno,
-                      dict_t *dict)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, dict);
-	return 0;
-}
-
-int32_t
-map_xattrop_cbk (call_frame_t *frame,
-		     void *cookie,
-		     xlator_t *this,
-		     int32_t op_ret,
-		     int32_t op_errno,
-		     dict_t *dict)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, dict);
-	return 0;
-}
-
-int32_t
-map_fxattrop_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno,
-		      dict_t *dict)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, dict);
-	return 0;
-}
-
-static int32_t
-map_removexattr_cbk (call_frame_t *frame,
-			 void *cookie,
-			 xlator_t *this,
-			 int32_t op_ret,
-			 int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-
-static int32_t
-map_lk_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		struct gf_flock *lock)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, lock);
-	return 0;
-}
-
-
-static int32_t
-map_inodelk_cbk (call_frame_t *frame, void *cookie,
-		     xlator_t *this, int32_t op_ret, int32_t op_errno)
-
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-
-
-static int32_t
-map_finodelk_cbk (call_frame_t *frame, void *cookie,
-		      xlator_t *this, int32_t op_ret, int32_t op_errno)
-
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-
-static int32_t
-map_entrylk_cbk (call_frame_t *frame, void *cookie,
-		     xlator_t *this, int32_t op_ret, int32_t op_errno)
-
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-
-static int32_t
-map_fentrylk_cbk (call_frame_t *frame, void *cookie,
-		      xlator_t *this, int32_t op_ret, int32_t op_errno)
-
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-static int32_t
-map_newentry_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  inode_t *inode,
-                  struct iatt *buf,
-                  struct iatt *preparent,
-                  struct iatt *postparent)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-	
-	map_itransform (this, prev->this, buf->ia_ino, &buf->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, inode, buf);
-	return 0;
-
-}
-
-
-static int32_t
-map_create_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		fd_t *fd,
-		inode_t *inode,
-		struct iatt *buf,
-                struct iatt *preparent,
-                struct iatt *postparent)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-	
-	map_itransform (this, prev->this, buf->ia_ino, &buf->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, fd, inode, buf);
-	return 0;
-}
-
-
-/*
- * map_normalize_stats -
- */
-void
-map_normalize_stats (struct statvfs *buf,
-		     unsigned long bsize,
-		     unsigned long frsize)
-{
-	double factor;
-
-	if (buf->f_bsize != bsize) {
-		factor = ((double) buf->f_bsize) / bsize;
-		buf->f_bsize  = bsize;
-		buf->f_bfree  = (fsblkcnt_t) (factor * buf->f_bfree);
-		buf->f_bavail = (fsblkcnt_t) (factor * buf->f_bavail);
-	}
-  
-	if (buf->f_frsize != frsize) {
-		factor = ((double) buf->f_frsize) / frsize;
-		buf->f_frsize = frsize;
-		buf->f_blocks = (fsblkcnt_t) (factor * buf->f_blocks);
-	}
-}
-
-
-int32_t
-map_statfs_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		struct statvfs *stbuf)
-{
-	struct statvfs *dict_buf = NULL;
-	map_local_t  *local = NULL;
-	int           this_call_cnt = 0;
-	unsigned long bsize;
-	unsigned long frsize;
-
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		this_call_cnt = --local->call_count;
-
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			goto unlock;
-		}
-		local->op_ret = 0;
-		
-		/* when a call is successfull, add it to local->dict */
-		dict_buf = &local->statvfs;
-		
-		if (dict_buf->f_bsize != 0) {
-			bsize  = max (dict_buf->f_bsize, 
-				      stbuf->f_bsize);
-			
-			frsize = max (dict_buf->f_frsize, 
-				      stbuf->f_frsize);
-			map_normalize_stats(dict_buf, bsize, frsize);
-			map_normalize_stats(stbuf, bsize, frsize);
-		} else {
-			dict_buf->f_bsize   = stbuf->f_bsize;
-			dict_buf->f_frsize  = stbuf->f_frsize;
-		}
-		
-		dict_buf->f_blocks += stbuf->f_blocks;
-		dict_buf->f_bfree  += stbuf->f_bfree;
-		dict_buf->f_bavail += stbuf->f_bavail;
-		dict_buf->f_files  += stbuf->f_files;
-		dict_buf->f_ffree  += stbuf->f_ffree;
-		dict_buf->f_favail += stbuf->f_favail;
-		dict_buf->f_fsid    = stbuf->f_fsid;
-		dict_buf->f_flag    = stbuf->f_flag;
-		dict_buf->f_namemax = stbuf->f_namemax;
-	}
-unlock:
-	UNLOCK (&frame->lock);
-
-	if (!this_call_cnt) {
-		STACK_UNWIND (frame, local->op_ret, local->op_errno,
-			      &local->statvfs);
-	}
-
-	return 0;
-}
-
-int32_t
-map_single_lookup_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno,
-		       inode_t *inode,
-		       struct iatt *buf,
-                       dict_t *dict,
-                       struct iatt *postparent)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-	
-	map_itransform (this, prev->this, buf->ia_ino, &buf->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, inode, buf, dict);
-
-	return 0;
-}
-
-int32_t
-map_root_lookup_cbk (call_frame_t *frame,
-                     void *cookie,
-                     xlator_t *this,
-                     int32_t op_ret,
-                     int32_t op_errno,
-                     inode_t *inode,
-                     struct iatt *buf,
-                     dict_t *dict,
-                     struct iatt *postparent)
-{
-	int          callcnt = 0;
-	map_local_t *local = NULL;
-	inode_t     *tmp_inode = NULL;
-	dict_t      *tmp_dict = NULL;
-
-	local = frame->local;
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-		if ((op_ret == 0) && (local->op_ret == -1)) {
-			local->op_ret = 0;
-			local->stbuf = *buf;
-			if (dict)
-				local->dict = dict_ref (dict);
-			local->inode = inode_ref (inode);
-		}
-		if (op_ret == -1)
-			local->op_errno = op_errno;
-		
-	}
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		tmp_dict = local->dict;
-		tmp_inode = local->inode;
-
-		STACK_UNWIND (frame, local->op_ret, 
-			      local->op_errno, local->inode, 
-			      &local->stbuf, local->dict);
-
-		inode_unref (local->inode);
-		if (tmp_dict)
-			dict_unref (tmp_dict);
-	}
-
-	return 0;
-}
-
-
-int32_t
-map_opendir_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 fd_t *fd)
-{
-	int callcnt = 0;
-	map_local_t *local = NULL;
-	fd_t *local_fd = NULL;
-
-	local = frame->local;
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			goto unlock;
-		}
-
-		local->op_ret = 0;
-	}
- unlock:
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		local_fd = local->fd;
-		local->fd = NULL;
-
-		STACK_UNWIND (frame, local->op_ret, 
-			      local->op_errno, local_fd);
-
-		fd_unref (local_fd);
-	}
-	return 0;
-}
-
-int32_t
-map_single_readdir_cbk (call_frame_t *frame,
-			void *cookie,
-			xlator_t *this,
-			int32_t op_ret,
-			int32_t op_errno,
-			gf_dirent_t *entries)
-{
-        call_frame_t *prev = NULL;
-	gf_dirent_t  *orig_entry = NULL;
-
-        prev  = cookie;
-
-	list_for_each_entry (orig_entry, &entries->list, list) {
-		map_itransform (this, prev->this, orig_entry->d_ino, 
-				&orig_entry->d_ino);
-	}
-	STACK_UNWIND (frame, op_ret, op_errno, entries);
-	
-	return 0;
-}
-
-
-int32_t
-map_single_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, gf_dirent_t *entries)
-{
-        call_frame_t *prev = NULL;
-	gf_dirent_t  *orig_entry = NULL;
-
-        prev  = cookie;
-
-	list_for_each_entry (orig_entry, &entries->list, list) {
-		map_itransform (this, prev->this, orig_entry->d_ino,
-				&orig_entry->d_ino);
-                orig_entry->d_stat.ia_ino = orig_entry->d_ino;
-	}
-	STACK_UNWIND (frame, op_ret, op_errno, entries);
-	return 0;
-}
-
-int
-map_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int op_ret, int op_errno, gf_dirent_t *orig_entries);
-
-int
-map_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int op_ret, int op_errno, gf_dirent_t *orig_entries);
-
-int
-map_generic_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int op_ret, int op_errno, gf_dirent_t *orig_entries,
-                         int whichop)
-{
-	map_local_t  *local = NULL;
-	gf_dirent_t   entries;
-	gf_dirent_t  *orig_entry = NULL;
-	gf_dirent_t  *entry = NULL;
-	call_frame_t *prev = NULL;
-	xlator_t     *subvol = NULL;
-	xlator_t     *next_subvol = NULL;
-        off_t         next_offset = 0;
-	int           count = 0;
-	fd_t         *local_fd = NULL;
-
-	INIT_LIST_HEAD (&entries.list);
-	prev = cookie;
-	local = frame->local;
-
-	if (op_ret < 0)
-		goto done;
-
-	list_for_each_entry (orig_entry, &orig_entries->list, list) {
-		subvol = prev->this;
-
-		entry = gf_dirent_for_name (orig_entry->d_name);
-		if (!entry) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"memory allocation failed :(");
-			goto unwind;
-		}
-		
-		map_itransform (this, subvol, orig_entry->d_ino,
-				&entry->d_ino);
-		map_itransform (this, subvol, orig_entry->d_off,
-				&entry->d_off);
-		
-                if (whichop == GF_FOP_READDIRP)
-                        entry->d_stat.ia_ino = entry->d_ino;
-		entry->d_type = orig_entry->d_type;
-		entry->d_len  = orig_entry->d_len;
-		
-		list_add_tail (&entry->list, &entries.list);
-		count++;
-                next_offset = orig_entry->d_off;
-	}
-
-	op_ret = count;
-
-done:
-	if (count == 0) {
-                /* non-zero next_offset means that
-                   EOF is not yet hit on the current subvol
-                */
-                if (next_offset == 0) {
-                        next_subvol = map_subvol_next (this, prev->this);
-                } else {
-                        next_subvol = prev->this;
-                }
-
-		if (!next_subvol) {
-			goto unwind;
-		}
-
-                if (whichop == GF_FOP_READDIR)
-                        STACK_WIND (frame, map_readdir_cbk, next_subvol,
-                                    next_subvol->fops->readdir, local->fd,
-                                    local->size, 0);
-                else
-                        STACK_WIND (frame, map_readdirp_cbk, next_subvol,
-                                    next_subvol->fops->readdirp, local->fd,
-                                    local->size, 0);
-		return 0;
-	}
-
-unwind:
-	if (op_ret < 0)
-		op_ret = 0;
-
-	local_fd = local->fd;
-	local->fd = NULL;
-
-	STACK_UNWIND (frame, op_ret, op_errno, &entries);
-
-	fd_unref (local_fd);
-
-	gf_dirent_free (&entries);
-
-        return 0;
-}
-
-
-int
-map_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int op_ret, int op_errno, gf_dirent_t *orig_entries)
-{
-        map_generic_readdir_cbk (frame, cookie, this, op_ret, op_errno,
-                                 orig_entries, GF_FOP_READDIR);
-        return 0;
-}
-
-
-int
-map_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int op_ret, int op_errno, gf_dirent_t *orig_entries)
-{
-        map_generic_readdir_cbk (frame, cookie, this, op_ret, op_errno,
-                                 orig_entries, GF_FOP_READDIRP);
-        return 0;
-}
-
-
-/* Management operations */
-
-static int32_t
-map_checksum_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno,
-		      uint8_t *file_checksum,
-		      uint8_t *dir_checksum)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, file_checksum, dir_checksum);
-	return 0;
-}
-
-
-/* Fops starts here */
-
-int32_t
-map_stat (call_frame_t *frame,
-	  xlator_t *this,
-	  loc_t *loc)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_stat_cbk, subvol, subvol->fops->stat, loc);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_setattr (call_frame_t *frame,
-             xlator_t *this,
-             loc_t *loc,
-             struct iatt *stbuf,
-             int32_t valid)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        GF_VALIDATE_OR_GOTO ("map", this, err);
-        GF_VALIDATE_OR_GOTO (this->name, frame, err);
-        GF_VALIDATE_OR_GOTO (this->name, loc, err);
-        GF_VALIDATE_OR_GOTO (this->name, loc->inode, err);
-        GF_VALIDATE_OR_GOTO (this->name, loc->path, err);
-        GF_VALIDATE_OR_GOTO (this->name, stbuf, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_setattr_cbk, subvol,
-		    subvol->fops->setattr, loc, stbuf, valid);
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_fsetattr (call_frame_t *frame,
-              xlator_t *this,
-              fd_t *fd,
-              struct iatt *stbuf,
-              int32_t valid)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        GF_VALIDATE_OR_GOTO ("map", this, err);
-        GF_VALIDATE_OR_GOTO (this->name, frame, err);
-        GF_VALIDATE_OR_GOTO (this->name, fd, err);
-        GF_VALIDATE_OR_GOTO (this->name, stbuf, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_fsetattr_cbk, subvol,
-		    subvol->fops->fsetattr, fd, stbuf, valid);
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_truncate (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc,
-	      off_t offset)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_truncate_cbk, subvol, 
-                    subvol->fops->truncate, loc, offset);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_ftruncate (call_frame_t *frame,
-	       xlator_t *this,
-	       fd_t *fd,
-	       off_t offset)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_ftruncate_cbk, subvol,
-		    subvol->fops->ftruncate, fd, offset);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_access (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *loc,
-	    int32_t mask)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_access_cbk, subvol,
-		    subvol->fops->access, loc, mask);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_readlink (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc,
-	      size_t size)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_readlink_cbk, subvol,
-		    subvol->fops->readlink, loc, size);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_unlink (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *loc)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_unlink_cbk, subvol, subvol->fops->unlink, loc);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_rmdir (call_frame_t *frame,
-	   xlator_t *this,
-	   loc_t *loc)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_rmdir_cbk, subvol, subvol->fops->rmdir, loc);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_rename (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *oldloc,
-	    loc_t *newloc)
-{
-	int32_t op_errno = 1;
-	xlator_t *old_subvol = NULL;
-	xlator_t *new_subvol = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (oldloc, err);
-        VALIDATE_OR_GOTO (oldloc->inode, err);
-        VALIDATE_OR_GOTO (oldloc->path, err);
-        VALIDATE_OR_GOTO (newloc, err);
-
-	old_subvol = get_mapping_subvol_from_ctx (this, oldloc->inode);
-	if (!old_subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	if (newloc->path) {
-		new_subvol = get_mapping_subvol_from_path (this, newloc->path);
-		if (new_subvol && (new_subvol != old_subvol)) {
-			op_errno = EXDEV;
-			goto err;
-		}
-	}
-
-	STACK_WIND (frame, map_rename_cbk, old_subvol,
-		    old_subvol->fops->rename, oldloc, newloc);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_link (call_frame_t *frame,
-	  xlator_t *this,
-	  loc_t *oldloc,
-	  loc_t *newloc)
-{
-	int32_t op_errno = 1;
-	xlator_t *old_subvol = NULL;
-	xlator_t *new_subvol = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (oldloc, err);
-        VALIDATE_OR_GOTO (oldloc->inode, err);
-        VALIDATE_OR_GOTO (oldloc->path, err);
-        VALIDATE_OR_GOTO (newloc, err);
-
-	old_subvol = get_mapping_subvol_from_ctx (this, oldloc->inode);
-	if (!old_subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	if (newloc->path) {
-		new_subvol = get_mapping_subvol_from_path (this, newloc->path);
-		if (new_subvol && (new_subvol != old_subvol)) {
-			op_errno = EXDEV;
-			goto err;
-		}
-	}
-
-	STACK_WIND (frame, map_link_cbk, old_subvol,
-		    old_subvol->fops->link, oldloc, newloc);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_open (call_frame_t *frame,
-	  xlator_t *this,
-	  loc_t *loc,
-	  int32_t flags, fd_t *fd, int wbflags)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_open_cbk, subvol, 
-                    subvol->fops->open, loc, flags, fd, wbflags);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_readv (call_frame_t *frame,
-	   xlator_t *this,
-	   fd_t *fd,
-	   size_t size,
-	   off_t offset)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_readv_cbk, subvol,
-		    subvol->fops->readv, fd, size, offset);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, 0, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_writev (call_frame_t *frame,
-	    xlator_t *this,
-	    fd_t *fd,
-	    struct iovec *vector,
-	    int32_t count,
-	    off_t off,
-            struct iobref *iobref)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_writev_cbk, subvol,
-		    subvol->fops->writev, fd, vector, count, off, iobref);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_flush (call_frame_t *frame,
-	   xlator_t *this,
-	   fd_t *fd)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_flush_cbk, subvol, subvol->fops->flush, fd);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_fsync (call_frame_t *frame,
-	   xlator_t *this,
-	   fd_t *fd,
-	   int32_t flags)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_fsync_cbk, subvol,
-		    subvol->fops->fsync, fd, flags);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_fstat (call_frame_t *frame,
-	   xlator_t *this,
-	   fd_t *fd)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_fstat_cbk, subvol, subvol->fops->fstat, fd);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_getdents (call_frame_t *frame,
-	      xlator_t *this,
-	      fd_t *fd,
-	      size_t size,
-	      off_t offset,
-	      int32_t flag)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_getdents_cbk, subvol,
-		    subvol->fops->getdents, fd, size, offset, flag);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_setdents (call_frame_t *frame,
-	      xlator_t *this,
-	      fd_t *fd,
-	      int32_t flags,
-	      dir_entry_t *entries,
-	      int32_t count)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_setdents_cbk, subvol,
-		    subvol->fops->setdents, fd, flags, entries, count);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_fsyncdir (call_frame_t *frame,
-	      xlator_t *this,
-	      fd_t *fd,
-	      int32_t flags)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_fsyncdir_cbk, subvol,
-		    subvol->fops->fsyncdir, fd, flags);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-
-
-int32_t
-map_setxattr (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc,
-	      dict_t *dict,
-	      int32_t flags)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_setxattr_cbk, subvol,
-		    subvol->fops->setxattr, loc, dict, flags);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_getxattr (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc,
-	      const char *name)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_getxattr_cbk, subvol,
-		    subvol->fops->getxattr, loc, name);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_fsetxattr (call_frame_t *frame,
-               xlator_t *this,
-               fd_t *fd,
-               dict_t *dict,
-               int32_t flags)
-{
-	int32_t   op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_fsetxattr_cbk, subvol, 
-                    subvol->fops->fsetxattr, fd, dict, flags);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_fgetxattr (call_frame_t *frame,
-               xlator_t *this,
-               fd_t *fd,
-               const char *name)
-{
-	int32_t   op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_fgetxattr_cbk, subvol,
-		    subvol->fops->fgetxattr, fd, name);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_xattrop (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc,
-	     gf_xattrop_flags_t flags,
-	     dict_t *dict)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_xattrop_cbk, subvol,
-		    subvol->fops->xattrop, loc, flags, dict);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_fxattrop (call_frame_t *frame,
-	      xlator_t *this,
-	      fd_t *fd,
-	      gf_xattrop_flags_t flags,
-	      dict_t *dict)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_fxattrop_cbk, subvol,
-		    subvol->fops->fxattrop, fd, flags, dict);
-
-        return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_removexattr (call_frame_t *frame,
-		 xlator_t *this,
-		 loc_t *loc,
-		 const char *name)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_removexattr_cbk, subvol,
-		    subvol->fops->removexattr, loc, name);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_lk (call_frame_t *frame,
-	xlator_t *this,
-	fd_t *fd,
-	int32_t cmd,
-	struct gf_flock *lock)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_lk_cbk, subvol,
-		    subvol->fops->lk, fd, cmd, lock);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_inodelk (call_frame_t *frame, xlator_t *this,
-	     const char *volume, loc_t *loc, int32_t cmd, struct gf_flock *lock)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_inodelk_cbk, subvol,
-		    subvol->fops->inodelk, volume, loc, cmd, lock);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_finodelk (call_frame_t *frame, xlator_t *this,
-	      const char *volume, fd_t *fd, int32_t cmd, struct gf_flock *lock)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_finodelk_cbk, subvol,
-		    subvol->fops->finodelk, volume, fd, cmd, lock);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_entrylk (call_frame_t *frame, xlator_t *this,
-	     const char *volume, loc_t *loc, const char *basename,
-	     entrylk_cmd cmd, entrylk_type type)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_entrylk_cbk, subvol,
-		    subvol->fops->entrylk, volume, loc, basename, cmd, type);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_fentrylk (call_frame_t *frame, xlator_t *this,
-	      const char *volume, fd_t *fd, const char *basename,
-	      entrylk_cmd cmd, entrylk_type type)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_fentrylk_cbk, subvol,
-		    subvol->fops->fentrylk, volume, fd, basename, cmd, type);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_checksum (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc,
-	      int32_t flag)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_checksum_cbk, subvol,
-		    subvol->fops->checksum, loc, flag);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_mknod (call_frame_t *frame,
-	   xlator_t *this,
-	   loc_t *loc,
-	   mode_t mode,
-	   dev_t rdev)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-	subvol = get_mapping_subvol_from_path (this, loc->path);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-	
-	op_errno = inode_ctx_put (loc->inode, this, (uint64_t)(long)subvol);
-	if (op_errno != 0) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"%s: failed to set subvolume ptr in inode ctx",
-			loc->path);
-	}
-
-	STACK_WIND (frame, map_newentry_cbk, subvol,
-		    subvol->fops->mknod, loc, mode, rdev);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_mkdir (call_frame_t *frame,
-	   xlator_t *this,
-	   loc_t *loc,
-	   mode_t mode)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-	subvol = get_mapping_subvol_from_path (this, loc->path);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	op_errno = inode_ctx_put (loc->inode, this, (uint64_t)(long)subvol);
-	if (op_errno != 0) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"%s: failed to set subvolume ptr in inode ctx",
-			loc->path);
-	}
-
-	STACK_WIND (frame, map_newentry_cbk, subvol,
-		    subvol->fops->mkdir, loc, mode);
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_symlink (call_frame_t *frame,
-	     xlator_t *this,
-	     const char *linkpath,
-	     loc_t *loc)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-	subvol = get_mapping_subvol_from_path (this, loc->path);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-	
-	op_errno = inode_ctx_put (loc->inode, this, (uint64_t)(long)subvol);
-	if (op_errno != 0) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"%s: failed to set subvolume ptr in inode ctx",
-			loc->path);
-	}
-
-	STACK_WIND (frame, map_newentry_cbk, subvol,
-		    subvol->fops->symlink, linkpath, loc);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_create (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *loc,
-	    int32_t flags,
-	    mode_t mode, fd_t *fd)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-	subvol = get_mapping_subvol_from_path (this, loc->path);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	op_errno = inode_ctx_put (loc->inode, this, (uint64_t)(long)subvol);
-	if (op_errno != 0) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"%s: failed to set subvolume ptr in inode ctx",
-			loc->path);
-	}
-
-	STACK_WIND (frame, map_create_cbk, subvol,
-		    subvol->fops->create, loc, flags, mode, fd);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_lookup (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *loc,
-	    dict_t *xattr_req)
-{
-	int32_t op_errno = EINVAL;
-	xlator_t *subvol   = NULL;
-	map_local_t *local = NULL;
-	map_private_t *priv = NULL;
-	xlator_list_t *trav = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-	priv = this->private;
-
-	if (loc->inode->ino == 1)
-		goto root_inode;
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		subvol = get_mapping_subvol_from_path (this, loc->path);
-		if (!subvol) {
-			goto err;
-		}
-
-		op_errno = inode_ctx_put (loc->inode, this, 
-					  (uint64_t)(long)subvol);
-		if (op_errno != 0) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"%s: failed to set subvolume in inode ctx",
-				loc->path);
-		}
-	}
-
-	/* Just one callback */
-	STACK_WIND (frame, map_single_lookup_cbk, subvol,
-		    subvol->fops->lookup, loc, xattr_req);
-
-	return 0;
-
- root_inode:
-	local = GF_CALLOC (1, sizeof (map_local_t),
-                           gf_map_mt_map_local_t);
-
-	frame->local = local;
-	local->call_count = priv->child_count;
-	local->op_ret = -1;
-
-	trav = this->children;
-	while (trav) {
-		STACK_WIND (frame, map_root_lookup_cbk, trav->xlator,
-			    trav->xlator->fops->lookup, loc, xattr_req);
-		trav = trav->next;
-	}
-
-	return 0;
-
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_statfs (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *loc)
-{
-	int32_t op_errno = EINVAL;
-	xlator_t *subvol   = NULL;
-	map_local_t *local = NULL;
-	map_private_t *priv = NULL;
-	xlator_list_t *trav = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-	if (loc->inode->ino == 1)
-		goto root_inode;
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		goto err;
-	}
-	
-	/* Just one callback */
-	STACK_WIND (frame, map_statfs_cbk, subvol, subvol->fops->statfs, loc);
-
-	return 0;
-
- root_inode:
-	local = GF_CALLOC (1, sizeof (map_local_t),
-                           gf_map_mt_map_local_t);
-
-	priv = this->private;
-	frame->local = local;
-	local->call_count = priv->child_count;
-	local->op_ret = -1;
-
-	trav = this->children;
-	while (trav) {
-		STACK_WIND (frame, map_statfs_cbk, trav->xlator,
-			    trav->xlator->fops->statfs, loc);
-		trav = trav->next;
-	}
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL);
-
-	return 0;
-}
-
-int32_t
-map_opendir (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc, fd_t *fd)
-{
-	int32_t op_errno = EINVAL;
-	xlator_t *subvol   = NULL;
-	map_local_t *local = NULL;
-	map_private_t *priv = NULL;
-	xlator_list_t *trav = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	if (loc->inode->ino == 1) 
-		goto root_inode;
-	
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		goto err;
-	}
-	
-	/* Just one callback */
-	STACK_WIND (frame, map_opendir_cbk, subvol,
-		    subvol->fops->opendir, loc, fd);
-
-	return 0;
-
- root_inode:
-	local = GF_CALLOC (1, sizeof (map_local_t),
-                           gf_map_mt_map_local_t);
-
-	priv = this->private;
-	frame->local = local;
-	local->call_count = priv->child_count;
-	local->op_ret = -1;
-	local->fd = fd_ref (fd);
-
-	trav = this->children;
-	while (trav) {
-		STACK_WIND (frame, map_opendir_cbk, trav->xlator,
-			    trav->xlator->fops->opendir, loc, fd);
-		trav = trav->next;
-	}
-	
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_do_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-                off_t yoff, int whichop)
-{
-	int32_t        op_errno = EINVAL;
-	xlator_t      *subvol = NULL;
-	map_local_t   *local = NULL;
-	map_private_t *priv = NULL;
-	xlator_t      *xvol = NULL;
-	off_t          xoff = 0;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	if (fd->inode->ino == 1) 
-		goto root_inode;
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		goto err;
-	}
-	
-	/* Just one callback */
-        if (whichop == GF_FOP_READDIR)
-                STACK_WIND (frame, map_single_readdir_cbk, subvol,
-                            subvol->fops->readdir, fd, size, yoff);
-        else
-                STACK_WIND (frame, map_single_readdirp_cbk, subvol,
-                            subvol->fops->readdirp, fd, size, yoff);
-
-	return 0;
-
- root_inode:
-	/* readdir on '/' */
-	local = GF_CALLOC (1, sizeof (map_local_t),
-                           gf_map_mt_map_local_t);
-	if (!local) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		op_errno = ENOMEM;
-		goto err;
-	}
-	
-	priv = this->private;
-	frame->local = local;
-	local->op_errno = ENOENT;
-	local->op_ret = -1;
-
-	local->fd = fd_ref (fd);
-	local->size = size;
-
-	map_deitransform (this, yoff, &xvol, (uint64_t *)&xoff);
-
-        if (whichop == GF_FOP_READDIR)
-                STACK_WIND (frame, map_readdir_cbk, xvol, xvol->fops->readdir,
-                            fd, size, xoff);
-        else
-                STACK_WIND (frame, map_readdirp_cbk, xvol, xvol->fops->readdirp,
-                            fd, size, xoff);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL);
-
-	return 0;
-}
-
-
-
-int32_t
-map_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-             off_t yoff)
-{
-        map_do_readdir (frame, this, fd, size, yoff, GF_FOP_READDIR);
-        return 0;
-}
-
-int32_t
-map_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-              off_t yoff)
-{
-        map_do_readdir (frame, this, fd, size, yoff, GF_FOP_READDIRP);
-        return 0;
-}
-
-
-void
-fini (xlator_t *this)
-{
-	map_private_t *priv = NULL;
-	struct map_pattern *trav_map = NULL;
-	struct map_pattern *tmp_map  = NULL;
-
-	priv = this->private;
-
-	if (priv) {
-		GF_FREE (priv->xlarray);
-
-		trav_map = priv->map;
-		while (trav_map) {
-			tmp_map = trav_map;
-			trav_map = trav_map->next;
-			GF_FREE (tmp_map);
-		}
-
-		GF_FREE(priv);
-	}
-
-	return;
-}
-
-int32_t
-mem_acct_init (xlator_t *this)
-{
-        int     ret = -1;
-
-        if (!this)
-                return ret;
-
-        ret = xlator_mem_acct_init (this, gf_map_mt_end + 1);
-        
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
-                                "failed");
-                return ret;
-        }
-
-        return ret;
-}
-
-int
-init (xlator_t *this)
-{
-	map_private_t *priv = NULL;
-	xlator_list_t *trav = NULL;
-	int   count = 0;
-	int   ret = -1;
-	char *pattern_string = NULL;
-	char *map_pair_str = NULL;
-	char *tmp_str = NULL;
-	char *tmp_str1 = NULL;
-	char *dup_map_pair = NULL;
-	char *dir_str = NULL;
-	char *subvol_str = NULL;
-	char *map_xl = NULL;
-
-
-	if (!this->children) {
-		gf_log (this->name,GF_LOG_ERROR,
-			"FATAL: map should have one or more child defined");
-		return -1;
-	}
-
-	if (!this->parents) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"dangling volume. check volfile ");
-	}
-
-	priv = GF_CALLOC (1, sizeof (map_private_t),
-                          gf_map_mt_map_private_t);
-	this->private = priv;
-
-	/* allocate xlator array */
-	trav = this->children;
-	while (trav) {
-		count++;
-		trav = trav->next;
-	}
-	priv->xlarray = GF_CALLOC (1, sizeof (struct map_xlator_array) * count,
-                                   gf_map_mt_map_xlator_array);
-	priv->child_count = count;
-
-	/* build xlator array */
-	count = 0;
-	trav = this->children;
-	while (trav) {
-		priv->xlarray[count++].xl = trav->xlator;
-		trav = trav->next;
-	}
-
-	/* map dir1:brick1;dir2:brick2;dir3:brick3;*:brick4 */
-	ret = dict_get_str (this->options, "map-directory", &pattern_string);
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"map.pattern not given, can't continue");
-		goto err;
-	}
-	map_pair_str = strtok_r (pattern_string, ";", &tmp_str);
-	while (map_pair_str) {
-		dup_map_pair = gf_strdup (map_pair_str);
-		dir_str = strtok_r (dup_map_pair, ":", &tmp_str1);
-		if (!dir_str) {
-			gf_log (this->name, GF_LOG_ERROR, 
-				"directory string invalid");
-			goto err;
-		}
-		subvol_str = strtok_r (NULL, ":", &tmp_str1);
-		if (!subvol_str) {
-			gf_log (this->name, GF_LOG_ERROR, 
-				"mapping subvolume string invalid");
-			goto err;
-		}
-		ret = verify_dir_and_assign_subvol (this, 
-						    dir_str, 
-						    subvol_str);
-		if (ret != 0) {
-			gf_log (this->name, GF_LOG_ERROR, 
-				"verification failed");
-			goto err;
-		}
-		
-		GF_FREE (dup_map_pair);
-
-		map_pair_str = strtok_r (NULL, ";", &tmp_str);
-	}
-
-	/* default-volume brick4 */
-	ret = dict_get_str (this->options, "default-volume", &map_xl);
-	if (ret == 0) {
-		ret = assign_default_subvol (this, map_xl);
-		if (ret != 0) {
-			gf_log (this->name, GF_LOG_ERROR, 
-				"assigning default failed");
-			goto err;
-		}
-	}
-
-	verify_if_all_subvolumes_got_used (this);
-	
-	return 0;
- err:
-	fini (this);
-	return -1;
-}
-
-
-struct xlator_fops fops = {
-	.lookup      = map_lookup,
-	.mknod       = map_mknod,
-	.create      = map_create,
-
-	.stat        = map_stat,
-	.fstat       = map_fstat,
-	.truncate    = map_truncate,
-	.ftruncate   = map_ftruncate,
-	.access      = map_access,
-	.readlink    = map_readlink,
-	.setxattr    = map_setxattr,
-	.getxattr    = map_getxattr,
-	.fsetxattr   = map_fsetxattr,
-	.fgetxattr   = map_fgetxattr,
-	.removexattr = map_removexattr,
-	.open        = map_open,
-	.readv       = map_readv,
-	.writev      = map_writev,
-	.flush       = map_flush,
-	.fsync       = map_fsync,
-	.statfs      = map_statfs,
-	.lk          = map_lk,
-	.opendir     = map_opendir,
-	.readdir     = map_readdir,
-	.readdirp    = map_readdirp,
-	.fsyncdir    = map_fsyncdir,
-	.symlink     = map_symlink,
-	.unlink      = map_unlink,
-	.link        = map_link,
-	.mkdir       = map_mkdir,
-	.rmdir       = map_rmdir,
-	.rename      = map_rename,
-	.inodelk     = map_inodelk,
-	.finodelk    = map_finodelk,
-	.entrylk     = map_entrylk,
-	.fentrylk    = map_fentrylk,
-	.xattrop     = map_xattrop,
-	.fxattrop    = map_fxattrop,
-	.setdents    = map_setdents,
-	.getdents    = map_getdents,
-	.checksum    = map_checksum,
-        .setattr     = map_setattr,
-        .fsetattr    = map_fsetattr,
-};
-
-struct xlator_cbks cbks = {
-};
-
-struct volume_options options[] = {
-	{ .key   = {"map-directory"},  
-	  .type  = GF_OPTION_TYPE_ANY 
-	},
-	{ .key   = {"default-volume"},  
-	  .type  = GF_OPTION_TYPE_XLATOR 
-	},
-	
-	{ .key = {NULL} }
-};
diff --git a/xlators/cluster/map/src/map.h b/xlators/cluster/map/src/map.h
deleted file mode 100644
index 7703a543e28..00000000000
--- a/xlators/cluster/map/src/map.h
+++ /dev/null
@@ -1,67 +0,0 @@
-/*
-   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
-   This file is part of GlusterFS.
-
-   This file is licensed to you under your choice of the GNU Lesser
-   General Public License, version 3 or any later version (LGPLv3 or
-   later), or the GNU General Public License, version 2 (GPLv2), in all
-   cases as published by the Free Software Foundation.
-*/
-#ifndef __MAP_H__
-#define __MAP_H__
-
-#include "xlator.h"
-#include "map-mem-types.h"
-
-struct map_pattern {
-	struct map_pattern *next;
-	xlator_t           *xl;
-	char               *directory;
-	int                 dir_len;
-};
-
-struct map_xlator_array {
-	xlator_t *xl;
-	int       mapped; /* yes/no */
-};
-
-typedef struct {
-	struct map_pattern      *map;
-	xlator_t                *default_xl;
-	struct map_xlator_array *xlarray;
-	int                      child_count;
-} map_private_t;
-
-typedef struct {
-	int32_t        op_ret;
-	int32_t        op_errno;
-	int            call_count;
-	struct statvfs statvfs;
-	struct iatt    stbuf;
-	inode_t       *inode;
-	dict_t        *dict;
-	fd_t          *fd;
-
-	size_t        size;
-} map_local_t;
-
-xlator_t *map_subvol_next (xlator_t *this, xlator_t *prev);
-int map_subvol_cnt (xlator_t *this, xlator_t *subvol);
-
-int map_itransform (xlator_t *this, xlator_t *subvol, 
-		    uint64_t x, uint64_t *y_p);
-int map_deitransform (xlator_t *this, uint64_t y, 
-		      xlator_t **subvol_p, uint64_t *x_p);
-
-
-xlator_t *get_mapping_subvol_from_path (xlator_t *this, const char *path);
-xlator_t *get_mapping_subvol_from_ctx (xlator_t *this, inode_t *inode);
-
-int check_multiple_volume_entry (xlator_t *this, xlator_t *subvol);
-int verify_dir_and_assign_subvol (xlator_t *this, 
-				  const char *directory, const char *subvol);
-int assign_default_subvol (xlator_t *this, const char *default_xl);
-void verify_if_all_subvolumes_got_used (xlator_t *this);
-
-
-#endif /* __MAP_H__ */
diff --git a/xlators/cluster/stripe/src/Makefile.am b/xlators/cluster/stripe/src/Makefile.am
deleted file mode 100644
index 2d151422ab6..00000000000
--- a/xlators/cluster/stripe/src/Makefile.am
+++ /dev/null
@@ -1,20 +0,0 @@
-
-xlator_LTLIBRARIES = stripe.la
-xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/cluster
-
-stripe_la_LDFLAGS = -module -avoid-version
-
-stripe_la_SOURCES = stripe.c stripe-helpers.c \
-	$(top_builddir)/xlators/lib/src/libxlator.c
-
-stripe_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
-
-noinst_HEADERS = stripe.h stripe-mem-types.h $(top_builddir)/xlators/lib/src/libxlator.h
-
-AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
-	-I$(top_srcdir)/xlators/lib/src
-
-AM_CFLAGS = -Wall $(GF_CFLAGS)
-
-CLEANFILES = 
-
diff --git a/xlators/cluster/stripe/src/stripe-helpers.c b/xlators/cluster/stripe/src/stripe-helpers.c
deleted file mode 100644
index a83abdc72c8..00000000000
--- a/xlators/cluster/stripe/src/stripe-helpers.c
+++ /dev/null
@@ -1,675 +0,0 @@
-/*
-  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
-  This file is part of GlusterFS.
-
-  This file is licensed to you under your choice of the GNU Lesser
-  General Public License, version 3 or any later version (LGPLv3 or
-  later), or the GNU General Public License, version 2 (GPLv2), in all
-  cases as published by the Free Software Foundation.
-*/
-
-#include <fnmatch.h>
-
-#include "stripe.h"
-#include "byte-order.h"
-#include "mem-types.h"
-
-void
-stripe_local_wipe (stripe_local_t *local)
-{
-        if (!local)
-                goto out;
-
-        loc_wipe (&local->loc);
-        loc_wipe (&local->loc2);
-
-        if (local->fd)
-                fd_unref (local->fd);
-
-        if (local->inode)
-                inode_unref (local->inode);
-
-        if (local->xattr)
-                dict_unref (local->xattr);
-
-        if (local->xdata)
-                dict_unref (local->xdata);
-
-out:
-        return;
-}
-
-
-
-int
-stripe_aggregate (dict_t *this, char *key, data_t *value, void *data)
-{
-        dict_t  *dst  = NULL;
-        int64_t *ptr  = 0, *size = NULL;
-        int32_t  ret  = -1;
-
-        dst = data;
-
-        if (strcmp (key, GF_XATTR_QUOTA_SIZE_KEY) == 0) {
-                ret = dict_get_bin (dst, key, (void **)&size);
-                if (ret < 0) {
-                        size = GF_CALLOC (1, sizeof (int64_t),
-                                          gf_common_mt_char);
-                        if (size == NULL) {
-                                gf_log ("stripe", GF_LOG_WARNING,
-                                        "memory allocation failed");
-                                goto out;
-                        }
-                        ret = dict_set_bin (dst, key, size, sizeof (int64_t));
-                        if (ret < 0) {
-                                gf_log ("stripe", GF_LOG_WARNING,
-                                        "stripe aggregate dict set failed");
-                                GF_FREE (size);
-                                goto out;
-                        }
-                }
-
-                ptr = data_to_bin (value);
-                if (ptr == NULL) {
-                        gf_log ("stripe", GF_LOG_WARNING, "data to bin failed");
-                        goto out;
-                }
-
-                *size = hton64 (ntoh64 (*size) + ntoh64 (*ptr));
-        } else if (strcmp (key, GF_CONTENT_KEY)) {
-                /* No need to aggregate 'CONTENT' data */
-                ret = dict_set (dst, key, value);
-                if (ret)
-                        gf_log ("stripe", GF_LOG_WARNING, "xattr dict set failed");
-        }
-
-out:
-        return 0;
-}
-
-
-void
-stripe_aggregate_xattr (dict_t *dst, dict_t *src)
-{
-        if ((dst == NULL) || (src == NULL)) {
-                goto out;
-        }
-
-        dict_foreach (src, stripe_aggregate, dst);
-out:
-        return;
-}
-
-
-int32_t
-stripe_xattr_aggregate (char *buffer, stripe_local_t *local, int32_t *total)
-{
-        int32_t              i     = 0;
-        int32_t              ret   = -1;
-        int32_t              len   = 0;
-        char                *sbuf  = NULL;
-        stripe_xattr_sort_t *xattr = NULL;
-
-        if (!buffer || !local || !local->xattr_list)
-                goto out;
-
-        sbuf = buffer;
-
-        for (i = 0; i < local->nallocs; i++) {
-                xattr = local->xattr_list + i;
-                len = xattr->xattr_len;
-
-                if (len && xattr && xattr->xattr_value) {
-                        memcpy (buffer, xattr->xattr_value, len);
-                        buffer += len;
-                        *buffer++ = ' ';
-                }
-        }
-
-        *--buffer = '\0';
-        if (total)
-                *total = buffer - sbuf;
-        ret = 0;
-
- out:
-        return ret;
-}
-
-int32_t
-stripe_free_xattr_str (stripe_local_t *local)
-{
-        int32_t              i     = 0;
-        int32_t              ret   = -1;
-        stripe_xattr_sort_t *xattr = NULL;
-
-        if (!local || !local->xattr_list)
-                goto out;
-
-        for (i = 0; i < local->nallocs; i++) {
-                xattr = local->xattr_list + i;
-
-                if (xattr && xattr->xattr_value)
-                        GF_FREE (xattr->xattr_value);
-        }
-
-        ret = 0;
- out:
-        return ret;
-}
-
-
-int32_t
-stripe_fill_lockinfo_xattr (xlator_t *this, stripe_local_t *local,
-                            void **xattr_serz)
-{
-        int32_t              ret   = -1, i = 0, len = 0;
-        dict_t              *tmp1  = NULL, *tmp2 = NULL;
-        char                *buf   = NULL;
-        stripe_xattr_sort_t *xattr = NULL;
-
-        if (xattr_serz == NULL) {
-                goto out;
-        }
-
-        tmp2 = dict_new ();
-
-        if (tmp2 == NULL) {
-                goto out;
-        }
-
-        for (i = 0; i < local->nallocs; i++) {
-                xattr = local->xattr_list + i;
-                len = xattr->xattr_len;
-
-                if (len && xattr && xattr->xattr_value) {
-                        ret = dict_reset (tmp2);
-                        if (ret < 0) {
-                                gf_log (this->name, GF_LOG_DEBUG,
-                                        "dict_reset failed (%s)",
-                                        strerror (-ret));
-                        }
-
-                        ret = dict_unserialize (xattr->xattr_value,
-                                                xattr->xattr_len,
-                                                &tmp2);
-                        if (ret < 0) {
-                                gf_log (this->name, GF_LOG_WARNING,
-                                        "dict_unserialize failed (%s)",
-                                        strerror (-ret));
-                                ret = -1;
-                                goto out;
-                        }
-
-                        tmp1 = dict_copy (tmp2, tmp1);
-                        if (tmp1 == NULL) {
-                                gf_log (this->name, GF_LOG_WARNING,
-                                        "dict_copy failed (%s)",
-                                        strerror (-ret));
-                                ret = -1;
-                                goto out;
-                        }
-                }
-        }
-
-        len = dict_serialized_length (tmp1);
-        if (len > 0) {
-                buf = GF_CALLOC (1, len, gf_common_mt_dict_t);
-                if (buf == NULL) {
-                        ret = -1;
-                        goto out;
-                }
-
-                ret = dict_serialize (tmp1, buf);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "dict_serialize failed (%s)", strerror (-ret));
-                        ret = -1;
-                        goto out;
-                }
-
-                *xattr_serz = buf;
-        }
-
-        ret = 0;
-out:
-        if (tmp1 != NULL) {
-                dict_unref (tmp1);
-        }
-
-        if (tmp2 != NULL) {
-                dict_unref (tmp2);
-        }
-
-        return ret;
-}
-
-
-int32_t
-stripe_fill_pathinfo_xattr (xlator_t *this, stripe_local_t *local,
-                            char **xattr_serz)
-{
-        int      ret             = -1;
-        int32_t  padding         = 0;
-        int32_t  tlen            = 0;
-        char stripe_size_str[20] = {0,};
-        char    *pathinfo_serz   = NULL;
-
-        if (!local) {
-                gf_log (this->name, GF_LOG_ERROR, "Possible NULL deref");
-                goto out;
-        }
-
-        (void) snprintf (stripe_size_str, 20, "%ld",
-                         (local->fctx) ? local->fctx->stripe_size : 0);
-
-        /* extra bytes for decorations (brackets and <>'s) */
-        padding = strlen (this->name) + strlen (STRIPE_PATHINFO_HEADER)
-                + strlen (stripe_size_str) + 7;
-        local->xattr_total_len += (padding + 2);
-
-        pathinfo_serz = GF_CALLOC (local->xattr_total_len, sizeof (char),
-                                   gf_common_mt_char);
-        if (!pathinfo_serz)
-                goto out;
-
-        /* xlator info */
-        (void) sprintf (pathinfo_serz, "(<"STRIPE_PATHINFO_HEADER"%s:[%s]> ",
-                        this->name, stripe_size_str);
-
-        ret = stripe_xattr_aggregate (pathinfo_serz + padding, local, &tlen);
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Cannot aggregate pathinfo list");
-                goto out;
-        }
-
-        *(pathinfo_serz + padding + tlen) = ')';
-        *(pathinfo_serz + padding + tlen + 1) = '\0';
-
-        *xattr_serz = pathinfo_serz;
-
-        ret = 0;
- out:
-        return ret;
-}
-
-/**
- * stripe_get_matching_bs - Get the matching block size for the given path.
- */
-int32_t
-stripe_get_matching_bs (const char *path, stripe_private_t *priv)
-{
-        struct stripe_options *trav       = NULL;
-        uint64_t               block_size = 0;
-
-        GF_VALIDATE_OR_GOTO ("stripe", priv, out);
-        GF_VALIDATE_OR_GOTO ("stripe", path, out);
-
-        LOCK (&priv->lock);
-        {
-                block_size = priv->block_size;
-                trav = priv->pattern;
-                while (trav) {
-                        if (!fnmatch (trav->path_pattern, path, FNM_NOESCAPE)) {
-                                block_size = trav->block_size;
-                                break;
-                        }
-                        trav = trav->next;
-                }
-        }
-        UNLOCK (&priv->lock);
-
-out:
-        return block_size;
-}
-
-int32_t
-stripe_ctx_handle (xlator_t *this, call_frame_t *prev, stripe_local_t *local,
-                   dict_t *dict)
-{
-        char            key[256]       = {0,};
-        data_t         *data            = NULL;
-        int32_t         index           = 0;
-        stripe_private_t *priv          = NULL;
-
-        priv = this->private;
-
-
-        if (!local->fctx) {
-                local->fctx =  GF_CALLOC (1, sizeof (stripe_fd_ctx_t),
-                                         gf_stripe_mt_stripe_fd_ctx_t);
-                if (!local->fctx) {
-                        local->op_errno = ENOMEM;
-                        local->op_ret = -1;
-                        goto out;
-                }
-
-                local->fctx->static_array = 0;
-        }
-        /* Stripe block size */
-        sprintf (key, "trusted.%s.stripe-size", this->name);
-        data = dict_get (dict, key);
-        if (!data) {
-                local->xattr_self_heal_needed = 1;
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Failed to get stripe-size");
-                goto out;
-        } else {
-                if (!local->fctx->stripe_size) {
-                        local->fctx->stripe_size =
-                                     data_to_int64 (data);
-                }
-
-                if (local->fctx->stripe_size != data_to_int64 (data)) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "stripe-size mismatch in blocks");
-                        local->xattr_self_heal_needed = 1;
-                }
-        }
-
-        /* Stripe count */
-        sprintf (key, "trusted.%s.stripe-count", this->name);
-        data = dict_get (dict, key);
-
-        if (!data) {
-                local->xattr_self_heal_needed = 1;
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Failed to get stripe-count");
-                goto out;
-        }
-        if (!local->fctx->xl_array) {
-                local->fctx->stripe_count = data_to_int32 (data);
-                if (!local->fctx->stripe_count) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "error with stripe-count xattr");
-                        local->op_ret   = -1;
-                        local->op_errno = EIO;
-                        goto out;
-                }
-
-                local->fctx->xl_array = GF_CALLOC (local->fctx->stripe_count,
-                                                   sizeof (xlator_t *),
-                                                   gf_stripe_mt_xlator_t);
-
-                if (!local->fctx->xl_array) {
-                        local->op_errno = ENOMEM;
-                        local->op_ret   = -1;
-                        goto out;
-                }
-        }
-        if (local->fctx->stripe_count != data_to_int32 (data)) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "error with stripe-count xattr (%d != %d)",
-                        local->fctx->stripe_count, data_to_int32 (data));
-                local->op_ret   = -1;
-                local->op_errno = EIO;
-                goto out;
-        }
-
-        /* index */
-        sprintf (key, "trusted.%s.stripe-index", this->name);
-        data = dict_get (dict, key);
-        if (!data) {
-                local->xattr_self_heal_needed = 1;
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Failed to get stripe-index");
-                goto out;
-        }
-        index = data_to_int32 (data);
-        if (index > priv->child_count) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "error with stripe-index xattr (%d)", index);
-                local->op_ret   = -1;
-                local->op_errno = EIO;
-                goto out;
-        }
-        if (local->fctx->xl_array) {
-                if (!local->fctx->xl_array[index])
-                        local->fctx->xl_array[index] = prev->this;
-        }
-
-	sprintf(key, "trusted.%s.stripe-coalesce", this->name);
-	data = dict_get(dict, key);
-	if (!data) {
-		/*
-		 * The file was probably created prior to coalesce support.
-		 * Assume non-coalesce mode for this file to maintain backwards
-		 * compatibility.
-		 */
-		gf_log(this->name, GF_LOG_DEBUG, "missing stripe-coalesce "
-			"attr, assume non-coalesce mode");
-		local->fctx->stripe_coalesce = 0;
-	} else {
-		local->fctx->stripe_coalesce = data_to_int32(data);
-	}
-
-
-out:
-        return 0;
-}
-
-int32_t
-stripe_xattr_request_build (xlator_t *this, dict_t *dict, uint64_t stripe_size,
-                            uint32_t stripe_count, uint32_t stripe_index,
-			    uint32_t stripe_coalesce)
-{
-        char            key[256]       = {0,};
-        int32_t         ret             = -1;
-
-        sprintf (key, "trusted.%s.stripe-size", this->name);
-        ret = dict_set_int64 (dict, key, stripe_size);
-        if (ret) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "failed to set %s in xattr_req dict", key);
-                goto out;
-        }
-
-        sprintf (key, "trusted.%s.stripe-count", this->name);
-        ret = dict_set_int32 (dict, key, stripe_count);
-        if (ret) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "failed to set %s in xattr_req dict", key);
-                goto out;
-        }
-
-        sprintf (key, "trusted.%s.stripe-index", this->name);
-        ret = dict_set_int32 (dict, key, stripe_index);
-        if (ret) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "failed to set %s in xattr_req dict", key);
-                goto out;
-        }
-
-	sprintf(key, "trusted.%s.stripe-coalesce", this->name);
-	ret = dict_set_int32(dict, key, stripe_coalesce);
-	if (ret) {
-		gf_log(this->name, GF_LOG_WARNING,
-			"failed to set %s in xattr_req_dict", key);
-		goto out;
-	}
-out:
-        return ret;
-}
-
-
-static int
-set_default_block_size (stripe_private_t *priv, char *num)
-{
-
-        int             ret = -1;
-        GF_VALIDATE_OR_GOTO ("stripe", THIS, out);
-        GF_VALIDATE_OR_GOTO (THIS->name, priv, out);
-        GF_VALIDATE_OR_GOTO (THIS->name, num, out);
-
-
-        if (gf_string2bytesize (num, &priv->block_size) != 0) {
-                gf_log (THIS->name, GF_LOG_ERROR,
-                        "invalid number format \"%s\"", num);
-                goto out;
-        }
-
-        ret = 0;
-
- out:
-        return ret;
-
-}
-
-
-int
-set_stripe_block_size (xlator_t *this, stripe_private_t *priv, char *data)
-{
-        int                    ret = -1;
-        char                  *tmp_str = NULL;
-        char                  *tmp_str1 = NULL;
-        char                  *dup_str = NULL;
-        char                  *stripe_str = NULL;
-        char                  *pattern = NULL;
-        char                  *num = NULL;
-        struct stripe_options *temp_stripeopt = NULL;
-        struct stripe_options *stripe_opt = NULL;
-
-        if (!this || !priv || !data)
-                goto out;
-
-        /* Get the pattern for striping.
-           "option block-size *avi:10MB" etc */
-        stripe_str = strtok_r (data, ",", &tmp_str);
-        while (stripe_str) {
-                dup_str = gf_strdup (stripe_str);
-                stripe_opt = GF_CALLOC (1, sizeof (struct stripe_options),
-                                        gf_stripe_mt_stripe_options);
-                if (!stripe_opt) {
-                        goto out;
-                }
-
-                pattern = strtok_r (dup_str, ":", &tmp_str1);
-                num = strtok_r (NULL, ":", &tmp_str1);
-                if (!num) {
-                        num = pattern;
-                        pattern = "*";
-                        ret = set_default_block_size (priv, num);
-                        if (ret)
-                                goto out;
-                }
-                if (gf_string2bytesize (num, &stripe_opt->block_size) != 0) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "invalid number format \"%s\"", num);
-                        goto out;
-                }
-
-                if (stripe_opt->block_size < STRIPE_MIN_BLOCK_SIZE) {
-                        gf_log (this->name, GF_LOG_ERROR, "Invalid Block-size: "
-                                "%s. Should be atleast %llu bytes", num,
-                                STRIPE_MIN_BLOCK_SIZE);
-                        goto out;
-                }
-                if (stripe_opt->block_size % 512) {
-                        gf_log (this->name, GF_LOG_ERROR, "Block-size: %s should"
-                                " be a multiple of 512 bytes", num);
-                        goto out;
-                }
-
-                memcpy (stripe_opt->path_pattern, pattern, strlen (pattern));
-
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "block-size : pattern %s : size %"PRId64,
-                        stripe_opt->path_pattern, stripe_opt->block_size);
-
-                if (priv->pattern)
-                        temp_stripeopt = NULL;
-                else
-                        temp_stripeopt = priv->pattern;
-
-                stripe_opt->next = temp_stripeopt;
-
-                priv->pattern = stripe_opt;
-                stripe_opt = NULL;
-
-                GF_FREE (dup_str);
-                dup_str = NULL;
-
-                stripe_str = strtok_r (NULL, ",", &tmp_str);
-        }
-
-        ret = 0;
-out:
-
-        GF_FREE (dup_str);
-
-        GF_FREE (stripe_opt);
-
-        return ret;
-}
-
-int32_t
-stripe_iatt_merge (struct iatt *from, struct iatt *to)
-{
-        if (to->ia_size < from->ia_size)
-                to->ia_size = from->ia_size;
-        if (to->ia_mtime < from->ia_mtime)
-                to->ia_mtime = from->ia_mtime;
-        if (to->ia_ctime < from->ia_ctime)
-                to->ia_ctime = from->ia_ctime;
-        if (to->ia_atime < from->ia_atime)
-                to->ia_atime = from->ia_atime;
-        return 0;
-}
-
-off_t
-coalesced_offset(off_t offset, uint64_t stripe_size, int stripe_count)
-{
-	size_t line_size = 0;
-	uint64_t stripe_num = 0;
-	off_t coalesced_offset = 0;
-
-	line_size = stripe_size * stripe_count;
-	stripe_num = offset / line_size;
-
-	coalesced_offset = (stripe_num * stripe_size) +
-		(offset % stripe_size);
-
-	return coalesced_offset;
-}
-
-off_t
-uncoalesced_size(off_t size, uint64_t stripe_size, int stripe_count,
-		int stripe_index)
-{
-        uint64_t nr_full_stripe_chunks = 0, mod = 0;
-
-        if (!size)
-                return size;
-
-	/*
-	 * Estimate the number of fully written stripes from the
-	 * local file size. Each stripe_size chunk corresponds to
-	 * a stripe.
-	 */
-        nr_full_stripe_chunks = (size / stripe_size) * stripe_count;
-        mod = size % stripe_size;
-
-        if (!mod) {
-                /*
-		 * There is no remainder, thus we could have overestimated
-		 * the size of the file in terms of chunks. Trim the number
-		 * of chunks by the following stripe members and leave it
-		 * up to those nodes to respond with a larger size (if
-		 * necessary).
-		 */
-                nr_full_stripe_chunks -= stripe_count -
-                        (stripe_index + 1);
-                size = nr_full_stripe_chunks * stripe_size;
-        } else {
-		/*
-		 * There is a remainder and thus we own the last chunk of the
-		 * file. Add the preceding stripe members of the final stripe
-		 * along with the remainder to calculate the exact size.
-		 */
-                nr_full_stripe_chunks += stripe_index;
-                size = nr_full_stripe_chunks * stripe_size + mod;
-        }
-
-        return size;
-}
-
diff --git a/xlators/cluster/stripe/src/stripe-mem-types.h b/xlators/cluster/stripe/src/stripe-mem-types.h
deleted file mode 100644
index e9ac9cf4648..00000000000
--- a/xlators/cluster/stripe/src/stripe-mem-types.h
+++ /dev/null
@@ -1,31 +0,0 @@
-/*
-  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
-  This file is part of GlusterFS.
-
-  This file is licensed to you under your choice of the GNU Lesser
-  General Public License, version 3 or any later version (LGPLv3 or
-  later), or the GNU General Public License, version 2 (GPLv2), in all
-  cases as published by the Free Software Foundation.
-*/
-
-
-#ifndef __STRIPE_MEM_TYPES_H__
-#define __STRIPE_MEM_TYPES_H__
-
-#include "mem-types.h"
-
-enum gf_stripe_mem_types_ {
-        gf_stripe_mt_iovec = gf_common_mt_end + 1,
-        gf_stripe_mt_stripe_replies,
-        gf_stripe_mt_stripe_fd_ctx_t,
-        gf_stripe_mt_char,
-        gf_stripe_mt_int8_t,
-        gf_stripe_mt_int32_t,
-        gf_stripe_mt_xlator_t,
-        gf_stripe_mt_stripe_private_t,
-        gf_stripe_mt_stripe_options,
-        gf_stripe_mt_xattr_sort_t,
-        gf_stripe_mt_end
-};
-#endif
-
diff --git a/xlators/cluster/stripe/src/stripe.c b/xlators/cluster/stripe/src/stripe.c
deleted file mode 100644
index 8faaec19673..00000000000
--- a/xlators/cluster/stripe/src/stripe.c
+++ /dev/null
@@ -1,5592 +0,0 @@
-/*
-  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
-  This file is part of GlusterFS.
-
-  This file is licensed to you under your choice of the GNU Lesser
-  General Public License, version 3 or any later version (LGPLv3 or
-  later), or the GNU General Public License, version 2 (GPLv2), in all
-  cases as published by the Free Software Foundation.
-*/
-
-/**
- * xlators/cluster/stripe:
- *    Stripe translator, stripes the data across its child nodes,
- *    as per the options given in the volfile. The striping works
- *    fairly simple. It writes files at different offset as per
- *    calculation. So, 'ls -l' output at the real posix level will
- *    show file size bigger than the actual size. But when one does
- *    'df' or 'du <file>', real size of the file on the server is shown.
- *
- * WARNING:
- *  Stripe translator can't regenerate data if a child node gets disconnected.
- *  So, no 'self-heal' for stripe. Hence the advice, use stripe only when its
- *  very much necessary, or else, use it in combination with AFR, to have a
- *  backup copy.
- */
-#include <fnmatch.h>
-
-#include "stripe.h"
-#include "libxlator.h"
-#include "byte-order.h"
-#include "statedump.h"
-
-struct volume_options options[];
-
-int32_t
-stripe_sh_chown_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno,
-                     struct iatt *preop, struct iatt *postop, dict_t *xdata)
-{
-        int             callcnt = -1;
-        stripe_local_t *local   = NULL;
-
-        if (!this || !frame || !frame->local) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                STRIPE_STACK_DESTROY (frame);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_sh_make_entry_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                          int32_t op_ret, int32_t op_errno, inode_t *inode,
-                          struct iatt *buf, struct iatt *preparent,
-                          struct iatt *postparent, dict_t *xdata)
-{
-        stripe_local_t *local = NULL;
-        call_frame_t    *prev = NULL;
-
-        if (!frame || !frame->local || !cookie || !this) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        STACK_WIND (frame, stripe_sh_chown_cbk, prev->this,
-                    prev->this->fops->setattr, &local->loc,
-                    &local->stbuf, (GF_SET_ATTR_UID | GF_SET_ATTR_GID), NULL);
-
-out:
-        return 0;
-}
-
-int32_t
-stripe_entry_self_heal (call_frame_t *frame, xlator_t *this,
-                        stripe_local_t *local)
-{
-        xlator_list_t    *trav   = NULL;
-        call_frame_t     *rframe = NULL;
-        stripe_local_t   *rlocal = NULL;
-        stripe_private_t *priv   = NULL;
-        dict_t           *xdata  = NULL;
-        int               ret    = 0;
-
-        if (!local || !this || !frame) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        if (!(IA_ISREG (local->stbuf.ia_type) ||
-              IA_ISDIR (local->stbuf.ia_type)))
-                return 0;
-
-        priv = this->private;
-        trav = this->children;
-        rframe = copy_frame (frame);
-        if (!rframe) {
-                goto out;
-        }
-        rlocal = mem_get0 (this->local_pool);
-        if (!rlocal) {
-                goto out;
-        }
-        rframe->local = rlocal;
-        rlocal->call_count = priv->child_count;
-        loc_copy (&rlocal->loc, &local->loc);
-        memcpy (&rlocal->stbuf, &local->stbuf, sizeof (struct iatt));
-
-        xdata = dict_new ();
-        if (!xdata)
-                goto out;
-
-        ret = dict_set_static_bin (xdata, "gfid-req", local->stbuf.ia_gfid, 16);
-        if (ret)
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s: failed to set gfid-req", local->loc.path);
-
-        while (trav) {
-                if (IA_ISREG (local->stbuf.ia_type)) {
-                        STACK_WIND (rframe, stripe_sh_make_entry_cbk,
-                                    trav->xlator, trav->xlator->fops->mknod,
-                                    &local->loc,
-                                    st_mode_from_ia (local->stbuf.ia_prot,
-                                                     local->stbuf.ia_type),
-                                    0, 0, xdata);
-                }
-                if (IA_ISDIR (local->stbuf.ia_type)) {
-                        STACK_WIND (rframe, stripe_sh_make_entry_cbk,
-                                    trav->xlator, trav->xlator->fops->mkdir,
-                                    &local->loc,
-                                    st_mode_from_ia (local->stbuf.ia_prot,
-                                                     local->stbuf.ia_type),
-                                    0, xdata);
-                }
-                trav = trav->next;
-        }
-
-        if (xdata)
-                dict_unref (xdata);
-        return 0;
-
-out:
-        if (rframe)
-                STRIPE_STACK_DESTROY (rframe);
-        if (xdata)
-                dict_unref (xdata);
-
-        return 0;
-}
-
-
-int32_t
-stripe_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, inode_t *inode,
-                   struct iatt *buf, dict_t *xdata, struct iatt *postparent)
-{
-        int32_t         callcnt     = 0;
-        stripe_local_t *local       = NULL;
-        call_frame_t   *prev        = NULL;
-        int             ret         = 0;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        if (op_errno != ENOENT)
-                                gf_log (this->name, GF_LOG_DEBUG,
-                                        "%s returned error %s",
-                                        prev->this->name,
-                                        strerror (op_errno));
-                        if (local->op_errno != ESTALE)
-                                local->op_errno = op_errno;
-                        if (((op_errno != ENOENT) && (op_errno != ENOTCONN)) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                        if (op_errno == ENOENT)
-                                local->entry_self_heal_needed = 1;
-                }
-
-                if (op_ret >= 0) {
-                        local->op_ret = 0;
-                        if (IA_ISREG (buf->ia_type)) {
-                                ret = stripe_ctx_handle (this, prev, local,
-                                                         xdata);
-                                if (ret)
-                                        gf_log (this->name, GF_LOG_ERROR,
-                                                 "Error getting fctx info from"
-                                                 " dict");
-                        }
-
-                        if (FIRST_CHILD(this) == prev->this) {
-                                local->stbuf      = *buf;
-                                local->postparent = *postparent;
-                                local->inode = inode_ref (inode);
-                                if (xdata)
-                                        local->xdata = dict_ref (xdata);
-                                if (local->xattr) {
-                                        stripe_aggregate_xattr (local->xdata,
-                                                                local->xattr);
-                                        dict_unref (local->xattr);
-                                        local->xattr = NULL;
-                                }
-                        }
-
-                        if (!local->xdata && !local->xattr) {
-                                local->xattr = dict_ref (xdata);
-                        } else if (local->xdata) {
-                                stripe_aggregate_xattr (local->xdata, xdata);
-                        } else if (local->xattr) {
-                                stripe_aggregate_xattr (local->xattr, xdata);
-                        }
-
-                        local->stbuf_blocks      += buf->ia_blocks;
-                        local->postparent_blocks += postparent->ia_blocks;
-
-			correct_file_size(buf, local->fctx, prev);
-
-                        if (local->stbuf_size < buf->ia_size)
-                                local->stbuf_size = buf->ia_size;
-                        if (local->postparent_size < postparent->ia_size)
-                                local->postparent_size = postparent->ia_size;
-
-                        if (uuid_is_null (local->ia_gfid))
-                                uuid_copy (local->ia_gfid, buf->ia_gfid);
-
-                        /* Make sure the gfid on all the nodes are same */
-                        if (uuid_compare (local->ia_gfid, buf->ia_gfid)) {
-                                gf_log (this->name, GF_LOG_WARNING,
-                                        "%s: gfid different on subvolume %s",
-                                        local->loc.path, prev->this->name);
-                        }
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->op_ret == 0 && local->entry_self_heal_needed &&
-                    !uuid_is_null (local->loc.inode->gfid))
-                        stripe_entry_self_heal (frame, this, local);
-
-                if (local->failed)
-                        local->op_ret = -1;
-
-                if (local->op_ret != -1) {
-                        local->stbuf.ia_blocks      = local->stbuf_blocks;
-                        local->stbuf.ia_size        = local->stbuf_size;
-                        local->postparent.ia_blocks = local->postparent_blocks;
-                        local->postparent.ia_size   = local->postparent_size;
-                        inode_ctx_put (local->inode, this,
-                                       (uint64_t) (long)local->fctx);
-                }
-
-                STRIPE_STACK_UNWIND (lookup, frame, local->op_ret,
-                                     local->op_errno, local->inode,
-                                     &local->stbuf, local->xdata,
-                                     &local->postparent);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc,
-               dict_t *xdata)
-{
-        stripe_local_t   *local    = NULL;
-        xlator_list_t    *trav     = NULL;
-        stripe_private_t *priv     = NULL;
-        int32_t           op_errno = EINVAL;
-        int64_t           filesize = 0;
-        int               ret      = 0;
-        uint64_t          tmpctx   = 0;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        /* Initialization */
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        loc_copy (&local->loc, loc);
-
-        inode_ctx_get (local->inode, this, &tmpctx);
-        if (tmpctx)
-                local->fctx = (stripe_fd_ctx_t*) (long)tmpctx;
-
-        /* quick-read friendly changes */
-        if (xdata && dict_get (xdata, GF_CONTENT_KEY)) {
-                ret = dict_get_int64 (xdata, GF_CONTENT_KEY, &filesize);
-                if (!ret && (filesize > priv->block_size))
-                        dict_del (xdata, GF_CONTENT_KEY);
-        }
-
-        /* get stripe-size xattr on lookup. This would be required for
-         * open/read/write/pathinfo calls. Hence we send down the request
-         * even when type == IA_INVAL */
-
-	/*
-	 * We aren't guaranteed to have xdata here. We need the format info for
-	 * the file, so allocate xdata if necessary.
-	 */
-	if (!xdata)
-		xdata = dict_new();
-	else
-		xdata = dict_ref(xdata);
-
-        if (xdata && (IA_ISREG (loc->inode->ia_type) ||
-            (loc->inode->ia_type == IA_INVAL))) {
-                ret = stripe_xattr_request_build (this, xdata, 8, 4, 4, 0);
-                if (ret)
-                        gf_log (this->name , GF_LOG_ERROR, "Failed to build"
-                                " xattr request for %s", loc->path);
-
-        }
-
-        /* Everytime in stripe lookup, all child nodes
-           should be looked up */
-        local->call_count = priv->child_count;
-        while (trav) {
-                STACK_WIND (frame, stripe_lookup_cbk, trav->xlator,
-                            trav->xlator->fops->lookup, loc, xdata);
-                trav = trav->next;
-        }
-
-	dict_unref(xdata);
-
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno, struct iatt *buf, dict_t *xdata)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-
-                if (op_ret == 0) {
-                        local->op_ret = 0;
-
-                        if (FIRST_CHILD(this) == prev->this) {
-                                local->stbuf = *buf;
-                        }
-
-                        local->stbuf_blocks += buf->ia_blocks;
-
-			correct_file_size(buf, local->fctx, prev);
-
-                        if (local->stbuf_size < buf->ia_size)
-                                local->stbuf_size = buf->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                if (local->op_ret != -1) {
-                        local->stbuf.ia_size   = local->stbuf_size;
-                        local->stbuf.ia_blocks = local->stbuf_blocks;
-                }
-
-                STRIPE_STACK_UNWIND (stat, frame, local->op_ret,
-                                     local->op_errno, &local->stbuf, NULL);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_stat (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
-{
-        xlator_list_t    *trav = NULL;
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-	stripe_fd_ctx_t  *fctx = NULL;
-        int32_t           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        if (priv->first_child_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Initialization */
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        local->call_count = priv->child_count;
-
-	if (IA_ISREG(loc->inode->ia_type)) {
-		inode_ctx_get(loc->inode, this, (uint64_t *) &fctx);
-		if (!fctx)
-			goto err;
-		local->fctx = fctx;
-	}
-
-        while (trav) {
-                STACK_WIND (frame, stripe_stat_cbk, trav->xlator,
-                            trav->xlator->fops->stat, loc, NULL);
-                trav = trav->next;
-        }
-
-        return 0;
-
-err:
-        STRIPE_STACK_UNWIND (stat, frame, -1, op_errno, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_statfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct statvfs *stbuf, dict_t *xdata)
-{
-        stripe_local_t *local = NULL;
-        int32_t         callcnt = 0;
-
-        if (!this || !frame || !frame->local) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-        local = frame->local;
-
-        LOCK(&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret && (op_errno != ENOTCONN)) {
-                        local->op_errno = op_errno;
-                }
-                if (op_ret == 0) {
-                        struct statvfs *dict_buf = &local->statvfs_buf;
-                        dict_buf->f_bsize   = stbuf->f_bsize;
-                        dict_buf->f_frsize  = stbuf->f_frsize;
-                        dict_buf->f_blocks += stbuf->f_blocks;
-                        dict_buf->f_bfree  += stbuf->f_bfree;
-                        dict_buf->f_bavail += stbuf->f_bavail;
-                        dict_buf->f_files  += stbuf->f_files;
-                        dict_buf->f_ffree  += stbuf->f_ffree;
-                        dict_buf->f_favail += stbuf->f_favail;
-                        dict_buf->f_fsid    = stbuf->f_fsid;
-                        dict_buf->f_flag    = stbuf->f_flag;
-                        dict_buf->f_namemax = stbuf->f_namemax;
-                        local->op_ret = 0;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                STRIPE_STACK_UNWIND (statfs, frame, local->op_ret,
-                                     local->op_errno, &local->statvfs_buf, NULL);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
-{
-        stripe_local_t   *local = NULL;
-        xlator_list_t    *trav = NULL;
-        stripe_private_t *priv = NULL;
-        int32_t           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-
-        trav = this->children;
-        priv = this->private;
-
-        /* Initialization */
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        local->op_errno = ENOTCONN;
-        frame->local = local;
-
-        local->call_count = priv->child_count;
-        while (trav) {
-                STACK_WIND (frame, stripe_statfs_cbk, trav->xlator,
-                            trav->xlator->fops->statfs, loc, NULL);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (statfs, frame, -1, op_errno, NULL, NULL);
-        return 0;
-}
-
-
-
-int32_t
-stripe_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                     struct iatt *postbuf, dict_t *xdata)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-
-                if (op_ret == 0) {
-                        local->op_ret = 0;
-                        if (FIRST_CHILD(this) == prev->this) {
-                                local->pre_buf  = *prebuf;
-                                local->post_buf = *postbuf;
-                        }
-
-                        local->prebuf_blocks  += prebuf->ia_blocks;
-                        local->postbuf_blocks += postbuf->ia_blocks;
-
-			correct_file_size(prebuf, local->fctx, prev);
-			correct_file_size(postbuf, local->fctx, prev);
-
-                        if (local->prebuf_size < prebuf->ia_size)
-                                local->prebuf_size = prebuf->ia_size;
-
-                        if (local->postbuf_size < postbuf->ia_size)
-                                local->postbuf_size = postbuf->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                if (local->op_ret != -1) {
-                        local->pre_buf.ia_blocks  = local->prebuf_blocks;
-                        local->pre_buf.ia_size    = local->prebuf_size;
-                        local->post_buf.ia_blocks = local->postbuf_blocks;
-                        local->post_buf.ia_size   = local->postbuf_size;
-                }
-
-                STRIPE_STACK_UNWIND (truncate, frame, local->op_ret,
-                                     local->op_errno, &local->pre_buf,
-                                     &local->post_buf, NULL);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset, dict_t *xdata)
-{
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-	stripe_fd_ctx_t  *fctx = NULL;
-        int32_t           op_errno = EINVAL;
-	int		  i, eof_idx;
-	off_t		  dest_offset, tmp_offset;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-
-        if (priv->first_child_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Initialization */
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        local->call_count = priv->child_count;
-
-	inode_ctx_get(loc->inode, this, (uint64_t *) &fctx);
-	if (!fctx) {
-		gf_log(this->name, GF_LOG_ERROR, "no stripe context");
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	local->fctx = fctx;
-	eof_idx = (offset / fctx->stripe_size) % fctx->stripe_count;
-
-	for (i = 0; i < fctx->stripe_count; i++) {
-		if (!fctx->xl_array[i]) {
-			gf_log(this->name, GF_LOG_ERROR,
-				"no xlator at index %d", i);
-			op_errno = EINVAL;
-			goto err;
-		}
-
-		if (fctx->stripe_coalesce) {
-			/*
-	 		 * The node that owns EOF is truncated to the exact
-	 		 * coalesced offset. Nodes prior to this index should
-	 		 * be rounded up to the size of the complete stripe,
-	 		 * while nodes after this index should be rounded down
-			 * to the size of the previous stripe.
-			 */
-			if (i < eof_idx)
-				tmp_offset = roof(offset, fctx->stripe_size *
-						fctx->stripe_count);
-			else if (i > eof_idx)
-				tmp_offset = floor(offset, fctx->stripe_size *
-						fctx->stripe_count);
-			else
-				tmp_offset = offset;
-
-			dest_offset = coalesced_offset(tmp_offset,
-					fctx->stripe_size, fctx->stripe_count);
-		} else {
-			dest_offset = offset;
-		}
-
-		STACK_WIND(frame, stripe_truncate_cbk, fctx->xl_array[i],
-			fctx->xl_array[i]->fops->truncate, loc, dest_offset,
-			NULL);
-	}
-
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (truncate, frame, -1, op_errno, NULL, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno,
-                    struct iatt *preop, struct iatt *postop, dict_t *xdata)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-
-                if (op_ret == 0) {
-                        local->op_ret = 0;
-
-                        if (FIRST_CHILD(this) == prev->this) {
-                                local->pre_buf  = *preop;
-                                local->post_buf = *postop;
-                        }
-
-                        local->prebuf_blocks  += preop->ia_blocks;
-                        local->postbuf_blocks += postop->ia_blocks;
-
-			correct_file_size(preop, local->fctx, prev);
-			correct_file_size(postop, local->fctx, prev);
-
-                        if (local->prebuf_size < preop->ia_size)
-                                local->prebuf_size = preop->ia_size;
-                        if (local->postbuf_size < postop->ia_size)
-                                local->postbuf_size = postop->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                if (local->op_ret != -1) {
-                        local->pre_buf.ia_blocks  = local->prebuf_blocks;
-                        local->pre_buf.ia_size    = local->prebuf_size;
-                        local->post_buf.ia_blocks = local->postbuf_blocks;
-                        local->post_buf.ia_size   = local->postbuf_size;
-                }
-
-                STRIPE_STACK_UNWIND (setattr, frame, local->op_ret,
-                                     local->op_errno, &local->pre_buf,
-                                     &local->post_buf, NULL);
-        }
-out:
-        return 0;
-}
-
-
-int32_t
-stripe_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                struct iatt *stbuf, int32_t valid, dict_t *xdata)
-{
-        xlator_list_t    *trav = NULL;
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-	stripe_fd_ctx_t	 *fctx = NULL;
-        int32_t           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        if (priv->first_child_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Initialization */
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        if (!IA_ISDIR (loc->inode->ia_type) &&
-            !IA_ISREG (loc->inode->ia_type)) {
-                local->call_count = 1;
-                STACK_WIND (frame, stripe_setattr_cbk, FIRST_CHILD (this),
-                            FIRST_CHILD (this)->fops->setattr,
-                            loc, stbuf, valid, NULL);
-                return 0;
-        }
-
-	if (IA_ISREG(loc->inode->ia_type)) {
-		inode_ctx_get(loc->inode, this, (uint64_t *) &fctx);
-		if (!fctx)
-			goto err;
-		local->fctx = fctx;
-	}
-
-        local->call_count = priv->child_count;
-        while (trav) {
-                STACK_WIND (frame, stripe_setattr_cbk,
-                            trav->xlator, trav->xlator->fops->setattr,
-                            loc, stbuf, valid, NULL);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (setattr, frame, -1, op_errno, NULL, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                 struct iatt *stbuf, int32_t valid, dict_t *xdata)
-{
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        xlator_list_t    *trav = NULL;
-        int32_t           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        /* Initialization */
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        local->call_count = priv->child_count;
-
-        while (trav) {
-                STACK_WIND (frame, stripe_setattr_cbk, trav->xlator,
-                            trav->xlator->fops->fsetattr, fd, stbuf, valid, NULL);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (fsetattr, frame, -1, op_errno, NULL, NULL, NULL);
-        return 0;
-}
-
-int32_t
-stripe_stack_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, struct iatt *buf,
-                         struct iatt *preoldparent, struct iatt *postoldparent,
-                         struct iatt *prenewparent, struct iatt *postnewparent,
-                         dict_t *xdata)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-
-                if (op_ret == 0) {
-                        local->op_ret = 0;
-
-                        local->stbuf.ia_blocks      += buf->ia_blocks;
-                        local->preparent.ia_blocks  += preoldparent->ia_blocks;
-                        local->postparent.ia_blocks += postoldparent->ia_blocks;
-                        local->pre_buf.ia_blocks    += prenewparent->ia_blocks;
-                        local->post_buf.ia_blocks   += postnewparent->ia_blocks;
-
-			correct_file_size(buf, local->fctx, prev);
-
-                        if (local->stbuf.ia_size < buf->ia_size)
-                                local->stbuf.ia_size =  buf->ia_size;
-
-                        if (local->preparent.ia_size < preoldparent->ia_size)
-                                local->preparent.ia_size = preoldparent->ia_size;
-
-                        if (local->postparent.ia_size < postoldparent->ia_size)
-                                local->postparent.ia_size = postoldparent->ia_size;
-
-                        if (local->pre_buf.ia_size < prenewparent->ia_size)
-                                local->pre_buf.ia_size = prenewparent->ia_size;
-
-                        if (local->post_buf.ia_size < postnewparent->ia_size)
-                                local->post_buf.ia_size = postnewparent->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                STRIPE_STACK_UNWIND (rename, frame, local->op_ret, local->op_errno,
-                                     &local->stbuf, &local->preparent,
-                                     &local->postparent,  &local->pre_buf,
-                                     &local->post_buf, NULL);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_first_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, struct iatt *buf,
-                         struct iatt *preoldparent, struct iatt *postoldparent,
-                         struct iatt *prenewparent, struct iatt *postnewparent,
-                         dict_t *xdata)
-{
-        stripe_local_t *local = NULL;
-        xlator_list_t  *trav = NULL;
-
-        if (!this || !frame || !frame->local) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                op_errno = EINVAL;
-                goto unwind;
-        }
-
-        if (op_ret == -1) {
-                goto unwind;
-        }
-
-        local = frame->local;
-        trav = this->children;
-
-        local->stbuf      = *buf;
-        local->preparent  = *preoldparent;
-        local->postparent = *postoldparent;
-        local->pre_buf    = *prenewparent;
-        local->post_buf   = *postnewparent;
-
-        local->op_ret = 0;
-        local->call_count--;
-
-        trav = trav->next; /* Skip first child */
-        while (trav) {
-                STACK_WIND (frame, stripe_stack_rename_cbk,
-                            trav->xlator, trav->xlator->fops->rename,
-                            &local->loc, &local->loc2, NULL);
-                trav = trav->next;
-        }
-        return 0;
-
-unwind:
-        STRIPE_STACK_UNWIND (rename, frame, -1, op_errno, buf, preoldparent,
-                             postoldparent, prenewparent, postnewparent, NULL);
-        return 0;
-}
-
-int32_t
-stripe_rename (call_frame_t *frame, xlator_t *this, loc_t *oldloc,
-               loc_t *newloc, dict_t *xdata)
-{
-        stripe_private_t *priv = NULL;
-        stripe_local_t   *local = NULL;
-        xlator_list_t    *trav = NULL;
-	stripe_fd_ctx_t	 *fctx = NULL;
-        int32_t           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (oldloc, err);
-        VALIDATE_OR_GOTO (oldloc->path, err);
-        VALIDATE_OR_GOTO (oldloc->inode, err);
-        VALIDATE_OR_GOTO (newloc, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        /* If any one node is down, don't allow rename */
-        if (priv->nodes_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Initialization */
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        loc_copy (&local->loc, oldloc);
-        loc_copy (&local->loc2, newloc);
-
-        local->call_count = priv->child_count;
-
-	if (IA_ISREG(oldloc->inode->ia_type)) {
-		inode_ctx_get(oldloc->inode, this, (uint64_t *) &fctx);
-		if (!fctx)
-			goto err;
-		local->fctx = fctx;
-	}
-
-        frame->local = local;
-
-        STACK_WIND (frame, stripe_first_rename_cbk, trav->xlator,
-                    trav->xlator->fops->rename, oldloc, newloc, NULL);
-
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (rename, frame, -1, op_errno, NULL, NULL, NULL,
-                             NULL, NULL, NULL);
-        return 0;
-}
-int32_t
-stripe_first_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, struct iatt *preparent,
-                         struct iatt *postparent, dict_t *xdata)
-{
-        stripe_local_t *local   = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_DEBUG, "%s returned %s",
-                        prev->this->name, strerror (op_errno));
-                goto out;
-        }
-        local->op_ret = 0;
-        local->preparent  = *preparent;
-        local->postparent = *postparent;
-        local->preparent_blocks  += preparent->ia_blocks;
-        local->postparent_blocks += postparent->ia_blocks;
-
-        STRIPE_STACK_UNWIND(unlink, frame, local->op_ret, local->op_errno,
-                            &local->preparent, &local->postparent, xdata);
-        return 0;
-out:
-        STRIPE_STACK_UNWIND (unlink, frame, -1, op_errno, NULL, NULL, NULL);
-
-        return 0;
-}
-
-
-
-
-int32_t
-stripe_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct iatt *preparent,
-                   struct iatt *postparent, dict_t *xdata)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local   = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG, "%s returned %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if (op_errno != ENOENT) {
-                                local->failed = 1;
-                                local->op_ret = op_ret;
-                        }
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (callcnt == 1) {
-                if (local->failed) {
-                        op_errno = local->op_errno;
-                        goto out;
-                }
-                STACK_WIND(frame, stripe_first_unlink_cbk, FIRST_CHILD (this),
-                           FIRST_CHILD (this)->fops->unlink, &local->loc,
-                           local->xflag, local->xdata);
-        }
-        return 0;
-out:
-        STRIPE_STACK_UNWIND (unlink, frame, -1, op_errno, NULL, NULL, NULL);
-
-        return 0;
-}
-
-int32_t
-stripe_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc,
-               int xflag, dict_t *xdata)
-{
-        xlator_list_t    *trav = NULL;
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        int32_t           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        if (priv->first_child_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Don't unlink a file if a node is down */
-        if (priv->nodes_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Initialization */
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        loc_copy (&local->loc, loc);
-        local->xflag = xflag;
-
-        if (xdata)
-            local->xdata = dict_ref (xdata);
-
-        frame->local = local;
-        local->call_count = priv->child_count;
-        trav = trav->next; /* Skip the first child */
-
-        while (trav) {
-                STACK_WIND (frame, stripe_unlink_cbk,
-                            trav->xlator, trav->xlator->fops->unlink,
-                            loc, xflag, xdata);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (unlink, frame, -1, op_errno, NULL, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_first_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno,struct iatt *preparent,
-                        struct iatt *postparent, dict_t *xdata)
-{
-        stripe_local_t *local = NULL;
-
-        if (!this || !frame || !frame->local) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                op_errno = EINVAL;
-                goto err;
-        }
-
-        if (op_ret == -1) {
-                goto err;
-        }
-
-        local = frame->local;
-        local->op_ret = 0;
-
-        local->call_count--; /* First child successful */
-
-        local->preparent  = *preparent;
-        local->postparent = *postparent;
-        local->preparent_size  = preparent->ia_size;
-        local->postparent_size = postparent->ia_size;
-        local->preparent_blocks  += preparent->ia_blocks;
-        local->postparent_blocks += postparent->ia_blocks;
-
-        STRIPE_STACK_UNWIND (rmdir, frame, local->op_ret, local->op_errno,
-                             &local->preparent, &local->postparent, xdata);
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (rmdir, frame, op_ret, op_errno, NULL, NULL, NULL);
-        return 0;
-
-}
-
-int32_t
-stripe_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct iatt *preparent,
-                   struct iatt *postparent, dict_t *xdata)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local   = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG, "%s returned %s",
-                                prev->this->name, strerror (op_errno));
-                        if (op_errno != ENOENT)
-                                local->failed = 1;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (callcnt == 1) {
-                if (local->failed)
-                        goto out;
-                STACK_WIND (frame, stripe_first_rmdir_cbk, FIRST_CHILD (this),
-                            FIRST_CHILD (this)->fops->rmdir, &local->loc,
-                            local->flags, NULL);
-        }
-        return 0;
-out:
-        STRIPE_STACK_UNWIND (rmdir, frame, -1, op_errno, NULL, NULL, NULL);
-        return 0;
-}
-
-int32_t
-stripe_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags, dict_t *xdata)
-{
-        xlator_list_t    *trav = NULL;
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        int32_t           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        /* don't delete a directory if any of the subvolume is down */
-        if (priv->nodes_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Initialization */
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        loc_copy (&local->loc, loc);
-        local->flags = flags;
-        local->call_count = priv->child_count;
-        trav = trav->next; /* skip the first child */
-
-        while (trav) {
-                STACK_WIND (frame, stripe_rmdir_cbk,  trav->xlator,
-                            trav->xlator->fops->rmdir, loc, flags, NULL);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (rmdir, frame, -1, op_errno, NULL, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_mknod_ifreg_fail_unlink_cbk (call_frame_t *frame, void *cookie,
-                                    xlator_t *this, int32_t op_ret,
-                                    int32_t op_errno, struct iatt *preparent,
-                                    struct iatt *postparent, dict_t *xdata)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-
-        if (!this || !frame || !frame->local) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                STRIPE_STACK_UNWIND (mknod, frame, local->op_ret, local->op_errno,
-                                     local->inode, &local->stbuf,
-                                     &local->preparent, &local->postparent, NULL);
-        }
-out:
-        return 0;
-}
-
-
-/**
- */
-int32_t
-stripe_mknod_ifreg_setxattr_cbk (call_frame_t *frame, void *cookie,
-                                 xlator_t *this, int32_t op_ret,
-                                 int32_t op_errno, dict_t *xdata)
-{
-        int32_t           callcnt = 0;
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        xlator_list_t    *trav = NULL;
-        call_frame_t     *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        priv  = this->private;
-        local = frame->local;
-
-	LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_ret = -1;
-                        local->op_errno = op_errno;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->op_ret == -1) {
-                        local->call_count = priv->child_count;
-                        while (trav) {
-                                STACK_WIND (frame,
-                                            stripe_mknod_ifreg_fail_unlink_cbk,
-                                            trav->xlator,
-                                            trav->xlator->fops->unlink,
-                                            &local->loc, 0, NULL);
-                                trav = trav->next;
-                        }
-                        return 0;
-                }
-
-                STRIPE_STACK_UNWIND (mknod, frame, local->op_ret, local->op_errno,
-                                     local->inode, &local->stbuf,
-                                     &local->preparent, &local->postparent, NULL);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_mknod_ifreg_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, inode_t *inode,
-                        struct iatt *buf, struct iatt *preparent,
-                        struct iatt *postparent, dict_t *xdata)
-{
-        int32_t           callcnt = 0;
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        call_frame_t     *prev = NULL;
-        xlator_list_t    *trav = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        priv  = this->private;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                        local->op_errno = op_errno;
-                }
-                if (op_ret >= 0) {
-                        local->op_ret = op_ret;
-
-                        /* Can be used as a mechanism to understand if mknod
-                           was successful in at least one place */
-                        if (uuid_is_null (local->ia_gfid))
-                                uuid_copy (local->ia_gfid, buf->ia_gfid);
-
-			if (stripe_ctx_handle(this, prev, local, xdata))
-				gf_log(this->name, GF_LOG_ERROR,
-					"Error getting fctx info from dict");
-
-                        local->stbuf_blocks += buf->ia_blocks;
-                        local->preparent_blocks  += preparent->ia_blocks;
-                        local->postparent_blocks += postparent->ia_blocks;
-
-			correct_file_size(buf, local->fctx, prev);
-
-                        if (local->stbuf_size < buf->ia_size)
-                                local->stbuf_size = buf->ia_size;
-                        if (local->preparent_size < preparent->ia_size)
-                                local->preparent_size = preparent->ia_size;
-                        if (local->postparent_size < postparent->ia_size)
-                                local->postparent_size = postparent->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                if ((local->op_ret == -1) && !uuid_is_null (local->ia_gfid)) {
-                        /* ia_gfid set means, at least on one node 'mknod'
-                           is successful */
-                        local->call_count = priv->child_count;
-                        trav = this->children;
-                        while (trav) {
-                                STACK_WIND (frame,
-                                            stripe_mknod_ifreg_fail_unlink_cbk,
-                                            trav->xlator,
-                                            trav->xlator->fops->unlink,
-                                            &local->loc, 0, NULL);
-                                trav = trav->next;
-                        }
-                        return 0;
-                }
-
-
-                if (local->op_ret != -1) {
-                        local->preparent.ia_blocks  = local->preparent_blocks;
-                        local->preparent.ia_size    = local->preparent_size;
-                        local->postparent.ia_blocks = local->postparent_blocks;
-                        local->postparent.ia_size   = local->postparent_size;
-                        local->stbuf.ia_size        = local->stbuf_size;
-                        local->stbuf.ia_blocks      = local->stbuf_blocks;
-                        inode_ctx_put (local->inode, this,
-                                       (uint64_t)(long) local->fctx);
-
-                }
-                STRIPE_STACK_UNWIND (mknod, frame, local->op_ret, local->op_errno,
-                                     local->inode, &local->stbuf,
-                                     &local->preparent, &local->postparent, NULL);
-        }
-out:
-        return 0;
-}
-
-
-int32_t
-stripe_mknod_first_ifreg_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, inode_t *inode,
-                        struct iatt *buf, struct iatt *preparent,
-                        struct iatt *postparent, dict_t *xdata)
-{
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        call_frame_t     *prev = NULL;
-        xlator_list_t    *trav = NULL;
-        int               i    = 1;
-        dict_t           *dict           = NULL;
-        int               ret            = 0;
-        int               need_unref     = 0;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        priv  = this->private;
-        local = frame->local;
-        trav = this->children;
-
-        local->call_count--;
-
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_DEBUG, "%s returned error %s",
-                        prev->this->name, strerror (op_errno));
-                local->failed = 1;
-                local->op_errno = op_errno;
-                goto out;
-        }
-
-        local->op_ret = op_ret;
-
-        local->stbuf      = *buf;
-        local->preparent  = *preparent;
-        local->postparent = *postparent;
-
-        if (uuid_is_null (local->ia_gfid))
-                uuid_copy (local->ia_gfid, buf->ia_gfid);
-        local->preparent.ia_blocks  = local->preparent_blocks;
-        local->preparent.ia_size    = local->preparent_size;
-        local->postparent.ia_blocks = local->postparent_blocks;
-        local->postparent.ia_size   = local->postparent_size;
-        local->stbuf.ia_size        = local->stbuf_size;
-        local->stbuf.ia_blocks      = local->stbuf_blocks;
-
-        trav = trav->next;
-        while (trav) {
-                if (priv->xattr_supported) {
-                        dict = dict_new ();
-                        if (!dict) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "failed to allocate dict %s", local->loc.path);
-                        }
-                        need_unref = 1;
-
-                        dict_copy (local->xattr, dict);
-
-                        ret = stripe_xattr_request_build (this, dict,
-                                                          local->stripe_size,
-                                                          priv->child_count, i,
-							  priv->coalesce);
-                        if (ret)
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "Failed to build xattr request");
-
-                } else {
-                        dict = local->xattr;
-                }
-
-                STACK_WIND (frame, stripe_mknod_ifreg_cbk,
-                            trav->xlator, trav->xlator->fops->mknod,
-                            &local->loc, local->mode, local->rdev, 0, dict);
-                trav = trav->next;
-                i++;
-
-                if (dict && need_unref)
-                        dict_unref (dict);
-        }
-
-        return 0;
-
-out:
-
-       STRIPE_STACK_UNWIND (mknod, frame, op_ret, op_errno, NULL, NULL, NULL, NULL, NULL);
-       return 0;
-}
-
-
-int32_t
-stripe_single_mknod_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, inode_t *inode,
-                         struct iatt *buf, struct iatt *preparent,
-                         struct iatt *postparent, dict_t *xdata)
-{
-        STRIPE_STACK_UNWIND (mknod, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent, xdata);
-        return 0;
-}
-
-
-int
-stripe_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
-              dev_t rdev, mode_t umask, dict_t *xdata)
-{
-        stripe_private_t *priv           = NULL;
-        stripe_local_t   *local          = NULL;
-        int32_t           op_errno       = EINVAL;
-        int32_t           i              = 0;
-        dict_t           *dict           = NULL;
-        int               ret            = 0;
-        int               need_unref     = 0;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-
-        if (priv->first_child_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        if (S_ISREG(mode)) {
-                /* NOTE: on older kernels (older than 2.6.9),
-                   creat() fops is sent as mknod() + open(). Hence handling
-                   S_IFREG files is necessary */
-                if (priv->nodes_down) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Some node down, returning EIO");
-                        op_errno = EIO;
-                        goto err;
-                }
-
-                /* Initialization */
-                local = mem_get0 (this->local_pool);
-                if (!local) {
-                        op_errno = ENOMEM;
-                        goto err;
-                }
-                local->op_ret = -1;
-                local->op_errno = ENOTCONN;
-                local->stripe_size = stripe_get_matching_bs (loc->path, priv);
-                frame->local = local;
-                local->inode = inode_ref (loc->inode);
-                loc_copy (&local->loc, loc);
-                local->xattr = dict_copy_with_ref (xdata, NULL);
-                local->mode = mode;
-                local->umask = umask;
-                local->rdev = rdev;
-
-                /* Everytime in stripe lookup, all child nodes should
-                   be looked up */
-                local->call_count = priv->child_count;
-
-                if (priv->xattr_supported) {
-                        dict = dict_new ();
-                        if (!dict) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "failed to allocate dict %s", loc->path);
-                        }
-                        need_unref = 1;
-
-                        dict_copy (xdata, dict);
-
-                        ret = stripe_xattr_request_build (this, dict,
-                                                          local->stripe_size,
-                                                          priv->child_count,
-                                                          i, priv->coalesce);
-                        if (ret)
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "failed to build xattr request");
-                } else {
-                        dict = xdata;
-                }
-
-                STACK_WIND (frame, stripe_mknod_first_ifreg_cbk,
-                            FIRST_CHILD (this), FIRST_CHILD (this)->fops->mknod,
-                            loc, mode, rdev, umask, dict);
-
-                        if (dict && need_unref)
-                                dict_unref (dict);
-                return 0;
-        }
-
-        STACK_WIND (frame, stripe_single_mknod_cbk,
-                    FIRST_CHILD(this), FIRST_CHILD(this)->fops->mknod,
-                    loc, mode, rdev, umask, xdata);
-
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno, inode_t *inode,
-                  struct iatt *buf, struct iatt *preparent,
-                  struct iatt *postparent, dict_t *xdata)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t  *local   = NULL;
-        call_frame_t    *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-
-                if (op_ret >= 0) {
-                        local->op_ret = 0;
-
-                        local->stbuf_blocks      += buf->ia_blocks;
-                        local->preparent_blocks  += preparent->ia_blocks;
-                        local->postparent_blocks += postparent->ia_blocks;
-
-                        if (local->stbuf_size < buf->ia_size)
-                                local->stbuf_size = buf->ia_size;
-                        if (local->preparent_size < preparent->ia_size)
-                                local->preparent_size = preparent->ia_size;
-                        if (local->postparent_size < postparent->ia_size)
-                                local->postparent_size = postparent->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed != -1) {
-                        local->preparent.ia_blocks  = local->preparent_blocks;
-                        local->preparent.ia_size    = local->preparent_size;
-                        local->postparent.ia_blocks = local->postparent_blocks;
-                        local->postparent.ia_size   = local->postparent_size;
-                        local->stbuf.ia_size        = local->stbuf_size;
-                        local->stbuf.ia_blocks      = local->stbuf_blocks;
-                }
-                STRIPE_STACK_UNWIND (mkdir, frame, local->op_ret,
-                                     local->op_errno, local->inode,
-                                     &local->stbuf, &local->preparent,
-                                     &local->postparent, NULL);
-        }
-out:
-        return 0;
-}
-
-
-int32_t
-stripe_first_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, inode_t *inode,
-                        struct iatt *buf, struct iatt *preparent,
-                        struct iatt *postparent, dict_t *xdata)
-{
-        stripe_local_t  *local   = NULL;
-        call_frame_t    *prev = NULL;
-        xlator_list_t        *trav = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-        trav = this->children;
-
-        local->call_count--; /* first child is successful */
-        trav = trav->next;   /* skip first child */
-
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_DEBUG, "%s returned error %s",
-                        prev->this->name, strerror (op_errno));
-                local->op_errno = op_errno;
-                goto out;
-        }
-
-        local->op_ret = 0;
-
-        local->inode      = inode_ref (inode);
-        local->stbuf      = *buf;
-        local->postparent = *postparent;
-        local->preparent  = *preparent;
-
-        local->stbuf_blocks      += buf->ia_blocks;
-        local->preparent_blocks  += preparent->ia_blocks;
-        local->postparent_blocks += postparent->ia_blocks;
-
-        local->stbuf_size = buf->ia_size;
-        local->preparent_size = preparent->ia_size;
-        local->postparent_size = postparent->ia_size;
-
-        while (trav) {
-                STACK_WIND (frame, stripe_mkdir_cbk, trav->xlator,
-                            trav->xlator->fops->mkdir, &local->loc, local->mode,
-                            local->umask, local->xdata);
-                trav = trav->next;
-        }
-        return 0;
-out:
-        STRIPE_STACK_UNWIND (mkdir, frame, -1, op_errno, NULL, NULL, NULL,
-                             NULL, NULL);
-
-        return 0;
-
-}
-
-
-int
-stripe_mkdir (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
-              mode_t umask, dict_t *xdata)
-{
-        stripe_private_t *priv = NULL;
-        stripe_local_t   *local = NULL;
-        xlator_list_t    *trav = NULL;
-        int32_t           op_errno = 1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        if (priv->first_child_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Initialization */
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        local->call_count = priv->child_count;
-        if (xdata)
-                local->xdata = dict_ref (xdata);
-        local->mode  = mode;
-        local->umask = umask;
-        loc_copy (&local->loc, loc);
-        frame->local = local;
-
-        /* Everytime in stripe lookup, all child nodes should be looked up */
-        STACK_WIND (frame, stripe_first_mkdir_cbk, trav->xlator,
-                    trav->xlator->fops->mkdir, loc, mode, umask, xdata);
-
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_link_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno, inode_t *inode,
-                 struct iatt *buf, struct iatt *preparent,
-                 struct iatt *postparent, dict_t *xdata)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t  *local   = NULL;
-        call_frame_t    *prev = NULL;
-	stripe_fd_ctx_t *fctx = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-
-                if (op_ret >= 0) {
-                        local->op_ret = 0;
-
-			if (IA_ISREG(inode->ia_type)) {
-				inode_ctx_get(inode, this, (uint64_t *) &fctx);
-				if (!fctx) {
-					gf_log(this->name, GF_LOG_ERROR,
-						"failed to get stripe context");
-					op_ret = -1;
-					op_errno = EINVAL;
-				}
-			}
-
-                        if (FIRST_CHILD(this) == prev->this) {
-                                local->inode      = inode_ref (inode);
-                                local->stbuf      = *buf;
-                                local->postparent = *postparent;
-                                local->preparent  = *preparent;
-                        }
-                        local->stbuf_blocks      += buf->ia_blocks;
-                        local->preparent_blocks  += preparent->ia_blocks;
-                        local->postparent_blocks += postparent->ia_blocks;
-
-			correct_file_size(buf, fctx, prev);
-
-                        if (local->stbuf_size < buf->ia_size)
-                                local->stbuf_size = buf->ia_size;
-                        if (local->preparent_size < preparent->ia_size)
-                                local->preparent_size = preparent->ia_size;
-                        if (local->postparent_size < postparent->ia_size)
-                                local->postparent_size = postparent->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                if (local->op_ret != -1) {
-                        local->preparent.ia_blocks  = local->preparent_blocks;
-                        local->preparent.ia_size    = local->preparent_size;
-                        local->postparent.ia_blocks = local->postparent_blocks;
-                        local->postparent.ia_size   = local->postparent_size;
-                        local->stbuf.ia_size        = local->stbuf_size;
-                        local->stbuf.ia_blocks      = local->stbuf_blocks;
-                }
-                STRIPE_STACK_UNWIND (link, frame, local->op_ret,
-                                     local->op_errno, local->inode,
-                                     &local->stbuf, &local->preparent,
-                                     &local->postparent, NULL);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_link (call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc, dict_t *xdata)
-{
-        xlator_list_t    *trav = NULL;
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        int32_t           op_errno = 1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (oldloc, err);
-        VALIDATE_OR_GOTO (oldloc->path, err);
-        VALIDATE_OR_GOTO (oldloc->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        /* If any one node is down, don't allow link operation */
-        if (priv->nodes_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Initialization */
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        local->call_count = priv->child_count;
-
-        /* Everytime in stripe lookup, all child
-           nodes should be looked up */
-        while (trav) {
-                STACK_WIND (frame, stripe_link_cbk,
-                            trav->xlator, trav->xlator->fops->link,
-                            oldloc, newloc, NULL);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (link, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
-        return 0;
-}
-
-int32_t
-stripe_create_fail_unlink_cbk (call_frame_t *frame, void *cookie,
-                               xlator_t *this, int32_t op_ret,
-                               int32_t op_errno, struct iatt *preparent,
-                               struct iatt *postparent, dict_t *xdata)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-
-        if (!this || !frame || !frame->local) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                STRIPE_STACK_UNWIND (create, frame, local->op_ret, local->op_errno,
-                                     local->fd, local->inode, &local->stbuf,
-                                     &local->preparent, &local->postparent, NULL);
-        }
-out:
-        return 0;
-}
-
-
-int32_t
-stripe_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, fd_t *fd,
-                   inode_t *inode, struct iatt *buf, struct iatt *preparent,
-                   struct iatt *postparent, dict_t *xdata)
-{
-        int32_t           callcnt = 0;
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        call_frame_t     *prev = NULL;
-        xlator_list_t    *trav = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        priv  = this->private;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->failed = 1;
-                        local->op_errno = op_errno;
-                }
-
-                if (op_ret >= 0) {
-			if (IA_ISREG(buf->ia_type)) {
-				if (stripe_ctx_handle(this, prev, local, xdata))
-					gf_log(this->name, GF_LOG_ERROR,
-						"Error getting fctx info from "
-						"dict");
-			}
-
-                        local->op_ret = op_ret;
-
-                        local->stbuf_blocks += buf->ia_blocks;
-                        local->preparent_blocks  += preparent->ia_blocks;
-                        local->postparent_blocks += postparent->ia_blocks;
-
-			correct_file_size(buf, local->fctx, prev);
-
-                        if (local->stbuf_size < buf->ia_size)
-                                local->stbuf_size = buf->ia_size;
-                        if (local->preparent_size < preparent->ia_size)
-                                local->preparent_size = preparent->ia_size;
-                        if (local->postparent_size < postparent->ia_size)
-                                local->postparent_size = postparent->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                if (local->op_ret == -1) {
-                        local->call_count = priv->child_count;
-                        trav = this->children;
-                        while (trav) {
-                                STACK_WIND (frame,
-                                            stripe_create_fail_unlink_cbk,
-                                            trav->xlator,
-                                            trav->xlator->fops->unlink,
-                                            &local->loc, 0, NULL);
-                                trav = trav->next;
-                        }
-
-                        return 0;
-                }
-
-                if (local->op_ret >= 0) {
-                        local->preparent.ia_blocks  = local->preparent_blocks;
-                        local->preparent.ia_size    = local->preparent_size;
-                        local->postparent.ia_blocks = local->postparent_blocks;
-                        local->postparent.ia_size   = local->postparent_size;
-                        local->stbuf.ia_size        = local->stbuf_size;
-                        local->stbuf.ia_blocks      = local->stbuf_blocks;
-
-			stripe_copy_xl_array(local->fctx->xl_array,
-				             priv->xl_array,
-					     local->fctx->stripe_count);
-			inode_ctx_put(local->inode, this,
-					(uint64_t) local->fctx);
-                }
-
-                /* Create itself has failed.. so return
-                   without setxattring */
-                STRIPE_STACK_UNWIND (create, frame, local->op_ret,
-                                     local->op_errno, local->fd,
-                                     local->inode, &local->stbuf,
-                                     &local->preparent, &local->postparent, NULL);
-        }
-
-out:
-        return 0;
-}
-
-
-
-int32_t
-stripe_first_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, fd_t *fd,
-                   inode_t *inode, struct iatt *buf, struct iatt *preparent,
-                   struct iatt *postparent, dict_t *xdata)
-{
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        call_frame_t     *prev = NULL;
-        xlator_list_t    *trav = NULL;
-        int               i    = 1;
-        dict_t           *dict = NULL;
-        loc_t            *loc  = NULL;
-        int32_t           need_unref = 0;
-        int32_t           ret  = -1;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        priv  = this->private;
-        local = frame->local;
-        trav = this->children;
-        loc = &local->loc;
-
-        --local->call_count;
-
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_DEBUG, "%s returned error %s",
-                        prev->this->name, strerror (op_errno));
-                 local->failed = 1;
-                 local->op_errno = op_errno;
-        }
-
-        local->op_ret = 0;
-        /* Get the mapping in inode private */
-        /* Get the stat buf right */
-        local->stbuf      = *buf;
-        local->preparent  = *preparent;
-        local->postparent = *postparent;
-
-        local->stbuf_blocks += buf->ia_blocks;
-        local->preparent_blocks  += preparent->ia_blocks;
-        local->postparent_blocks += postparent->ia_blocks;
-
-        if (local->stbuf_size < buf->ia_size)
-              local->stbuf_size = buf->ia_size;
-        if (local->preparent_size < preparent->ia_size)
-              local->preparent_size = preparent->ia_size;
-        if (local->postparent_size < postparent->ia_size)
-              local->postparent_size = postparent->ia_size;
-
-        if (local->failed)
-                local->op_ret = -1;
-
-        if (local->op_ret == -1) {
-                local->call_count = 1;
-                STACK_WIND (frame, stripe_create_fail_unlink_cbk,
-                            FIRST_CHILD (this), FIRST_CHILD (this)->fops->unlink,
-                            &local->loc, 0, NULL);
-                return 0;
-        }
-
-        if (local->op_ret >= 0) {
-                local->preparent.ia_blocks  = local->preparent_blocks;
-                local->preparent.ia_size    = local->preparent_size;
-                local->postparent.ia_blocks = local->postparent_blocks;
-                local->postparent.ia_size   = local->postparent_size;
-                local->stbuf.ia_size        = local->stbuf_size;
-                local->stbuf.ia_blocks      = local->stbuf_blocks;
-        }
-
-        /* Send a setxattr request to nodes where the
-           files are created */
-        trav = trav->next;
-        while (trav) {
-                if (priv->xattr_supported) {
-                        dict = dict_new ();
-                        if (!dict) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "failed to allocate dict %s", loc->path);
-                        }
-                        need_unref = 1;
-
-                        dict_copy (local->xattr, dict);
-
-                        ret = stripe_xattr_request_build (this, dict,
-                                                          local->stripe_size,
-                                                          priv->child_count,
-                                                          i, priv->coalesce);
-                        if (ret)
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "failed to build xattr request");
-                } else {
-                        dict = local->xattr;
-                }
-
-                STACK_WIND (frame, stripe_create_cbk, trav->xlator,
-                            trav->xlator->fops->create, &local->loc,
-                            local->flags, local->mode, local->umask, local->fd,
-                            dict);
-                trav = trav->next;
-                if (need_unref && dict)
-                        dict_unref (dict);
-                i++;
-        }
-
-out:
-        return 0;
-}
-
-
-
-/**
- * stripe_create - If a block-size is specified for the 'name', create the
- *    file in all the child nodes. If not, create it in only first child.
- *
- * @name- complete path of the file to be created.
- */
-int32_t
-stripe_create (call_frame_t *frame, xlator_t *this, loc_t *loc,
-               int32_t flags, mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
-{
-        stripe_private_t *priv = NULL;
-        stripe_local_t   *local = NULL;
-        int32_t           op_errno = EINVAL;
-        int               ret            = 0;
-        int               need_unref     = 0;
-        int               i              = 0;
-        dict_t           *dict           = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-
-        /* files created in O_APPEND mode does not allow lseek() on fd */
-        flags &= ~O_APPEND;
-
-        if (priv->first_child_down || priv->nodes_down) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "First node down, returning EIO");
-                op_errno = EIO;
-                goto err;
-        }
-
-        /* Initialization */
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        local->op_errno = ENOTCONN;
-        local->stripe_size = stripe_get_matching_bs (loc->path, priv);
-        frame->local = local;
-        local->inode = inode_ref (loc->inode);
-        loc_copy (&local->loc, loc);
-        local->fd = fd_ref (fd);
-        local->flags = flags;
-        local->mode = mode;
-        local->umask = umask;
-        if (xdata)
-                local->xattr = dict_ref (xdata);
-
-        local->call_count = priv->child_count;
-        /* Send a setxattr request to nodes where the
-           files are created */
-
-        if (priv->xattr_supported) {
-                dict = dict_new ();
-                if (!dict) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "failed to allocate dict %s", loc->path);
-                }
-                need_unref = 1;
-
-                dict_copy (xdata, dict);
-
-                ret = stripe_xattr_request_build (this, dict,
-                                                  local->stripe_size,
-                                                  priv->child_count,
-                                                  i, priv->coalesce);
-                if (ret)
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "failed to build xattr request");
-        } else {
-                        dict = xdata;
-        }
-
-
-        STACK_WIND (frame, stripe_first_create_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->create, loc, flags, mode,
-                    umask, fd, dict);
-
-        if (need_unref && dict)
-                dict_unref (dict);
-
-
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (create, frame, -1, op_errno, NULL, NULL, NULL,
-                             NULL, NULL, xdata);
-        return 0;
-}
-
-int32_t
-stripe_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                        local->op_errno = op_errno;
-                }
-
-                if (op_ret >= 0)
-                        local->op_ret = op_ret;
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                STRIPE_STACK_UNWIND (open, frame, local->op_ret,
-                                     local->op_errno, local->fd, xdata);
-        }
-out:
-        return 0;
-}
-
-
-/**
- * stripe_open -
- */
-int32_t
-stripe_open (call_frame_t *frame, xlator_t *this, loc_t *loc,
-             int32_t flags, fd_t *fd, dict_t *xdata)
-{
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        xlator_list_t    *trav = NULL;
-        int32_t           op_errno = 1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        if (priv->first_child_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Initialization */
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-        /* files opened in O_APPEND mode does not allow lseek() on fd */
-        flags &= ~O_APPEND;
-
-        local->fd = fd_ref (fd);
-        frame->local = local;
-        loc_copy (&local->loc, loc);
-
-        /* Striped files */
-        local->flags = flags;
-        local->call_count = priv->child_count;
-        local->stripe_size = stripe_get_matching_bs (loc->path, priv);
-
-        while (trav) {
-                STACK_WIND (frame, stripe_open_cbk, trav->xlator,
-                            trav->xlator->fops->open,
-                            &local->loc, local->flags, local->fd,
-                            xdata);
-                trav = trav->next;
-        }
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (open, frame, -1, op_errno, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_ret = -1;
-                        local->op_errno = op_errno;
-                }
-
-                if (op_ret >= 0)
-                        local->op_ret = op_ret;
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                STRIPE_STACK_UNWIND (opendir, frame, local->op_ret,
-                                     local->op_errno, local->fd, NULL);
-        }
-out:
-        return 0;
-}
-
-
-int32_t
-stripe_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd, dict_t *xdata)
-{
-        xlator_list_t    *trav = NULL;
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        int32_t           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        if (priv->first_child_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Initialization */
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        frame->local = local;
-        local->call_count = priv->child_count;
-        local->fd = fd_ref (fd);
-
-        while (trav) {
-                STACK_WIND (frame, stripe_opendir_cbk, trav->xlator,
-                            trav->xlator->fops->opendir, loc, fd, NULL);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (opendir, frame, -1, op_errno, NULL, NULL);
-        return 0;
-}
-
-int32_t
-stripe_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-               int32_t op_ret, int32_t op_errno, struct gf_flock *lock, dict_t *xdata)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-                if (op_ret >= 0) {
-                        if (FIRST_CHILD(this) == prev->this) {
-                                /* First successful call, copy the *lock */
-                                local->op_ret = op_ret;
-                                local->lock = *lock;
-                        }
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-                STRIPE_STACK_UNWIND (lk, frame, local->op_ret,
-                                     local->op_errno, &local->lock, NULL);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_lk (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
-           struct gf_flock *lock, dict_t *xdata)
-{
-        stripe_local_t   *local = NULL;
-        xlator_list_t    *trav = NULL;
-        stripe_private_t *priv = NULL;
-        int32_t           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-        trav = this->children;
-        priv = this->private;
-
-        /* Initialization */
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        local->call_count = priv->child_count;
-
-        while (trav) {
-                STACK_WIND (frame, stripe_lk_cbk, trav->xlator,
-                            trav->xlator->fops->lk, fd, cmd, lock, NULL);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (lk, frame, -1, op_errno, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local   = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-                if (op_ret >= 0)
-                        local->op_ret = op_ret;
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                STRIPE_STACK_UNWIND (flush, frame, local->op_ret,
-                                     local->op_errno, NULL);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_flush (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
-{
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        xlator_list_t    *trav = NULL;
-        int32_t           op_errno = 1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        if (priv->first_child_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-        /* Initialization */
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        local->call_count = priv->child_count;
-
-        while (trav) {
-                STACK_WIND (frame, stripe_flush_cbk,  trav->xlator,
-                            trav->xlator->fops->flush, fd, NULL);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (flush, frame, -1, op_errno, NULL);
-        return 0;
-}
-
-
-
-int32_t
-stripe_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                  struct iatt *postbuf, dict_t *xdata)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local   = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-                if (op_ret >= 0) {
-                        local->op_ret = op_ret;
-                        if (FIRST_CHILD(this) == prev->this) {
-                                local->pre_buf  = *prebuf;
-                                local->post_buf = *postbuf;
-                        }
-                        local->prebuf_blocks  += prebuf->ia_blocks;
-                        local->postbuf_blocks += postbuf->ia_blocks;
-
-			correct_file_size(prebuf, local->fctx, prev);
-			correct_file_size(postbuf, local->fctx, prev);
-
-                        if (local->prebuf_size < prebuf->ia_size)
-                                local->prebuf_size = prebuf->ia_size;
-
-                        if (local->postbuf_size < postbuf->ia_size)
-                                local->postbuf_size = postbuf->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                if (local->op_ret != -1) {
-                        local->pre_buf.ia_blocks  = local->prebuf_blocks;
-                        local->pre_buf.ia_size    = local->prebuf_size;
-                        local->post_buf.ia_blocks = local->postbuf_blocks;
-                        local->post_buf.ia_size   = local->postbuf_size;
-                }
-
-                STRIPE_STACK_UNWIND (fsync, frame, local->op_ret,
-                                     local->op_errno, &local->pre_buf,
-                                     &local->post_buf, NULL);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags, dict_t *xdata)
-{
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        xlator_list_t    *trav = NULL;
-	stripe_fd_ctx_t  *fctx = NULL;
-        int32_t           op_errno = 1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        /* Initialization */
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	inode_ctx_get(fd->inode, this, (uint64_t *) &fctx);
-	if (!fctx) {
-		op_errno = EINVAL;
-		goto err;
-	}
-	local->fctx = fctx;
-
-        local->op_ret = -1;
-        frame->local = local;
-        local->call_count = priv->child_count;
-
-        while (trav) {
-                STACK_WIND (frame, stripe_fsync_cbk, trav->xlator,
-                            trav->xlator->fops->fsync, fd, flags, NULL);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (fsync, frame, -1, op_errno, NULL, NULL, NULL);
-        return 0;
-}
-
-int32_t
-stripe_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno, struct iatt *buf, dict_t *xdata)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-
-                if (op_ret == 0) {
-                        local->op_ret = 0;
-
-                        if (FIRST_CHILD(this) == prev->this)
-                                local->stbuf = *buf;
-
-                        local->stbuf_blocks += buf->ia_blocks;
-
-			correct_file_size(buf, local->fctx, prev);
-
-                        if (local->stbuf_size < buf->ia_size)
-                                local->stbuf_size = buf->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                if (local->op_ret != -1) {
-                        local->stbuf.ia_size   = local->stbuf_size;
-                        local->stbuf.ia_blocks = local->stbuf_blocks;
-                }
-
-                STRIPE_STACK_UNWIND (fstat, frame, local->op_ret,
-                                     local->op_errno, &local->stbuf, NULL);
-        }
-
-out:
-        return 0;
-}
-
-int32_t
-stripe_fstat (call_frame_t *frame,
-              xlator_t *this,
-              fd_t *fd, dict_t *xdata)
-{
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        xlator_list_t    *trav = NULL;
-	stripe_fd_ctx_t  *fctx = NULL;
-        int32_t           op_errno = 1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        /* Initialization */
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        local->call_count = priv->child_count;
-
-	if (IA_ISREG(fd->inode->ia_type)) {
-		inode_ctx_get(fd->inode, this, (uint64_t *) &fctx);
-		if (!fctx)
-			goto err;
-		local->fctx = fctx;
-	}
-
-        while (trav) {
-                STACK_WIND (frame, stripe_fstat_cbk, trav->xlator,
-                            trav->xlator->fops->fstat, fd, NULL);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (fstat, frame, -1, op_errno, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_ftruncate (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset, dict_t *xdata)
-{
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-	stripe_fd_ctx_t  *fctx = NULL;
-	int		  i, eof_idx;
-	off_t		  dest_offset, tmp_offset;
-        int32_t		  op_errno = 1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-        priv = this->private;
-
-        /* Initialization */
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        local->call_count = priv->child_count;
-
-	inode_ctx_get(fd->inode, this, (uint64_t *) &fctx);
-	if (!fctx) {
-		gf_log(this->name, GF_LOG_ERROR, "no stripe context");
-		op_errno = EINVAL;
-		goto err;
-	}
-	if (!fctx->stripe_count) {
-		gf_log(this->name, GF_LOG_ERROR, "no stripe count");
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	local->fctx = fctx;
-	eof_idx = (offset / fctx->stripe_size) % fctx->stripe_count;
-
-	for (i = 0; i < fctx->stripe_count; i++) {
-		if (!fctx->xl_array[i]) {
-			gf_log(this->name, GF_LOG_ERROR, "no xlator at index "
-				"%d", i);
-			op_errno = EINVAL;
-			goto err;
-		}
-
-		if (fctx->stripe_coalesce) {
-			if (i < eof_idx)
-				tmp_offset = roof(offset, fctx->stripe_size *
-						fctx->stripe_count);
-			else if (i > eof_idx)
-				tmp_offset = floor(offset, fctx->stripe_size *
-						fctx->stripe_count);
-			else
-				tmp_offset = offset;
-
-			dest_offset = coalesced_offset(tmp_offset,
-				fctx->stripe_size, fctx->stripe_count);
-		} else {
-			dest_offset = offset;
-		}
-
-		STACK_WIND(frame, stripe_truncate_cbk, fctx->xl_array[i],
-			fctx->xl_array[i]->fops->ftruncate, fd, dest_offset,
-			NULL);
-	}
-
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (ftruncate, frame, -1, op_errno, NULL, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_fsyncdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local   = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-                if (op_ret >= 0)
-                        local->op_ret = op_ret;
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                STRIPE_STACK_UNWIND (fsyncdir, frame, local->op_ret,
-                                     local->op_errno, NULL);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags, dict_t *xdata)
-{
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        xlator_list_t    *trav = NULL;
-        int32_t           op_errno = 1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        /* Initialization */
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        local->call_count = priv->child_count;
-
-        while (trav) {
-                STACK_WIND (frame, stripe_fsyncdir_cbk, trav->xlator,
-                            trav->xlator->fops->fsyncdir, fd, flags, NULL);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (fsyncdir, frame, -1, op_errno, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_readv_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, struct iatt *buf, dict_t *xdata)
-{
-        int32_t         i = 0;
-        int32_t         callcnt = 0;
-        int32_t         count = 0;
-        stripe_local_t *local = NULL;
-        struct iovec   *vec = NULL;
-        struct iatt     tmp_stbuf = {0,};
-        struct iobref  *tmp_iobref = NULL;
-        struct iobuf   *iobuf = NULL;
-	call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        local = frame->local;
-	prev = cookie;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-                if (op_ret != -1) {
-			correct_file_size(buf, local->fctx, prev);
-                        if (local->stbuf_size < buf->ia_size)
-                                local->stbuf_size = buf->ia_size;
-		}
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                op_ret = 0;
-
-                /* Keep extra space for filling in '\0's */
-                vec = GF_CALLOC ((local->count * 2), sizeof (struct iovec),
-                                 gf_stripe_mt_iovec);
-                if (!vec) {
-                        op_ret = -1;
-                        goto done;
-                }
-
-                for (i = 0; i < local->wind_count; i++) {
-                        if (local->replies[i].op_ret) {
-                                memcpy ((vec + count), local->replies[i].vector,
-                                        (local->replies[i].count * sizeof (struct iovec)));
-                                count +=  local->replies[i].count;
-                                op_ret += local->replies[i].op_ret;
-                        }
-                        if ((local->replies[i].op_ret <
-                             local->replies[i].requested_size) &&
-                            (local->stbuf_size > (local->offset + op_ret))) {
-                                /* Fill in 0s here */
-                                vec[count].iov_len  =
-                                        (local->replies[i].requested_size -
-                                         local->replies[i].op_ret);
-                                iobuf = iobuf_get2 (this->ctx->iobuf_pool,
-                                                    vec[count].iov_len);
-                                if (!iobuf) {
-                                        gf_log (this->name, GF_LOG_ERROR,
-                                                "Out of memory.");
-                                        op_ret = -1;
-                                        op_errno = ENOMEM;
-                                        goto done;
-                                }
-                                memset (iobuf->ptr, 0, vec[count].iov_len);
-                                vec[count].iov_base = iobuf->ptr;
-
-                                iobref_add (local->iobref, iobuf);
-                                iobuf_unref(iobuf);
-
-                                op_ret += vec[count].iov_len;
-                                count++;
-                        }
-                        GF_FREE (local->replies[i].vector);
-                }
-
-                /* FIXME: notice that st_ino, and st_dev (gen) will be
-                 * different than what inode will have. Make sure this doesn't
-                 * cause any bugs at higher levels */
-                memcpy (&tmp_stbuf, &local->replies[0].stbuf,
-                        sizeof (struct iatt));
-                tmp_stbuf.ia_size = local->stbuf_size;
-
-        done:
-                GF_FREE (local->replies);
-                tmp_iobref = local->iobref;
-                STRIPE_STACK_UNWIND (readv, frame, op_ret, op_errno, vec,
-                                     count, &tmp_stbuf, tmp_iobref, NULL);
-
-                iobref_unref (tmp_iobref);
-                GF_FREE (vec);
-        }
-out:
-        return 0;
-}
-
-/**
- * stripe_readv_cbk - get all the striped reads, and order it properly, send it
- *        to above layer after putting it in a single vector.
- */
-int32_t
-stripe_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno, struct iovec *vector,
-                  int32_t count, struct iatt *stbuf, struct iobref *iobref, dict_t *xdata)
-{
-        int32_t         index = 0;
-        int32_t         callcnt = 0;
-        int32_t         final_count = 0;
-        int32_t         need_to_check_proper_size = 0;
-        call_frame_t   *mframe = NULL;
-        stripe_local_t *mlocal = NULL;
-        stripe_local_t *local = NULL;
-        struct iovec   *final_vec = NULL;
-        struct iatt     tmp_stbuf = {0,};
-        struct iatt    *tmp_stbuf_p = NULL; //need it for a warning
-        struct iobref  *tmp_iobref = NULL;
-        stripe_fd_ctx_t  *fctx = NULL;
-	call_frame_t	*prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto end;
-        }
-
-        local  = frame->local;
-        index  = local->node_index;
-	prev = cookie;
-        mframe = local->orig_frame;
-        if (!mframe)
-                goto out;
-
-        mlocal = mframe->local;
-        if (!mlocal)
-                goto out;
-
-        fctx = mlocal->fctx;
-
-        LOCK (&mframe->lock);
-        {
-                mlocal->replies[index].op_ret = op_ret;
-                mlocal->replies[index].op_errno = op_errno;
-                mlocal->replies[index].requested_size = local->readv_size;
-                if (op_ret >= 0) {
-                        mlocal->replies[index].stbuf  = *stbuf;
-                        mlocal->replies[index].count  = count;
-                        mlocal->replies[index].vector = iov_dup (vector, count);
-
-			correct_file_size(stbuf, fctx, prev);
-
-                        if (local->stbuf_size < stbuf->ia_size)
-                                local->stbuf_size = stbuf->ia_size;
-                        local->stbuf_blocks += stbuf->ia_blocks;
-
-                        if (!mlocal->iobref)
-                                mlocal->iobref = iobref_new ();
-                        iobref_merge (mlocal->iobref, iobref);
-                }
-                callcnt = ++mlocal->call_count;
-        }
-        UNLOCK(&mframe->lock);
-
-        if (callcnt == mlocal->wind_count) {
-                op_ret = 0;
-
-                for (index=0; index < mlocal->wind_count; index++) {
-                        /* check whether each stripe returned
-                         * 'expected' number of bytes */
-                        if (mlocal->replies[index].op_ret == -1) {
-                                op_ret = -1;
-                                op_errno = mlocal->replies[index].op_errno;
-                                break;
-                        }
-                        /* TODO: handle the 'holes' within the read range
-                           properly */
-                        if (mlocal->replies[index].op_ret <
-                            mlocal->replies[index].requested_size) {
-                                need_to_check_proper_size = 1;
-                        }
-
-                        op_ret       += mlocal->replies[index].op_ret;
-                        mlocal->count += mlocal->replies[index].count;
-                }
-                if (op_ret == -1)
-                        goto done;
-                if (need_to_check_proper_size)
-                        goto check_size;
-
-                final_vec = GF_CALLOC (mlocal->count, sizeof (struct iovec),
-                                       gf_stripe_mt_iovec);
-
-                if (!final_vec) {
-                        op_ret = -1;
-                        goto done;
-                }
-
-                for (index = 0; index < mlocal->wind_count; index++) {
-                        memcpy ((final_vec + final_count),
-                                mlocal->replies[index].vector,
-                                (mlocal->replies[index].count *
-                                 sizeof (struct iovec)));
-                        final_count +=  mlocal->replies[index].count;
-                        GF_FREE (mlocal->replies[index].vector);
-                }
-
-                /* FIXME: notice that st_ino, and st_dev (gen) will be
-                 * different than what inode will have. Make sure this doesn't
-                 * cause any bugs at higher levels */
-                memcpy (&tmp_stbuf, &mlocal->replies[0].stbuf,
-                        sizeof (struct iatt));
-                tmp_stbuf.ia_size = local->stbuf_size;
-                tmp_stbuf.ia_blocks = local->stbuf_blocks;
-
-        done:
-                /* */
-                GF_FREE (mlocal->replies);
-                tmp_iobref = mlocal->iobref;
-                /* work around for nfs truncated read. Bug 3774 */
-                tmp_stbuf_p = &tmp_stbuf;
-                WIPE (tmp_stbuf_p);
-                STRIPE_STACK_UNWIND (readv, mframe, op_ret, op_errno, final_vec,
-                                     final_count, &tmp_stbuf, tmp_iobref, NULL);
-
-                iobref_unref (tmp_iobref);
-                GF_FREE (final_vec);
-        }
-
-        goto out;
-
-check_size:
-        mlocal->call_count = fctx->stripe_count;
-
-        for (index = 0; index < fctx->stripe_count; index++) {
-                STACK_WIND (mframe, stripe_readv_fstat_cbk,
-                            (fctx->xl_array[index]),
-                            (fctx->xl_array[index])->fops->fstat,
-                            mlocal->fd, NULL);
-        }
-
-out:
-        STRIPE_STACK_DESTROY (frame);
-end:
-        return 0;
-}
-
-
-int32_t
-stripe_readv (call_frame_t *frame, xlator_t *this, fd_t *fd,
-              size_t size, off_t offset, uint32_t flags, dict_t *xdata)
-{
-        int32_t           op_errno = EINVAL;
-        int32_t           idx = 0;
-        int32_t           index = 0;
-        int32_t           num_stripe = 0;
-        int32_t           off_index = 0;
-        size_t            frame_size = 0;
-        off_t             rounded_end = 0;
-        uint64_t          tmp_fctx = 0;
-        uint64_t          stripe_size = 0;
-        off_t             rounded_start = 0;
-        off_t             frame_offset = offset;
-	off_t		  dest_offset = 0;
-        stripe_local_t   *local = NULL;
-        call_frame_t     *rframe = NULL;
-        stripe_local_t   *rlocal = NULL;
-        stripe_fd_ctx_t  *fctx = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-        inode_ctx_get (fd->inode, this, &tmp_fctx);
-        if (!tmp_fctx) {
-                op_errno = EBADFD;
-                goto err;
-        }
-        fctx = (stripe_fd_ctx_t *)(long)tmp_fctx;
-        stripe_size = fctx->stripe_size;
-
-        STRIPE_VALIDATE_FCTX (fctx, err);
-
-        if (!stripe_size) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Wrong stripe size for the file");
-                goto err;
-        }
-        /* The file is stripe across the child nodes. Send the read request
-         * to the child nodes appropriately after checking which region of
-         * the file is in which child node. Always '0-<stripe_size>' part of
-         * the file resides in the first child.
-         */
-        rounded_start = floor (offset, stripe_size);
-        rounded_end = roof (offset+size, stripe_size);
-        num_stripe = (rounded_end- rounded_start)/stripe_size;
-
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        frame->local = local;
-
-        /* This is where all the vectors should be copied. */
-        local->replies = GF_CALLOC (num_stripe, sizeof (struct stripe_replies),
-                                    gf_stripe_mt_stripe_replies);
-        if (!local->replies) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-        off_index = (offset / stripe_size) % fctx->stripe_count;
-        local->wind_count = num_stripe;
-        local->readv_size = size;
-        local->offset     = offset;
-        local->fd         = fd_ref (fd);
-        local->fctx       = fctx;
-
-        for (index = off_index; index < (num_stripe + off_index); index++) {
-                rframe = copy_frame (frame);
-                rlocal = mem_get0 (this->local_pool);
-                if (!rlocal) {
-                        op_errno = ENOMEM;
-                        goto err;
-                }
-
-                frame_size = min (roof (frame_offset+1, stripe_size),
-                                  (offset + size)) - frame_offset;
-
-                rlocal->node_index = index - off_index;
-                rlocal->orig_frame = frame;
-                rlocal->readv_size = frame_size;
-                rframe->local = rlocal;
-                idx = (index % fctx->stripe_count);
-
-		if (fctx->stripe_coalesce)
-			dest_offset = coalesced_offset(frame_offset,
-				stripe_size, fctx->stripe_count);
-		else
-			dest_offset = frame_offset;
-
-                STACK_WIND (rframe, stripe_readv_cbk, fctx->xl_array[idx],
-                            fctx->xl_array[idx]->fops->readv,
-                            fd, frame_size, dest_offset, flags, xdata);
-
-                frame_offset += frame_size;
-        }
-
-        return 0;
-err:
-        if (rframe)
-                STRIPE_STACK_DESTROY (rframe);
-
-        STRIPE_STACK_UNWIND (readv, frame, -1, op_errno, NULL, 0, NULL, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                   struct iatt *postbuf, dict_t *xdata)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-	stripe_local_t *mlocal = NULL;
-        call_frame_t   *prev = NULL;
-	call_frame_t   *mframe = NULL;
-	struct stripe_replies *reply = NULL;
-	int32_t		i = 0;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-	mframe = local->orig_frame;
-	mlocal = mframe->local;
-
-        LOCK(&frame->lock);
-        {
-                callcnt = ++mlocal->call_count;
-
-		mlocal->replies[local->node_index].op_ret = op_ret;
-		mlocal->replies[local->node_index].op_errno = op_errno;
-
-                if (op_ret >= 0) {
-                        mlocal->post_buf = *postbuf;
-                        mlocal->pre_buf = *prebuf;
-
-			mlocal->prebuf_blocks  += prebuf->ia_blocks;
-			mlocal->postbuf_blocks += postbuf->ia_blocks;
-
-			correct_file_size(prebuf, mlocal->fctx, prev);
-			correct_file_size(postbuf, mlocal->fctx, prev);
-
-			if (mlocal->prebuf_size < prebuf->ia_size)
-				mlocal->prebuf_size = prebuf->ia_size;
-			if (mlocal->postbuf_size < postbuf->ia_size)
-				mlocal->postbuf_size = postbuf->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if ((callcnt == mlocal->wind_count) && mlocal->unwind) {
-		mlocal->pre_buf.ia_size = mlocal->prebuf_size;
-		mlocal->pre_buf.ia_blocks = mlocal->prebuf_blocks;
-		mlocal->post_buf.ia_size = mlocal->postbuf_size;
-		mlocal->post_buf.ia_blocks = mlocal->postbuf_blocks;
-
-		/*
-		 * Only return the number of consecutively written bytes up until
-		 * the first error. Only return an error if it occurs first.
-		 *
-		 * When a short write occurs, the application should retry at the
-		 * appropriate offset, at which point we'll potentially pass back
-		 * the error.
-		 */
-		for (i = 0, reply = mlocal->replies; i < mlocal->wind_count;
-			i++, reply++) {
-			if (reply->op_ret == -1) {
-				gf_log(this->name, GF_LOG_DEBUG, "reply %d "
-					"returned error %s", i,
-					strerror(reply->op_errno));
-				if (!mlocal->op_ret) {
-					mlocal->op_ret = -1;
-					mlocal->op_errno = reply->op_errno;
-				}
-				break;
-			}
-
-			mlocal->op_ret += reply->op_ret;
-
-			if (reply->op_ret < reply->requested_size)
-				break;
-		}
-
-		GF_FREE(mlocal->replies);
-
-                STRIPE_STACK_UNWIND (writev, mframe, mlocal->op_ret,
-                                     mlocal->op_errno, &mlocal->pre_buf,
-                                     &mlocal->post_buf, NULL);
-        }
-out:
-	STRIPE_STACK_DESTROY(frame);
-        return 0;
-}
-
-int32_t
-stripe_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
-               struct iovec *vector, int32_t count, off_t offset,
-               uint32_t flags, struct iobref *iobref, dict_t *xdata)
-{
-        struct iovec     *tmp_vec = NULL;
-        stripe_local_t   *local = NULL;
-        stripe_fd_ctx_t  *fctx = NULL;
-        int32_t           op_errno = 1;
-        int32_t           idx = 0;
-        int32_t           total_size = 0;
-        int32_t           offset_offset = 0;
-        int32_t           remaining_size = 0;
-        int32_t           tmp_count = count;
-        off_t             fill_size = 0;
-        uint64_t          stripe_size = 0;
-        uint64_t          tmp_fctx = 0;
-	off_t		  dest_offset = 0;
-	off_t		  rounded_start = 0;
-	off_t		  rounded_end = 0;
-	int32_t		  total_chunks = 0;
-	call_frame_t	  *wframe = NULL;
-	stripe_local_t	  *wlocal = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-        inode_ctx_get (fd->inode, this, &tmp_fctx);
-        if (!tmp_fctx) {
-                op_errno = EINVAL;
-                goto err;
-        }
-        fctx = (stripe_fd_ctx_t *)(long)tmp_fctx;
-        stripe_size = fctx->stripe_size;
-
-        STRIPE_VALIDATE_FCTX (fctx, err);
-
-        /* File has to be stripped across the child nodes */
-        for (idx = 0; idx< count; idx ++) {
-                total_size += vector[idx].iov_len;
-        }
-        remaining_size = total_size;
-
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        frame->local = local;
-        local->stripe_size = stripe_size;
-	local->fctx = fctx;
-
-        if (!stripe_size) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Wrong stripe size for the file");
-                op_errno = EINVAL;
-                goto err;
-        }
-
-	rounded_start = floor(offset, stripe_size);
-	rounded_end = roof(offset + total_size, stripe_size);
-	total_chunks = (rounded_end - rounded_start) / stripe_size;
-	local->replies = GF_CALLOC(total_chunks, sizeof(struct stripe_replies),
-				gf_stripe_mt_stripe_replies);
-	if (!local->replies) {
-		op_errno = ENOMEM;
-		goto err;
-	}
-
-	total_chunks = 0;
-        while (1) {
-		wframe = copy_frame(frame);
-		wlocal = mem_get0(this->local_pool);
-		if (!wlocal) {
-			op_errno = ENOMEM;
-			goto err;
-		}
-		wlocal->orig_frame = frame;
-		wframe->local = wlocal;
-
-                /* Send striped chunk of the vector to child
-                   nodes appropriately. */
-                idx = (((offset + offset_offset) /
-                        local->stripe_size) % fctx->stripe_count);
-
-                fill_size = (local->stripe_size -
-                             ((offset + offset_offset) % local->stripe_size));
-                if (fill_size > remaining_size)
-                        fill_size = remaining_size;
-
-                remaining_size -= fill_size;
-
-                tmp_count = iov_subset (vector, count, offset_offset,
-                                        offset_offset + fill_size, NULL);
-                tmp_vec = GF_CALLOC (tmp_count, sizeof (struct iovec),
-                                     gf_stripe_mt_iovec);
-                if (!tmp_vec) {
-                        op_errno = ENOMEM;
-                        goto err;
-                }
-                tmp_count = iov_subset (vector, count, offset_offset,
-                                        offset_offset + fill_size, tmp_vec);
-
-                local->wind_count++;
-                if (remaining_size == 0)
-                        local->unwind = 1;
-
-		/*
-		 * Store off the request index (with respect to the chunk of the
-		 * initial offset) and the size of the request. This is required
-		 * in the callback to calculate an appropriate return value in
-		 * the event of a write failure in one or more requests.
-		 */
-		wlocal->node_index = total_chunks;
-		local->replies[total_chunks].requested_size = fill_size;
-
-		dest_offset = offset + offset_offset;
-		if (fctx->stripe_coalesce)
-			dest_offset = coalesced_offset(dest_offset,
-					local->stripe_size, fctx->stripe_count);
-
-                STACK_WIND (wframe, stripe_writev_cbk, fctx->xl_array[idx],
-                            fctx->xl_array[idx]->fops->writev, fd, tmp_vec,
-                            tmp_count, dest_offset, flags, iobref,
-                            xdata);
-
-                GF_FREE (tmp_vec);
-                offset_offset += fill_size;
-		total_chunks++;
-                if (remaining_size == 0)
-                        break;
-        }
-
-        return 0;
-err:
-	if (wframe)
-		STRIPE_STACK_DESTROY(wframe);
-
-        STRIPE_STACK_UNWIND (writev, frame, -1, op_errno, NULL, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                     struct iatt *postbuf, dict_t *xdata)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-	stripe_local_t *mlocal = NULL;
-        call_frame_t   *prev = NULL;
-	call_frame_t   *mframe = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-	mframe = local->orig_frame;
-	mlocal = mframe->local;
-
-        LOCK(&frame->lock);
-        {
-                callcnt = ++mlocal->call_count;
-
-                if (op_ret == 0) {
-                        mlocal->post_buf = *postbuf;
-                        mlocal->pre_buf = *prebuf;
-
-			mlocal->prebuf_blocks  += prebuf->ia_blocks;
-			mlocal->postbuf_blocks += postbuf->ia_blocks;
-
-			correct_file_size(prebuf, mlocal->fctx, prev);
-			correct_file_size(postbuf, mlocal->fctx, prev);
-
-			if (mlocal->prebuf_size < prebuf->ia_size)
-				mlocal->prebuf_size = prebuf->ia_size;
-			if (mlocal->postbuf_size < postbuf->ia_size)
-				mlocal->postbuf_size = postbuf->ia_size;
-                }
-
-		/* return the first failure */
-		if (mlocal->op_ret == 0) {
-			mlocal->op_ret = op_ret;
-			mlocal->op_errno = op_errno;
-		}
-        }
-        UNLOCK (&frame->lock);
-
-        if ((callcnt == mlocal->wind_count) && mlocal->unwind) {
-		mlocal->pre_buf.ia_size = mlocal->prebuf_size;
-		mlocal->pre_buf.ia_blocks = mlocal->prebuf_blocks;
-		mlocal->post_buf.ia_size = mlocal->postbuf_size;
-		mlocal->post_buf.ia_blocks = mlocal->postbuf_blocks;
-
-                STRIPE_STACK_UNWIND (fallocate, mframe, mlocal->op_ret,
-                                     mlocal->op_errno, &mlocal->pre_buf,
-                                     &mlocal->post_buf, NULL);
-        }
-out:
-	STRIPE_STACK_DESTROY(frame);
-        return 0;
-}
-
-int32_t
-stripe_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
-		 off_t offset, size_t len, dict_t *xdata)
-{
-        stripe_local_t   *local = NULL;
-        stripe_fd_ctx_t  *fctx = NULL;
-        int32_t           op_errno = 1;
-        int32_t           idx = 0;
-        int32_t           offset_offset = 0;
-        int32_t           remaining_size = 0;
-        off_t             fill_size = 0;
-        uint64_t          stripe_size = 0;
-        uint64_t          tmp_fctx = 0;
-	off_t		  dest_offset = 0;
-	call_frame_t	  *fframe = NULL;
-	stripe_local_t	  *flocal = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-        inode_ctx_get (fd->inode, this, &tmp_fctx);
-        if (!tmp_fctx) {
-                op_errno = EINVAL;
-                goto err;
-        }
-        fctx = (stripe_fd_ctx_t *)(long)tmp_fctx;
-        stripe_size = fctx->stripe_size;
-
-        STRIPE_VALIDATE_FCTX (fctx, err);
-
-        remaining_size = len;
-
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        frame->local = local;
-        local->stripe_size = stripe_size;
-	local->fctx = fctx;
-
-        if (!stripe_size) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Wrong stripe size for the file");
-                op_errno = EINVAL;
-                goto err;
-        }
-
-        while (1) {
-		fframe = copy_frame(frame);
-		flocal = mem_get0(this->local_pool);
-		if (!flocal) {
-			op_errno = ENOMEM;
-			goto err;
-		}
-		flocal->orig_frame = frame;
-		fframe->local = flocal;
-
-		/* send fallocate request to the associated child node */
-                idx = (((offset + offset_offset) /
-                        local->stripe_size) % fctx->stripe_count);
-
-                fill_size = (local->stripe_size -
-                             ((offset + offset_offset) % local->stripe_size));
-                if (fill_size > remaining_size)
-                        fill_size = remaining_size;
-
-                remaining_size -= fill_size;
-
-                local->wind_count++;
-                if (remaining_size == 0)
-                        local->unwind = 1;
-
-		dest_offset = offset + offset_offset;
-		if (fctx->stripe_coalesce)
-			dest_offset = coalesced_offset(dest_offset,
-					local->stripe_size, fctx->stripe_count);
-
-		/*
-		 * TODO: Create a separate handler for coalesce mode that sends a
-		 * single fallocate per-child (since the ranges are linear).
-		 */
-		STACK_WIND(fframe, stripe_fallocate_cbk, fctx->xl_array[idx],
-			   fctx->xl_array[idx]->fops->fallocate, fd, mode,
-			   dest_offset, fill_size, xdata);
-
-                offset_offset += fill_size;
-                if (remaining_size == 0)
-                        break;
-        }
-
-        return 0;
-err:
-	if (fframe)
-		STRIPE_STACK_DESTROY(fframe);
-
-        STRIPE_STACK_UNWIND (fallocate, frame, -1, op_errno, NULL, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                   struct iatt *postbuf, dict_t *xdata)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-	stripe_local_t *mlocal = NULL;
-        call_frame_t   *prev = NULL;
-	call_frame_t   *mframe = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-	mframe = local->orig_frame;
-	mlocal = mframe->local;
-
-        LOCK(&frame->lock);
-        {
-                callcnt = ++mlocal->call_count;
-
-                if (op_ret == 0) {
-                        mlocal->post_buf = *postbuf;
-                        mlocal->pre_buf = *prebuf;
-
-			mlocal->prebuf_blocks  += prebuf->ia_blocks;
-			mlocal->postbuf_blocks += postbuf->ia_blocks;
-
-			correct_file_size(prebuf, mlocal->fctx, prev);
-			correct_file_size(postbuf, mlocal->fctx, prev);
-
-			if (mlocal->prebuf_size < prebuf->ia_size)
-				mlocal->prebuf_size = prebuf->ia_size;
-			if (mlocal->postbuf_size < postbuf->ia_size)
-				mlocal->postbuf_size = postbuf->ia_size;
-                }
-
-		/* return the first failure */
-		if (mlocal->op_ret == 0) {
-			mlocal->op_ret = op_ret;
-			mlocal->op_errno = op_errno;
-		}
-        }
-        UNLOCK (&frame->lock);
-
-        if ((callcnt == mlocal->wind_count) && mlocal->unwind) {
-		mlocal->pre_buf.ia_size = mlocal->prebuf_size;
-		mlocal->pre_buf.ia_blocks = mlocal->prebuf_blocks;
-		mlocal->post_buf.ia_size = mlocal->postbuf_size;
-		mlocal->post_buf.ia_blocks = mlocal->postbuf_blocks;
-
-                STRIPE_STACK_UNWIND (discard, mframe, mlocal->op_ret,
-                                     mlocal->op_errno, &mlocal->pre_buf,
-                                     &mlocal->post_buf, NULL);
-        }
-out:
-	STRIPE_STACK_DESTROY(frame);
-        return 0;
-}
-
-int32_t
-stripe_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
-	       size_t len, dict_t *xdata)
-{
-        stripe_local_t   *local = NULL;
-        stripe_fd_ctx_t  *fctx = NULL;
-        int32_t           op_errno = 1;
-        int32_t           idx = 0;
-        int32_t           offset_offset = 0;
-        int32_t           remaining_size = 0;
-        off_t             fill_size = 0;
-        uint64_t          stripe_size = 0;
-        uint64_t          tmp_fctx = 0;
-	off_t		  dest_offset = 0;
-	call_frame_t	  *fframe = NULL;
-	stripe_local_t	  *flocal = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-        inode_ctx_get (fd->inode, this, &tmp_fctx);
-        if (!tmp_fctx) {
-                op_errno = EINVAL;
-                goto err;
-        }
-        fctx = (stripe_fd_ctx_t *)(long)tmp_fctx;
-        stripe_size = fctx->stripe_size;
-
-        STRIPE_VALIDATE_FCTX (fctx, err);
-
-        remaining_size = len;
-
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        frame->local = local;
-        local->stripe_size = stripe_size;
-	local->fctx = fctx;
-
-        if (!stripe_size) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Wrong stripe size for the file");
-                op_errno = EINVAL;
-                goto err;
-        }
-
-        while (1) {
-		fframe = copy_frame(frame);
-		flocal = mem_get0(this->local_pool);
-		if (!flocal) {
-			op_errno = ENOMEM;
-			goto err;
-		}
-		flocal->orig_frame = frame;
-		fframe->local = flocal;
-
-		/* send discard request to the associated child node */
-                idx = (((offset + offset_offset) /
-                        local->stripe_size) % fctx->stripe_count);
-
-                fill_size = (local->stripe_size -
-                             ((offset + offset_offset) % local->stripe_size));
-                if (fill_size > remaining_size)
-                        fill_size = remaining_size;
-
-                remaining_size -= fill_size;
-
-                local->wind_count++;
-                if (remaining_size == 0)
-                        local->unwind = 1;
-
-		dest_offset = offset + offset_offset;
-		if (fctx->stripe_coalesce)
-			dest_offset = coalesced_offset(dest_offset,
-					local->stripe_size, fctx->stripe_count);
-
-		/*
-		 * TODO: Create a separate handler for coalesce mode that sends a
-		 * single discard per-child (since the ranges are linear).
-		 */
-		STACK_WIND(fframe, stripe_discard_cbk, fctx->xl_array[idx],
-			   fctx->xl_array[idx]->fops->discard, fd, dest_offset,
-			   fill_size, xdata);
-
-                offset_offset += fill_size;
-                if (remaining_size == 0)
-                        break;
-        }
-
-        return 0;
-err:
-	if (fframe)
-		STRIPE_STACK_DESTROY(fframe);
-
-        STRIPE_STACK_UNWIND (discard, frame, -1, op_errno, NULL, NULL, NULL);
-        return 0;
-}
-
-int32_t
-stripe_release (xlator_t *this, fd_t *fd)
-{
-	return 0;
-}
-
-int
-stripe_forget (xlator_t *this, inode_t *inode)
-{
-        uint64_t          tmp_fctx = 0;
-        stripe_fd_ctx_t  *fctx = NULL;
-
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (inode, err);
-
-        (void) inode_ctx_del (inode, this, &tmp_fctx);
-        if (!tmp_fctx) {
-                goto err;
-        }
-
-        fctx = (stripe_fd_ctx_t *)(long)tmp_fctx;
-
-        if (!fctx->static_array)
-                GF_FREE (fctx->xl_array);
-
-        GF_FREE (fctx);
-err:
-        return 0;
-}
-
-int32_t
-notify (xlator_t *this, int32_t event, void *data, ...)
-{
-        stripe_private_t *priv = NULL;
-        int               down_client = 0;
-        int               i = 0;
-        gf_boolean_t      heard_from_all_children = _gf_false;
-
-        if (!this)
-                return 0;
-
-        priv = this->private;
-        if (!priv)
-                return 0;
-
-        switch (event)
-        {
-        case GF_EVENT_CHILD_UP:
-        {
-                /* get an index number to set */
-                for (i = 0; i < priv->child_count; i++) {
-                        if (data == priv->xl_array[i])
-                                break;
-                }
-
-                if (priv->child_count == i) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "got GF_EVENT_CHILD_UP bad subvolume %s",
-                                data? ((xlator_t *)data)->name: NULL);
-                        break;
-                }
-
-                LOCK (&priv->lock);
-                {
-                        if (data == FIRST_CHILD (this))
-                                priv->first_child_down = 0;
-                        priv->last_event[i] = event;
-                }
-                UNLOCK (&priv->lock);
-        }
-        break;
-        case GF_EVENT_CHILD_CONNECTING:
-        {
-                // 'CONNECTING' doesn't ensure its CHILD_UP, so do nothing
-                goto out;
-        }
-        case GF_EVENT_CHILD_DOWN:
-        {
-                /* get an index number to set */
-                for (i = 0; i < priv->child_count; i++) {
-                        if (data == priv->xl_array[i])
-                                break;
-                }
-
-                if (priv->child_count == i) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "got GF_EVENT_CHILD_DOWN bad subvolume %s",
-                                data? ((xlator_t *)data)->name: NULL);
-                        break;
-                }
-
-                LOCK (&priv->lock);
-                {
-                        if (data == FIRST_CHILD (this))
-                                priv->first_child_down = 1;
-                        priv->last_event[i] = event;
-                }
-                UNLOCK (&priv->lock);
-        }
-        break;
-
-        default:
-        {
-                /* */
-                default_notify (this, event, data);
-                goto out;
-        }
-        break;
-        }
-
-        // Consider child as down if it's last_event is not CHILD_UP
-        for (i = 0, down_client = 0; i < priv->child_count; i++)
-                if (priv->last_event[i] != GF_EVENT_CHILD_UP)
-                        down_client++;
-
-        LOCK (&priv->lock);
-        {
-                priv->nodes_down = down_client;
-        }
-        UNLOCK (&priv->lock);
-
-        heard_from_all_children = _gf_true;
-        for (i = 0; i < priv->child_count; i++)
-                if (!priv->last_event[i])
-                        heard_from_all_children = _gf_false;
-
-        if (heard_from_all_children)
-                default_notify (this, event, data);
-out:
-        return 0;
-}
-
-int
-stripe_setxattr_cbk (call_frame_t *frame, void *cookie,
-                     xlator_t *this, int op_ret, int op_errno, dict_t *xdata)
-{
-        int ret = -1;
-        int call_cnt = 0;
-        stripe_local_t *local = NULL;
-
-        if (!frame || !frame->local || !this) {
-                gf_log ("", GF_LOG_ERROR, "Possible NULL deref");
-                return ret;
-        }
-
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                call_cnt = --local->wind_count;
-
-                /**
-                 * We overwrite ->op_* values here for subsequent faliure
-                 * conditions, hence we propogate the last errno down the
-                 * stack.
-                 */
-                if (op_ret < 0) {
-                        local->op_ret = op_ret;
-                        local->op_errno = op_errno;
-                        goto unlock;
-                }
-        }
-
- unlock:
-        UNLOCK (&frame->lock);
-
-        if (!call_cnt) {
-                STRIPE_STACK_UNWIND (setxattr, frame, local->op_ret,
-                                     local->op_errno, xdata);
-        }
-
-        return 0;
-}
-
-int
-stripe_setxattr (call_frame_t *frame, xlator_t *this,
-                 loc_t *loc, dict_t *dict, int flags, dict_t *xdata)
-{
-        int32_t           op_errno = EINVAL;
-        xlator_list_t    *trav     = NULL;
-        stripe_private_t *priv     = NULL;
-        stripe_local_t   *local    = NULL;
-        int               i        = 0;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        GF_IF_INTERNAL_XATTR_GOTO ("trusted.*stripe*", dict,
-                                   op_errno, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-        frame->local = local;
-        local->wind_count = priv->child_count;
-        local->op_ret = local->op_errno = 0;
-
-        /**
-         * Set xattrs for directories on all subvolumes. Additionally
-         * this power is only given to a special client.
-         */
-        if ((frame->root->pid == GF_CLIENT_PID_GSYNCD) && IA_ISDIR (loc->inode->ia_type)) {
-                for (i = 0; i < priv->child_count; i++, trav = trav->next) {
-                        STACK_WIND (frame, stripe_setxattr_cbk,
-                                    trav->xlator, trav->xlator->fops->setxattr,
-                                    loc, dict, flags, xdata);
-                }
-        } else {
-                local->wind_count = 1;
-                STACK_WIND (frame, stripe_setxattr_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->setxattr,
-                            loc, dict, flags, xdata);
-        }
-
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (setxattr, frame, -1,  op_errno, NULL);
-        return 0;
-}
-
-
-int
-stripe_fsetxattr_cbk (call_frame_t *frame, void *cookie,
-                      xlator_t *this, int op_ret, int op_errno, dict_t *xdata)
-{
-        STRIPE_STACK_UNWIND (fsetxattr, frame, op_ret, op_errno, xdata);
-        return 0;
-}
-
-
-int
-stripe_is_lockinfo (dict_t *this,
-                    char *key,
-                    data_t *value,
-                    void *data)
-{
-        gf_boolean_t *is_lockinfo = NULL;
-
-        if (data == NULL) {
-                goto out;
-        }
-
-        is_lockinfo = data;
-
-        if (XATTR_IS_LOCKINFO (key))
-                *is_lockinfo = _gf_true;
-
-out:
-        return 0;
-}
-
-int32_t
-stripe_fsetxattr_everyone_cbk (call_frame_t *frame, void *cookie,
-                               xlator_t *this, int32_t op_ret, int32_t op_errno,
-                               dict_t *xdata)
-{
-        int             call_count = 0;
-        stripe_local_t *local      = NULL;
-
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                call_count = --local->wind_count;
-
-                if (op_ret < 0) {
-                        local->op_ret = op_ret;
-                        local->op_errno = op_errno;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (call_count == 0) {
-                STRIPE_STACK_UNWIND (fsetxattr, frame, local->op_ret,
-                                     local->op_errno, NULL);
-        }
-        return 0;
-}
-
-int
-stripe_fsetxattr_to_everyone (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                              dict_t *dict, int flags, dict_t *xdata)
-{
-        xlator_list_t    *trav  = NULL;
-        stripe_private_t *priv  = NULL;
-        int               ret   = -1;
-        stripe_local_t   *local = NULL;
-
-        priv = this->private;
-
-        local = mem_get0 (this->local_pool);
-        if (local == NULL) {
-                goto out;
-        }
-
-        frame->local = local;
-
-        local->wind_count = priv->child_count;
-
-        trav = this->children;
-
-        while (trav) {
-                STACK_WIND (frame, stripe_fsetxattr_everyone_cbk,
-                            trav->xlator, trav->xlator->fops->fsetxattr,
-                            fd, dict, flags, xdata);
-                trav = trav->next;
-        }
-
-        ret = 0;
-out:
-        return ret;
-}
-
-inline gf_boolean_t
-stripe_fsetxattr_is_special (dict_t *dict)
-{
-        gf_boolean_t is_spl = _gf_false;
-
-        if (dict == NULL) {
-                goto out;
-        }
-
-        dict_foreach (dict, stripe_is_lockinfo, &is_spl);
-
-out:
-        return is_spl;
-}
-
-int
-stripe_fsetxattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                  dict_t *dict, int flags, dict_t *xdata)
-{
-        int32_t      op_ret = -1, ret = -1, op_errno = EINVAL;
-        gf_boolean_t is_spl = _gf_false;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        GF_IF_INTERNAL_XATTR_GOTO ("trusted.*stripe*", dict,
-                                   op_errno, err);
-
-        is_spl = stripe_fsetxattr_is_special (dict);
-        if (is_spl) {
-                ret = stripe_fsetxattr_to_everyone (frame, this, fd, dict,
-                                                    flags, xdata);
-                if (ret < 0) {
-                        op_errno = ENOMEM;
-                        goto err;
-                }
-
-                goto out;
-        }
-
-        STACK_WIND (frame, stripe_fsetxattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fsetxattr,
-                    fd, dict, flags, xdata);
-out:
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (fsetxattr, frame, op_ret, op_errno, NULL);
-        return 0;
-}
-
-int
-stripe_removexattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-        STRIPE_STACK_UNWIND (removexattr, frame, op_ret, op_errno, xdata);
-        return 0;
-}
-
-int
-stripe_removexattr (call_frame_t *frame, xlator_t *this,
-                    loc_t *loc, const char *name, dict_t *xdata)
-{
-        int32_t         op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (this, err);
-
-        GF_IF_NATIVE_XATTR_GOTO ("trusted.*stripe*",
-                                 name, op_errno, err);
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (loc, err);
-
-        STACK_WIND (frame, stripe_removexattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->removexattr,
-                    loc, name, xdata);
-        return 0;
-err:
-        STRIPE_STACK_UNWIND (removexattr, frame, -1,  op_errno, NULL);
-        return 0;
-}
-
-
-int
-stripe_fremovexattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-        STRIPE_STACK_UNWIND (fremovexattr, frame, op_ret, op_errno, xdata);
-        return 0;
-}
-
-int
-stripe_fremovexattr (call_frame_t *frame, xlator_t *this,
-                     fd_t *fd, const char *name, dict_t *xdata)
-{
-        int32_t         op_ret   = -1;
-        int32_t         op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        GF_IF_NATIVE_XATTR_GOTO ("trusted.*stripe*",
-                                 name, op_errno, err);
-
-        STACK_WIND (frame, stripe_fremovexattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fremovexattr,
-                    fd, name, xdata);
-        return 0;
- err:
-        STRIPE_STACK_UNWIND (fremovexattr, frame, op_ret, op_errno, xdata);
-        return 0;
-}
-
-int32_t
-stripe_readdirp_lookup_cbk (call_frame_t *frame, void *cookie,
-                            xlator_t *this, int op_ret, int op_errno,
-                            inode_t *inode, struct iatt *stbuf,
-                            dict_t *xattr, struct iatt *parent)
-{
-        stripe_local_t          *local          = NULL;
-        call_frame_t            *main_frame     = NULL;
-        stripe_local_t          *main_local     = NULL;
-        gf_dirent_t             *entry          = NULL;
-        call_frame_t            *prev           = NULL;
-        int                      done           = 0;
-
-        local = frame->local;
-        prev = cookie;
-
-        entry = local->dirent;
-
-        main_frame = local->orig_frame;
-        main_local = main_frame->local;
-        LOCK (&frame->lock);
-        {
-
-                local->call_count--;
-                if (!local->call_count)
-                        done = 1;
-                if (op_ret == -1) {
-                        local->op_errno = op_errno;
-                        local->op_ret = op_ret;
-                        goto unlock;
-                }
-
-		if (stripe_ctx_handle(this, prev, local, xattr))
-			gf_log(this->name, GF_LOG_ERROR,
-				"Error getting fctx info from dict.");
-
-		correct_file_size(stbuf, local->fctx, prev);
-
-                stripe_iatt_merge (stbuf, &entry->d_stat);
-                local->stbuf_blocks += stbuf->ia_blocks;
-        }
-unlock:
-        UNLOCK(&frame->lock);
-
-        if (done) {
-                inode_ctx_put (entry->inode, this,
-                               (uint64_t) (long)local->fctx);
-
-                done = 0;
-                LOCK (&main_frame->lock);
-                {
-                        main_local->wind_count--;
-                        if (!main_local->wind_count)
-                                done = 1;
-                        if (local->op_ret == -1) {
-                                main_local->op_errno = local->op_errno;
-                                main_local->op_ret = local->op_ret;
-                        }
-                        entry->d_stat.ia_blocks = local->stbuf_blocks;
-                }
-                UNLOCK (&main_frame->lock);
-                if (done) {
-                        main_frame->local = NULL;
-                        STRIPE_STACK_UNWIND (readdir, main_frame,
-                                             main_local->op_ret,
-                                             main_local->op_errno,
-                                             &main_local->entries, NULL);
-                        gf_dirent_free (&main_local->entries);
-                        stripe_local_wipe (main_local);
-                        mem_put (main_local);
-                }
-                frame->local = NULL;
-                stripe_local_wipe (local);
-                mem_put (local);
-                STRIPE_STACK_DESTROY (frame);
-        }
-
-        return 0;
-}
-
-int32_t
-stripe_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno,
-                     gf_dirent_t *orig_entries, dict_t *xdata)
-{
-        stripe_local_t *local = NULL;
-        call_frame_t   *prev = NULL;
-        gf_dirent_t    *local_entry = NULL;
-        gf_dirent_t    *tmp_entry = NULL;
-        xlator_list_t  *trav = NULL;
-        loc_t          loc = {0, };
-        int32_t        count = 0;
-        stripe_private_t *priv = NULL;
-        int32_t        subvols = 0;
-        dict_t         *xattrs = NULL;
-        call_frame_t   *local_frame = NULL;
-        stripe_local_t *local_ent = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-        prev  = cookie;
-        local = frame->local;
-        trav = this->children;
-        priv = this->private;
-
-        subvols = priv->child_count;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        local->op_ret = op_ret;
-                        goto unlock;
-                } else {
-                        local->op_ret = op_ret;
-                        list_splice_init (&orig_entries->list,
-                                          &local->entries.list);
-                        local->wind_count = op_ret;
-                }
-
-        }
-unlock:
-        UNLOCK (&frame->lock);
-
-        if (op_ret == -1)
-                goto out;
-
-        xattrs = dict_new ();
-        if (xattrs)
-                (void) stripe_xattr_request_build (this, xattrs, 0, 0, 0, 0);
-        count = op_ret;
-        list_for_each_entry_safe (local_entry, tmp_entry,
-                                  (&local->entries.list), list) {
-
-                if (!local_entry)
-                        break;
-                if (!IA_ISREG (local_entry->d_stat.ia_type)) {
-                        LOCK (&frame->lock);
-                        {
-                                local->wind_count--;
-                                count = local->wind_count;
-                        }
-                        UNLOCK (&frame->lock);
-                        continue;
-                }
-
-                local_frame = copy_frame (frame);
-
-                if (!local_frame) {
-                        op_errno = ENOMEM;
-                        op_ret = -1;
-                        goto out;
-                }
-
-                local_ent = mem_get0 (this->local_pool);
-                if (!local_ent) {
-                        op_errno = ENOMEM;
-                        op_ret = -1;
-                        goto out;
-                }
-
-                loc.inode = inode_ref (local_entry->inode);
-
-                uuid_copy (loc.gfid, local_entry->d_stat.ia_gfid);
-
-                local_ent->orig_frame = frame;
-
-                local_ent->call_count = subvols;
-
-                local_ent->dirent = local_entry;
-
-                local_frame->local = local_ent;
-
-                trav = this->children;
-                while (trav) {
-                        STACK_WIND (local_frame, stripe_readdirp_lookup_cbk,
-                                    trav->xlator, trav->xlator->fops->lookup,
-                                    &loc, xattrs);
-                        trav = trav->next;
-                }
-                loc_wipe (&loc);
-        }
-out:
-        if (!count) {
-                /* all entries are directories */
-                frame->local = NULL;
-                STRIPE_STACK_UNWIND (readdir, frame, local->op_ret,
-                                     local->op_errno, &local->entries, NULL);
-                gf_dirent_free (&local->entries);
-                stripe_local_wipe (local);
-                mem_put (local);
-        }
-        if (xattrs)
-                dict_unref (xattrs);
-        return 0;
-
-}
-int32_t
-stripe_readdirp (call_frame_t *frame, xlator_t *this,
-                 fd_t *fd, size_t size, off_t off, dict_t *xdata)
-{
-        stripe_local_t  *local  = NULL;
-        stripe_private_t *priv = NULL;
-        xlator_list_t   *trav = NULL;
-        int             op_errno = -1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        if (priv->first_child_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Initialization */
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-        frame->local = local;
-
-        local->fd = fd_ref (fd);
-
-        local->wind_count = 0;
-
-        local->count = 0;
-        local->op_ret = -1;
-        INIT_LIST_HEAD(&local->entries);
-
-        if (!trav)
-                goto err;
-
-        STACK_WIND (frame, stripe_readdirp_cbk, trav->xlator,
-                    trav->xlator->fops->readdirp, fd, size, off, xdata);
-        return 0;
-err:
-        op_errno = (op_errno == -1) ? errno : op_errno;
-        STRIPE_STACK_UNWIND (readdir, frame, -1, op_errno, NULL, NULL);
-
-        return 0;
-
-}
-
-int32_t
-mem_acct_init (xlator_t *this)
-{
-        int     ret = -1;
-
-        if (!this)
-                goto out;
-
-        ret = xlator_mem_acct_init (this, gf_stripe_mt_end + 1);
-
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
-                        "failed");
-                goto out;
-        }
-
-out:
-        return ret;
-}
-
-static int
-clear_pattern_list (stripe_private_t *priv)
-{
-        struct stripe_options *prev = NULL;
-        struct stripe_options *trav = NULL;
-        int                    ret = -1;
-
-        GF_VALIDATE_OR_GOTO ("stripe", priv, out);
-
-        trav = priv->pattern;
-        priv->pattern = NULL;
-        while (trav) {
-                prev = trav;
-                trav = trav->next;
-                GF_FREE (prev);
-        }
-
-        ret = 0;
- out:
-        return ret;
-
-
-}
-
-
-int
-reconfigure (xlator_t *this, dict_t *options)
-{
-
-        stripe_private_t *priv = NULL;
-        data_t           *data = NULL;
-        int               ret = -1;
-        volume_option_t  *opt = NULL;
-
-        GF_ASSERT (this);
-        GF_ASSERT (this->private);
-
-        priv = this->private;
-
-
-        ret = 0;
-        LOCK (&priv->lock);
-        {
-                ret = clear_pattern_list (priv);
-                if (ret)
-                         goto unlock;
-
-                data = dict_get (options, "block-size");
-                if (data) {
-                        ret = set_stripe_block_size (this, priv, data->data);
-                        if (ret)
-                                goto unlock;
-                } else {
-                        opt = xlator_volume_option_get (this, "block-size");
-                        if (!opt) {
-                                gf_log (this->name, GF_LOG_WARNING,
-                                        "option 'block-size' not found");
-                                ret = -1;
-                                goto unlock;
-                        }
-
-                        if (gf_string2bytesize (opt->default_value, &priv->block_size)){
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "Unable to set default block-size ");
-                                ret = -1;
-                                goto unlock;
-                        }
-                }
-
-		GF_OPTION_RECONF("coalesce", priv->coalesce, options, bool,
-				unlock);
-        }
- unlock:
-        UNLOCK (&priv->lock);
-        if (ret)
-                goto out;
-
-        ret = 0;
- out:
-        return ret;
-
-}
-
-/**
- * init - This function is called when xlator-graph gets initialized.
- *     The option given in volfiles are parsed here.
- * @this -
- */
-int32_t
-init (xlator_t *this)
-{
-        stripe_private_t *priv = NULL;
-        volume_option_t  *opt = NULL;
-        xlator_list_t    *trav = NULL;
-        data_t           *data = NULL;
-        int32_t           count = 0;
-        int               ret = -1;
-
-        if (!this)
-                goto out;
-
-        trav = this->children;
-        while (trav) {
-                count++;
-                trav = trav->next;
-        }
-
-        if (!count) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "stripe configured without \"subvolumes\" option. "
-                        "exiting");
-                goto out;
-        }
-
-        if (!this->parents) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "dangling volume. check volfile ");
-        }
-
-        if (count == 1) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "stripe configured with only one \"subvolumes\" option."
-                        " please check the volume. exiting");
-                goto out;
-        }
-
-        priv = GF_CALLOC (1, sizeof (stripe_private_t),
-                          gf_stripe_mt_stripe_private_t);
-
-        if (!priv)
-                goto out;
-        priv->xl_array = GF_CALLOC (count, sizeof (xlator_t *),
-                                    gf_stripe_mt_xlator_t);
-        if (!priv->xl_array)
-                goto out;
-
-        priv->last_event = GF_CALLOC (count, sizeof (int),
-                                      gf_stripe_mt_int32_t);
-        if (!priv->last_event)
-                goto out;
-
-        priv->child_count = count;
-        LOCK_INIT (&priv->lock);
-
-        trav = this->children;
-        count = 0;
-        while (trav) {
-                priv->xl_array[count++] = trav->xlator;
-                trav = trav->next;
-        }
-
-        if (count > 256) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "maximum number of stripe subvolumes supported "
-                        "is 256");
-                goto out;
-        }
-
-        ret = 0;
-        LOCK (&priv->lock);
-        {
-                opt = xlator_volume_option_get (this, "block-size");
-                if (!opt) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "option 'block-size' not found");
-                        ret = -1;
-                        goto unlock;
-                }
-                if (gf_string2bytesize (opt->default_value, &priv->block_size)){
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Unable to set default block-size ");
-                        ret = -1;
-                        goto unlock;
-                }
-                /* option stripe-pattern *avi:1GB,*pdf:16K */
-                data = dict_get (this->options, "block-size");
-                if (data) {
-                        ret = set_stripe_block_size (this, priv, data->data);
-                        if (ret)
-                                goto unlock;
-                }
-        }
- unlock:
-        UNLOCK (&priv->lock);
-        if (ret)
-                goto out;
-
-        GF_OPTION_INIT ("use-xattr", priv->xattr_supported, bool, out);
-        /* notify related */
-        priv->nodes_down = priv->child_count;
-
-	GF_OPTION_INIT("coalesce", priv->coalesce, bool, out);
-
-        this->local_pool = mem_pool_new (stripe_local_t, 128);
-        if (!this->local_pool) {
-                ret = -1;
-                gf_log (this->name, GF_LOG_ERROR,
-                        "failed to create local_t's memory pool");
-                goto out;
-        }
-
-        this->private = priv;
-
-        ret = 0;
-out:
-        if (ret) {
-                if (priv) {
-                        GF_FREE (priv->xl_array);
-                        GF_FREE (priv);
-                }
-        }
-        return ret;
-}
-
-/**
- * fini -   Free all the private variables
- * @this -
- */
-void
-fini (xlator_t *this)
-{
-        stripe_private_t      *priv = NULL;
-        struct stripe_options *prev = NULL;
-        struct stripe_options *trav = NULL;
-
-        if (!this)
-                goto out;
-
-        priv = this->private;
-        if (priv) {
-                this->private = NULL;
-                GF_FREE (priv->xl_array);
-
-                trav = priv->pattern;
-                while (trav) {
-                        prev = trav;
-                        trav = trav->next;
-                        GF_FREE (prev);
-                }
-                GF_FREE (priv->last_event);
-                LOCK_DESTROY (&priv->lock);
-                GF_FREE (priv);
-        }
-
-out:
-        return;
-}
-
-int32_t
-stripe_getxattr_unwind (call_frame_t *frame,
-                        int op_ret, int op_errno, dict_t *dict, dict_t *xdata)
-
-{
-        STRIPE_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict, xdata);
-        return 0;
-}
-
-int
-stripe_internal_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                              int op_ret, int op_errno, dict_t *xattr,
-                              dict_t *xdata)
-{
-
-        char        size_key[256]  = {0,};
-        char        index_key[256] = {0,};
-        char        count_key[256] = {0,};
-        char        coalesce_key[256] = {0,};
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (frame->local, out);
-
-        if (!xattr || (op_ret == -1))
-            goto out;
-
-        sprintf (size_key, "trusted.%s.stripe-size", this->name);
-        sprintf (count_key, "trusted.%s.stripe-count", this->name);
-        sprintf (index_key, "trusted.%s.stripe-index", this->name);
-	sprintf (coalesce_key, "trusted.%s.stripe-coalesce", this->name);
-
-        dict_del (xattr, size_key);
-        dict_del (xattr, count_key);
-        dict_del (xattr, index_key);
-        dict_del (xattr, coalesce_key);
-
-out:
-        STRIPE_STACK_UNWIND (getxattr, frame, op_ret, op_errno, xattr, xdata);
-
-        return 0;
-
-}
-
-int
-stripe_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
-{
-        int                     call_cnt = 0;
-        stripe_local_t         *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (frame->local, out);
-
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                call_cnt = --local->wind_count;
-        }
-        UNLOCK (&frame->lock);
-
-        if (!xattr || (op_ret < 0))
-                goto out;
-
-        local->op_ret = 0;
-
-        if (!local->xattr) {
-                local->xattr = dict_ref (xattr);
-        } else {
-                stripe_aggregate_xattr (local->xattr, xattr);
-        }
-
-out:
-        if (!call_cnt) {
-                STRIPE_STACK_UNWIND (getxattr, frame, local->op_ret, op_errno,
-                                     local->xattr, xdata);
-        }
-
-        return 0;
-}
-
-int32_t
-stripe_vgetxattr_cbk (call_frame_t *frame, void *cookie,
-                      xlator_t *this, int32_t op_ret, int32_t op_errno,
-                      dict_t *dict, dict_t *xdata)
-{
-        stripe_local_t      *local         = NULL;
-        int32_t              callcnt       = 0;
-        int32_t              ret           = -1;
-        long                 cky           = 0;
-        void                *xattr_val     = NULL;
-        void                *xattr_serz    = NULL;
-        stripe_xattr_sort_t *xattr         = NULL;
-        dict_t              *stripe_xattr  = NULL;
-
-        if (!frame || !frame->local || !this) {
-                gf_log ("", GF_LOG_ERROR, "Possible NULL deref");
-                return ret;
-        }
-
-        local = frame->local;
-        cky = (long) cookie;
-
-        if (local->xsel[0] == '\0') {
-                gf_log (this->name, GF_LOG_ERROR, "Empty xattr in cbk");
-                return ret;
-        }
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->wind_count;
-
-                if (!dict || (op_ret < 0))
-                        goto out;
-
-                if (!local->xattr_list)
-                        local->xattr_list = (stripe_xattr_sort_t *)
-                                GF_CALLOC (local->nallocs,
-                                           sizeof (stripe_xattr_sort_t),
-                                           gf_stripe_mt_xattr_sort_t);
-
-                if (local->xattr_list) {
-                        xattr = local->xattr_list + (int32_t) cky;
-
-                        ret = dict_get_ptr_and_len (dict, local->xsel,
-                                                    &xattr_val,
-                                                    &xattr->xattr_len);
-                        if (xattr->xattr_len == 0)
-                                goto out;
-
-                        xattr->pos = cky;
-                        xattr->xattr_value = gf_memdup (xattr_val,
-                                                        xattr->xattr_len);
-
-                        if (xattr->xattr_value != NULL)
-                                local->xattr_total_len += xattr->xattr_len + 1;
-                }
-        }
- out:
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (!local->xattr_total_len)
-                        goto unwind;
-
-                stripe_xattr = dict_new ();
-                if (!stripe_xattr)
-                        goto unwind;
-
-                /* select filler based on ->xsel */
-                if (XATTR_IS_PATHINFO (local->xsel))
-                        ret = stripe_fill_pathinfo_xattr (this, local,
-                                                          (char **)&xattr_serz);
-                else if (XATTR_IS_LOCKINFO (local->xsel)) {
-                        ret = stripe_fill_lockinfo_xattr (this, local,
-                                                          &xattr_serz);
-                } else {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Unknown xattr in xattr request");
-                        goto unwind;
-                }
-
-                if (!ret) {
-                        ret = dict_set_dynptr (stripe_xattr, local->xsel,
-                                               xattr_serz,
-                                               local->xattr_total_len);
-                        if (ret)
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "Can't set %s key in dict",
-                                        local->xsel);
-                }
-
-        unwind:
-                STRIPE_STACK_UNWIND (getxattr, frame, op_ret, op_errno,
-                                     stripe_xattr, NULL);
-
-                ret = stripe_free_xattr_str (local);
-
-                GF_FREE (local->xattr_list);
-
-                if (stripe_xattr)
-                        dict_unref (stripe_xattr);
-        }
-
-        return ret;
-}
-
-int32_t
-stripe_getxattr (call_frame_t *frame, xlator_t *this,
-                 loc_t *loc, const char *name, dict_t *xdata)
-{
-        stripe_local_t    *local    = NULL;
-        xlator_list_t     *trav     = NULL;
-        stripe_private_t  *priv     = NULL;
-        int32_t            op_errno = EINVAL;
-        int                i        = 0;
-        xlator_t         **sub_volumes;
-        int                ret      = 0;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        /* Initialization */
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        loc_copy (&local->loc, loc);
-
-
-        if (name && (strcmp (GF_XATTR_MARKER_KEY, name) == 0)
-            && (GF_CLIENT_PID_GSYNCD == frame->root->pid)) {
-                local->marker.call_count = priv->child_count;
-
-                sub_volumes = alloca ( priv->child_count *
-                                       sizeof (xlator_t *));
-                for (i = 0, trav = this->children; trav ;
-                     trav = trav->next, i++) {
-
-                        *(sub_volumes + i)  = trav->xlator;
-
-                }
-
-                if (cluster_getmarkerattr (frame, this, loc, name,
-                                           local, stripe_getxattr_unwind,
-                                           sub_volumes, priv->child_count,
-                                           MARKER_UUID_TYPE, marker_uuid_default_gauge,
-                                           priv->vol_uuid)) {
-                        op_errno = EINVAL;
-                        goto err;
-                }
-
-                return 0;
-        }
-
-        if (name && strncmp (name, GF_XATTR_QUOTA_SIZE_KEY,
-                             strlen (GF_XATTR_QUOTA_SIZE_KEY)) == 0) {
-                local->wind_count = priv->child_count;
-
-                for (i = 0, trav=this->children; i < priv->child_count; i++,
-                             trav = trav->next) {
-                        STACK_WIND (frame, stripe_getxattr_cbk,
-                                    trav->xlator, trav->xlator->fops->getxattr,
-                                    loc, name, xdata);
-                }
-
-                return 0;
-        }
-
-        if (name &&
-            ((strncmp (name, GF_XATTR_PATHINFO_KEY,
-                       strlen (GF_XATTR_PATHINFO_KEY)) == 0))) {
-                if (IA_ISREG (loc->inode->ia_type)) {
-                        ret = inode_ctx_get (loc->inode, this,
-                                             (uint64_t *) &local->fctx);
-                        if (ret)
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "stripe size unavailable from fctx"
-                                        " relying on pathinfo could lead to"
-                                        " wrong results");
-                }
-
-                local->nallocs = local->wind_count = priv->child_count;
-                (void) strncpy (local->xsel, name, strlen (name));
-
-                /**
-                 * for xattrs that need info from all childs, fill ->xsel
-                 * as above and call the filler function in cbk based on
-                 * it
-                 */
-                for (i = 0, trav = this->children; i < priv->child_count; i++,
-                     trav = trav->next) {
-                        STACK_WIND_COOKIE (frame, stripe_vgetxattr_cbk,
-                                           (void *) (long) i, trav->xlator,
-                                           trav->xlator->fops->getxattr,
-                                           loc, name, xdata);
-                }
-
-                return 0;
-        }
-
-        if (name &&(*priv->vol_uuid)) {
-                if ((match_uuid_local (name, priv->vol_uuid) == 0)
-                    && (GF_CLIENT_PID_GSYNCD == frame->root->pid)) {
-
-                        if (!IA_FILE_OR_DIR (loc->inode->ia_type))
-                                local->marker.call_count = 1;
-                        else
-                                local->marker.call_count = priv->child_count;
-
-                        sub_volumes = alloca (local->marker.call_count *
-                                              sizeof (xlator_t *));
-
-                        for (i = 0, trav = this->children;
-                             i < local->marker.call_count;
-                             i++, trav = trav->next) {
-                                *(sub_volumes + i) = trav->xlator;
-
-                        }
-
-                        if (cluster_getmarkerattr (frame, this, loc, name,
-                                                   local,
-                                                   stripe_getxattr_unwind,
-                                                   sub_volumes,
-                                                   local->marker.call_count,
-                                                   MARKER_XTIME_TYPE,
-                                                   marker_xtime_default_gauge,
-                                                   priv->vol_uuid)) {
-                                op_errno = EINVAL;
-                                goto err;
-                        }
-
-                        return 0;
-                }
-        }
-
-
-        STACK_WIND (frame, stripe_internal_getxattr_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->getxattr, loc, name, xdata);
-
-        return 0;
-
-err:
-        STRIPE_STACK_UNWIND (getxattr, frame, -1, op_errno, NULL, NULL);
-        return 0;
-}
-
-inline gf_boolean_t
-stripe_is_special_xattr (const char *name)
-{
-        gf_boolean_t    is_spl = _gf_false;
-
-        if (!name) {
-                goto out;
-        }
-
-        if (!strncmp (name, GF_XATTR_LOCKINFO_KEY,
-                      strlen (GF_XATTR_LOCKINFO_KEY))
-            || !strncmp (name, GF_XATTR_PATHINFO_KEY,
-                         strlen (GF_XATTR_PATHINFO_KEY)))
-                is_spl = _gf_true;
-out:
-        return is_spl;
-}
-
-int32_t
-stripe_fgetxattr_from_everyone (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                                const char *name, dict_t *xdata)
-{
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv  = NULL;
-        int32_t           ret   = -1, op_errno = 0;
-        int               i     = 0;
-        xlator_list_t    *trav  = NULL;
-
-        priv = this->private;
-
-        local = mem_get0 (this->local_pool);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-        local->op_ret = -1;
-        frame->local = local;
-
-        strncpy (local->xsel, name, strlen (name));
-        local->nallocs = local->wind_count = priv->child_count;
-
-        for (i = 0, trav = this->children; i < priv->child_count; i++,
-                     trav = trav->next) {
-                STACK_WIND_COOKIE (frame, stripe_vgetxattr_cbk,
-                                   (void *) (long) i, trav->xlator,
-                                   trav->xlator->fops->fgetxattr,
-                                   fd, name, xdata);
-        }
-
-        return 0;
-
-err:
-        STACK_UNWIND_STRICT (fgetxattr, frame, -1, op_errno, NULL, NULL);
-        return ret;
-}
-
-int32_t
-stripe_fgetxattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                  const char *name, dict_t *xdata)
-{
-        if (stripe_is_special_xattr (name)) {
-                stripe_fgetxattr_from_everyone (frame, this, fd, name, xdata);
-                goto out;
-        }
-
-        STACK_WIND (frame, stripe_internal_getxattr_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fgetxattr, fd, name, xdata);
-
-out:
-        return 0;
-}
-
-
-
-int32_t
-stripe_priv_dump (xlator_t *this)
-{
-        char                    key[GF_DUMP_MAX_BUF_LEN];
-        int                     i = 0;
-        stripe_private_t       *priv = NULL;
-        int                     ret = -1;
-        struct stripe_options  *options = NULL;
-
-        GF_VALIDATE_OR_GOTO ("stripe", this, out);
-
-        priv = this->private;
-        if (!priv)
-                goto out;
-
-        ret = TRY_LOCK (&priv->lock);
-        if (ret != 0)
-                goto out;
-
-        gf_proc_dump_add_section("xlator.cluster.stripe.%s.priv", this->name);
-        gf_proc_dump_write("child_count","%d", priv->child_count);
-
-        for (i = 0; i < priv->child_count; i++) {
-                sprintf (key, "subvolumes[%d]", i);
-                gf_proc_dump_write (key, "%s.%s", priv->xl_array[i]->type,
-                                    priv->xl_array[i]->name);
-        }
-
-        options = priv->pattern;
-        while (options != NULL) {
-                gf_proc_dump_write ("path_pattern", "%s", priv->pattern->path_pattern);
-                gf_proc_dump_write ("options_block_size", "%ul", options->block_size);
-
-                options = options->next;
-        }
-
-        gf_proc_dump_write ("block_size", "%ul", priv->block_size);
-        gf_proc_dump_write ("nodes-down", "%d", priv->nodes_down);
-        gf_proc_dump_write ("first-child_down", "%d", priv->first_child_down);
-        gf_proc_dump_write ("xattr_supported", "%d", priv->xattr_supported);
-
-        UNLOCK (&priv->lock);
-
-out:
-        return ret;
-}
-
-struct xlator_fops fops = {
-        .stat           = stripe_stat,
-        .unlink         = stripe_unlink,
-        .rename         = stripe_rename,
-        .link           = stripe_link,
-        .truncate       = stripe_truncate,
-        .create         = stripe_create,
-        .open           = stripe_open,
-        .readv          = stripe_readv,
-        .writev         = stripe_writev,
-        .statfs         = stripe_statfs,
-        .flush          = stripe_flush,
-        .fsync          = stripe_fsync,
-        .ftruncate      = stripe_ftruncate,
-        .fstat          = stripe_fstat,
-        .mkdir          = stripe_mkdir,
-        .rmdir          = stripe_rmdir,
-        .lk             = stripe_lk,
-        .opendir        = stripe_opendir,
-        .fsyncdir       = stripe_fsyncdir,
-        .setattr        = stripe_setattr,
-        .fsetattr       = stripe_fsetattr,
-        .lookup         = stripe_lookup,
-        .mknod          = stripe_mknod,
-        .setxattr       = stripe_setxattr,
-        .fsetxattr      = stripe_fsetxattr,
-        .getxattr       = stripe_getxattr,
-        .fgetxattr      = stripe_fgetxattr,
-        .removexattr    = stripe_removexattr,
-        .fremovexattr   = stripe_fremovexattr,
-        .readdirp       = stripe_readdirp,
-	.fallocate	= stripe_fallocate,
-	.discard	= stripe_discard,
-};
-
-struct xlator_cbks cbks = {
-        .release = stripe_release,
-        .forget  = stripe_forget,
-};
-
-struct xlator_dumpops dumpops = {
-        .priv = stripe_priv_dump,
-};
-
-struct volume_options options[] = {
-        { .key  = {"block-size"},
-          .type = GF_OPTION_TYPE_SIZE_LIST,
-          .default_value = "128KB",
-          .min = STRIPE_MIN_BLOCK_SIZE,
-          .description = "Size of the stripe unit that would be read "
-                         "from or written to the striped servers."
-        },
-        { .key  = {"use-xattr"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "true"
-        },
-	{ .key = {"coalesce"},
-	  .type = GF_OPTION_TYPE_BOOL,
-	  .default_value = "true",
-	  .description = "Enable/Disable coalesce mode to flatten striped "
-			 "files as stored on the server (i.e., eliminate holes "
-			 "caused by the traditional format)."
-	},
-        { .key  = {NULL} },
-};
diff --git a/xlators/cluster/stripe/src/stripe.h b/xlators/cluster/stripe/src/stripe.h
deleted file mode 100644
index 5673d18f390..00000000000
--- a/xlators/cluster/stripe/src/stripe.h
+++ /dev/null
@@ -1,288 +0,0 @@
-/*
-  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
-  This file is part of GlusterFS.
-
-  This file is licensed to you under your choice of the GNU Lesser
-  General Public License, version 3 or any later version (LGPLv3 or
-  later), or the GNU General Public License, version 2 (GPLv2), in all
-  cases as published by the Free Software Foundation.
-*/
-
-
-#ifndef _STRIPE_H_
-#define _STRIPE_H_
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "xlator.h"
-#include "logging.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat.h"
-#include "compat-errno.h"
-#include "stripe-mem-types.h"
-#include "libxlator.h"
-#include <fnmatch.h>
-#include <signal.h>
-
-#define STRIPE_PATHINFO_HEADER "STRIPE:"
-#define STRIPE_MIN_BLOCK_SIZE  (16*GF_UNIT_KB)
-
-#define STRIPE_STACK_UNWIND(fop, frame, params ...) do {           \
-                stripe_local_t *__local = NULL;                    \
-                if (frame) {                                       \
-                        __local = frame->local;                    \
-                        frame->local = NULL;                       \
-                }                                                  \
-                STACK_UNWIND_STRICT (fop, frame, params);          \
-                if (__local) {                                     \
-                        stripe_local_wipe(__local);                \
-                        mem_put (__local);       \
-                }                                                  \
-        } while (0)
-
-#define STRIPE_STACK_DESTROY(frame) do {                        \
-                stripe_local_t *__local = NULL;                 \
-                __local = frame->local;                         \
-                frame->local = NULL;                            \
-                STACK_DESTROY (frame->root);                    \
-                if (__local) {                                  \
-                        stripe_local_wipe (__local);            \
-                        mem_put (__local);    \
-                }                                               \
-        } while (0)
-
-#define STRIPE_VALIDATE_FCTX(fctx, label) do {                  \
-        int     idx = 0;                                        \
-        if (!fctx) {                                            \
-                op_errno = EINVAL;                              \
-                goto label;                                     \
-        }                                                       \
-        for (idx = 0; idx < fctx->stripe_count; idx++) {        \
-                if (!fctx->xl_array[idx]) {                     \
-                        gf_log (this->name, GF_LOG_ERROR,       \
-                                "fctx->xl_array[%d] is NULL",   \
-                                idx);                           \
-                        op_errno = ESTALE;                      \
-                        goto label;                             \
-                }                                               \
-        }                                                       \
-       } while (0)
-
-typedef struct stripe_xattr_sort {
-        int   pos;
-        int   xattr_len;
-        char *xattr_value;
-} stripe_xattr_sort_t;
-
-/**
- * struct stripe_options : This keeps the pattern and the block-size
- *     information, which is used for striping on a file.
- */
-struct stripe_options {
-        struct stripe_options *next;
-        char                   path_pattern[256];
-        uint64_t               block_size;
-};
-
-/**
- * Private structure for stripe translator
- */
-struct stripe_private {
-        struct stripe_options  *pattern;
-        xlator_t              **xl_array;
-        uint64_t                block_size;
-        gf_lock_t               lock;
-        uint8_t                 nodes_down;
-        int8_t                  first_child_down;
-        int                    *last_event;
-        int8_t                  child_count;
-        gf_boolean_t            xattr_supported;  /* default yes */
-	gf_boolean_t		coalesce;
-        char                    vol_uuid[UUID_SIZE + 1];
-};
-
-/**
- * Used to keep info about the replies received from readv/writev calls
- */
-struct stripe_replies {
-        struct iovec *vector;
-        int32_t       count;    //count of vector
-        int32_t       op_ret;   //op_ret of readv
-        int32_t       op_errno;
-        int32_t       requested_size;
-        struct iatt   stbuf;    /* 'stbuf' is also a part of reply */
-};
-
-typedef struct _stripe_fd_ctx {
-        off_t      stripe_size;
-        int        stripe_count;
-	int	   stripe_coalesce;
-        int        static_array;
-        xlator_t **xl_array;
-} stripe_fd_ctx_t;
-
-
-/**
- * Local structure to be passed with all the frames in case of STACK_WIND
- */
-struct stripe_local; /* this itself is used inside the structure; */
-
-struct stripe_local {
-        struct stripe_local *next;
-        call_frame_t        *orig_frame;
-
-        stripe_fd_ctx_t     *fctx;
-
-        /* Used by _cbk functions */
-        struct iatt          stbuf;
-        struct iatt          pre_buf;
-        struct iatt          post_buf;
-        struct iatt          preparent;
-        struct iatt          postparent;
-
-        off_t                stbuf_size;
-        off_t                prebuf_size;
-        off_t                postbuf_size;
-        off_t                preparent_size;
-        off_t                postparent_size;
-
-        blkcnt_t             stbuf_blocks;
-        blkcnt_t             prebuf_blocks;
-        blkcnt_t             postbuf_blocks;
-        blkcnt_t             preparent_blocks;
-        blkcnt_t             postparent_blocks;
-
-        struct stripe_replies *replies;
-        struct statvfs        statvfs_buf;
-        dir_entry_t          *entry;
-
-        int8_t               revalidate;
-        int8_t               failed;
-        int8_t               unwind;
-
-        size_t               readv_size;
-        int32_t              entry_count;
-        int32_t              node_index;
-        int32_t              call_count;
-        int32_t              wind_count; /* used instead of child_cound
-                                            in case of read and write */
-        int32_t              op_ret;
-        int32_t              op_errno;
-        int32_t              count;
-        int32_t              flags;
-        char                *name;
-        inode_t             *inode;
-
-        loc_t                loc;
-        loc_t                loc2;
-
-        mode_t               mode;
-        dev_t                rdev;
-        /* For File I/O fops */
-        dict_t              *xdata;
-
-        stripe_xattr_sort_t *xattr_list;
-        int32_t              xattr_total_len;
-        int32_t              nallocs;
-        char xsel[256];
-
-        struct marker_str    marker;
-
-        /* General usage */
-        off_t                offset;
-        off_t                stripe_size;
-
-        int xattr_self_heal_needed;
-        int entry_self_heal_needed;
-
-        int8_t              *list;
-        struct gf_flock         lock;
-        fd_t                *fd;
-        void                *value;
-        struct iobref       *iobref;
-        gf_dirent_t          entries;
-        gf_dirent_t         *dirent;
-        dict_t              *xattr;
-        uuid_t               ia_gfid;
-
-        int                  xflag;
-        mode_t               umask;
-};
-
-typedef struct stripe_local   stripe_local_t;
-typedef struct stripe_private stripe_private_t;
-
-/*
- * Determine the stripe index of a particular frame based on the translator.
- */
-static inline int32_t stripe_get_frame_index(stripe_fd_ctx_t *fctx,
-					     call_frame_t *prev)
-{
-	int32_t i, idx = -1;
-
-	for (i = 0; i < fctx->stripe_count; i++) {
-		if (fctx->xl_array[i] == prev->this) {
-			idx = i;
-			break;
-		}
-	}
-
-	return idx;
-}
-
-static inline void stripe_copy_xl_array(xlator_t **dst, xlator_t **src,
-					int count)
-{
-	int i;
-
-	for (i = 0; i < count; i++)
-		dst[i] = src[i];
-}
-
-void stripe_local_wipe (stripe_local_t *local);
-int32_t stripe_ctx_handle (xlator_t *this, call_frame_t *prev,
-                           stripe_local_t *local, dict_t *dict);
-void stripe_aggregate_xattr (dict_t *dst, dict_t *src);
-int32_t stripe_xattr_request_build (xlator_t *this, dict_t *dict,
-                                    uint64_t stripe_size, uint32_t stripe_count,
-                                    uint32_t stripe_index,
-				    uint32_t stripe_coalesce);
-int32_t stripe_get_matching_bs (const char *path, stripe_private_t *priv);
-int set_stripe_block_size (xlator_t *this, stripe_private_t *priv, char *data);
-int32_t stripe_iatt_merge (struct iatt *from, struct iatt *to);
-int32_t stripe_fill_pathinfo_xattr (xlator_t *this, stripe_local_t *local,
-                                    char **xattr_serz);
-int32_t stripe_free_xattr_str (stripe_local_t *local);
-int32_t stripe_xattr_aggregate (char *buffer, stripe_local_t *local,
-                                int32_t *total);
-off_t coalesced_offset(off_t offset, uint64_t stripe_size, int stripe_count);
-off_t uncoalesced_size(off_t size, uint64_t stripe_size, int stripe_count,
-			int stripe_index);
-int32_t
-stripe_fill_lockinfo_xattr (xlator_t *this, stripe_local_t *local,
-                            void **xattr_serz);
-
-/*
- * Adjust the size attribute for files if coalesce is enabled.
- */
-static inline void correct_file_size(struct iatt *buf, stripe_fd_ctx_t *fctx,
-	call_frame_t *prev)
-{
-	int index;
-
-	if (!IA_ISREG(buf->ia_type))
-		return;
-
-	if (!fctx || !fctx->stripe_coalesce)
-		return;
-
-	index = stripe_get_frame_index(fctx, prev);
-	buf->ia_size = uncoalesced_size(buf->ia_size, fctx->stripe_size,
-		fctx->stripe_count, index);
-}
-
-#endif /* _STRIPE_H_ */
diff --git a/xlators/debug/Makefile.am b/xlators/debug/Makefile.am
index b655554efec..88fac1c6d9e 100644
--- a/xlators/debug/Makefile.am
+++ b/xlators/debug/Makefile.am
@@ -1,3 +1,3 @@
-SUBDIRS = trace error-gen io-stats
+SUBDIRS = error-gen io-stats sink trace delay-gen
 
 CLEANFILES = 
diff --git a/xlators/features/glupy/Makefile.am b/xlators/debug/delay-gen/Makefile.am
index a985f42a877..a985f42a877 100644
--- a/xlators/features/glupy/Makefile.am
+++ b/xlators/debug/delay-gen/Makefile.am
diff --git a/xlators/debug/delay-gen/src/Makefile.am b/xlators/debug/delay-gen/src/Makefile.am
new file mode 100644
index 00000000000..8f758dec199
--- /dev/null
+++ b/xlators/debug/delay-gen/src/Makefile.am
@@ -0,0 +1,11 @@
+
+xlator_LTLIBRARIES = delay-gen.la
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/debug
+delay_gen_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+delay_gen_la_SOURCES = delay-gen.c
+delay_gen_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+noinst_HEADERS = delay-gen.h delay-gen-mem-types.h delay-gen-messages.h
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	   -I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
+AM_CFLAGS = -Wall -fno-strict-aliasing $(GF_CFLAGS)
+CLEANFILES =
diff --git a/xlators/debug/delay-gen/src/delay-gen-mem-types.h b/xlators/debug/delay-gen/src/delay-gen-mem-types.h
new file mode 100644
index 00000000000..c89a9217193
--- /dev/null
+++ b/xlators/debug/delay-gen/src/delay-gen-mem-types.h
@@ -0,0 +1,21 @@
+/*
+ *   Copyright (c) 2017 Red Hat, Inc. <http://www.redhat.com>
+ *   This file is part of GlusterFS.
+ *
+ *   This file is licensed to you under your choice of the GNU Lesser
+ *   General Public License, version 3 or any later version (LGPLv3 or
+ *   later), or the GNU General Public License, version 2 (GPLv2), in all
+ *   cases as published by the Free Software Foundation.
+ */
+
+#ifndef __DELAY_GEN_MEM_TYPES_H__
+#define __DELAY_GEN_MEM_TYPES_H__
+
+#include <glusterfs/mem-types.h>
+
+enum gf_delay_gen_mem_types_ {
+    gf_delay_gen_mt_dg_t = gf_common_mt_end + 1,
+    gf_delay_gen_mt_end
+};
+
+#endif /* __DELAY_GEN_MEM_TYPES_H__ */
diff --git a/xlators/debug/delay-gen/src/delay-gen-messages.h b/xlators/debug/delay-gen/src/delay-gen-messages.h
new file mode 100644
index 00000000000..bc98cec2885
--- /dev/null
+++ b/xlators/debug/delay-gen/src/delay-gen-messages.h
@@ -0,0 +1,26 @@
+/*
+ *   Copyright (c) 2017 Red Hat, Inc. <http://www.redhat.com>
+ *   This file is part of GlusterFS.
+ *
+ *   This file is licensed to you under your choice of the GNU Lesser
+ *   General Public License, version 3 or any later version (LGPLv3 or
+ *   later), or the GNU General Public License, version 2 (GPLv2), in all
+ *   cases as published by the Free Software Foundation.
+ */
+
+#ifndef __DELAY_GEN_MESSAGES_H__
+#define __DELAY_GEN_MESSAGES_H__
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+#endif /* __DELAY_GEN_MESSAGES_H__ */
diff --git a/xlators/debug/delay-gen/src/delay-gen.c b/xlators/debug/delay-gen/src/delay-gen.c
new file mode 100644
index 00000000000..4698f1fd785
--- /dev/null
+++ b/xlators/debug/delay-gen/src/delay-gen.c
@@ -0,0 +1,697 @@
+/*
+ *  Copyright (c) 2017 Red Hat, Inc. <http://www.redhat.com>
+ *  This file is part of GlusterFS.
+ *
+ *  This file is licensed to you under your choice of the GNU Lesser
+ *  General Public License, version 3 or any later version (LGPLv3 or
+ *  later), or the GNU General Public License, version 2 (GPLv2), in all
+ *  cases as published by the Free Software Foundation.
+ */
+
+#include "delay-gen.h"
+
+#define DELAY_GRANULARITY (1 << 20)
+
+#define DG_FOP(fop, name, frame, this, args...)                                \
+    do {                                                                       \
+        delay_gen(this, fop);                                                  \
+        default_##name(frame, this, args);                                     \
+    } while (0)
+
+int
+delay_gen(xlator_t *this, int fop)
+{
+    dg_t *dg = this->private;
+
+    if (!dg->enable[fop] || !dg->delay_ppm)
+        return 0;
+
+    if ((rand() % DELAY_GRANULARITY) < dg->delay_ppm)
+        gf_nanosleep(dg->delay_duration * GF_US_IN_NS);
+
+    return 0;
+}
+
+int32_t
+dg_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+          dict_t *xdata)
+{
+    DG_FOP(GF_FOP_RENAME, rename, frame, this, oldloc, newloc, xdata);
+    return 0;
+}
+
+int32_t
+dg_ipc(call_frame_t *frame, xlator_t *this, int32_t op, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_IPC, ipc, frame, this, op, xdata);
+    return 0;
+}
+
+int32_t
+dg_setactivelk(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               lock_migration_info_t *locklist, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_SETACTIVELK, setactivelk, frame, this, loc, locklist, xdata);
+    return 0;
+}
+
+int32_t
+dg_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FLUSH, flush, frame, this, fd, xdata);
+    return 0;
+}
+
+int32_t
+dg_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+           off_t off, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_READDIR, readdir, frame, this, fd, size, off, xdata);
+    return 0;
+}
+
+int32_t
+dg_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+            int32_t flags, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_SETXATTR, setxattr, frame, this, loc, dict, flags, xdata);
+    return 0;
+}
+
+int32_t
+dg_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+         dev_t rdev, mode_t umask, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_MKNOD, mknod, frame, this, loc, mode, rdev, umask, xdata);
+    return 0;
+}
+
+int32_t
+dg_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+             int32_t flags, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FSETXATTR, fsetxattr, frame, this, fd, dict, flags, xdata);
+    return 0;
+}
+
+int32_t
+dg_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+         off_t offset, uint32_t flags, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_READ, readv, frame, this, fd, size, offset, flags, xdata);
+    return 0;
+}
+
+int32_t
+dg_inodelk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+           int32_t cmd, struct gf_flock *lock, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_INODELK, inodelk, frame, this, volume, loc, cmd, lock, xdata);
+    return 0;
+}
+
+int32_t
+dg_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+                dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FREMOVEXATTR, fremovexattr, frame, this, fd, name, xdata);
+    return 0;
+}
+
+int32_t
+dg_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+        fd_t *fd, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_OPEN, open, frame, this, loc, flags, fd, xdata);
+    return 0;
+}
+
+int32_t
+dg_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+           gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_XATTROP, xattrop, frame, this, loc, flags, dict, xdata);
+    return 0;
+}
+
+int32_t
+dg_entrylk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+           const char *basename, entrylk_cmd cmd, entrylk_type type,
+           dict_t *xdata)
+{
+    DG_FOP(GF_FOP_ENTRYLK, entrylk, frame, this, volume, loc, basename, cmd,
+           type, xdata);
+    return 0;
+}
+
+int32_t
+dg_getactivelk(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_GETACTIVELK, getactivelk, frame, this, loc, xdata);
+    return 0;
+}
+
+int32_t
+dg_finodelk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+            int32_t cmd, struct gf_flock *lock, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FINODELK, finodelk, frame, this, volume, fd, cmd, lock,
+           xdata);
+    return 0;
+}
+
+int32_t
+dg_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+          mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_CREATE, create, frame, this, loc, flags, mode, umask, fd,
+           xdata);
+    return 0;
+}
+
+int32_t
+dg_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+           size_t len, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_DISCARD, discard, frame, this, fd, offset, len, xdata);
+    return 0;
+}
+
+int32_t
+dg_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+         mode_t umask, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_MKDIR, mkdir, frame, this, loc, mode, umask, xdata);
+    return 0;
+}
+
+int32_t
+dg_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
+      struct gf_flock *lock, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_LK, lk, frame, this, fd, cmd, lock, xdata);
+    return 0;
+}
+
+int32_t
+dg_writev(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,
+          int32_t count, off_t off, uint32_t flags, struct iobref *iobref,
+          dict_t *xdata)
+{
+    DG_FOP(GF_FOP_WRITE, writev, frame, this, fd, vector, count, off, flags,
+           iobref, xdata);
+    return 0;
+}
+
+int32_t
+dg_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+          dict_t *xdata)
+{
+    DG_FOP(GF_FOP_ACCESS, access, frame, this, loc, mask, xdata);
+    return 0;
+}
+
+int32_t
+dg_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_LOOKUP, lookup, frame, this, loc, xdata);
+    return 0;
+}
+
+int32_t
+dg_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+         dict_t *xdata)
+{
+    DG_FOP(GF_FOP_RMDIR, rmdir, frame, this, loc, flags, xdata);
+    return 0;
+}
+
+int32_t
+dg_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t keep_size,
+             off_t offset, size_t len, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FALLOCATE, fallocate, frame, this, fd, keep_size, offset, len,
+           xdata);
+    return 0;
+}
+
+int32_t
+dg_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FSTAT, fstat, frame, this, fd, xdata);
+    return 0;
+}
+
+int32_t
+dg_lease(call_frame_t *frame, xlator_t *this, loc_t *loc,
+         struct gf_lease *lease, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_LEASE, lease, frame, this, loc, lease, xdata);
+    return 0;
+}
+
+int32_t
+dg_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_STAT, stat, frame, this, loc, xdata);
+    return 0;
+}
+
+int32_t
+dg_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+            dict_t *xdata)
+{
+    DG_FOP(GF_FOP_TRUNCATE, truncate, frame, this, loc, offset, xdata);
+    return 0;
+}
+
+int32_t
+dg_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, const char *name,
+            dict_t *xdata)
+{
+    DG_FOP(GF_FOP_GETXATTR, getxattr, frame, this, loc, name, xdata);
+    return 0;
+}
+
+int32_t
+dg_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath,
+           loc_t *loc, mode_t umask, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_SYMLINK, symlink, frame, this, linkpath, loc, umask, xdata);
+    return 0;
+}
+
+int32_t
+dg_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+            off_t len, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_ZEROFILL, zerofill, frame, this, fd, offset, len, xdata);
+    return 0;
+}
+
+int32_t
+dg_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+            dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FSYNCDIR, fsyncdir, frame, this, fd, flags, xdata);
+    return 0;
+}
+
+int32_t
+dg_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+             dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FGETXATTR, fgetxattr, frame, this, fd, name, xdata);
+    return 0;
+}
+
+int32_t
+dg_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t off, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_READDIRP, readdirp, frame, this, fd, size, off, xdata);
+    return 0;
+}
+
+int32_t
+dg_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+        dict_t *xdata)
+{
+    DG_FOP(GF_FOP_LINK, link, frame, this, oldloc, newloc, xdata);
+    return 0;
+}
+
+int32_t
+dg_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+            gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FXATTROP, fxattrop, frame, this, fd, flags, dict, xdata);
+    return 0;
+}
+
+int32_t
+dg_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FTRUNCATE, ftruncate, frame, this, fd, offset, xdata);
+    return 0;
+}
+
+int32_t
+dg_rchecksum(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             int32_t len, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_RCHECKSUM, rchecksum, frame, this, fd, offset, len, xdata);
+    return 0;
+}
+
+int32_t
+dg_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+          dict_t *xdata)
+{
+    DG_FOP(GF_FOP_UNLINK, unlink, frame, this, loc, flags, xdata);
+    return 0;
+}
+
+int32_t
+dg_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+            const char *basename, entrylk_cmd cmd, entrylk_type type,
+            dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FENTRYLK, fentrylk, frame, this, volume, fd, basename, cmd,
+           type, xdata);
+    return 0;
+}
+
+int32_t
+dg_getspec(call_frame_t *frame, xlator_t *this, const char *key, int32_t flags)
+{
+    DG_FOP(GF_FOP_GETSPEC, getspec, frame, this, key, flags);
+    return 0;
+}
+
+int32_t
+dg_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *stbuf,
+           int32_t valid, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_SETATTR, setattr, frame, this, loc, stbuf, valid, xdata);
+    return 0;
+}
+
+int32_t
+dg_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+         dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FSYNC, fsync, frame, this, fd, flags, xdata);
+    return 0;
+}
+
+int32_t
+dg_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_STATFS, statfs, frame, this, loc, xdata);
+    return 0;
+}
+
+int32_t
+dg_seek(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+        gf_seek_what_t what, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_SEEK, seek, frame, this, fd, offset, what, xdata);
+    return 0;
+}
+
+int32_t
+dg_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf,
+            int32_t valid, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FSETATTR, fsetattr, frame, this, fd, stbuf, valid, xdata);
+    return 0;
+}
+
+int32_t
+dg_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+           dict_t *xdata)
+{
+    DG_FOP(GF_FOP_OPENDIR, opendir, frame, this, loc, fd, xdata);
+    return 0;
+}
+
+int32_t
+dg_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+            dict_t *xdata)
+{
+    DG_FOP(GF_FOP_READLINK, readlink, frame, this, loc, size, xdata);
+    return 0;
+}
+
+int32_t
+dg_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               const char *name, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_REMOVEXATTR, removexattr, frame, this, loc, name, xdata);
+    return 0;
+}
+
+int32_t
+dg_forget(xlator_t *this, inode_t *inode)
+{
+    return 0;
+}
+
+int32_t
+dg_release(xlator_t *this, fd_t *fd)
+{
+    return 0;
+}
+
+int32_t
+dg_releasedir(xlator_t *this, fd_t *fd)
+{
+    return 0;
+}
+
+static int
+delay_gen_parse_fill_fops(dg_t *dg, char *enable_fops)
+{
+    char *op_no_str = NULL;
+    int op_no = -1;
+    int i = 0;
+    int ret = 0;
+    xlator_t *this = THIS;
+    char *saveptr = NULL;
+    char *dup_enable_fops = NULL;
+
+    if (strlen(enable_fops) == 0) {
+        for (i = GF_FOP_NULL + 1; i < GF_FOP_MAXVALUE; i++)
+            dg->enable[i] = 1;
+    } else {
+        dup_enable_fops = gf_strdup(enable_fops);
+        if (!dup_enable_fops) {
+            ret = -1;
+            goto out;
+        }
+        op_no_str = strtok_r(dup_enable_fops, ",", &saveptr);
+        while (op_no_str) {
+            op_no = gf_fop_int(op_no_str);
+            if (op_no == -1) {
+                gf_log(this->name, GF_LOG_WARNING, "Wrong option value %s",
+                       op_no_str);
+                ret = -1;
+                goto out;
+            } else {
+                dg->enable[op_no] = 1;
+            }
+
+            op_no_str = strtok_r(NULL, ",", &saveptr);
+        }
+    }
+out:
+    GF_FREE(dup_enable_fops);
+    return ret;
+}
+
+void
+delay_gen_set_delay_ppm(dg_t *dg, double percent)
+{
+    double ppm;
+
+    ppm = (percent / 100.0) * (double)DELAY_GRANULARITY;
+    dg->delay_ppm = ppm;
+}
+
+int32_t
+init(xlator_t *this)
+{
+    dg_t *dg = NULL;
+    int32_t ret = 0;
+    double delay_percent = 0;
+    char *delay_enable_fops = NULL;
+
+    if (!this->children || this->children->next) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "delay-gen not configured with one subvolume");
+        ret = -1;
+        goto out;
+    }
+
+    if (!this->parents) {
+        gf_log(this->name, GF_LOG_WARNING, "dangling volume. check volfile ");
+    }
+
+    dg = GF_CALLOC(1, sizeof(*dg), gf_delay_gen_mt_dg_t);
+
+    if (!dg) {
+        ret = -1;
+        goto out;
+    }
+
+    ret = -1;
+
+    GF_OPTION_INIT("delay-percentage", delay_percent, percent, out);
+    GF_OPTION_INIT("enable", delay_enable_fops, str, out);
+    GF_OPTION_INIT("delay-duration", dg->delay_duration, int32, out);
+
+    delay_gen_set_delay_ppm(dg, delay_percent);
+
+    ret = delay_gen_parse_fill_fops(dg, delay_enable_fops);
+    if (ret)
+        goto out;
+
+    this->private = dg;
+
+    ret = 0;
+out:
+    if (ret)
+        GF_FREE(dg);
+    return ret;
+}
+
+void
+fini(xlator_t *this)
+{
+    GF_FREE(this->private);
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    if (!this)
+        return ret;
+
+    ret = xlator_mem_acct_init(this, gf_delay_gen_mt_end + 1);
+
+    if (ret != 0) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Memory accounting init"
+               " failed");
+        return ret;
+    }
+
+    return ret;
+}
+
+int32_t
+reconfigure(xlator_t *this, dict_t *dict)
+{
+    /*At the moment I don't see any need to implement this. In future
+     *if this is needed we can add code here.
+     */
+    return 0;
+}
+
+int
+notify(xlator_t *this, int event, void *data, ...)
+{
+    return default_notify(this, event, data);
+}
+
+struct xlator_fops fops = {
+    .rename = dg_rename,
+    .ipc = dg_ipc,
+    .setactivelk = dg_setactivelk,
+    .flush = dg_flush,
+    .readdir = dg_readdir,
+    .setxattr = dg_setxattr,
+    .mknod = dg_mknod,
+    .fsetxattr = dg_fsetxattr,
+    .readv = dg_readv,
+    .inodelk = dg_inodelk,
+    .fremovexattr = dg_fremovexattr,
+    .open = dg_open,
+    .xattrop = dg_xattrop,
+    .entrylk = dg_entrylk,
+    .getactivelk = dg_getactivelk,
+    .finodelk = dg_finodelk,
+    .create = dg_create,
+    .discard = dg_discard,
+    .mkdir = dg_mkdir,
+    .lk = dg_lk,
+    .writev = dg_writev,
+    .access = dg_access,
+    .lookup = dg_lookup,
+    .rmdir = dg_rmdir,
+    .fallocate = dg_fallocate,
+    .fstat = dg_fstat,
+    .lease = dg_lease,
+    .stat = dg_stat,
+    .truncate = dg_truncate,
+    .getxattr = dg_getxattr,
+    .symlink = dg_symlink,
+    .zerofill = dg_zerofill,
+    .fsyncdir = dg_fsyncdir,
+    .fgetxattr = dg_fgetxattr,
+    .readdirp = dg_readdirp,
+    .link = dg_link,
+    .fxattrop = dg_fxattrop,
+    .ftruncate = dg_ftruncate,
+    .rchecksum = dg_rchecksum,
+    .unlink = dg_unlink,
+    .fentrylk = dg_fentrylk,
+    .getspec = dg_getspec,
+    .setattr = dg_setattr,
+    .fsync = dg_fsync,
+    .statfs = dg_statfs,
+    .seek = dg_seek,
+    .fsetattr = dg_fsetattr,
+    .opendir = dg_opendir,
+    .readlink = dg_readlink,
+    .removexattr = dg_removexattr,
+};
+
+struct xlator_cbks cbks = {
+    .forget = dg_forget,
+    .release = dg_release,
+    .releasedir = dg_releasedir,
+};
+
+struct volume_options options[] = {
+    {
+        .key = {"delay-percentage"},
+        .type = GF_OPTION_TYPE_PERCENT,
+        .default_value = "10%",
+        .description = "Percentage delay of operations when enabled.",
+        .op_version = {GD_OP_VERSION_3_13_0},
+        .flags = OPT_FLAG_SETTABLE,
+        .tags = {"delay-gen"},
+    },
+
+    {
+        .key = {"delay-duration"},
+        .type = GF_OPTION_TYPE_INT,
+        .description = "Delay duration in micro seconds",
+        .default_value = "100000",
+        .op_version = {GD_OP_VERSION_3_13_0},
+        .flags = OPT_FLAG_SETTABLE,
+        .tags = {"delay-gen"},
+    },
+
+    {
+        .key = {"enable"},
+        .type = GF_OPTION_TYPE_STR,
+        .description = "Accepts a string which takes ',' separated fop "
+                       "strings to denote which fops are enabled for delay",
+        .op_version = {GD_OP_VERSION_3_13_0},
+        .flags = OPT_FLAG_SETTABLE,
+        .tags = {"delay-gen"},
+        .default_value = "",
+    },
+
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {GD_OP_VERSION_3_12_0},
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "delay-gen",
+    .category = GF_TECH_PREVIEW,
+};
diff --git a/xlators/debug/delay-gen/src/delay-gen.h b/xlators/debug/delay-gen/src/delay-gen.h
new file mode 100644
index 00000000000..afa95e5eb2d
--- /dev/null
+++ b/xlators/debug/delay-gen/src/delay-gen.h
@@ -0,0 +1,27 @@
+/*
+ *   Copyright (c) 2017 Red Hat, Inc. <http://www.redhat.com>
+ *   This file is part of GlusterFS.
+ *
+ *   This file is licensed to you under your choice of the GNU Lesser
+ *   General Public License, version 3 or any later version (LGPLv3 or
+ *   later), or the GNU General Public License, version 2 (GPLv2), in all
+ *   cases as published by the Free Software Foundation.
+ */
+
+#ifndef __DELAY_GEN_H__
+#define __DELAY_GEN_H__
+
+#include "delay-gen-mem-types.h"
+#include "delay-gen-messages.h"
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+typedef struct {
+    int enable[GF_FOP_MAXVALUE];
+    int op_count;
+    int delay_ppm;
+    int delay_duration;
+} dg_t;
+
+#endif /* __DELAY_GEN_H__ */
diff --git a/xlators/debug/error-gen/src/Makefile.am b/xlators/debug/error-gen/src/Makefile.am
index 5075c59a861..038d2e8e66d 100644
--- a/xlators/debug/error-gen/src/Makefile.am
+++ b/xlators/debug/error-gen/src/Makefile.am
@@ -2,14 +2,15 @@
 xlator_LTLIBRARIES = error-gen.la
 xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/debug
 
-error_gen_la_LDFLAGS = -module -avoid-version
+error_gen_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
 
 error_gen_la_SOURCES = error-gen.c
 error_gen_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
 noinst_HEADERS = error-gen.h error-gen-mem-types.h
 
-AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
 
 AM_CFLAGS = -Wall $(GF_CFLAGS)
 
diff --git a/xlators/debug/error-gen/src/error-gen-mem-types.h b/xlators/debug/error-gen/src/error-gen-mem-types.h
index f02280535df..b9b713af8fc 100644
--- a/xlators/debug/error-gen/src/error-gen-mem-types.h
+++ b/xlators/debug/error-gen/src/error-gen-mem-types.h
@@ -11,10 +11,10 @@
 #ifndef __ERROR_GEN_MEM_TYPES_H__
 #define __ERROR_GEN_MEM_TYPES_H__
 
-#include "mem-types.h"
+#include <glusterfs/mem-types.h>
 
 enum gf_error_gen_mem_types_ {
-        gf_error_gen_mt_eg_t = gf_common_mt_end + 1,
-        gf_error_gen_mt_end
+    gf_error_gen_mt_eg_t = gf_common_mt_end + 1,
+    gf_error_gen_mt_end
 };
 #endif
diff --git a/xlators/debug/error-gen/src/error-gen.c b/xlators/debug/error-gen/src/error-gen.c
index ec0874b354a..d45655ef4c3 100644
--- a/xlators/debug/error-gen/src/error-gen.c
+++ b/xlators/debug/error-gen/src/error-gen.c
@@ -7,2226 +7,1657 @@
    later), or the GNU General Public License, version 2 (GPLv2), in all
    cases as published by the Free Software Foundation.
 */
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "xlator.h"
+#include <glusterfs/xlator.h>
 #include "error-gen.h"
-#include "statedump.h"
-
-sys_error_t error_no_list[] = {
-        [GF_FOP_LOOKUP]            = { .error_no_count = 4,
-                                    .error_no = {ENOENT,ENOTDIR,
-                                                 ENAMETOOLONG,EAGAIN}},
-        [GF_FOP_STAT]              = { .error_no_count = 7,
-                                    .error_no = {EACCES,EBADF,EFAULT,
-                                                 ENAMETOOLONG,ENOENT,
-                                                 ENOMEM,ENOTDIR}},
-        [GF_FOP_READLINK]          = { .error_no_count = 8,
-                                    .error_no = {EACCES,EFAULT,EINVAL,EIO,
-                                                 ENAMETOOLONG,ENOENT,ENOMEM,
-                                                 ENOTDIR}},
-        [GF_FOP_MKNOD]             = { .error_no_count = 11,
-                                    .error_no = {EACCES,EEXIST,EFAULT,
-                                                 EINVAL,ENAMETOOLONG,
-                                                 ENOENT,ENOMEM,ENOSPC,
-                                                 ENOTDIR,EPERM,EROFS}},
-        [GF_FOP_MKDIR]             = { .error_no_count = 10,
-                                    .error_no = {EACCES,EEXIST,EFAULT,
-                                                 ENAMETOOLONG,ENOENT,
-                                                 ENOMEM,ENOSPC,ENOTDIR,
-						 EPERM,EROFS}},
-        [GF_FOP_UNLINK]            = { .error_no_count = 10,
-                                    .error_no = {EACCES,EBUSY,EFAULT,EIO,
-                                                 EISDIR,ENAMETOOLONG,
-                                                 ENOENT,ENOMEM,ENOTDIR,
-                                                 EPERM,EROFS}},
-        [GF_FOP_RMDIR]             = { .error_no_count = 8,
-                                    .error_no = {EACCES,EBUSY,EFAULT,
-                                                 ENOMEM,ENOTDIR,ENOTEMPTY,
-                                                 EPERM,EROFS}},
-        [GF_FOP_SYMLINK]           = { .error_no_count = 11,
-                                    .error_no = {EACCES,EEXIST,EFAULT,EIO,
-                                                 ENAMETOOLONG,ENOENT,ENOMEM,
-                                                 ENOSPC,ENOTDIR,EPERM,
-                                                 EROFS}},
-        [GF_FOP_RENAME]            = { .error_no_count = 13,
-                                    .error_no = {EACCES,EBUSY,EFAULT,
-                                                 EINVAL,EISDIR,EMLINK,
-                                                 ENAMETOOLONG,ENOENT,ENOMEM,
-                                                 ENOSPC,ENOTDIR,EEXIST,
-                                                 EXDEV}},
-        [GF_FOP_LINK]              = { .error_no_count = 13,
-                                    .error_no = {EACCES,EFAULT,EEXIST,EIO,
-                                                 EMLINK,ENAMETOOLONG,
-                                                 ENOENT,ENOMEM,ENOSPC,
-                                                 ENOTDIR,EPERM,EROFS,
-                                                 EXDEV}},
-        [GF_FOP_TRUNCATE]          = { .error_no_count = 10,
-                                    .error_no = {EACCES,EFAULT,EFBIG,
-                                                 EINTR,EINVAL,EIO,EISDIR,
-                                                 ENAMETOOLONG,ENOENT,
-                                                 EISDIR}},
-        [GF_FOP_CREATE]            = {.error_no_count = 10,
-                                   .error_no = {EACCES,EEXIST,EFAULT,
-                                                EISDIR,EMFILE,ENAMETOOLONG,
-                                                ENFILE,ENODEV,ENOENT,
-                                                ENODEV}},
-        [GF_FOP_OPEN]              = { .error_no_count = 10,
-                                    .error_no = {EACCES,EEXIST,EFAULT,
-                                                 EISDIR,EMFILE,
-                                                 ENAMETOOLONG,ENFILE,
-                                                 ENODEV,ENOENT,ENOMEM}},
-        [GF_FOP_READ]              = { .error_no_count = 5,
-                                    .error_no = {EINVAL,EBADF,EFAULT,EISDIR,
-                                                 ENAMETOOLONG}},
-        [GF_FOP_WRITE]             = { .error_no_count = 7,
-                                    .error_no = {EINVAL,EBADF,EFAULT,EISDIR,
-                                                 ENAMETOOLONG,ENOSPC,
-						 GF_ERROR_SHORT_WRITE}},
-        [GF_FOP_STATFS]            = {.error_no_count = 10,
-                                   .error_no = {EACCES,EBADF,EFAULT,EINTR,
-                                                EIO,ENAMETOOLONG,ENOENT,
-                                                ENOMEM,ENOSYS,ENOTDIR}},
-        [GF_FOP_FLUSH]             = { .error_no_count = 5,
-                                    .error_no = {EACCES,EFAULT,
-                                                 ENAMETOOLONG,ENOSYS,
-                                                 ENOENT}},
-        [GF_FOP_FSYNC]             = { .error_no_count = 4,
-                                    .error_no = {EBADF,EIO,EROFS,EINVAL}},
-        [GF_FOP_SETXATTR]          = { .error_no_count = 4,
-                                    .error_no = {EACCES,EBADF,EINTR,
-                                                 ENAMETOOLONG}},
-        [GF_FOP_GETXATTR]          = { .error_no_count = 4,
-                                    .error_no = {EACCES,EBADF,ENAMETOOLONG,
-                                                 EINTR}},
-        [GF_FOP_REMOVEXATTR]       = { .error_no_count = 4,
-                                    .error_no = {EACCES,EBADF,ENAMETOOLONG,
-                                                 EINTR}},
-        [GF_FOP_FSETXATTR]          = { .error_no_count = 4,
-                                        .error_no = {EACCES,EBADF,EINTR,
-                                                     ENAMETOOLONG}},
-        [GF_FOP_FGETXATTR]          = { .error_no_count = 4,
-                                        .error_no = {EACCES,EBADF,ENAMETOOLONG,
-                                                     EINTR}},
-        [GF_FOP_FREMOVEXATTR]       = { .error_no_count = 4,
-                                        .error_no = {EACCES,EBADF,ENAMETOOLONG,
-                                                     EINTR}},
-        [GF_FOP_OPENDIR]           = { .error_no_count = 8,
-                                    .error_no = {EACCES,EEXIST,EFAULT,
-                                                 EISDIR,EMFILE,
-                                                 ENAMETOOLONG,ENFILE,
-                                                 ENODEV}},
-        [GF_FOP_READDIR]           = { .error_no_count = 5,
-                                    .error_no = {EINVAL,EACCES,EBADF,
-                                                 EMFILE,ENOENT}},
-        [GF_FOP_READDIRP]          = { .error_no_count = 5,
-                                    .error_no = {EINVAL,EACCES,EBADF,
-                                                 EMFILE,ENOENT}},
-        [GF_FOP_FSYNCDIR]          = { .error_no_count = 4,
-                                    .error_no = {EBADF,EIO,EROFS,EINVAL}},
-        [GF_FOP_ACCESS]            = { .error_no_count = 8,
-                                    .error_no = {EACCES,ENAMETOOLONG,
-                                                 ENOENT,ENOTDIR,EROFS,
-                                                 EFAULT,EINVAL,EIO}},
-        [GF_FOP_FTRUNCATE]         = { .error_no_count = 9,
-                                    .error_no = {EACCES,EFAULT,EFBIG,
-                                                 EINTR,EINVAL,EIO,EISDIR,
-                                                 ENAMETOOLONG,ENOENT}},
-        [GF_FOP_FSTAT]             = { .error_no_count = 7,
-                                    .error_no = {EACCES,EBADF,EFAULT,
-                                                 ENAMETOOLONG,ENOENT,
-                                                 ENOMEM,ENOTDIR}},
-        [GF_FOP_LK]                = { .error_no_count = 4,
-                                    .error_no = {EACCES,EFAULT,ENOENT,
-                                                 EINTR}},
-        [GF_FOP_XATTROP]           = { .error_no_count = 5,
-                                    .error_no = {EACCES,EFAULT,
-                                                 ENAMETOOLONG,ENOSYS,
-                                                 ENOENT}},
-        [GF_FOP_FXATTROP]          = { .error_no_count = 4,
-                                    .error_no = {EBADF,EIO,EROFS,EINVAL}},
-        [GF_FOP_INODELK]           = { .error_no_count = 4,
-                                    .error_no = {EACCES,EBADF,EINTR,
-                                                 ENAMETOOLONG}},
-        [GF_FOP_FINODELK]          = { .error_no_count = 4,
-                                    .error_no = {EACCES,EBADF,EINTR,
-                                                 ENAMETOOLONG}},
-        [GF_FOP_ENTRYLK]           = { .error_no_count = 4,
-                                    .error_no = {EACCES,EBADF,
-                                                 ENAMETOOLONG,EINTR}},
-        [GF_FOP_FENTRYLK]          = { .error_no_count = 10,
-                                    .error_no = {EACCES,EEXIST,EFAULT,
-                                                 EISDIR,EMFILE,
-                                                 ENAMETOOLONG,ENFILE,
-                                                 ENODEV,ENOENT,ENOMEM}},
-        [GF_FOP_SETATTR]           = {.error_no_count = 11,
-                                    .error_no = {EACCES,EFAULT,EIO,
-                                                 ENAMETOOLONG,ENOENT,
-                                                 ENOMEM,ENOTDIR,EPERM,
-                                                 EROFS,EBADF,EIO}},
-        [GF_FOP_FSETATTR]          = { .error_no_count = 11,
-                                    .error_no = {EACCES,EFAULT,EIO,
-                                                 ENAMETOOLONG,ENOENT,
-                                                 ENOMEM,ENOTDIR,EPERM,
-                                                 EROFS,EBADF,EIO}},
-        [GF_FOP_GETSPEC]           = { .error_no_count = 4,
-                                    .error_no = {EACCES,EBADF,ENAMETOOLONG,
-                                                 EINTR}}
-};
-
-int
-generate_rand_no (int op_no)
-{
-        int             rand_no = 0;
-
-        if (op_no < GF_FOP_MAXVALUE)
-                rand_no = rand () % error_no_list[op_no].error_no_count;
-        return rand_no;
-}
+#include <glusterfs/statedump.h>
+#include <glusterfs/defaults.h>
 
-int
-conv_errno_to_int (char **error_no)
-{
-        if (!strcmp ((*error_no), "ENOENT"))
-                return ENOENT;
-        else if (!strcmp ((*error_no), "ENOTDIR"))
-                return ENOTDIR;
-        else if (!strcmp ((*error_no), "ENAMETOOLONG"))
-                return ENAMETOOLONG;
-        else if (!strcmp ((*error_no), "EACCES"))
-                return EACCES;
-        else if (!strcmp ((*error_no), "EBADF"))
-                return EBADF;
-        else if (!strcmp ((*error_no), "EFAULT"))
-                return EFAULT;
-        else if (!strcmp ((*error_no), "ENOMEM"))
-                return ENOMEM;
-        else if (!strcmp ((*error_no), "EINVAL"))
-                return EINVAL;
-        else if (!strcmp ((*error_no), "EIO"))
-                return EIO;
-        else if (!strcmp ((*error_no), "EEXIST"))
-                return EEXIST;
-        else if (!strcmp ((*error_no), "ENOSPC"))
-                return ENOSPC;
-        else if (!strcmp ((*error_no), "EPERM"))
-                return EPERM;
-        else if (!strcmp ((*error_no), "EROFS"))
-                return EROFS;
-        else if (!strcmp ((*error_no), "EBUSY"))
-                return EBUSY;
-        else if (!strcmp ((*error_no), "EISDIR"))
-                return EISDIR;
-        else if (!strcmp ((*error_no), "ENOTEMPTY"))
-                return ENOTEMPTY;
-        else if (!strcmp ((*error_no), "EMLINK"))
-                return EMLINK;
-        else if (!strcmp ((*error_no), "ENODEV"))
-                return ENODEV;
-        else if (!strcmp ((*error_no), "EXDEV"))
-                return EXDEV;
-        else if (!strcmp ((*error_no), "EMFILE"))
-                return EMFILE;
-        else if (!strcmp ((*error_no), "ENFILE"))
-                return ENFILE;
-        else if (!strcmp ((*error_no), "ENOSYS"))
-                return ENOSYS;
-        else if (!strcmp ((*error_no), "EINTR"))
-                return EINTR;
-        else if (!strcmp ((*error_no), "EFBIG"))
-                return EFBIG;
-	else if (!strcmp((*error_no), "GF_ERROR_SHORT_WRITE"))
-		return GF_ERROR_SHORT_WRITE;
-        else
-                return EAGAIN;
-}
-
-int
-get_fop_int (char **op_no_str)
-{
-        if (!strcmp ((*op_no_str), "lookup"))
-                return GF_FOP_LOOKUP;
-        else if (!strcmp ((*op_no_str), "stat"))
-                return GF_FOP_STAT;
-        else if (!strcmp ((*op_no_str), "readlink"))
-                return GF_FOP_READLINK;
-        else if (!strcmp ((*op_no_str), "mknod"))
-                return GF_FOP_MKNOD;
-        else if (!strcmp ((*op_no_str), "mkdir"))
-                return GF_FOP_MKDIR;
-        else if (!strcmp ((*op_no_str), "unlink"))
-                return GF_FOP_UNLINK;
-        else if (!strcmp ((*op_no_str), "rmdir"))
-                return GF_FOP_RMDIR;
-        else if (!strcmp ((*op_no_str), "symlink"))
-                return GF_FOP_SYMLINK;
-        else if (!strcmp ((*op_no_str), "rename"))
-                return GF_FOP_RENAME;
-        else if (!strcmp ((*op_no_str), "link"))
-                return GF_FOP_LINK;
-        else if (!strcmp ((*op_no_str), "truncate"))
-                return GF_FOP_TRUNCATE;
-        else if (!strcmp ((*op_no_str), "create"))
-                return GF_FOP_CREATE;
-        else if (!strcmp ((*op_no_str), "open"))
-                return GF_FOP_OPEN;
-        else if (!strcmp ((*op_no_str), "readv"))
-                return GF_FOP_READ;
-        else if (!strcmp ((*op_no_str), "writev"))
-                return GF_FOP_WRITE;
-        else if (!strcmp ((*op_no_str), "statfs"))
-                return GF_FOP_STATFS;
-        else if (!strcmp ((*op_no_str), "flush"))
-                return GF_FOP_FLUSH;
-        else if (!strcmp ((*op_no_str), "fsync"))
-                return GF_FOP_FSYNC;
-        else if (!strcmp ((*op_no_str), "setxattr"))
-                return GF_FOP_SETXATTR;
-        else if (!strcmp ((*op_no_str), "getxattr"))
-                return GF_FOP_GETXATTR;
-        else if (!strcmp ((*op_no_str), "removexattr"))
-                return GF_FOP_REMOVEXATTR;
-        else if (!strcmp ((*op_no_str), "fsetxattr"))
-                return GF_FOP_FSETXATTR;
-        else if (!strcmp ((*op_no_str), "fgetxattr"))
-                return GF_FOP_FGETXATTR;
-        else if (!strcmp ((*op_no_str), "fremovexattr"))
-                return GF_FOP_FREMOVEXATTR;
-        else if (!strcmp ((*op_no_str), "opendir"))
-                return GF_FOP_OPENDIR;
-        else if (!strcmp ((*op_no_str), "readdir"))
-                return GF_FOP_READDIR;
-        else if (!strcmp ((*op_no_str), "readdirp"))
-                return GF_FOP_READDIRP;
-        else if (!strcmp ((*op_no_str), "fsyncdir"))
-                return GF_FOP_FSYNCDIR;
-        else if (!strcmp ((*op_no_str), "access"))
-                return GF_FOP_ACCESS;
-        else if (!strcmp ((*op_no_str), "ftruncate"))
-                return GF_FOP_FTRUNCATE;
-        else if (!strcmp ((*op_no_str), "fstat"))
-                return GF_FOP_FSTAT;
-        else if (!strcmp ((*op_no_str), "lk"))
-                return GF_FOP_LK;
-        else if (!strcmp ((*op_no_str), "xattrop"))
-                return GF_FOP_XATTROP;
-        else if (!strcmp ((*op_no_str), "fxattrop"))
-                return GF_FOP_FXATTROP;
-        else if (!strcmp ((*op_no_str), "inodelk"))
-                return GF_FOP_INODELK;
-        else if (!strcmp ((*op_no_str), "finodelk"))
-                return GF_FOP_FINODELK;
-        else if (!strcmp ((*op_no_str), "etrylk"))
-                return GF_FOP_ENTRYLK;
-        else if (!strcmp ((*op_no_str), "fentrylk"))
-                return GF_FOP_FENTRYLK;
-        else if (!strcmp ((*op_no_str), "setattr"))
-                return GF_FOP_SETATTR;
-        else if (!strcmp ((*op_no_str), "fsetattr"))
-                return GF_FOP_FSETATTR;
-        else if (!strcmp ((*op_no_str), "getspec"))
-                return GF_FOP_GETSPEC;
-	else
-                return -1;
-}
-
-int
-error_gen (xlator_t *this, int op_no)
-{
-        eg_t             *egp = NULL;
-        int              count = 0;
-        int              failure_iter_no = GF_FAILURE_DEFAULT;
-        char             *error_no = NULL;
-        int              rand_no = 0;
-        int              ret = 0;
-
-        egp = this->private;
+/*
+ * The user can specify an error probability as a float percentage, but we
+ * store it internally as a numerator with this as the denominator.  When it's
+ * used, it's like this:
+ *
+ *    (rand() % FAILURE_GRANULARITY) < error_rate
+ *
+ * To minimize rounding errors from the modulo operation, it's good for this to
+ * be a power of two.
+ *
+ * (BTW this is just the normal case.  If "random-failure" is set, that does
+ * something completely different and this number is irrelevant.  See error_gen
+ * for the legacy code.)
+ */
+#define FAILURE_GRANULARITY (1 << 20)
 
-        LOCK (&egp->lock);
+sys_error_t error_no_list[] = {
+    [GF_FOP_LOOKUP] = {.error_no_count = 4,
+                       .error_no = {ENOENT, ENOTDIR, ENAMETOOLONG, EAGAIN}},
+    [GF_FOP_STAT] = {.error_no_count = 6,
+                     .error_no = {EACCES, EFAULT, ENAMETOOLONG, ENOENT, ENOMEM,
+                                  ENOTDIR}},
+    [GF_FOP_READLINK] = {.error_no_count = 8,
+                         .error_no = {EACCES, EFAULT, EINVAL, EIO, ENAMETOOLONG,
+                                      ENOENT, ENOMEM, ENOTDIR}},
+    [GF_FOP_MKNOD] = {.error_no_count = 11,
+                      .error_no = {EACCES, EEXIST, EFAULT, EINVAL, ENAMETOOLONG,
+                                   ENOENT, ENOMEM, ENOSPC, ENOTDIR, EPERM,
+                                   EROFS}},
+    [GF_FOP_MKDIR] = {.error_no_count = 10,
+                      .error_no = {EACCES, EEXIST, EFAULT, ENAMETOOLONG, ENOENT,
+                                   ENOMEM, ENOSPC, ENOTDIR, EPERM, EROFS}},
+    [GF_FOP_UNLINK] = {.error_no_count = 10,
+                       .error_no = {EACCES, EBUSY, EFAULT, EIO, EISDIR,
+                                    ENAMETOOLONG, ENOENT, ENOMEM, ENOTDIR,
+                                    EPERM, EROFS}},
+    [GF_FOP_RMDIR] = {.error_no_count = 8,
+                      .error_no = {EACCES, EBUSY, EFAULT, ENOMEM, ENOTDIR,
+                                   ENOTEMPTY, EPERM, EROFS}},
+    [GF_FOP_SYMLINK] = {.error_no_count = 11,
+                        .error_no = {EACCES, EEXIST, EFAULT, EIO, ENAMETOOLONG,
+                                     ENOENT, ENOMEM, ENOSPC, ENOTDIR, EPERM,
+                                     EROFS}},
+    [GF_FOP_RENAME] = {.error_no_count = 13,
+                       .error_no = {EACCES, EBUSY, EFAULT, EINVAL, EISDIR,
+                                    EMLINK, ENAMETOOLONG, ENOENT, ENOMEM,
+                                    ENOSPC, ENOTDIR, EEXIST, EXDEV}},
+    [GF_FOP_LINK] = {.error_no_count = 13,
+                     .error_no = {EACCES, EFAULT, EEXIST, EIO, EMLINK,
+                                  ENAMETOOLONG, ENOENT, ENOMEM, ENOSPC, ENOTDIR,
+                                  EPERM, EROFS, EXDEV}},
+    [GF_FOP_TRUNCATE] = {.error_no_count = 10,
+                         .error_no = {EACCES, EFAULT, EFBIG, EINTR, EINVAL, EIO,
+                                      EISDIR, ENAMETOOLONG, ENOENT, EISDIR}},
+    [GF_FOP_CREATE] = {.error_no_count = 10,
+                       .error_no = {EACCES, EEXIST, EFAULT, EISDIR, EMFILE,
+                                    ENAMETOOLONG, ENFILE, ENODEV, ENOENT,
+                                    ENODEV}},
+    [GF_FOP_OPEN] = {.error_no_count = 10,
+                     .error_no = {EACCES, EEXIST, EFAULT, EISDIR, EMFILE,
+                                  ENAMETOOLONG, ENFILE, ENODEV, ENOENT,
+                                  ENOMEM}},
+    [GF_FOP_READ] = {.error_no_count = 5,
+                     .error_no = {EINVAL, EBADF, EFAULT, EISDIR, ENAMETOOLONG}},
+    [GF_FOP_WRITE] = {.error_no_count = 7,
+                      .error_no = {EINVAL, EBADF, EFAULT, EISDIR, ENAMETOOLONG,
+                                   ENOSPC, GF_ERROR_SHORT_WRITE}},
+    [GF_FOP_STATFS] = {.error_no_count = 9,
+                       .error_no = {EACCES, EFAULT, EINTR, EIO, ENAMETOOLONG,
+                                    ENOENT, ENOMEM, ENOSYS, ENOTDIR}},
+    [GF_FOP_FLUSH] = {.error_no_count = 5,
+                      .error_no = {EACCES, EFAULT, ENAMETOOLONG, ENOSYS,
+                                   ENOENT}},
+    [GF_FOP_FSYNC] = {.error_no_count = 4,
+                      .error_no = {EBADF, EIO, EROFS, EINVAL}},
+    [GF_FOP_SETXATTR] = {.error_no_count = 3,
+                         .error_no = {EACCES, EINTR, ENAMETOOLONG}},
+    [GF_FOP_GETXATTR] = {.error_no_count = 3,
+                         .error_no = {EACCES, ENAMETOOLONG, EINTR}},
+    [GF_FOP_REMOVEXATTR] = {.error_no_count = 3,
+                            .error_no = {EACCES, ENAMETOOLONG, EINTR}},
+    [GF_FOP_FSETXATTR] = {.error_no_count = 4,
+                          .error_no = {EACCES, EBADF, EINTR, ENAMETOOLONG}},
+    [GF_FOP_FGETXATTR] = {.error_no_count = 4,
+                          .error_no = {EACCES, EBADF, ENAMETOOLONG, EINTR}},
+    [GF_FOP_FREMOVEXATTR] = {.error_no_count = 4,
+                             .error_no = {EACCES, EBADF, ENAMETOOLONG, EINTR}},
+    [GF_FOP_OPENDIR] = {.error_no_count = 8,
+                        .error_no = {EACCES, EEXIST, EFAULT, EISDIR, EMFILE,
+                                     ENAMETOOLONG, ENFILE, ENODEV}},
+    [GF_FOP_READDIR] = {.error_no_count = 5,
+                        .error_no = {EINVAL, EACCES, EBADF, EMFILE, ENOENT}},
+    [GF_FOP_READDIRP] = {.error_no_count = 5,
+                         .error_no = {EINVAL, EACCES, EBADF, EMFILE, ENOENT}},
+    [GF_FOP_FSYNCDIR] = {.error_no_count = 4,
+                         .error_no = {EBADF, EIO, EROFS, EINVAL}},
+    [GF_FOP_ACCESS] = {.error_no_count = 8,
+                       .error_no = {EACCES, ENAMETOOLONG, ENOENT, ENOTDIR,
+                                    EROFS, EFAULT, EINVAL, EIO}},
+    [GF_FOP_FTRUNCATE] = {.error_no_count = 9,
+                          .error_no = {EACCES, EFAULT, EFBIG, EINTR, EINVAL,
+                                       EIO, EISDIR, ENAMETOOLONG, ENOENT}},
+    [GF_FOP_FSTAT] = {.error_no_count = 7,
+                      .error_no = {EACCES, EBADF, EFAULT, ENAMETOOLONG, ENOENT,
+                                   ENOMEM, ENOTDIR}},
+    [GF_FOP_LK] = {.error_no_count = 4,
+                   .error_no = {EACCES, EFAULT, ENOENT, EINTR}},
+    [GF_FOP_XATTROP] = {.error_no_count = 5,
+                        .error_no = {EACCES, EFAULT, ENAMETOOLONG, ENOSYS,
+                                     ENOENT}},
+    [GF_FOP_FXATTROP] = {.error_no_count = 4,
+                         .error_no = {EBADF, EIO, EROFS, EINVAL}},
+    [GF_FOP_INODELK] = {.error_no_count = 3,
+                        .error_no = {EACCES, EINTR, ENAMETOOLONG}},
+    [GF_FOP_FINODELK] = {.error_no_count = 4,
+                         .error_no = {EACCES, EBADF, EINTR, ENAMETOOLONG}},
+    [GF_FOP_ENTRYLK] = {.error_no_count = 3,
+                        .error_no = {EACCES, ENAMETOOLONG, EINTR}},
+    [GF_FOP_FENTRYLK] = {.error_no_count = 10,
+                         .error_no = {EACCES, EEXIST, EFAULT, EISDIR, EMFILE,
+                                      ENAMETOOLONG, ENFILE, ENODEV, ENOENT,
+                                      ENOMEM}},
+    [GF_FOP_SETATTR] = {.error_no_count = 10,
+                        .error_no = {EACCES, EFAULT, EIO, ENAMETOOLONG, ENOENT,
+                                     ENOMEM, ENOTDIR, EPERM, EROFS, EIO}},
+    [GF_FOP_FSETATTR] = {.error_no_count = 11,
+                         .error_no = {EACCES, EFAULT, EIO, ENAMETOOLONG, ENOENT,
+                                      ENOMEM, ENOTDIR, EPERM, EROFS, EBADF,
+                                      EIO}},
+    [GF_FOP_GETSPEC] = {.error_no_count = 3,
+                        .error_no = {EACCES, ENAMETOOLONG, EINTR}}};
+
+int
+generate_rand_no(int op_no)
+{
+    int rand_no = 0;
+    int error_no_list_size = 0;
+
+    error_no_list_size = sizeof(error_no_list) / sizeof(error_no_list[0]);
+
+    if (op_no < error_no_list_size)
+        /* coverity[DC.WEAK_CRYPTO] */
+        rand_no = rand() % error_no_list[op_no].error_no_count;
+    return rand_no;
+}
+
+int
+conv_errno_to_int(char **error_no)
+{
+    if (!strcmp((*error_no), "ENOENT"))
+        return ENOENT;
+    else if (!strcmp((*error_no), "ENOTDIR"))
+        return ENOTDIR;
+    else if (!strcmp((*error_no), "ENAMETOOLONG"))
+        return ENAMETOOLONG;
+    else if (!strcmp((*error_no), "EACCES"))
+        return EACCES;
+    else if (!strcmp((*error_no), "EBADF"))
+        return EBADF;
+    else if (!strcmp((*error_no), "EFAULT"))
+        return EFAULT;
+    else if (!strcmp((*error_no), "ENOMEM"))
+        return ENOMEM;
+    else if (!strcmp((*error_no), "EINVAL"))
+        return EINVAL;
+    else if (!strcmp((*error_no), "EIO"))
+        return EIO;
+    else if (!strcmp((*error_no), "EEXIST"))
+        return EEXIST;
+    else if (!strcmp((*error_no), "ENOSPC"))
+        return ENOSPC;
+    else if (!strcmp((*error_no), "EPERM"))
+        return EPERM;
+    else if (!strcmp((*error_no), "EROFS"))
+        return EROFS;
+    else if (!strcmp((*error_no), "EBUSY"))
+        return EBUSY;
+    else if (!strcmp((*error_no), "EISDIR"))
+        return EISDIR;
+    else if (!strcmp((*error_no), "ENOTEMPTY"))
+        return ENOTEMPTY;
+    else if (!strcmp((*error_no), "EMLINK"))
+        return EMLINK;
+    else if (!strcmp((*error_no), "ENODEV"))
+        return ENODEV;
+    else if (!strcmp((*error_no), "EXDEV"))
+        return EXDEV;
+    else if (!strcmp((*error_no), "EMFILE"))
+        return EMFILE;
+    else if (!strcmp((*error_no), "ENFILE"))
+        return ENFILE;
+    else if (!strcmp((*error_no), "ENOSYS"))
+        return ENOSYS;
+    else if (!strcmp((*error_no), "EINTR"))
+        return EINTR;
+    else if (!strcmp((*error_no), "EFBIG"))
+        return EFBIG;
+    else if (!strcmp((*error_no), "GF_ERROR_SHORT_WRITE"))
+        return GF_ERROR_SHORT_WRITE;
+    else
+        return EAGAIN;
+}
+
+int
+error_gen(xlator_t *this, int op_no)
+{
+    eg_t *egp = NULL;
+    int count = 0;
+    int error_no_int = 0;
+    int rand_no = 0;
+    int ret = 0;
+    gf_boolean_t should_err = _gf_false;
+    int error_no_list_size = 0;
+
+    egp = this->private;
+
+    if (egp->random_failure) {
+        /*
+         * I honestly don't know why anyone would use this "feature"
+         * but I'll try to preserve its functionality anyway.  Without
+         * locking twice to update failure_iter_no and egp->op_count
+         * separately, then not locking at all to update
+         * egp->failure_iter_no.  That's not needed for compatibility,
+         * and it's abhorrently wrong.  I have *some* standards.
+         */
+        LOCK(&egp->lock);
         {
-                count = ++egp->op_count;
-                failure_iter_no = egp->failure_iter_no;
-                error_no = egp->error_no;
+            count = ++(egp->op_count);
+            error_no_int = egp->error_no_int;
+            if ((count % egp->failure_iter_no) == 0) {
+                egp->op_count = 0;
+                /* coverity[DC.WEAK_CRYPTO] */
+                egp->failure_iter_no = 3 + (rand() % GF_UNIVERSAL_ANSWER);
+                should_err = _gf_true;
+            }
         }
-        UNLOCK (&egp->lock);
-
-        if((count % failure_iter_no) == 0) {
-                LOCK (&egp->lock);
-                {
-                        egp->op_count = 0;
-                }
-                UNLOCK (&egp->lock);
-
-                if (error_no)
-                        ret = conv_errno_to_int (&error_no);
-                else {
-
-                        rand_no = generate_rand_no (op_no);
-                        if (op_no >= GF_FOP_MAXVALUE)
-                                op_no = 0;
-                        if (rand_no >= error_no_list[op_no].error_no_count)
-                                rand_no = 0;
-                        ret = error_no_list[op_no].error_no[rand_no];
-                }
-                if (egp->random_failure == _gf_true)
-                        egp->failure_iter_no = 3 + (rand () % GF_UNIVERSAL_ANSWER);
+        UNLOCK(&egp->lock);
+    } else {
+        /*
+         * It turns out that rand() is almost universally implemented
+         * as a linear congruential PRNG, which is about as cheap as
+         * it gets.  This gets us real random behavior, including
+         * phenomena like streaks and dry spells, with controllable
+         * long-term probability, cheaply.
+         */
+        if ((rand() % FAILURE_GRANULARITY) < egp->failure_iter_no) {
+            should_err = _gf_true;
         }
-        return ret;
-}
-
+    }
+
+    error_no_list_size = sizeof(error_no_list) / sizeof(error_no_list[0]);
+    if (should_err) {
+        if (error_no_int)
+            ret = error_no_int;
+        else {
+            rand_no = generate_rand_no(op_no);
+            if (op_no >= error_no_list_size)
+                op_no = 0;
+            if (rand_no >= error_no_list[op_no].error_no_count)
+                rand_no = 0;
+            ret = error_no_list[op_no].error_no[rand_no];
+        }
+    }
 
-int
-error_gen_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno, inode_t *inode,
-		      struct iatt *buf, dict_t *xdata, struct iatt *postparent)
-{
-	STACK_UNWIND_STRICT (lookup, frame, op_ret, op_errno, inode,
-                             buf, xdata, postparent);
-        return 0;
+    return ret;
 }
 
-
 int
-error_gen_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                  dict_t *xdata)
+error_gen_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_LOOKUP];
+    egp = this->private;
+    enable = egp->enable[GF_FOP_LOOKUP];
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_LOOKUP);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_LOOKUP);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (lookup, frame, -1, op_errno, NULL, NULL, NULL,
-                                     NULL);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(lookup, frame, -1, op_errno, NULL, NULL, NULL,
+                            NULL);
         return 0;
-	}
-
-	STACK_WIND (frame, error_gen_lookup_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->lookup,
-		    loc, xdata);
-        return 0;
-}
+    }
 
-
-int
-error_gen_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		    int32_t op_ret, int32_t op_errno, struct iatt *buf, dict_t *xdata)
-{
-	STACK_UNWIND_STRICT (stat, frame, op_ret, op_errno, buf, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->lookup,
+                    loc, xdata);
+    return 0;
 }
 
 int
-error_gen_stat (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+error_gen_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_STAT];
+    egp = this->private;
+    enable = egp->enable[GF_FOP_STAT];
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_STAT);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_STAT);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (stat, frame, -1, op_errno, NULL, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(stat, frame, -1, op_errno, NULL, xdata);
         return 0;
-	}
+    }
 
-	STACK_WIND (frame, error_gen_stat_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->stat,
-		    loc, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->stat,
+                    loc, xdata);
+    return 0;
 }
 
-
 int
-error_gen_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno,
-                       struct iatt *preop, struct iatt *postop, dict_t *xdata)
+error_gen_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                  struct iatt *stbuf, int32_t valid, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (setattr, frame, op_ret, op_errno, preop, postop, xdata);
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
+
+    egp = this->private;
+    enable = egp->enable[GF_FOP_SETATTR];
+
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_SETATTR);
+
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(setattr, frame, -1, op_errno, NULL, NULL, xdata);
         return 0;
-}
+    }
 
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->setattr,
+                    loc, stbuf, valid, xdata);
+    return 0;
+}
 
 int
-error_gen_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
+error_gen_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
                    struct iatt *stbuf, int32_t valid, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_SETATTR];
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FSETATTR];
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_SETATTR);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FSETATTR);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (setattr, frame, -1, op_errno, NULL, NULL, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(fsetattr, frame, -1, op_errno, NULL, NULL, xdata);
         return 0;
-	}
+    }
 
-	STACK_WIND (frame, error_gen_setattr_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->setattr,
-		    loc, stbuf, valid, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->fsetattr,
+                    fd, stbuf, valid, xdata);
+    return 0;
 }
 
-
 int
-error_gen_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                    struct iatt *stbuf, int32_t valid, dict_t *xdata)
+error_gen_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                   off_t offset, dict_t *xdata)
 {
-	int             op_errno = 0;
-        eg_t            *egp = NULL;
-        int             enable = 1;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_FSETATTR];
+    egp = this->private;
+    enable = egp->enable[GF_FOP_TRUNCATE];
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_FSETATTR);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_TRUNCATE);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (fsetattr, frame, -1, op_errno, NULL, NULL, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(truncate, frame, -1, op_errno, NULL, NULL, xdata);
         return 0;
-	}
+    }
 
-	STACK_WIND (frame, error_gen_setattr_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->fsetattr,
-		    fd, stbuf, valid, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->truncate,
+                    loc, offset, xdata);
+    return 0;
 }
 
-
 int
-error_gen_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int32_t op_ret, int32_t op_errno,
-			struct iatt *prebuf, struct iatt *postbuf, dict_t *xdata)
+error_gen_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                    dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (truncate, frame, op_ret, op_errno,
-                             prebuf, postbuf, xdata);
-        return 0;
-}
-
-
-int
-error_gen_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                    off_t offset, dict_t *xdata)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_TRUNCATE];
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_TRUNCATE);
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FTRUNCATE];
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (truncate, frame, -1, op_errno,
-                                     NULL, NULL, xdata);
-        return 0;
-	}
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FTRUNCATE);
 
-	STACK_WIND (frame, error_gen_truncate_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->truncate,
-		    loc, offset, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(ftruncate, frame, -1, op_errno, NULL, NULL, xdata);
         return 0;
-}
-
+    }
 
-int
-error_gen_ftruncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			 int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                         struct iatt *postbuf, dict_t *xdata)
-{
-	STACK_UNWIND_STRICT (ftruncate, frame, op_ret, op_errno,
-                             prebuf, postbuf, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+    return 0;
 }
 
-
 int
-error_gen_ftruncate (call_frame_t *frame, xlator_t *this, fd_t *fd,
-		     off_t offset, dict_t *xdata)
+error_gen_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+                 dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp =NULL;
-        int              enable = 1;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_FTRUNCATE];
+    egp = this->private;
+    enable = egp->enable[GF_FOP_ACCESS];
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_FTRUNCATE);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_ACCESS);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (ftruncate, frame, -1, op_errno,
-                                     NULL, NULL, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(access, frame, -1, op_errno, xdata);
         return 0;
-	}
-
-	STACK_WIND (frame, error_gen_ftruncate_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->ftruncate,
-		    fd, offset, xdata);
-        return 0;
-}
+    }
 
-
-int
-error_gen_access_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-	STACK_UNWIND_STRICT (access, frame, op_ret, op_errno, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->access,
+                    loc, mask, xdata);
+    return 0;
 }
 
-
 int
-error_gen_access (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		  int32_t mask, dict_t *xdata)
+error_gen_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+                   dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_ACCESS];
+    egp = this->private;
+    enable = egp->enable[GF_FOP_READLINK];
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_ACCESS);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_READLINK);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (access, frame, -1, op_errno, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(readlink, frame, -1, op_errno, NULL, NULL, xdata);
         return 0;
-	}
+    }
 
-	STACK_WIND (frame, error_gen_access_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->access,
-		    loc, mask, xdata);
-        return 0;
-}
-
-
-int
-error_gen_readlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int32_t op_ret, int32_t op_errno,
-			const char *path, struct iatt *sbuf, dict_t *xdata)
-{
-	STACK_UNWIND_STRICT (readlink, frame, op_ret, op_errno, path, sbuf, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->readlink,
+                    loc, size, xdata);
+    return 0;
 }
 
-
 int
-error_gen_readlink (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                    size_t size, dict_t *xdata)
+error_gen_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+                dev_t rdev, mode_t umask, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_READLINK];
+    egp = this->private;
+    enable = egp->enable[GF_FOP_MKNOD];
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_READLINK);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_MKNOD);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (readlink, frame, -1, op_errno, NULL, NULL, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                            xdata);
         return 0;
-	}
+    }
 
-	STACK_WIND (frame, error_gen_readlink_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->readlink,
-		    loc, size, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->mknod,
+                    loc, mode, rdev, umask, xdata);
+    return 0;
 }
 
-
 int
-error_gen_mknod_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		     int32_t op_ret, int32_t op_errno, inode_t *inode,
-                     struct iatt *buf, struct iatt *preparent,
-                     struct iatt *postparent, dict_t *xdata)
+error_gen_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+                mode_t umask, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (mknod, frame, op_ret, op_errno,
-                             inode, buf,
-                             preparent, postparent, xdata);
-        return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_MKDIR];
 
-int
-error_gen_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		 mode_t mode, dev_t rdev, mode_t umask, dict_t *xdata)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_MKNOD];
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_MKDIR);
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_MKNOD);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (mknod, frame, -1, op_errno, NULL, NULL,
-                                     NULL, NULL, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                            xdata);
         return 0;
-	}
+    }
 
-	STACK_WIND (frame, error_gen_mknod_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->mknod,
-		    loc, mode, rdev, umask, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->mkdir,
+                    loc, mode, umask, xdata);
+    return 0;
 }
 
-
 int
-error_gen_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		     int32_t op_ret, int32_t op_errno, inode_t *inode,
-                     struct iatt *buf, struct iatt *preparent,
-                     struct iatt *postparent, dict_t *xdata)
+error_gen_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+                 dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (mkdir, frame, op_ret, op_errno,
-                             inode, buf,
-                             preparent, postparent, xdata);
-        return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-int
-error_gen_mkdir (call_frame_t *frame, xlator_t *this,
-		 loc_t *loc, mode_t mode, mode_t umask, dict_t *xdata)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_MKDIR];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_MKDIR);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (mkdir, frame, -1, op_errno, NULL, NULL,
-                                     NULL, NULL, xdata);
-                return 0;
-	}
-
-	STACK_WIND (frame, error_gen_mkdir_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->mkdir,
-		    loc, mode, umask, xdata);
-        return 0;
-}
+    egp = this->private;
+    enable = egp->enable[GF_FOP_UNLINK];
 
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_UNLINK);
 
-int
-error_gen_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno,
-                      struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
-{
-	STACK_UNWIND_STRICT (unlink, frame, op_ret, op_errno,
-                             preparent, postparent, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(unlink, frame, -1, op_errno, NULL, NULL, xdata);
         return 0;
-}
+    }
 
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->unlink,
+                    loc, xflag, xdata);
+    return 0;
+}
 
 int
-error_gen_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
-                  dict_t *xdata)
+error_gen_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+                dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_UNLINK];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_UNLINK);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (unlink, frame, -1, op_errno, NULL, NULL,
-                                     xdata);
-                return 0;
-	}
-
-	STACK_WIND (frame, error_gen_unlink_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->unlink,
-		    loc, xflag, xdata);
-        return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_RMDIR];
 
-int
-error_gen_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		     int32_t op_ret, int32_t op_errno,
-                     struct iatt *preparent, struct iatt *postparent,
-                     dict_t *xdata)
-{
-	STACK_UNWIND_STRICT (rmdir, frame, op_ret, op_errno,
-                             preparent, postparent, xdata);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_RMDIR);
+
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(rmdir, frame, -1, op_errno, NULL, NULL, xdata);
         return 0;
-}
+    }
 
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->rmdir,
+                    loc, flags, xdata);
+    return 0;
+}
 
 int
-error_gen_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
-                 dict_t *xdata)
+error_gen_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath,
+                  loc_t *loc, mode_t umask, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_RMDIR];
+    egp = this->private;
+    enable = egp->enable[GF_FOP_SYMLINK];
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_RMDIR);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_SYMLINK);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (rmdir, frame, -1, op_errno, NULL, NULL, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(symlink, frame, -1, op_errno, NULL, NULL, NULL,
+                            NULL, NULL); /* pre & post parent attr */
         return 0;
-	}
+    }
 
-	STACK_WIND (frame, error_gen_rmdir_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->rmdir,
-		    loc, flags, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->symlink,
+                    linkpath, loc, umask, xdata);
+    return 0;
 }
 
-
 int
-error_gen_symlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int32_t op_ret, int32_t op_errno, inode_t *inode,
-                       struct iatt *buf, struct iatt *preparent,
-                       struct iatt *postparent, dict_t *xdata)
+error_gen_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc,
+                 loc_t *newloc, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (symlink, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent, xdata);
-        return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_RENAME];
 
-int
-error_gen_symlink (call_frame_t *frame, xlator_t *this, const char *linkpath,
-		   loc_t *loc, mode_t umask, dict_t *xdata)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_SYMLINK];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_SYMLINK);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (symlink, frame, -1, op_errno, NULL, NULL,
-                                     NULL, NULL, NULL); /* pre & post parent attr */
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_symlink_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->symlink,
-		    linkpath, loc, umask, xdata);
-        return 0;
-}
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_RENAME);
 
-
-int
-error_gen_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno, struct iatt *buf,
-                      struct iatt *preoldparent, struct iatt *postoldparent,
-                      struct iatt *prenewparent, struct iatt *postnewparent,
-                      dict_t *xdata)
-{
-	STACK_UNWIND_STRICT (rename, frame, op_ret, op_errno, buf,
-                             preoldparent, postoldparent,
-                             prenewparent, postnewparent, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(rename, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                            NULL, NULL);
         return 0;
-}
-
+    }
 
-int
-error_gen_rename (call_frame_t *frame, xlator_t *this,
-		  loc_t *oldloc, loc_t *newloc, dict_t *xdata)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_RENAME];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_RENAME);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (rename, frame, -1, op_errno, NULL,
-                                     NULL, NULL, NULL, NULL, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_rename_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->rename,
-		    oldloc, newloc, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->rename,
+                    oldloc, newloc, xdata);
+    return 0;
 }
 
-
 int
-error_gen_link_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		    int32_t op_ret, int32_t op_errno, inode_t *inode,
-                    struct iatt *buf, struct iatt *preparent,
-                    struct iatt *postparent, dict_t *xdata)
+error_gen_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc,
+               loc_t *newloc, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (link, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent, xdata);
-        return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_LINK];
 
-int
-error_gen_link (call_frame_t *frame, xlator_t *this,
-		loc_t *oldloc, loc_t *newloc, dict_t *xdata)
-{
-	int             op_errno = 0;
-        eg_t            *egp = NULL;
-        int             enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_LINK];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_LINK);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (link, frame, -1, op_errno, NULL, NULL,
-                                     NULL, NULL, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_link_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->link,
-		    oldloc, newloc, xdata);
-        return 0;
-}
-
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_LINK);
 
-int
-error_gen_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno,
-		      fd_t *fd, inode_t *inode, struct iatt *buf,
-                      struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
-{
-	STACK_UNWIND_STRICT (create, frame, op_ret, op_errno, fd, inode, buf,
-                             preparent, postparent, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(link, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                            NULL);
         return 0;
-}
+    }
 
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->link,
+                    oldloc, newloc, xdata);
+    return 0;
+}
 
 int
-error_gen_create (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		  int32_t flags, mode_t mode, mode_t umask, fd_t *fd,
-                  dict_t *xdata)
+error_gen_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+                 mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
 {
-	int             op_errno = 0;
-        eg_t            *egp = NULL;
-        int             enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_CREATE];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_CREATE);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (create, frame, -1, op_errno, NULL, NULL,
-                                     NULL, NULL, NULL, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_create_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->create,
-		    loc, flags, mode, umask, fd, xdata);
-        return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_CREATE];
 
-int
-error_gen_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		    int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
-{
-	STACK_UNWIND_STRICT (open, frame, op_ret, op_errno, fd, xdata);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_CREATE);
+
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(create, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                            NULL, NULL);
         return 0;
-}
+    }
 
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->create,
+                    loc, flags, mode, umask, fd, xdata);
+    return 0;
+}
 
 int
-error_gen_open (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		int32_t flags, fd_t *fd, dict_t *xdata)
+error_gen_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+               fd_t *fd, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_OPEN];
+    egp = this->private;
+    enable = egp->enable[GF_FOP_OPEN];
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_OPEN);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_OPEN);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (open, frame, -1, op_errno, NULL, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(open, frame, -1, op_errno, NULL, xdata);
         return 0;
-	}
+    }
 
-	STACK_WIND (frame, error_gen_open_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->open,
-		    loc, flags, fd, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->open,
+                    loc, flags, fd, xdata);
+    return 0;
 }
 
-
 int
-error_gen_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		     int32_t op_ret, int32_t op_errno,
-		     struct iovec *vector, int32_t count,
-		     struct iatt *stbuf, struct iobref *iobref, dict_t *xdata)
+error_gen_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                off_t offset, uint32_t flags, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (readv, frame, op_ret, op_errno,
-                             vector, count, stbuf, iobref, xdata);
-        return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_READ];
 
-int
-error_gen_readv (call_frame_t *frame, xlator_t *this,
-		 fd_t *fd, size_t size, off_t offset, uint32_t flags, dict_t *xdata)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_READ];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_READ);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_READ);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (readv, frame, -1, op_errno, NULL, 0,
-                                     NULL, NULL, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(readv, frame, -1, op_errno, NULL, 0, NULL, NULL,
+                            xdata);
         return 0;
-	}
+    }
 
-
-	STACK_WIND (frame, error_gen_readv_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->readv,
-		    fd, size, offset, flags, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->readv,
+                    fd, size, offset, flags, xdata);
+    return 0;
 }
 
-
 int
-error_gen_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno,
-                      struct iatt *prebuf, struct iatt *postbuf, dict_t *xdata)
+error_gen_writev(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                 struct iovec *vector, int32_t count, off_t off, uint32_t flags,
+                 struct iobref *iobref, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno, prebuf, postbuf, xdata);
-        return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
+    struct iovec *shortvec = NULL;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_WRITE];
 
-int
-error_gen_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
-		  struct iovec *vector, int32_t count,
-		  off_t off, uint32_t flags, struct iobref *iobref, dict_t *xdata)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_WRITE];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_WRITE);
-
-	if (op_errno == GF_ERROR_SHORT_WRITE) {
-		struct iovec *shortvec;
-
-		/*
-		 * A short write error returns some value less than what was
-		 * requested from a write. To simulate this, replace the vector
-		 * with one half the size;
-		 */
-		shortvec = iov_dup(vector, 1);
-		shortvec->iov_len /= 2;
-
-		STACK_WIND(frame, error_gen_writev_cbk, FIRST_CHILD(this),
-			   FIRST_CHILD(this)->fops->writev, fd, shortvec, count,
-			   off, flags, iobref, xdata);
-		GF_FREE(shortvec);
-		return 0;
-	} else if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (writev, frame, -1, op_errno, NULL, NULL, xdata);
-        	return 0;
-	}
-
-	STACK_WIND (frame, error_gen_writev_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->writev,
-		    fd, vector, count, off, flags, iobref, xdata);
-        return 0;
-}
-
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_WRITE);
 
-int
-error_gen_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		     int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-	STACK_UNWIND_STRICT (flush, frame, op_ret, op_errno, xdata);
+    if (op_errno == GF_ERROR_SHORT_WRITE) {
+        /*
+         * A short write error returns some value less than what was
+         * requested from a write. To simulate this, replace the vector
+         * with one half the size;
+         */
+        shortvec = iov_dup(vector, 1);
+        shortvec->iov_len /= 2;
+        count = 1;
+        goto wind;
+    } else if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(writev, frame, -1, op_errno, NULL, NULL, xdata);
         return 0;
-}
+    }
+wind:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->writev,
+                    fd, shortvec ? shortvec : vector, count, off, flags, iobref,
+                    xdata);
 
+    if (shortvec)
+        GF_FREE(shortvec);
+    return 0;
+}
 
 int
-error_gen_flush (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+error_gen_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_FLUSH];
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FLUSH];
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_FLUSH);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FLUSH);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (flush, frame, -1, op_errno, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(flush, frame, -1, op_errno, xdata);
         return 0;
-	}
+    }
 
-	STACK_WIND (frame, error_gen_flush_cbk,
-                    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->flush,
-		    fd, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->flush,
+                    fd, xdata);
+    return 0;
 }
 
-
 int
-error_gen_fsync_cbk (call_frame_t *frame, void *cookie,
-		     xlator_t *this, int32_t op_ret,
-		     int32_t op_errno, struct iatt *prebuf,
-                     struct iatt *postbuf, dict_t *xdata)
+error_gen_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+                dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (fsync, frame, op_ret, op_errno, prebuf, postbuf, xdata);
-        return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FSYNC];
 
-int
-error_gen_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags, dict_t *xdata)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_FSYNC];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_FSYNC);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FSYNC);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (fsync, frame, -1, op_errno, NULL, NULL, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(fsync, frame, -1, op_errno, NULL, NULL, xdata);
         return 0;
-	}
+    }
 
-	STACK_WIND (frame, error_gen_fsync_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->fsync,
-		    fd, flags, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->fsync,
+                    fd, flags, xdata);
+    return 0;
 }
 
-
 int
-error_gen_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, struct iatt *buf, dict_t *xdata)
+error_gen_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (fstat, frame, op_ret, op_errno, buf, xdata);
-        return 0;
-}
-
-
-int
-error_gen_fstat (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_FSTAT];
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FSTAT];
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_FSTAT);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FSTAT);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (fstat, frame, -1, op_errno, NULL, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(fstat, frame, -1, op_errno, NULL, xdata);
         return 0;
-	}
+    }
 
-	STACK_WIND (frame, error_gen_fstat_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->fstat,
-		    fd, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->fstat,
+                    fd, xdata);
+    return 0;
 }
 
-
 int
-error_gen_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
-{
-	STACK_UNWIND_STRICT (opendir, frame, op_ret, op_errno, fd, xdata);
-        return 0;
-}
-
-
-int
-error_gen_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd, dict_t *xdata)
+error_gen_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+                  dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_OPENDIR];
+    egp = this->private;
+    enable = egp->enable[GF_FOP_OPENDIR];
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_OPENDIR);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_OPENDIR);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (opendir, frame, -1, op_errno, NULL, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(opendir, frame, -1, op_errno, NULL, xdata);
         return 0;
-	}
+    }
 
-	STACK_WIND (frame, error_gen_opendir_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->opendir,
-		    loc, fd, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->opendir,
+                    loc, fd, xdata);
+    return 0;
 }
 
 int
-error_gen_fsyncdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-	STACK_UNWIND_STRICT (fsyncdir, frame, op_ret, op_errno, xdata);
-        return 0;
-}
-
-
-int
-error_gen_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                    int32_t flags, dict_t *xdata)
+error_gen_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+                   dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_FSYNCDIR];
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FSYNCDIR];
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_FSYNCDIR);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FSYNCDIR);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (fsyncdir, frame, -1, op_errno, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(fsyncdir, frame, -1, op_errno, xdata);
         return 0;
-	}
+    }
 
-	STACK_WIND (frame, error_gen_fsyncdir_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->fsyncdir,
-		    fd, flags, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->fsyncdir,
+                    fd, flags, xdata);
+    return 0;
 }
 
-
 int
-error_gen_statfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno, struct statvfs *buf, dict_t *xdata)
+error_gen_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (statfs, frame, op_ret, op_errno, buf, xdata);
-        return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_STATFS];
 
-int
-error_gen_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_STATFS);
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_STATFS];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_STATFS);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (statfs, frame, -1, op_errno, NULL, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(statfs, frame, -1, op_errno, NULL, xdata);
         return 0;
-	}
-
-	STACK_WIND (frame, error_gen_statfs_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->statfs,
-		    loc, xdata);
-        return 0;
-}
+    }
 
-
-int
-error_gen_setxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-	STACK_UNWIND_STRICT (setxattr, frame, op_ret, op_errno, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->statfs,
+                    loc, xdata);
+    return 0;
 }
 
-
 int
-error_gen_setxattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		    dict_t *dict, int32_t flags, dict_t *xdata)
+error_gen_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                   dict_t *dict, int32_t flags, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_SETXATTR];
+    egp = this->private;
+    enable = egp->enable[GF_FOP_SETXATTR];
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_SETXATTR);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_SETXATTR);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (setxattr, frame, -1, op_errno, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(setxattr, frame, -1, op_errno, xdata);
         return 0;
-	}
+    }
 
-	STACK_WIND (frame, error_gen_setxattr_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->setxattr,
-		    loc, dict, flags, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->setxattr,
+                    loc, dict, flags, xdata);
+    return 0;
 }
 
-
 int
-error_gen_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+error_gen_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                   const char *name, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (getxattr, frame, op_ret, op_errno, dict, xdata);
-        return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_GETXATTR];
 
-int
-error_gen_getxattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		    const char *name, dict_t *xdata)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_GETXATTR);
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_GETXATTR];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_GETXATTR);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (getxattr, frame, -1, op_errno, NULL, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(getxattr, frame, -1, op_errno, NULL, xdata);
         return 0;
-	}
+    }
 
-	STACK_WIND (frame, error_gen_getxattr_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->getxattr,
-		    loc, name, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->getxattr,
+                    loc, name, xdata);
+    return 0;
 }
 
 int
-error_gen_fsetxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-	STACK_UNWIND_STRICT (fsetxattr, frame, op_ret, op_errno, xdata);
-        return 0;
-}
-
-
-int
-error_gen_fsetxattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                     dict_t *dict, int32_t flags, dict_t *xdata)
+error_gen_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+                    int32_t flags, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_FSETXATTR];
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_FSETXATTR);
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FSETXATTR];
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (fsetxattr, frame, -1, op_errno, xdata);
-        return 0;
-	}
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FSETXATTR);
 
-	STACK_WIND (frame, error_gen_fsetxattr_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->fsetxattr,
-		    fd, dict, flags, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(fsetxattr, frame, -1, op_errno, xdata);
         return 0;
-}
-
+    }
 
-int
-error_gen_fgetxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
-{
-	STACK_UNWIND_STRICT (fgetxattr, frame, op_ret, op_errno, dict, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags, xdata);
+    return 0;
 }
 
-
 int
-error_gen_fgetxattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                     const char *name, dict_t *xdata)
+error_gen_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                    const char *name, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_FGETXATTR];
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_FGETXATTR);
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FGETXATTR];
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (fgetxattr, frame, -1, op_errno, NULL, xdata);
-        return 0;
-	}
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FGETXATTR);
 
-	STACK_WIND (frame, error_gen_fgetxattr_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->fgetxattr,
-		    fd, name, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(fgetxattr, frame, -1, op_errno, NULL, xdata);
         return 0;
-}
-
+    }
 
-int
-error_gen_xattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
-{
-	STACK_UNWIND_STRICT (xattrop, frame, op_ret, op_errno, dict, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->fgetxattr, fd, name, xdata);
+    return 0;
 }
 
-
 int
-error_gen_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		   gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+error_gen_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                  gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_XATTROP];
+    egp = this->private;
+    enable = egp->enable[GF_FOP_XATTROP];
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_XATTROP);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_XATTROP);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (xattrop, frame, -1, op_errno, NULL, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(xattrop, frame, -1, op_errno, NULL, xdata);
         return 0;
-	}
+    }
 
-	STACK_WIND (frame, error_gen_xattrop_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->xattrop,
-		    loc, flags, dict, xdata);
-        return 0;
-}
-
-
-int
-error_gen_fxattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
-{
-	STACK_UNWIND_STRICT (fxattrop, frame, op_ret, op_errno, dict, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->xattrop,
+                    loc, flags, dict, xdata);
+    return 0;
 }
 
-
 int
-error_gen_fxattrop (call_frame_t *frame, xlator_t *this, fd_t *fd,
-		    gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+error_gen_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                   gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_FXATTROP];
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FXATTROP];
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_FXATTROP);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FXATTROP);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (fxattrop, frame, -1, op_errno, NULL, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(fxattrop, frame, -1, op_errno, NULL, xdata);
         return 0;
-	}
+    }
 
-	STACK_WIND (frame, error_gen_fxattrop_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->fxattrop,
-		    fd, flags, dict, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->fxattrop,
+                    fd, flags, dict, xdata);
+    return 0;
 }
 
-
 int
-error_gen_removexattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			   int32_t op_ret, int32_t op_errno, dict_t *xdata)
+error_gen_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                      const char *name, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (removexattr, frame, op_ret, op_errno, xdata);
-        return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_REMOVEXATTR];
 
-int
-error_gen_removexattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		       const char *name, dict_t *xdata)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_REMOVEXATTR];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_REMOVEXATTR);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_REMOVEXATTR);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (removexattr, frame, -1, op_errno, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(removexattr, frame, -1, op_errno, xdata);
         return 0;
-	}
+    }
 
-	STACK_WIND (frame, error_gen_removexattr_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->removexattr,
-		    loc, name, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->removexattr, loc, name, xdata);
+    return 0;
 }
 
 int
-error_gen_fremovexattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			   int32_t op_ret, int32_t op_errno, dict_t *xdata)
+error_gen_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                       const char *name, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (fremovexattr, frame, op_ret, op_errno, xdata);
-        return 0;
-}
-
-
-int
-error_gen_fremovexattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                        const char *name, dict_t *xdata)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_FREMOVEXATTR];
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FREMOVEXATTR];
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_FREMOVEXATTR);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FREMOVEXATTR);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (fremovexattr, frame, -1, op_errno, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(fremovexattr, frame, -1, op_errno, xdata);
         return 0;
-	}
+    }
 
-	STACK_WIND (frame, error_gen_fremovexattr_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->fremovexattr,
-		    fd, name, xdata);
-        return 0;
-}
-
-
-int
-error_gen_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		  int32_t op_ret, int32_t op_errno, struct gf_flock *lock, dict_t *xdata)
-{
-	STACK_UNWIND_STRICT (lk, frame, op_ret, op_errno, lock, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->fremovexattr, fd, name, xdata);
+    return 0;
 }
 
-
 int
-error_gen_lk (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
-	      struct gf_flock *lock, dict_t *xdata)
+error_gen_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
+             struct gf_flock *lock, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_LK];
+    egp = this->private;
+    enable = egp->enable[GF_FOP_LK];
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_LK);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_LK);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (lk, frame, -1, op_errno, NULL, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(lk, frame, -1, op_errno, NULL, xdata);
         return 0;
-	}
+    }
 
-	STACK_WIND (frame, error_gen_lk_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->lk,
-		    fd, cmd, lock, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->lk, fd,
+                    cmd, lock, xdata);
+    return 0;
 }
 
-
 int
-error_gen_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
+error_gen_inodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+                  loc_t *loc, int32_t cmd, struct gf_flock *lock, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (inodelk, frame, op_ret, op_errno, xdata);
-        return 0;
-}
-
-
-int
-error_gen_inodelk (call_frame_t *frame, xlator_t *this,
-		   const char *volume, loc_t *loc, int32_t cmd,
-                   struct gf_flock *lock, dict_t *xdata)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_INODELK];
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_INODELK);
+    egp = this->private;
+    enable = egp->enable[GF_FOP_INODELK];
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (inodelk, frame, -1, op_errno, xdata);
-        return 0;
-	}
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_INODELK);
 
-	STACK_WIND (frame, error_gen_inodelk_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->inodelk,
-		    volume, loc, cmd, lock, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(inodelk, frame, -1, op_errno, xdata);
         return 0;
-}
-
+    }
 
-int
-error_gen_finodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-	STACK_UNWIND_STRICT (finodelk, frame, op_ret, op_errno, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->inodelk,
+                    volume, loc, cmd, lock, xdata);
+    return 0;
 }
 
-
 int
-error_gen_finodelk (call_frame_t *frame, xlator_t *this,
-		    const char *volume, fd_t *fd, int32_t cmd,
-                    struct gf_flock *lock, dict_t *xdata)
+error_gen_finodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+                   fd_t *fd, int32_t cmd, struct gf_flock *lock, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_FINODELK];
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_FINODELK);
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FINODELK];
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (finodelk, frame, -1, op_errno, xdata);
-        return 0;
-	}
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FINODELK);
 
-	STACK_WIND (frame, error_gen_finodelk_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->finodelk,
-		    volume, fd, cmd, lock, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(finodelk, frame, -1, op_errno, xdata);
         return 0;
-}
-
+    }
 
-int
-error_gen_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-	STACK_UNWIND_STRICT (entrylk, frame, op_ret, op_errno, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->finodelk,
+                    volume, fd, cmd, lock, xdata);
+    return 0;
 }
 
-
 int
-error_gen_entrylk (call_frame_t *frame, xlator_t *this,
-		   const char *volume, loc_t *loc, const char *basename,
-		   entrylk_cmd cmd, entrylk_type type, dict_t *xdata)
+error_gen_entrylk(call_frame_t *frame, xlator_t *this, const char *volume,
+                  loc_t *loc, const char *basename, entrylk_cmd cmd,
+                  entrylk_type type, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_ENTRYLK];
+    egp = this->private;
+    enable = egp->enable[GF_FOP_ENTRYLK];
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_ENTRYLK);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_ENTRYLK);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (entrylk, frame, -1, op_errno, xdata);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(entrylk, frame, -1, op_errno, xdata);
         return 0;
-	}
+    }
 
-	STACK_WIND (frame, error_gen_entrylk_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->entrylk,
-		    volume, loc, basename, cmd, type, xdata);
-        return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->entrylk,
+                    volume, loc, basename, cmd, type, xdata);
+    return 0;
 }
 
-
 int
-error_gen_fentrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+error_gen_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume,
+                   fd_t *fd, const char *basename, entrylk_cmd cmd,
+                   entrylk_type type, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (fentrylk, frame, op_ret, op_errno, xdata);
-        return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FENTRYLK];
 
-int
-error_gen_fentrylk (call_frame_t *frame, xlator_t *this,
-		    const char *volume, fd_t *fd, const char *basename,
-		    entrylk_cmd cmd, entrylk_type type, dict_t *xdata)
-{
-	int             op_errno = 0;
-        eg_t            *egp = NULL;
-        int             enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_FENTRYLK];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_FENTRYLK);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (fentrylk, frame, -1, op_errno, xdata);
-                return 0;
-	}
-
-	STACK_WIND (frame, error_gen_fentrylk_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->fentrylk,
-		    volume, fd, basename, cmd, type, xdata);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FENTRYLK);
+
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(fentrylk, frame, -1, op_errno, xdata);
         return 0;
+    }
+
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->fentrylk,
+                    volume, fd, basename, cmd, type, xdata);
+    return 0;
 }
 
+int
+error_gen_getspec(call_frame_t *frame, xlator_t *this, const char *key,
+                  int32_t flags)
+{
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-/* Management operations */
+    egp = this->private;
+    enable = egp->enable[GF_FOP_GETSPEC];
 
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_GETSPEC);
 
-int
-error_gen_getspec_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int32_t op_ret, int32_t op_errno, char *spec_data)
-{
-	STACK_UNWIND_STRICT (getspec, frame, op_ret, op_errno, spec_data);
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(getspec, frame, -1, op_errno, NULL);
         return 0;
-}
+    }
 
-
-int
-error_gen_getspec (call_frame_t *frame, xlator_t *this, const char *key,
-		   int32_t flags)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_GETSPEC];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_GETSPEC);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (getspec, frame, -1, op_errno, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_getspec_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->getspec,
-		    key, flags);
-	return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->getspec,
+                    key, flags);
+    return 0;
 }
 
-
 int
-error_gen_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int32_t op_ret, int32_t op_errno, gf_dirent_t *entries,
-                       dict_t *xdata)
+error_gen_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                  off_t off, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (readdir, frame, op_ret, op_errno, entries, xdata);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_READDIR];
 
-int
-error_gen_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd,
-		   size_t size, off_t off, dict_t *xdata)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_READDIR];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_READDIR);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (readdir, frame, -1, op_errno, NULL, xdata);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_readdir_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->readdir,
-		    fd, size, off, xdata);
-	return 0;
-}
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_READDIR);
 
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(readdir, frame, -1, op_errno, NULL, xdata);
+        return 0;
+    }
 
-int
-error_gen_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, gf_dirent_t *entries,
-                        dict_t *xdata)
-{
-	STACK_UNWIND_STRICT (readdirp, frame, op_ret, op_errno, entries, xdata);
-	return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->readdir,
+                    fd, size, off, xdata);
+    return 0;
 }
 
-
 int
-error_gen_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-                    off_t off, dict_t *dict)
+error_gen_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                   off_t off, dict_t *dict)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_READDIRP];
+    egp = this->private;
+    enable = egp->enable[GF_FOP_READDIRP];
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_READDIRP);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_READDIRP);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (readdirp, frame, -1, op_errno, NULL, NULL);
-		return 0;
-	}
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(readdirp, frame, -1, op_errno, NULL, NULL);
+        return 0;
+    }
 
-	STACK_WIND (frame, error_gen_readdirp_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->readdirp,
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->readdirp,
                     fd, size, off, dict);
-	return 0;
+    return 0;
 }
 
 static void
-error_gen_set_failure (eg_t *pvt, int percent)
+error_gen_set_failure(eg_t *pvt, double percent)
 {
-        GF_ASSERT (pvt);
+    double ppm;
+
+    GF_ASSERT(pvt);
 
-        if (percent)
-                pvt->failure_iter_no = 100/percent;
-        else
-                pvt->failure_iter_no = 100/GF_FAILURE_DEFAULT;
+    ppm = (percent / 100.0) * (double)FAILURE_GRANULARITY;
+    pvt->failure_iter_no = (int)ppm;
 }
 
 static void
-error_gen_parse_fill_fops (eg_t *pvt, char *enable_fops)
+error_gen_parse_fill_fops(eg_t *pvt, char *enable_fops)
 {
-        char            *op_no_str = NULL;
-        int              op_no = -1;
-        int              i = 0;
-        xlator_t        *this = THIS;
-        char            *saveptr = NULL;
+    char *op_no_str = NULL;
+    int op_no = -1;
+    int i = 0;
+    xlator_t *this = THIS;
+    char *saveptr = NULL;
 
-        GF_ASSERT (pvt);
-        GF_ASSERT (this);
+    GF_ASSERT(pvt);
+    GF_ASSERT(this);
 
+    for (i = 0; i < GF_FOP_MAXVALUE; i++)
+        pvt->enable[i] = 0;
+
+    if (!enable_fops) {
+        gf_log(this->name, GF_LOG_WARNING, "All fops are enabled.");
         for (i = 0; i < GF_FOP_MAXVALUE; i++)
-                pvt->enable[i] = 0;
-
-        if (!enable_fops) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "All fops are enabled.");
-                for (i = 0; i < GF_FOP_MAXVALUE; i++)
-                        pvt->enable[i] = 1;
-        } else {
-                op_no_str = strtok_r (enable_fops, ",", &saveptr);
-                while (op_no_str) {
-                        op_no = get_fop_int (&op_no_str);
-                        if (op_no == -1) {
-                                gf_log (this->name, GF_LOG_WARNING,
-                                        "Wrong option value %s", op_no_str);
-                        } else
-                                pvt->enable[op_no] = 1;
-
-                        op_no_str = strtok_r (NULL, ",", &saveptr);
-                }
+            pvt->enable[i] = 1;
+    } else {
+        op_no_str = strtok_r(enable_fops, ",", &saveptr);
+        while (op_no_str) {
+            op_no = gf_fop_int(op_no_str);
+            if (op_no == -1) {
+                gf_log(this->name, GF_LOG_WARNING, "Wrong option value %s",
+                       op_no_str);
+            } else
+                pvt->enable[op_no] = 1;
+
+            op_no_str = strtok_r(NULL, ",", &saveptr);
         }
+    }
 }
 
 int32_t
-error_gen_priv_dump (xlator_t *this)
+error_gen_priv_dump(xlator_t *this)
 {
-        char            key_prefix[GF_DUMP_MAX_BUF_LEN];
-        int             ret = -1;
-        eg_t            *conf = NULL;
+    char key_prefix[GF_DUMP_MAX_BUF_LEN];
+    int ret = -1;
+    eg_t *conf = NULL;
 
-        if (!this)
-            goto out;
+    if (!this)
+        goto out;
 
-        conf = this->private;
-        if (!conf)
-            goto out;
+    conf = this->private;
+    if (!conf)
+        goto out;
 
-        ret = TRY_LOCK(&conf->lock);
-        if (ret != 0) {
-                return ret;
-        }
+    ret = TRY_LOCK(&conf->lock);
+    if (ret != 0) {
+        return ret;
+    }
 
-        gf_proc_dump_add_section("xlator.debug.error-gen.%s.priv", this->name);
-        gf_proc_dump_build_key(key_prefix,"xlator.debug.error-gen","%s.priv",
-                               this->name);
+    gf_proc_dump_add_section("xlator.debug.error-gen.%s.priv", this->name);
+    gf_proc_dump_build_key(key_prefix, "xlator.debug.error-gen", "%s.priv",
+                           this->name);
 
-        gf_proc_dump_write("op_count", "%d", conf->op_count);
-        gf_proc_dump_write("failure_iter_no", "%d", conf->failure_iter_no);
-        gf_proc_dump_write("error_no", "%s", conf->error_no);
-        gf_proc_dump_write("random_failure", "%d", conf->random_failure);
+    gf_proc_dump_write("op_count", "%d", conf->op_count);
+    gf_proc_dump_write("failure_iter_no", "%d", conf->failure_iter_no);
+    gf_proc_dump_write("error_no", "%d", conf->error_no_int);
+    gf_proc_dump_write("random_failure", "%d", conf->random_failure);
 
-        UNLOCK(&conf->lock);
+    UNLOCK(&conf->lock);
 out:
-        return ret;
+    return ret;
 }
 
 int32_t
-mem_acct_init (xlator_t *this)
+mem_acct_init(xlator_t *this)
 {
-        int     ret = -1;
-
-        if (!this)
-                return ret;
+    int ret = -1;
 
-        ret = xlator_mem_acct_init (this, gf_error_gen_mt_end + 1);
+    if (!this)
+        return ret;
 
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
-                        " failed");
-                return ret;
-        }
+    ret = xlator_mem_acct_init(this, gf_error_gen_mt_end + 1);
 
+    if (ret != 0) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Memory accounting init"
+               " failed");
         return ret;
+    }
+
+    return ret;
 }
 
 int
-reconfigure (xlator_t *this, dict_t *options)
+reconfigure(xlator_t *this, dict_t *options)
 {
-        eg_t            *pvt = NULL;
-        int32_t          ret = 0;
-        char            *error_enable_fops = NULL;
-        int32_t          failure_percent_int = 0;
+    eg_t *pvt = NULL;
+    int32_t ret = 0;
+    char *error_enable_fops = NULL;
+    char *error_no = NULL;
+    double failure_percent_dbl = 0.0;
 
-        if (!this || !this->private)
-                goto out;
+    if (!this || !this->private)
+        goto out;
 
-        pvt = this->private;
+    pvt = this->private;
 
-        GF_OPTION_RECONF ("error-no", pvt->error_no, options, str, out);
+    ret = -1;
 
-        GF_OPTION_RECONF ("failure", failure_percent_int, options, int32,
-                          out);
+    GF_OPTION_RECONF("error-no", error_no, options, str, out);
 
-        GF_OPTION_RECONF ("enable", error_enable_fops, options, str, out);
+    if (error_no)
+        pvt->error_no_int = conv_errno_to_int(&error_no);
 
-        GF_OPTION_RECONF ("random-failure", pvt->random_failure, options,
-                          bool, out);
+    GF_OPTION_RECONF("failure", failure_percent_dbl, options, percent, out);
 
-        error_gen_parse_fill_fops (pvt, error_enable_fops);
-        error_gen_set_failure (pvt, failure_percent_int);
+    GF_OPTION_RECONF("enable", error_enable_fops, options, str, out);
 
-        ret = 0;
+    GF_OPTION_RECONF("random-failure", pvt->random_failure, options, bool, out);
+
+    error_gen_parse_fill_fops(pvt, error_enable_fops);
+    error_gen_set_failure(pvt, failure_percent_dbl);
+
+    ret = 0;
 out:
-        gf_log (this->name, GF_LOG_DEBUG, "reconfigure returning %d", ret);
-        return ret;
+    gf_log(this ? this->name : "error-gen", GF_LOG_DEBUG,
+           "reconfigure returning %d", ret);
+    return ret;
 }
 
 int
-init (xlator_t *this)
+init(xlator_t *this)
 {
-        eg_t            *pvt = NULL;
-        int32_t          ret = 0;
-        char            *error_enable_fops = NULL;
-        int32_t          failure_percent_int = 0;
-
-        if (!this->children || this->children->next) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "error-gen not configured with one subvolume");
-                ret = -1;
-                goto out;
-        }
+    eg_t *pvt = NULL;
+    int32_t ret = 0;
+    char *error_enable_fops = NULL;
+    char *error_no = NULL;
+    double failure_percent_dbl = 0.0;
 
-        if (!this->parents) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "dangling volume. check volfile ");
-        }
+    if (!this->children || this->children->next) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "error-gen not configured with one subvolume");
+        ret = -1;
+        goto out;
+    }
 
-        pvt = GF_CALLOC (1, sizeof (eg_t), gf_error_gen_mt_eg_t);
+    if (!this->parents) {
+        gf_log(this->name, GF_LOG_WARNING, "dangling volume. check volfile ");
+    }
 
-        if (!pvt) {
-                ret = -1;
-                goto out;
-        }
+    pvt = GF_CALLOC(1, sizeof(eg_t), gf_error_gen_mt_eg_t);
+
+    if (!pvt) {
+        ret = -1;
+        goto out;
+    }
+
+    LOCK_INIT(&pvt->lock);
 
-        LOCK_INIT (&pvt->lock);
+    ret = -1;
 
-        GF_OPTION_INIT ("error-no", pvt->error_no, str, out);
+    GF_OPTION_INIT("error-no", error_no, str, out);
 
-        GF_OPTION_INIT ("failure", failure_percent_int, int32, out);
+    if (error_no)
+        pvt->error_no_int = conv_errno_to_int(&error_no);
 
-        GF_OPTION_INIT ("enable", error_enable_fops, str, out);
+    GF_OPTION_INIT("failure", failure_percent_dbl, percent, out);
 
-        GF_OPTION_INIT ("random-failure", pvt->random_failure, bool, out);
+    GF_OPTION_INIT("enable", error_enable_fops, str, out);
 
+    GF_OPTION_INIT("random-failure", pvt->random_failure, bool, out);
 
-        error_gen_parse_fill_fops (pvt, error_enable_fops);
-        error_gen_set_failure (pvt, failure_percent_int);
+    error_gen_parse_fill_fops(pvt, error_enable_fops);
+    error_gen_set_failure(pvt, failure_percent_dbl);
 
-        this->private = pvt;
+    this->private = pvt;
 
-        /* Give some seed value here */
-        srand (time(NULL));
+    /* Give some seed value here. */
+    srand(gf_time());
+
+    ret = 0;
 out:
-        if (ret)
-                GF_FREE (pvt);
-        return ret;
+    if (ret)
+        GF_FREE(pvt);
+    return ret;
 }
 
-
 void
-fini (xlator_t *this)
+fini(xlator_t *this)
 {
-        eg_t            *pvt = NULL;
+    eg_t *pvt = NULL;
 
-        if (!this)
-                 return;
-        pvt = this->private;
-
-        if (pvt) {
-                LOCK_DESTROY (&pvt->lock);
-                GF_FREE (pvt);
-                gf_log (this->name, GF_LOG_DEBUG, "fini called");
-        }
+    if (!this)
         return;
+    pvt = this->private;
+
+    if (pvt) {
+        LOCK_DESTROY(&pvt->lock);
+        GF_FREE(pvt);
+        gf_log(this->name, GF_LOG_DEBUG, "fini called");
+    }
+    return;
 }
 
 struct xlator_dumpops dumpops = {
-        .priv = error_gen_priv_dump,
+    .priv = error_gen_priv_dump,
 };
 
-struct xlator_fops cbks;
+struct xlator_cbks cbks;
 
 struct xlator_fops fops = {
-	.lookup      = error_gen_lookup,
-	.stat        = error_gen_stat,
-	.readlink    = error_gen_readlink,
-	.mknod       = error_gen_mknod,
-	.mkdir       = error_gen_mkdir,
-	.unlink      = error_gen_unlink,
-	.rmdir       = error_gen_rmdir,
-	.symlink     = error_gen_symlink,
-	.rename      = error_gen_rename,
-	.link        = error_gen_link,
-	.truncate    = error_gen_truncate,
-	.create      = error_gen_create,
-	.open        = error_gen_open,
-	.readv       = error_gen_readv,
-	.writev      = error_gen_writev,
-	.statfs      = error_gen_statfs,
-	.flush       = error_gen_flush,
-	.fsync       = error_gen_fsync,
-	.setxattr    = error_gen_setxattr,
-	.getxattr    = error_gen_getxattr,
-	.removexattr = error_gen_removexattr,
-	.fsetxattr    = error_gen_fsetxattr,
-	.fgetxattr    = error_gen_fgetxattr,
-	.fremovexattr = error_gen_fremovexattr,
-	.opendir     = error_gen_opendir,
-	.readdir     = error_gen_readdir,
-	.readdirp    = error_gen_readdirp,
-	.fsyncdir    = error_gen_fsyncdir,
-	.access      = error_gen_access,
-	.ftruncate   = error_gen_ftruncate,
-	.fstat       = error_gen_fstat,
-	.lk          = error_gen_lk,
-	.lookup_cbk  = error_gen_lookup_cbk,
-	.xattrop     = error_gen_xattrop,
-	.fxattrop    = error_gen_fxattrop,
-	.inodelk     = error_gen_inodelk,
-	.finodelk    = error_gen_finodelk,
-	.entrylk     = error_gen_entrylk,
-	.fentrylk    = error_gen_fentrylk,
-        .setattr     = error_gen_setattr,
-        .fsetattr    = error_gen_fsetattr,
-	.getspec     = error_gen_getspec,
+    .lookup = error_gen_lookup,
+    .stat = error_gen_stat,
+    .readlink = error_gen_readlink,
+    .mknod = error_gen_mknod,
+    .mkdir = error_gen_mkdir,
+    .unlink = error_gen_unlink,
+    .rmdir = error_gen_rmdir,
+    .symlink = error_gen_symlink,
+    .rename = error_gen_rename,
+    .link = error_gen_link,
+    .truncate = error_gen_truncate,
+    .create = error_gen_create,
+    .open = error_gen_open,
+    .readv = error_gen_readv,
+    .writev = error_gen_writev,
+    .statfs = error_gen_statfs,
+    .flush = error_gen_flush,
+    .fsync = error_gen_fsync,
+    .setxattr = error_gen_setxattr,
+    .getxattr = error_gen_getxattr,
+    .removexattr = error_gen_removexattr,
+    .fsetxattr = error_gen_fsetxattr,
+    .fgetxattr = error_gen_fgetxattr,
+    .fremovexattr = error_gen_fremovexattr,
+    .opendir = error_gen_opendir,
+    .readdir = error_gen_readdir,
+    .readdirp = error_gen_readdirp,
+    .fsyncdir = error_gen_fsyncdir,
+    .access = error_gen_access,
+    .ftruncate = error_gen_ftruncate,
+    .fstat = error_gen_fstat,
+    .lk = error_gen_lk,
+    .xattrop = error_gen_xattrop,
+    .fxattrop = error_gen_fxattrop,
+    .inodelk = error_gen_inodelk,
+    .finodelk = error_gen_finodelk,
+    .entrylk = error_gen_entrylk,
+    .fentrylk = error_gen_fentrylk,
+    .setattr = error_gen_setattr,
+    .fsetattr = error_gen_fsetattr,
+    .getspec = error_gen_getspec,
 };
 
 struct volume_options options[] = {
-        { .key  = {"failure"},
-          .type = GF_OPTION_TYPE_INT,
-          .description = "Percentage failure of operations when enabled.",
-        },
-
-        { .key  = {"error-no"},
-          .value = {"ENOENT","ENOTDIR","ENAMETOOLONG","EACCES","EBADF",
-                    "EFAULT","ENOMEM","EINVAL","EIO","EEXIST","ENOSPC",
-                    "EPERM","EROFS","EBUSY","EISDIR","ENOTEMPTY","EMLINK"
-                    "ENODEV","EXDEV","EMFILE","ENFILE","ENOSYS","EINTR",
-                    "EFBIG","EAGAIN","GF_ERROR_SHORT_WRITE"},
-          .type = GF_OPTION_TYPE_STR,
-        },
-
-        { .key = {"random-failure"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "off",
-        },
-
-        { .key  = {"enable"},
-          .type = GF_OPTION_TYPE_STR,
-        },
-
-        { .key  = {NULL} }
+    {
+        .key = {"failure"},
+        .type = GF_OPTION_TYPE_PERCENT,
+        .description = "Percentage failure of operations when enabled.",
+    },
+
+    {
+        .key = {"error-no"},
+        .value = {"ENOENT",
+                  "ENOTDIR",
+                  "ENAMETOOLONG",
+                  "EACCES",
+                  "EBADF",
+                  "EFAULT",
+                  "ENOMEM",
+                  "EINVAL",
+                  "EIO",
+                  "EEXIST",
+                  "ENOSPC",
+                  "EPERM",
+                  "EROFS",
+                  "EBUSY",
+                  "EISDIR",
+                  "ENOTEMPTY",
+                  "EMLINK"
+                  "ENODEV",
+                  "EXDEV",
+                  "EMFILE",
+                  "ENFILE",
+                  "ENOSYS",
+                  "EINTR",
+                  "EFBIG",
+                  "EAGAIN",
+                  "GF_ERROR_SHORT_WRITE"},
+        .type = GF_OPTION_TYPE_STR,
+        .op_version = {3},
+        .tags = {"error-gen"},
+        .flags = OPT_FLAG_SETTABLE,
+
+    },
+
+    {
+        .key = {"random-failure"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "off",
+        .op_version = {3},
+        .tags = {"error-gen"},
+        .flags = OPT_FLAG_SETTABLE,
+    },
+
+    {
+        .key = {"enable", "error-fops"},
+        .type = GF_OPTION_TYPE_STR,
+        .description = "Accepts a string which takes ',' separated fop "
+                       "strings to denote which fops are enabled for error",
+        .op_version = {3},
+        .tags = {"error-gen"},
+        .flags = OPT_FLAG_SETTABLE,
+    },
+
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1},
+    .dumpops = &dumpops,
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "error-gen",
+    .category = GF_TECH_PREVIEW,
 };
diff --git a/xlators/debug/error-gen/src/error-gen.h b/xlators/debug/error-gen/src/error-gen.h
index d92c2306240..2478cd5b21c 100644
--- a/xlators/debug/error-gen/src/error-gen.h
+++ b/xlators/debug/error-gen/src/error-gen.h
@@ -10,11 +10,6 @@
 #ifndef _ERROR_GEN_H
 #define _ERROR_GEN_H
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
 #include "error-gen-mem-types.h"
 
 #define GF_FAILURE_DEFAULT 10
@@ -27,22 +22,28 @@
  * range.
  */
 enum GF_PSEUDO_ERRORS {
-	GF_ERROR_SHORT_WRITE = 1000,	/* short writev return value */
-	GF_ERROR_MAX
+    GF_ERROR_SHORT_WRITE = 1000, /* short writev return value */
+    GF_ERROR_MAX
 };
 
 typedef struct {
-        int enable[GF_FOP_MAXVALUE];
-        int op_count;
-        int failure_iter_no;
-        char *error_no;
-        gf_boolean_t random_failure;
-        gf_lock_t lock;
+    int enable[GF_FOP_MAXVALUE];
+    int op_count;
+    /*
+     * This is only an iteration number in the random-failure case.  For
+     * the normal controlled-probability case, it's actually a numerator
+     * for the failure probability (see FAILURE_GRANULARITY declaration).
+     * It's just not worth blowing up the diff by changing it.
+     */
+    int failure_iter_no;
+    int error_no_int;
+    gf_boolean_t random_failure;
+    gf_lock_t lock;
 } eg_t;
 
 typedef struct {
-        int error_no_count;
-	int error_no[20];
+    int error_no_count;
+    int error_no[20];
 } sys_error_t;
 
 #endif
diff --git a/xlators/debug/io-stats/src/Makefile.am b/xlators/debug/io-stats/src/Makefile.am
index 332d79015e9..c69f3caf0fe 100644
--- a/xlators/debug/io-stats/src/Makefile.am
+++ b/xlators/debug/io-stats/src/Makefile.am
@@ -2,14 +2,17 @@
 xlator_LTLIBRARIES = io-stats.la
 xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/debug
 
-io_stats_la_LDFLAGS = -module -avoid-version
+io_stats_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
 
 io_stats_la_SOURCES = io-stats.c
 io_stats_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
 noinst_HEADERS = io-stats-mem-types.h
 
-AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src \
+	-I$(top_srcdir)/rpc/rpc-lib/src \
+	-DDATADIR=\"$(localstatedir)\"
 
 AM_CFLAGS = -Wall $(GF_CFLAGS)
 
diff --git a/xlators/debug/io-stats/src/io-stats-mem-types.h b/xlators/debug/io-stats/src/io-stats-mem-types.h
index c30dfb17e7c..51d38d8b97c 100644
--- a/xlators/debug/io-stats/src/io-stats-mem-types.h
+++ b/xlators/debug/io-stats/src/io-stats-mem-types.h
@@ -11,14 +11,17 @@
 #ifndef __IO_STATS_MEM_TYPES_H__
 #define __IO_STATS_MEM_TYPES_H__
 
-#include "mem-types.h"
+#include <glusterfs/mem-types.h>
+
+extern const char *__progname;
 
 enum gf_io_stats_mem_types_ {
-        gf_io_stats_mt_ios_conf = gf_common_mt_end + 1,
-        gf_io_stats_mt_ios_fd,
-        gf_io_stats_mt_ios_stat,
-        gf_io_stats_mt_ios_stat_list,
-        gf_io_stats_mt_end
+    gf_io_stats_mt_ios_conf = gf_common_mt_end + 1,
+    gf_io_stats_mt_ios_fd,
+    gf_io_stats_mt_ios_stat,
+    gf_io_stats_mt_ios_stat_list,
+    gf_io_stats_mt_ios_sample_buf,
+    gf_io_stats_mt_ios_sample,
+    gf_io_stats_mt_end
 };
 #endif
-
diff --git a/xlators/debug/io-stats/src/io-stats.c b/xlators/debug/io-stats/src/io-stats.c
index 4f27a2e418f..aa00c446e5a 100644
--- a/xlators/debug/io-stats/src/io-stats.c
+++ b/xlators/debug/io-stats/src/io-stats.c
@@ -7,11 +7,8 @@
    later), or the GNU General Public License, version 2 (GPLv2), in all
    cases as published by the Free Software Foundation.
 */
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#include "xlator.h"
-#endif
+#include <glusterfs/xlator.h>
+#include <glusterfs/syscall.h>
 
 /**
  * xlators/debug/io_stats :
@@ -20,2908 +17,4464 @@
  *
  *  a) total read data - since process start, last interval and per fd
  *  b) total write data - since process start, last interval and per fd
- *  c) counts of read IO block size - since process start, last interval and per fd
- *  d) counts of write IO block size - since process start, last interval and per fd
- *  e) counts of all FOP types passing through it
+ *  c) counts of read IO block size - since process start, last interval and per
+ * fd d) counts of write IO block size - since process start, last interval and
+ * per fd e) counts of all FOP types passing through it
  *
- *  Usage: setfattr -n io-stats-dump /tmp/filename /mnt/gluster
+ *  Usage: setfattr -n trusted.io-stats-dump /tmp/filename /mnt/gluster
+ *      output is written to /tmp/filename.<iostats xlator instance name>
  *
  */
 
 #include <fnmatch.h>
 #include <errno.h>
-#include "glusterfs.h"
-#include "xlator.h"
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/xlator.h>
 #include "io-stats-mem-types.h"
 #include <stdarg.h>
-#include "defaults.h"
-#include "logging.h"
+#include <glusterfs/defaults.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/statedump.h>
+#include <glusterfs/syncop.h>
+#include <pwd.h>
+#include <grp.h>
+#include <glusterfs/upcall-utils.h>
+#include <glusterfs/async.h>
 
 #define MAX_LIST_MEMBERS 100
+#define DEFAULT_PWD_BUF_SZ 16384
+#define DEFAULT_GRP_BUF_SZ 16384
+#define IOS_BLOCK_COUNT_SIZE 32
+
+#define IOS_STATS_DUMP_DIR DEFAULT_VAR_RUN_DIRECTORY
 
 typedef enum {
-        IOS_STATS_TYPE_NONE,
-        IOS_STATS_TYPE_OPEN,
-        IOS_STATS_TYPE_READ,
-        IOS_STATS_TYPE_WRITE,
-        IOS_STATS_TYPE_OPENDIR,
-        IOS_STATS_TYPE_READDIRP,
-        IOS_STATS_TYPE_READ_THROUGHPUT,
-        IOS_STATS_TYPE_WRITE_THROUGHPUT,
-        IOS_STATS_TYPE_MAX
-}ios_stats_type_t;
+    IOS_STATS_TYPE_NONE,
+    IOS_STATS_TYPE_OPEN,
+    IOS_STATS_TYPE_READ,
+    IOS_STATS_TYPE_WRITE,
+    IOS_STATS_TYPE_OPENDIR,
+    IOS_STATS_TYPE_READDIRP,
+    IOS_STATS_TYPE_READ_THROUGHPUT,
+    IOS_STATS_TYPE_WRITE_THROUGHPUT,
+    IOS_STATS_TYPE_MAX
+} ios_stats_type_t;
 
 typedef enum {
-        IOS_STATS_THRU_READ,
-        IOS_STATS_THRU_WRITE,
-        IOS_STATS_THRU_MAX,
-}ios_stats_thru_t;
+    IOS_STATS_THRU_READ,
+    IOS_STATS_THRU_WRITE,
+    IOS_STATS_THRU_MAX,
+} ios_stats_thru_t;
+
+/* This is same as gf1_cli_info_op */
+/* had to be defined here again, so we have modularity between
+ xdr, xlator, and library functions */
+typedef enum ios_info_op {
+    GF_IOS_INFO_NONE = 0,
+    GF_IOS_INFO_ALL = 1,
+    GF_IOS_INFO_INCREMENTAL = 2,
+    GF_IOS_INFO_CUMULATIVE = 3,
+    GF_IOS_INFO_CLEAR = 4,
+} ios_info_op_t;
 
 struct ios_stat_lat {
-        struct timeval  time;
-        double          throughput;
+    struct timeval time;
+    double throughput;
 };
 
 struct ios_stat {
-        gf_lock_t       lock;
-        uuid_t          gfid;
-        char           *filename;
-        uint64_t        counters [IOS_STATS_TYPE_MAX];
-        struct ios_stat_lat thru_counters [IOS_STATS_THRU_MAX];
-        int             refcnt;
+    gf_lock_t lock;
+    uuid_t gfid;
+    char *filename;
+    gf_atomic_t counters[IOS_STATS_TYPE_MAX];
+    struct ios_stat_lat thru_counters[IOS_STATS_THRU_MAX];
+    gf_atomic_t refcnt;
 };
 
 struct ios_stat_list {
-        struct list_head  list;
-        struct ios_stat  *iosstat;
-        double            value;
+    struct list_head list;
+    struct ios_stat *iosstat;
+    double value;
 };
 
 struct ios_stat_head {
-       gf_lock_t                lock;
-       double                   min_cnt;
-       uint64_t                 members;
-       struct ios_stat_list    *iosstats;
+    gf_lock_t lock;
+    double min_cnt;
+    uint64_t members;
+    struct ios_stat_list *iosstats;
 };
 
+typedef struct _ios_sample_t {
+    uid_t uid;
+    gid_t gid;
+    char identifier[UNIX_PATH_MAX];
+    glusterfs_fop_t fop_type;
+    struct timeval timestamp;
+    double elapsed;
+} ios_sample_t;
+
+typedef struct _ios_sample_buf_t {
+    uint64_t pos;              /* Position in write buffer */
+    uint64_t size;             /* Size of ring buffer */
+    uint64_t collected;        /* Number of samples we've collected */
+    uint64_t observed;         /* Number of FOPs we've observed */
+    ios_sample_t *ios_samples; /* Our list of samples */
+} ios_sample_buf_t;
+
 struct ios_lat {
-        double  min;
-        double  max;
-        double  avg;
+    double min;
+    double max;
+    double avg;
+    uint64_t total;
 };
 
 struct ios_global_stats {
-        uint64_t        data_written;
-        uint64_t        data_read;
-        uint64_t        block_count_write[32];
-        uint64_t        block_count_read[32];
-        uint64_t        fop_hits[GF_FOP_MAXVALUE];
-        struct timeval  started_at;
-        struct ios_lat  latency[GF_FOP_MAXVALUE];
-        uint64_t        nr_opens;
-        uint64_t        max_nr_opens;
-        struct timeval  max_openfd_time;
+    gf_atomic_t data_written;
+    gf_atomic_t data_read;
+    gf_atomic_t block_count_write[IOS_BLOCK_COUNT_SIZE];
+    gf_atomic_t block_count_read[IOS_BLOCK_COUNT_SIZE];
+    gf_atomic_t fop_hits[GF_FOP_MAXVALUE];
+    gf_atomic_t upcall_hits[GF_UPCALL_FLAGS_MAXVALUE];
+    time_t started_at;
+    struct ios_lat latency[GF_FOP_MAXVALUE];
+    uint64_t nr_opens;
+    uint64_t max_nr_opens;
+    struct timeval max_openfd_time;
 };
 
+typedef enum {
+    IOS_DUMP_TYPE_NONE = 0,
+    IOS_DUMP_TYPE_FILE = 1,
+    IOS_DUMP_TYPE_DICT = 2,
+    IOS_DUMP_TYPE_JSON_FILE = 3,
+    IOS_DUMP_TYPE_SAMPLES = 4,
+    IOS_DUMP_TYPE_MAX = 5
+} ios_dump_type_t;
 
 struct ios_conf {
-        gf_lock_t                 lock;
-        struct ios_global_stats   cumulative;
-        uint64_t                  increment;
-        struct ios_global_stats   incremental;
-        gf_boolean_t              dump_fd_stats;
-        gf_boolean_t              count_fop_hits;
-        gf_boolean_t              measure_latency;
-        struct ios_stat_head      list[IOS_STATS_TYPE_MAX];
-        struct ios_stat_head      thru_list[IOS_STATS_THRU_MAX];
+    gf_lock_t lock;
+    struct ios_global_stats cumulative;
+    uint64_t increment;
+    struct ios_global_stats incremental;
+    gf_boolean_t dump_fd_stats;
+    gf_boolean_t count_fop_hits;
+    gf_boolean_t measure_latency;
+    struct ios_stat_head list[IOS_STATS_TYPE_MAX];
+    struct ios_stat_head thru_list[IOS_STATS_THRU_MAX];
+    int32_t ios_dump_interval;
+    pthread_t dump_thread;
+    gf_boolean_t dump_thread_should_die;
+    gf_boolean_t dump_thread_running;
+    gf_lock_t ios_sampling_lock;
+    int32_t ios_sample_interval;
+    int32_t ios_sample_buf_size;
+    ios_sample_buf_t *ios_sample_buf;
+    struct dnscache *dnscache;
+    int32_t ios_dnscache_ttl_sec;
+    /*
+     * What we really need here is just a unique value to keep files
+     * created by this instance distinct from those created by any other.
+     * On the client side this isn't a problem, so we just use the
+     * translator name.  On the server side conflicts can occur, so the
+     * volfile-generation code automatically sets this (via an option)
+     * to be the brick path.
+     *
+     * NB While the *field* name has changed, it didn't seem worth changing
+     * all of the cases where "xlator_name" is used as a *variable* name.
+     */
+    char *unique_id;
+    ios_dump_type_t dump_format;
 };
 
-
 struct ios_fd {
-        char           *filename;
-        uint64_t        data_written;
-        uint64_t        data_read;
-        uint64_t        block_count_write[32];
-        uint64_t        block_count_read[32];
-        struct timeval  opened_at;
+    char *filename;
+    gf_atomic_t data_written;
+    gf_atomic_t data_read;
+    gf_atomic_t block_count_write[IOS_BLOCK_COUNT_SIZE];
+    gf_atomic_t block_count_read[IOS_BLOCK_COUNT_SIZE];
+    struct timeval opened_at;
 };
 
-typedef enum {
-        IOS_DUMP_TYPE_NONE = 0,
-        IOS_DUMP_TYPE_FILE = 1,
-        IOS_DUMP_TYPE_DICT = 2,
-        IOS_DUMP_TYPE_MAX  = 3
-} ios_dump_type_t;
-
 struct ios_dump_args {
-        ios_dump_type_t type;
-        union {
-                FILE *logfp;
-                dict_t *dict;
-        } u;
+    ios_dump_type_t type;
+    union {
+        FILE *logfp;
+        dict_t *dict;
+    } u;
 };
 
-typedef int (*block_dump_func) (xlator_t *, struct ios_dump_args*,
-                                    int , int , uint64_t ) ;
+typedef int (*block_dump_func)(xlator_t *, struct ios_dump_args *, int, int,
+                               uint64_t);
 
 struct ios_local {
-        struct timeval  wind_at;
-        struct timeval  unwind_at;
+    struct timeval wind_at;
+    struct timeval unwind_at;
 };
 
 struct volume_options options[];
 
-inline static int
-is_fop_latency_started (call_frame_t *frame)
-{
-        GF_ASSERT (frame);
-        struct timeval epoch = {0,};
-        return memcmp (&frame->begin, &epoch, sizeof (epoch));
-}
-
-#define END_FOP_LATENCY(frame, op)                                      \
-        do {                                                            \
-                struct ios_conf  *conf = NULL;                          \
-                                                                        \
-                conf = this->private;                                   \
-                if (conf && conf->measure_latency) {                    \
-                        gettimeofday (&frame->end, NULL);               \
-                        update_ios_latency (conf, frame, GF_FOP_##op);  \
-                }                                                       \
-        } while (0)
-
-#define START_FOP_LATENCY(frame)                                         \
-        do {                                                             \
-                struct ios_conf  *conf = NULL;                           \
-                                                                         \
-                conf = this->private;                                    \
-                if (conf && conf->measure_latency) {                     \
-                        gettimeofday (&frame->begin, NULL);              \
-                } else {                                                 \
-                        memset (&frame->begin, 0, sizeof (frame->begin));\
-                }                                                        \
-        } while (0)
-
-
-#define BUMP_FOP(op)                                                    \
-        do {                                                            \
-                struct ios_conf  *conf = NULL;                          \
-                                                                        \
-                conf = this->private;                                   \
-                if (!conf)                                              \
-                        break;                                          \
-                conf->cumulative.fop_hits[GF_FOP_##op]++;               \
-                conf->incremental.fop_hits[GF_FOP_##op]++;              \
-        } while (0)
-
-#define UPDATE_PROFILE_STATS(frame, op)                                       \
-        do {                                                                  \
-                struct ios_conf  *conf = NULL;                                \
-                                                                              \
-                if (!is_fop_latency_started (frame))                          \
-                        break;                                                \
-                conf = this->private;                                         \
-                LOCK (&conf->lock);                                           \
-                {                                                             \
-                        if (conf && conf->measure_latency &&                  \
-                            conf->count_fop_hits) {                           \
-                                BUMP_FOP(op);                                 \
-                                gettimeofday (&frame->end, NULL);             \
-                                update_ios_latency (conf, frame, GF_FOP_##op);\
-                        }                                                     \
-                }                                                             \
-                UNLOCK (&conf->lock);                                         \
-        } while (0)
-
-#define BUMP_READ(fd, len)                                              \
-        do {                                                            \
-                struct ios_conf  *conf = NULL;                          \
-                struct ios_fd    *iosfd = NULL;                         \
-                int               lb2 = 0;                              \
-                                                                        \
-                conf = this->private;                                   \
-                lb2 = log_base2 (len);                                  \
-                ios_fd_ctx_get (fd, this, &iosfd);                      \
-                if (!conf)                                              \
-                        break;                                          \
-                                                                        \
-                LOCK (&conf->lock);                                     \
-                {                                                       \
-                        conf->cumulative.data_read += len;              \
-                        conf->incremental.data_read += len;             \
-                        conf->cumulative.block_count_read[lb2]++;       \
-                        conf->incremental.block_count_read[lb2]++;      \
-                                                                        \
-                        if (iosfd) {                                    \
-                                iosfd->data_read += len;                \
-                                iosfd->block_count_read[lb2]++;         \
-                        }                                               \
-                }                                                       \
-                UNLOCK (&conf->lock);                                   \
-        } while (0)
-
-
-#define BUMP_WRITE(fd, len)                                             \
-        do {                                                            \
-                struct ios_conf  *conf = NULL;                          \
-                struct ios_fd    *iosfd = NULL;                         \
-                int               lb2 = 0;                              \
-                                                                        \
-                conf = this->private;                                   \
-                lb2 = log_base2 (len);                                  \
-                ios_fd_ctx_get (fd, this, &iosfd);                      \
-                if (!conf)                                              \
-                        break;                                          \
-                                                                        \
-                LOCK (&conf->lock);                                     \
-                {                                                       \
-                        conf->cumulative.data_written += len;           \
-                        conf->incremental.data_written += len;          \
-                        conf->cumulative.block_count_write[lb2]++;      \
-                        conf->incremental.block_count_write[lb2]++;     \
-                                                                        \
-                        if (iosfd) {                                    \
-                                iosfd->data_written += len;             \
-                                iosfd->block_count_write[lb2]++;        \
-                        }                                               \
-                }                                                       \
-                UNLOCK (&conf->lock);                                   \
-        } while (0)
-
-
-#define BUMP_STATS(iosstat, type)                                               \
-        do {                                                                    \
-                struct ios_conf         *conf = NULL;                           \
-                uint64_t                 value = 0;                             \
-                                                                                \
-                conf = this->private;                                           \
-                                                                                \
-                LOCK(&iosstat->lock);                                           \
-                {                                                               \
-                        iosstat->counters[type]++;                              \
-                        value = iosstat->counters[type];                        \
-                }                                                               \
-                UNLOCK (&iosstat->lock);                                        \
-                ios_stat_add_to_list (&conf->list[type],                        \
-                                     value, iosstat);                           \
-                                                                                \
-        } while (0)
-
-
-#define BUMP_THROUGHPUT(iosstat, type)						\
-        do {									\
-                struct ios_conf         *conf = NULL;				\
-                double                   elapsed;				\
-                struct timeval          *begin, *end;				\
-                double                   throughput;				\
-               int                      flag = 0;                              \
-                                                                                \
-                begin = &frame->begin;						\
-                end   = &frame->end;						\
-                                                                                \
-                elapsed = (end->tv_sec - begin->tv_sec) * 1e6			\
-                        + (end->tv_usec - begin->tv_usec);			\
-                throughput = op_ret / elapsed;					\
-                                                                                \
-                conf = this->private;						\
-                LOCK(&iosstat->lock);						\
-                {								\
-                        if (iosstat->thru_counters[type].throughput             \
-                                <= throughput) {                                \
-                                iosstat->thru_counters[type].throughput =       \
-                                                                throughput;     \
-                                gettimeofday (&iosstat->                        \
-                                             thru_counters[type].time, NULL);   \
-                               flag = 1;                                       \
-                        }							\
-                }								\
-                UNLOCK (&iosstat->lock);					\
-               if (flag)                                                       \
-                       ios_stat_add_to_list (&conf->thru_list[type],           \
-                                               throughput, iosstat);           \
-        } while (0)
-
-int
-ios_fd_ctx_get (fd_t *fd, xlator_t *this, struct ios_fd **iosfd)
-{
-        uint64_t      iosfd64 = 0;
-        unsigned long iosfdlong = 0;
-        int           ret = 0;
-
-        ret = fd_ctx_get (fd, this, &iosfd64);
-        iosfdlong = iosfd64;
-        if (ret != -1)
-                *iosfd = (void *) iosfdlong;
-
-        return ret;
+static int
+is_fop_latency_started(call_frame_t *frame)
+{
+    GF_ASSERT(frame);
+    struct timeval epoch = {
+        0,
+    };
+    return memcmp(&frame->begin, &epoch, sizeof(epoch));
 }
 
+#define _IOS_SAMP_DIR DEFAULT_LOG_FILE_DIRECTORY "/samples"
+#ifdef GF_LINUX_HOST_OS
+#define _IOS_DUMP_DIR DATADIR "/lib/glusterd/stats"
+#else
+#define _IOS_DUMP_DIR DATADIR "/db/glusterd/stats"
+#endif
 
+#define END_FOP_LATENCY(frame, op)                                             \
+    do {                                                                       \
+        struct ios_conf *conf = NULL;                                          \
+                                                                               \
+        conf = this->private;                                                  \
+        if (conf && conf->measure_latency) {                                   \
+            timespec_now(&frame->end);                                         \
+            update_ios_latency(conf, frame, GF_FOP_##op);                      \
+        }                                                                      \
+    } while (0)
+
+#define START_FOP_LATENCY(frame)                                               \
+    do {                                                                       \
+        struct ios_conf *conf = NULL;                                          \
+                                                                               \
+        conf = this->private;                                                  \
+        if (conf && conf->measure_latency) {                                   \
+            timespec_now(&frame->begin);                                       \
+        } else {                                                               \
+            memset(&frame->begin, 0, sizeof(frame->begin));                    \
+        }                                                                      \
+    } while (0)
+
+#define BUMP_FOP(op)                                                           \
+    do {                                                                       \
+        struct ios_conf *conf = NULL;                                          \
+                                                                               \
+        conf = this->private;                                                  \
+        if (!conf)                                                             \
+            break;                                                             \
+        GF_ATOMIC_INC(conf->cumulative.fop_hits[GF_FOP_##op]);                 \
+        GF_ATOMIC_INC(conf->incremental.fop_hits[GF_FOP_##op]);                \
+    } while (0)
+
+#define UPDATE_PROFILE_STATS(frame, op)                                        \
+    do {                                                                       \
+        struct ios_conf *conf = NULL;                                          \
+                                                                               \
+        if (!is_fop_latency_started(frame))                                    \
+            break;                                                             \
+        conf = this->private;                                                  \
+        if (conf && conf->measure_latency && conf->count_fop_hits) {           \
+            BUMP_FOP(op);                                                      \
+            timespec_now(&frame->end);                                         \
+            update_ios_latency(conf, frame, GF_FOP_##op);                      \
+        }                                                                      \
+    } while (0)
+
+#define BUMP_THROUGHPUT(iosstat, type)                                         \
+    do {                                                                       \
+        struct ios_conf *conf = NULL;                                          \
+        double elapsed;                                                        \
+        struct timespec *begin, *end;                                          \
+        double throughput;                                                     \
+        int flag = 0;                                                          \
+        struct timeval tv = {                                                  \
+            0,                                                                 \
+        };                                                                     \
+                                                                               \
+        begin = &frame->begin;                                                 \
+        end = &frame->end;                                                     \
+                                                                               \
+        elapsed = gf_tsdiff(begin, end) / 1000.0;                              \
+        throughput = op_ret / elapsed;                                         \
+                                                                               \
+        conf = this->private;                                                  \
+        gettimeofday(&tv, NULL);                                               \
+        LOCK(&iosstat->lock);                                                  \
+        {                                                                      \
+            if (iosstat->thru_counters[type].throughput <= throughput) {       \
+                iosstat->thru_counters[type].throughput = throughput;          \
+                memcpy(&iosstat->thru_counters[type].time, &tv,                \
+                       sizeof(struct timeval));                                \
+                flag = 1;                                                      \
+            }                                                                  \
+        }                                                                      \
+        UNLOCK(&iosstat->lock);                                                \
+        if (flag)                                                              \
+            ios_stat_add_to_list(&conf->thru_list[type], throughput, iosstat); \
+    } while (0)
 
-int
-ios_fd_ctx_set (fd_t *fd, xlator_t *this, struct ios_fd *iosfd)
+static int
+ios_fd_ctx_get(fd_t *fd, xlator_t *this, struct ios_fd **iosfd)
 {
-        uint64_t   iosfd64 = 0;
-        int        ret = 0;
+    uint64_t iosfd64 = 0;
+    unsigned long iosfdlong = 0;
+    int ret = 0;
 
-        iosfd64 = (unsigned long) iosfd;
-        ret = fd_ctx_set (fd, this, iosfd64);
+    ret = fd_ctx_get(fd, this, &iosfd64);
+    iosfdlong = iosfd64;
+    if (ret != -1)
+        *iosfd = (void *)iosfdlong;
 
-        return ret;
+    return ret;
 }
 
-int
-ios_stat_ref (struct ios_stat *iosstat)
+static int
+ios_fd_ctx_set(fd_t *fd, xlator_t *this, struct ios_fd *iosfd)
 {
-        LOCK (&iosstat->lock);
-        {
-                iosstat->refcnt++;
+    uint64_t iosfd64 = 0;
+    int ret = 0;
+
+    iosfd64 = (unsigned long)iosfd;
+    ret = fd_ctx_set(fd, this, iosfd64);
+
+    return ret;
+}
+
+static int
+ios_stat_ref(struct ios_stat *iosstat)
+{
+    uint64_t refcnt = 0;
+    refcnt = GF_ATOMIC_INC(iosstat->refcnt);
+
+    return refcnt;
+}
+
+static int
+ios_stat_unref(struct ios_stat *iosstat)
+{
+    int cleanup = 0;
+    uint64_t refcnt = 0;
+
+    refcnt = GF_ATOMIC_DEC(iosstat->refcnt);
+    if (refcnt == 0) {
+        if (iosstat->filename) {
+            GF_FREE(iosstat->filename);
+            iosstat->filename = NULL;
         }
-        UNLOCK (&iosstat->lock);
+        cleanup = 1;
+    }
+
+    if (cleanup) {
+        LOCK_DESTROY(&iosstat->lock);
+        GF_FREE(iosstat);
+        iosstat = NULL;
+    }
 
-        return iosstat->refcnt;
+    return 0;
 }
 
-int
-ios_stat_unref (struct ios_stat *iosstat)
+static int
+ios_stat_add_to_list(struct ios_stat_head *list_head, uint64_t value,
+                     struct ios_stat *iosstat)
 {
-        int cleanup = 0;
-        LOCK (&iosstat->lock);
+    struct ios_stat_list *new = NULL;
+    struct ios_stat_list *entry = NULL;
+    struct ios_stat_list *t = NULL;
+    struct ios_stat_list *list_entry = NULL;
+    struct ios_stat_list *tmp = NULL;
+    struct ios_stat_list *last = NULL;
+    struct ios_stat *stat = NULL;
+    int cnt = 0;
+    int found = 0;
+    int reposition = 0;
+    double min_count = 0;
+
+    LOCK(&list_head->lock);
+    {
+        if (list_head->min_cnt == 0)
+            list_head->min_cnt = value;
+        if ((list_head->members == MAX_LIST_MEMBERS) &&
+            (list_head->min_cnt > value))
+            goto out;
+
+        list_for_each_entry_safe(entry, t, &list_head->iosstats->list, list)
         {
-                iosstat->refcnt--;
-                if (iosstat->refcnt == 0) {
-                        if (iosstat->filename) {
-                                GF_FREE (iosstat->filename);
-                                iosstat->filename = NULL;
-                        }
-                        cleanup = 1;
+            cnt++;
+            if (cnt == list_head->members)
+                last = entry;
+
+            if (!gf_uuid_compare(iosstat->gfid, entry->iosstat->gfid)) {
+                list_entry = entry;
+                found = cnt;
+                entry->value = value;
+                if (!reposition) {
+                    if (cnt == list_head->members)
+                        list_head->min_cnt = value;
+                    goto out;
                 }
+                break;
+            } else if (entry->value <= value && !reposition) {
+                reposition = cnt;
+                tmp = entry;
+                if (cnt == list_head->members - 1)
+                    min_count = entry->value;
+            }
         }
-        UNLOCK (&iosstat->lock);
-
-        if (cleanup) {
-                GF_FREE (iosstat);
-                iosstat = NULL;
+        if (found) {
+            list_del(&list_entry->list);
+            list_add_tail(&list_entry->list, &tmp->list);
+            if (min_count)
+                list_head->min_cnt = min_count;
+            goto out;
+        } else if (list_head->members == MAX_LIST_MEMBERS && reposition) {
+            new = GF_CALLOC(1, sizeof(*new), gf_io_stats_mt_ios_stat_list);
+            new->iosstat = iosstat;
+            new->value = value;
+            ios_stat_ref(iosstat);
+            list_add_tail(&new->list, &tmp->list);
+            if (last) {
+                stat = last->iosstat;
+                last->iosstat = NULL;
+                ios_stat_unref(stat);
+                list_del(&last->list);
+                GF_FREE(last);
+            }
+            if (reposition == MAX_LIST_MEMBERS)
+                list_head->min_cnt = value;
+            else if (min_count) {
+                list_head->min_cnt = min_count;
+            }
+        } else if (list_head->members < MAX_LIST_MEMBERS) {
+            new = GF_CALLOC(1, sizeof(*new), gf_io_stats_mt_ios_stat_list);
+            new->iosstat = iosstat;
+            new->value = value;
+            ios_stat_ref(iosstat);
+            if (reposition) {
+                list_add_tail(&new->list, &tmp->list);
+            } else {
+                list_add_tail(&new->list, &entry->list);
+            }
+            list_head->members++;
+            if (list_head->min_cnt > value)
+                list_head->min_cnt = value;
         }
+    }
+out:
+    UNLOCK(&list_head->lock);
+    return 0;
+}
 
-        return 0;
+static void
+ios_bump_read(xlator_t *this, fd_t *fd, size_t len)
+{
+    struct ios_conf *conf = NULL;
+    struct ios_fd *iosfd = NULL;
+    int lb2 = 0;
+
+    conf = this->private;
+    lb2 = log_base2(len);
+    ios_fd_ctx_get(fd, this, &iosfd);
+    if (!conf)
+        return;
+
+    GF_ATOMIC_ADD(conf->cumulative.data_read, len);
+    GF_ATOMIC_ADD(conf->incremental.data_read, len);
+    GF_ATOMIC_INC(conf->cumulative.block_count_read[lb2]);
+    GF_ATOMIC_INC(conf->incremental.block_count_read[lb2]);
+
+    if (iosfd) {
+        GF_ATOMIC_ADD(iosfd->data_read, len);
+        GF_ATOMIC_INC(iosfd->block_count_read[lb2]);
+    }
 }
 
-int
-ios_inode_ctx_set (inode_t *inode, xlator_t *this, struct ios_stat *iosstat)
+static void
+ios_bump_write(xlator_t *this, fd_t *fd, size_t len)
 {
-        uint64_t   iosstat64 = 0;
-        int        ret     = 0;
+    struct ios_conf *conf = NULL;
+    struct ios_fd *iosfd = NULL;
+    int lb2 = 0;
+
+    conf = this->private;
+    lb2 = log_base2(len);
+    ios_fd_ctx_get(fd, this, &iosfd);
+    if (!conf)
+        return;
 
-        ios_stat_ref (iosstat);
-        iosstat64 = (unsigned long )iosstat;
-        ret = inode_ctx_put (inode, this, iosstat64);
-        return ret;
+    GF_ATOMIC_ADD(conf->cumulative.data_written, len);
+    GF_ATOMIC_ADD(conf->incremental.data_written, len);
+    GF_ATOMIC_INC(conf->cumulative.block_count_write[lb2]);
+    GF_ATOMIC_INC(conf->incremental.block_count_write[lb2]);
+
+    if (iosfd) {
+        GF_ATOMIC_ADD(iosfd->data_written, len);
+        GF_ATOMIC_INC(iosfd->block_count_write[lb2]);
+    }
 }
 
-int
-ios_inode_ctx_get (inode_t *inode, xlator_t *this, struct ios_stat **iosstat)
+static void
+ios_bump_upcall(xlator_t *this, gf_upcall_flags_t event)
 {
-        uint64_t      iosstat64 = 0;
-        unsigned long iosstatlong = 0;
-        int           ret = 0;
+    struct ios_conf *conf = NULL;
 
-        ret = inode_ctx_get (inode, this, &iosstat64);
-        iosstatlong = iosstat64;
-        if (ret != -1)
-                *iosstat = (void *) iosstatlong;
+    conf = this->private;
+    if (!conf)
+        return;
+    if (conf->count_fop_hits) {
+        GF_ATOMIC_INC(conf->cumulative.upcall_hits[event]);
+        GF_ATOMIC_INC(conf->incremental.upcall_hits[event]);
+    }
+}
 
-        return ret;
+static void
+ios_bump_stats(xlator_t *this, struct ios_stat *iosstat, ios_stats_type_t type)
+{
+    struct ios_conf *conf = NULL;
+    uint64_t value = 0;
+
+    conf = this->private;
 
+    value = GF_ATOMIC_INC(iosstat->counters[type]);
+    ios_stat_add_to_list(&conf->list[type], value, iosstat);
 }
 
 int
-ios_stat_add_to_list (struct ios_stat_head *list_head, uint64_t value,
-                            struct ios_stat *iosstat)
+ios_inode_ctx_set(inode_t *inode, xlator_t *this, struct ios_stat *iosstat)
 {
-        struct ios_stat_list *new = NULL;
-        struct ios_stat_list *entry = NULL;
-        struct ios_stat_list *t = NULL;
-        struct ios_stat_list *list_entry = NULL;
-        struct ios_stat_list *tmp = NULL;
-        struct ios_stat_list *last = NULL;
-        struct ios_stat      *stat = NULL;
-        int                   cnt = 0;
-        int                   found = 0;
-        int                   reposition = 0;
-        double                min_count = 0;
+    uint64_t iosstat64 = 0;
+    int ret = 0;
 
-        LOCK (&list_head->lock);
-        {
+    ios_stat_ref(iosstat);
+    iosstat64 = (unsigned long)iosstat;
+    ret = inode_ctx_put(inode, this, iosstat64);
+    return ret;
+}
 
-                if (list_head->min_cnt == 0)
-                        list_head->min_cnt = value;
-                if ((list_head->members == MAX_LIST_MEMBERS) &&
-                     (list_head->min_cnt > value))
-                        goto out;
-
-                list_for_each_entry_safe (entry, t,
-                              &list_head->iosstats->list, list) {
-                        cnt++;
-                        if (cnt == list_head->members)
-                                last = entry;
-
-                        if (!uuid_compare (iosstat->gfid,
-                            entry->iosstat->gfid)) {
-                                list_entry = entry;
-                                found = cnt;
-                                entry->value = value;
-                                if (!reposition) {
-                                        if (cnt == list_head->members)
-                                                list_head->min_cnt = value;
-                                        goto out;
-                                }
-                                break;
-                        } else if (entry->value <= value && !reposition) {
-                                reposition = cnt;
-                                tmp = entry;
-                                if (cnt == list_head->members - 1)
-                                        min_count = entry->value;
-                        }
-                }
-                if (found) {
-                        list_del (&list_entry->list);
-                        list_add_tail (&list_entry->list, &tmp->list);
-                        if (min_count)
-                                list_head->min_cnt = min_count;
-                        goto out;
-                } else if (list_head->members == MAX_LIST_MEMBERS && reposition) {
-                        new = GF_CALLOC (1, sizeof (*new),
-                                        gf_io_stats_mt_ios_stat_list);
-                        new->iosstat = iosstat;
-                        new->value = value;
-                        ios_stat_ref (iosstat);
-                        list_add_tail (&new->list, &tmp->list);
-                        stat = last->iosstat;
-                        last->iosstat = NULL;
-                        ios_stat_unref (stat);
-                        list_del (&last->list);
-                        GF_FREE (last);
-                        if (reposition == MAX_LIST_MEMBERS)
-                                list_head->min_cnt = value;
-                        else if (min_count) {
-                                list_head->min_cnt = min_count;
-                        }
-                } else if (list_head->members < MAX_LIST_MEMBERS) {
-                        new  = GF_CALLOC (1, sizeof (*new),
-                                          gf_io_stats_mt_ios_stat_list);
-                        new->iosstat = iosstat;
-                        new->value = value;
-                        ios_stat_ref (iosstat);
-                        if (reposition) {
-                                list_add_tail (&new->list, &tmp->list);
-                        } else {
-                                list_add_tail (&new->list, &entry->list);
-                        }
-                        list_head->members++;
-                        if (list_head->min_cnt > value)
-                                list_head->min_cnt = value;
-                }
-        }
-out:
-        UNLOCK (&list_head->lock);
-        return 0;
+int
+ios_inode_ctx_get(inode_t *inode, xlator_t *this, struct ios_stat **iosstat)
+{
+    uint64_t iosstat64 = 0;
+    unsigned long iosstatlong = 0;
+    int ret = 0;
+
+    ret = inode_ctx_get(inode, this, &iosstat64);
+    iosstatlong = iosstat64;
+    if (ret != -1)
+        *iosstat = (void *)iosstatlong;
+
+    return ret;
 }
 
-inline int
-ios_stats_cleanup (xlator_t *this, inode_t *inode)
+/*
+ * So why goto all this trouble?  Why not just queue up some samples in
+ * a big list and malloc away?  Well malloc is expensive relative
+ * to what we are measuring, so cannot have any malloc's (or worse
+ * callocs) in our measurement code paths.  Instead, we are going to
+ * pre-allocate a circular buffer and collect a maximum number of samples.
+ * Prior to dumping them all we'll create a new buffer and swap the
+ * old buffer with the new, and then proceed to dump the statistics
+ * in our dump thread.
+ *
+ */
+ios_sample_buf_t *
+ios_create_sample_buf(size_t buf_size)
 {
+    ios_sample_buf_t *ios_sample_buf = NULL;
+    ios_sample_t *ios_samples = NULL;
+
+    ios_sample_buf = GF_CALLOC(1, sizeof(*ios_sample_buf),
+                               gf_io_stats_mt_ios_sample_buf);
+    if (!ios_sample_buf)
+        goto err;
+
+    ios_samples = GF_CALLOC(buf_size, sizeof(*ios_samples),
+                            gf_io_stats_mt_ios_sample);
+
+    if (!ios_samples)
+        goto err;
+
+    ios_sample_buf->ios_samples = ios_samples;
+    ios_sample_buf->size = buf_size;
+    ios_sample_buf->pos = 0;
+    ios_sample_buf->observed = 0;
+    ios_sample_buf->collected = 0;
+
+    return ios_sample_buf;
+err:
+    GF_FREE(ios_sample_buf);
+    return NULL;
+}
 
-        struct ios_stat *iosstat = NULL;
-        uint64_t         iosstat64 = 0;
+void
+ios_destroy_sample_buf(ios_sample_buf_t *ios_sample_buf)
+{
+    GF_FREE(ios_sample_buf->ios_samples);
+    GF_FREE(ios_sample_buf);
+}
 
-        inode_ctx_del (inode, this, &iosstat64);
-        if (!iosstat64) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "could not get inode ctx");
-                return 0;
-        }
-        iosstat = (void *) (long)iosstat64;
-        if (iosstat) {
-                ios_stat_unref (iosstat);
-        }
+static int
+ios_init_sample_buf(struct ios_conf *conf)
+{
+    int32_t ret = -1;
+
+    GF_ASSERT(conf);
+    LOCK(&conf->lock);
+    conf->ios_sample_buf = ios_create_sample_buf(conf->ios_sample_buf_size);
+    if (!conf->ios_sample_buf)
+        goto out;
+    ret = 0;
+out:
+    UNLOCK(&conf->lock);
+    return ret;
+}
+
+static int
+ios_stats_cleanup(xlator_t *this, inode_t *inode)
+{
+    struct ios_stat *iosstat = NULL;
+    uint64_t iosstat64 = 0;
+
+    inode_ctx_del(inode, this, &iosstat64);
+    if (!iosstat64) {
+        gf_log(this->name, GF_LOG_WARNING, "could not get inode ctx");
         return 0;
+    }
+    iosstat = (void *)(long)iosstat64;
+    if (iosstat) {
+        ios_stat_unref(iosstat);
+    }
+    return 0;
 }
 
-#define ios_log(this, logfp, fmt ...)                           \
-        do {                                                    \
-                if (logfp) {                                    \
-                        fprintf (logfp, fmt);                   \
-                        fprintf (logfp, "\n");                  \
-                }                                               \
-                gf_log (this->name, GF_LOG_INFO, fmt);        \
-        } while (0)
+#define ios_log(this, logfp, fmt...)                                           \
+    do {                                                                       \
+        if (logfp) {                                                           \
+            fprintf(logfp, fmt);                                               \
+            fprintf(logfp, "\n");                                              \
+        }                                                                      \
+        gf_log(this->name, GF_LOG_DEBUG, fmt);                                 \
+    } while (0)
 
 int
-ios_dump_file_stats (struct ios_stat_head *list_head, xlator_t *this, FILE* logfp)
+ios_dump_file_stats(struct ios_stat_head *list_head, xlator_t *this,
+                    FILE *logfp)
 {
-        struct ios_stat_list *entry = NULL;
+    struct ios_stat_list *entry = NULL;
 
-        LOCK (&list_head->lock);
+    LOCK(&list_head->lock);
+    {
+        list_for_each_entry(entry, &list_head->iosstats->list, list)
         {
-                list_for_each_entry (entry, &list_head->iosstats->list, list) {
-                        ios_log (this, logfp, "%-12.0f %s",
-                                entry->value, entry->iosstat->filename);
-                }
+            ios_log(this, logfp, "%-12.0f %s", entry->value,
+                    entry->iosstat->filename);
         }
-        UNLOCK (&list_head->lock);
-        return 0;
+    }
+    UNLOCK(&list_head->lock);
+    return 0;
 }
 
 int
-ios_dump_throughput_stats (struct ios_stat_head *list_head, xlator_t *this,
-                            FILE* logfp, ios_stats_type_t type)
+ios_dump_throughput_stats(struct ios_stat_head *list_head, xlator_t *this,
+                          FILE *logfp, ios_stats_thru_t type)
 {
-        struct ios_stat_list *entry = NULL;
-        struct timeval        time  = {0, };
-        char                  timestr[256] = {0, };
-
-        LOCK (&list_head->lock);
+    struct ios_stat_list *entry = NULL;
+    char timestr[GF_TIMESTR_SIZE] = {
+        0,
+    };
+
+    LOCK(&list_head->lock);
+    {
+        list_for_each_entry(entry, &list_head->iosstats->list, list)
         {
-                list_for_each_entry (entry, &list_head->iosstats->list, list) {
-                        gf_time_fmt (timestr, sizeof timestr,
-                                     entry->iosstat->thru_counters[type].time.tv_sec,
-                                     gf_timefmt_FT);
-                        snprintf (timestr + strlen (timestr), sizeof timestr - strlen (timestr),
-                          ".%"GF_PRI_SUSECONDS, time.tv_usec);
-
-                        ios_log (this, logfp, "%s \t %-10.2f  \t  %s",
-                                 timestr, entry->value, entry->iosstat->filename);
-                }
+            gf_time_fmt_tv(timestr, sizeof timestr,
+                           &entry->iosstat->thru_counters[type].time,
+                           gf_timefmt_FT);
+
+            ios_log(this, logfp, "%s \t %-10.2f  \t  %s", timestr, entry->value,
+                    entry->iosstat->filename);
         }
-        UNLOCK (&list_head->lock);
-        return 0;
+    }
+    UNLOCK(&list_head->lock);
+    return 0;
 }
 
 int
-io_stats_dump_global_to_logfp (xlator_t *this, struct ios_global_stats *stats,
-                               struct timeval *now, int interval, FILE* logfp)
+_io_stats_get_key_prefix(xlator_t *this, char **key_prefix)
 {
-        int                   i = 0;
-        int                   per_line = 0;
-        int                   index = 0;
-        struct ios_stat_head *list_head = NULL;
-        struct ios_conf      *conf = NULL;
-        char                  timestr[256] = {0, };
-        char                  str_header[128] = {0};
-        char                  str_read[128] = {0};
-        char                  str_write[128] = {0};
-
-        conf = this->private;
+    char *key_root = "gluster";
+    char *xlator_name = NULL;
+    char *instance_name = NULL;
+    size_t key_len = 0;
+    int bytes_written = 0;
+    int i = 0;
+    int ret = 0;
+    struct ios_conf *conf = this->private;
+
+    xlator_name = strdupa(conf->unique_id);
+    for (i = 0; i < strlen(xlator_name); i++) {
+        if (xlator_name[i] == '/')
+            xlator_name[i] = '_';
+    }
+
+    instance_name = this->instance_name;
+    if (this->name && strcmp(this->name, "glustershd") == 0) {
+        xlator_name = "shd";
+    } else if (this->prev && strcmp(this->prev->name, "nfs-server") == 0) {
+        xlator_name = "nfsd";
+        if (this->prev->instance_name)
+            instance_name = strdupa(this->prev->instance_name);
+    }
+
+    if (strcmp(__progname, "glusterfsd") == 0)
+        key_root = "gluster.brick";
+
+    if (instance_name) {
+        /* +3 for 2 x "." + NULL */
+        key_len = strlen(key_root) + strlen(xlator_name) +
+                  strlen(instance_name) + 3;
+        *key_prefix = GF_CALLOC(key_len, sizeof(char), gf_common_mt_char);
+        if (!*key_prefix) {
+            ret = -ENOMEM;
+            goto err;
+        }
+        bytes_written = snprintf(*key_prefix, key_len, "%s.%s.%s", key_root,
+                                 xlator_name, instance_name);
+        if (bytes_written != key_len - 1) {
+            ret = -EINVAL;
+            goto err;
+        }
+    } else {
+        /* +2 for 1 x "." + NULL */
+        key_len = strlen(key_root) + strlen(xlator_name) + 2;
+        *key_prefix = GF_CALLOC(key_len, sizeof(char), gf_common_mt_char);
+        if (!*key_prefix) {
+            ret = -ENOMEM;
+            goto err;
+        }
+        bytes_written = snprintf(*key_prefix, key_len, "%s.%s", key_root,
+                                 xlator_name);
+        if (bytes_written != key_len - 1) {
+            ret = -EINVAL;
+            goto err;
+        }
+    }
+    return 0;
+err:
+    GF_FREE(*key_prefix);
+    *key_prefix = NULL;
+    return ret;
+}
 
-        if (interval == -1)
-                ios_log (this, logfp, "\n=== Cumulative stats ===");
-        else
-                ios_log (this, logfp, "\n=== Interval %d stats ===",
-                         interval);
-        ios_log (this, logfp, "      Duration : %"PRId64" secs",
-                 (uint64_t) (now->tv_sec - stats->started_at.tv_sec));
-        ios_log (this, logfp, "     BytesRead : %"PRId64,
-                 stats->data_read);
-        ios_log (this, logfp, "  BytesWritten : %"PRId64"\n",
-                 stats->data_written);
-
-        snprintf (str_header, sizeof (str_header), "%-12s %c", "Block Size", ':');
-        snprintf (str_read, sizeof (str_read), "%-12s %c", "Read Count", ':');
-        snprintf (str_write, sizeof (str_write), "%-12s %c", "Write Count", ':');
-        index = 14;
-        for (i = 0; i < 32; i++) {
-                if ((stats->block_count_read[i] == 0) &&
-                    (stats->block_count_write[i] == 0))
-                        continue;
-                per_line++;
-
-                snprintf (str_header+index, sizeof (str_header)-index,
-                          "%16dB+", (1<<i));
-                if (stats->block_count_read[i])
-                        snprintf (str_read+index, sizeof (str_read)-index,
-                                  "%18"PRId64, stats->block_count_read[i]);
-                else    snprintf (str_read+index, sizeof (str_read)-index,
-                                  "%18s", "0");
-                if (stats->block_count_write[i])
-                        snprintf (str_write+index, sizeof (str_write)-index,
-                                  "%18"PRId64, stats->block_count_write[i]);
-                else    snprintf (str_write+index, sizeof (str_write)-index,
-                                  "%18s", "0");
-
-                index += 18;
-                if (per_line == 3) {
-                        ios_log (this, logfp, "%s", str_header);
-                        ios_log (this, logfp, "%s", str_read);
-                        ios_log (this, logfp, "%s\n", str_write);
-
-                        memset (str_header, 0, sizeof (str_header));
-                        memset (str_read, 0, sizeof (str_read));
-                        memset (str_write, 0, sizeof (str_write));
-
-                        snprintf (str_header, sizeof (str_header), "%-12s %c",
-                                  "Block Size", ':');
-                        snprintf (str_read, sizeof (str_read), "%-12s %c",
-                                  "Read Count", ':');
-                        snprintf (str_write, sizeof (str_write), "%-12s %c",
-                                  "Write Count", ':');
-
-                        index = 14;
-                        per_line = 0;
-                }
+int
+io_stats_dump_global_to_json_logfp(xlator_t *this,
+                                   struct ios_global_stats *stats, time_t now,
+                                   int interval, FILE *logfp)
+{
+    int i = 0;
+    int j = 0;
+    struct ios_conf *conf = NULL;
+    char *key_prefix = NULL;
+    char *str_prefix = NULL;
+    char *lc_fop_name = NULL;
+    int ret = 1; /* Default to error */
+    int rw_size;
+    char *rw_unit = NULL;
+    uint64_t fop_hits;
+    float fop_lat_ave;
+    float fop_lat_min;
+    float fop_lat_max;
+    double interval_sec;
+    double fop_ave_usec = 0.0;
+    double fop_ave_usec_sum = 0.0;
+    double weighted_fop_ave_usec = 0.0;
+    double weighted_fop_ave_usec_sum = 0.0;
+    long total_fop_hits = 0;
+    loc_t unused_loc = {
+        0,
+    };
+    dict_t *xattr = NULL;
+
+    interval_sec = (double)(now - stats->started_at);
+
+    conf = this->private;
+
+    ret = _io_stats_get_key_prefix(this, &key_prefix);
+    if (ret) {
+        goto out;
+    }
+
+    if (interval == -1) {
+        str_prefix = "aggr";
+
+    } else {
+        str_prefix = "inter";
+    }
+    ios_log(this, logfp, "{");
+
+    for (i = 0; i < 31; i++) {
+        rw_size = (1 << i);
+        if (rw_size >= 1024 * 1024) {
+            rw_size = rw_size / (1024 * 1024);
+            rw_unit = "mb";
+        } else if (rw_size >= 1024) {
+            rw_size = rw_size / 1024;
+            rw_unit = "kb";
+        } else {
+            rw_unit = "b";
         }
 
-        if (per_line != 0) {
-                ios_log (this, logfp, "%s", str_header);
-                ios_log (this, logfp, "%s", str_read);
-                ios_log (this, logfp, "%s\n", str_write);
+        if (interval == -1) {
+            ios_log(this, logfp, "\"%s.%s.read_%d%s\": %" GF_PRI_ATOMIC ",",
+                    key_prefix, str_prefix, rw_size, rw_unit,
+                    GF_ATOMIC_GET(stats->block_count_read[i]));
+            ios_log(this, logfp, "\"%s.%s.write_%d%s\": %" GF_PRI_ATOMIC ",",
+                    key_prefix, str_prefix, rw_size, rw_unit,
+                    GF_ATOMIC_GET(stats->block_count_write[i]));
+        } else {
+            ios_log(this, logfp, "\"%s.%s.read_%d%s_per_sec\": %0.2lf,",
+                    key_prefix, str_prefix, rw_size, rw_unit,
+                    (double)(GF_ATOMIC_GET(stats->block_count_read[i]) /
+                             interval_sec));
+            ios_log(this, logfp, "\"%s.%s.write_%d%s_per_sec\": %0.2lf,",
+                    key_prefix, str_prefix, rw_size, rw_unit,
+                    (double)(GF_ATOMIC_GET(stats->block_count_write[i]) /
+                             interval_sec));
+        }
+    }
+
+    if (interval == -1) {
+        ios_log(this, logfp, "\"%s.%s.fds.open_count\": %" PRId64 ",",
+                key_prefix, str_prefix, conf->cumulative.nr_opens);
+        ios_log(this, logfp, "\"%s.%s.fds.max_open_count\": %" PRId64 ",",
+                key_prefix, str_prefix, conf->cumulative.max_nr_opens);
+    }
+
+    for (i = 0; i < GF_FOP_MAXVALUE; i++) {
+        lc_fop_name = strdupa(gf_fop_list[i]);
+        for (j = 0; lc_fop_name[j]; j++) {
+            lc_fop_name[j] = tolower(lc_fop_name[j]);
         }
 
-        ios_log (this, logfp, "%-13s %10s %14s %14s %14s", "Fop",
-                 "Call Count", "Avg-Latency", "Min-Latency",
-                 "Max-Latency");
-        ios_log (this, logfp, "%-13s %10s %14s %14s %14s", "---", "----------",
-                 "-----------", "-----------", "-----------");
-
-        for (i = 0; i < GF_FOP_MAXVALUE; i++) {
-                if (stats->fop_hits[i] && !stats->latency[i].avg)
-                        ios_log (this, logfp, "%-13s %10"PRId64" %11s "
-                                 "us %11s us %11s us", gf_fop_list[i],
-                                 stats->fop_hits[i], "0", "0", "0");
-                else if (stats->fop_hits[i] && stats->latency[i].avg)
-                        ios_log (this, logfp, "%-13s %10"PRId64" %11.2lf us "
-                                 "%11.2lf us %11.2lf us", gf_fop_list[i],
-                                 stats->fop_hits[i], stats->latency[i].avg,
-                                 stats->latency[i].min, stats->latency[i].max);
+        fop_hits = GF_ATOMIC_GET(stats->fop_hits[i]);
+        fop_lat_ave = 0.0;
+        fop_lat_min = 0.0;
+        fop_lat_max = 0.0;
+        if (fop_hits) {
+            if (stats->latency[i].avg) {
+                fop_lat_ave = stats->latency[i].avg;
+                fop_lat_min = stats->latency[i].min;
+                fop_lat_max = stats->latency[i].max;
+            }
+        }
+        if (interval == -1) {
+            ios_log(this, logfp, "\"%s.%s.fop.%s.count\": %" GF_PRI_ATOMIC ",",
+                    key_prefix, str_prefix, lc_fop_name, fop_hits);
+        } else {
+            ios_log(this, logfp, "\"%s.%s.fop.%s.per_sec\": %0.2lf,",
+                    key_prefix, str_prefix, lc_fop_name,
+                    (double)(fop_hits / interval_sec));
         }
-        ios_log (this, logfp, "------ ----- ----- ----- ----- ----- ----- ----- "
-                 " ----- ----- ----- -----\n");
 
+        ios_log(this, logfp, "\"%s.%s.fop.%s.latency_ave_usec\": %0.2lf,",
+                key_prefix, str_prefix, lc_fop_name, fop_lat_ave);
+        ios_log(this, logfp, "\"%s.%s.fop.%s.latency_min_usec\": %0.2lf,",
+                key_prefix, str_prefix, lc_fop_name, fop_lat_min);
+        ios_log(this, logfp, "\"%s.%s.fop.%s.latency_max_usec\": %0.2lf,",
+                key_prefix, str_prefix, lc_fop_name, fop_lat_max);
+
+        fop_ave_usec_sum += fop_lat_ave;
+        weighted_fop_ave_usec_sum += fop_hits * fop_lat_ave;
+        total_fop_hits += fop_hits;
+    }
+
+    if (total_fop_hits) {
+        weighted_fop_ave_usec = weighted_fop_ave_usec_sum / total_fop_hits;
+        /* Extra key that does not print out an entry w/ 0.00 for
+         * intervals with no data
+         */
+        ios_log(this, logfp,
+                "\"%s.%s.fop.weighted_latency_ave_usec_nozerofill\": "
+                "%0.4lf,",
+                key_prefix, str_prefix, weighted_fop_ave_usec);
+    }
+    ios_log(this, logfp, "\"%s.%s.fop.weighted_latency_ave_usec\": %0.4lf,",
+            key_prefix, str_prefix, weighted_fop_ave_usec);
+    ios_log(this, logfp, "\"%s.%s.fop.weighted_fop_count\": %ld,", key_prefix,
+            str_prefix, total_fop_hits);
+
+    fop_ave_usec = fop_ave_usec_sum / GF_FOP_MAXVALUE;
+    ios_log(this, logfp, "\"%s.%s.fop.unweighted_latency_ave_usec\":%0.4lf,",
+            key_prefix, str_prefix, fop_ave_usec);
+
+    for (i = 0; i < GF_UPCALL_FLAGS_MAXVALUE; i++) {
+        lc_fop_name = strdupa(gf_upcall_list[i]);
+        for (j = 0; lc_fop_name[j]; j++) {
+            lc_fop_name[j] = tolower(lc_fop_name[j]);
+        }
+        fop_hits = GF_ATOMIC_GET(stats->upcall_hits[i]);
         if (interval == -1) {
-                LOCK (&conf->lock);
-                {
-                        gf_time_fmt (timestr, sizeof timestr,
-                                     conf->cumulative.max_openfd_time.tv_sec,
-                                     gf_timefmt_FT);
-                        snprintf (timestr + strlen (timestr), sizeof timestr - strlen (timestr),
-                                  ".%"GF_PRI_SUSECONDS,
-                                  conf->cumulative.max_openfd_time.tv_usec);
-                        ios_log (this, logfp, "Current open fd's: %"PRId64
-                                 " Max open fd's: %"PRId64" time %s",
-                                 conf->cumulative.nr_opens,
-                                 conf->cumulative.max_nr_opens, timestr);
-                }
-                UNLOCK (&conf->lock);
-                ios_log (this, logfp, "\n==========Open File Stats========");
-                ios_log (this, logfp, "\nCOUNT:  \t  FILE NAME");
-                list_head = &conf->list[IOS_STATS_TYPE_OPEN];
-                ios_dump_file_stats (list_head, this, logfp);
-
-
-                ios_log (this, logfp, "\n==========Read File Stats========");
-                ios_log (this, logfp, "\nCOUNT:  \t  FILE NAME");
-                list_head = &conf->list[IOS_STATS_TYPE_READ];
-                ios_dump_file_stats (list_head, this, logfp);
-
-                ios_log (this, logfp, "\n==========Write File Stats========");
-                ios_log (this, logfp, "\nCOUNT:  \t  FILE NAME");
-                list_head = &conf->list[IOS_STATS_TYPE_WRITE];
-                ios_dump_file_stats (list_head, this, logfp);
-
-                ios_log (this, logfp, "\n==========Directory open stats========");
-                ios_log (this, logfp, "\nCOUNT:  \t  DIRECTORY NAME");
-                list_head = &conf->list[IOS_STATS_TYPE_OPENDIR];
-                ios_dump_file_stats (list_head, this, logfp);
-
-                ios_log (this, logfp, "\n========Directory readdirp Stats=======");
-                ios_log (this, logfp, "\nCOUNT:  \t  DIRECTORY NAME");
-                list_head = &conf->list[IOS_STATS_TYPE_READDIRP];
-                ios_dump_file_stats (list_head, this, logfp);
-
-                ios_log (this, logfp, "\n========Read Throughput File Stats=====");
-                ios_log (this, logfp, "\nTIMESTAMP \t\t\t THROUGHPUT(KBPS)"
-                         "\tFILE NAME");
-                list_head = &conf->thru_list[IOS_STATS_THRU_READ];
-                ios_dump_throughput_stats(list_head, this, logfp, IOS_STATS_THRU_READ);
-
-                ios_log (this, logfp, "\n======Write Throughput File Stats======");
-                ios_log (this, logfp, "\nTIMESTAMP \t\t\t THROUGHPUT(KBPS)"
-                         "\tFILE NAME");
-                list_head = &conf->thru_list[IOS_STATS_THRU_WRITE];
-                ios_dump_throughput_stats (list_head, this, logfp, IOS_STATS_THRU_WRITE);
+            ios_log(this, logfp, "\"%s.%s.fop.%s.count\": %" GF_PRI_ATOMIC ",",
+                    key_prefix, str_prefix, lc_fop_name, fop_hits);
+        } else {
+            ios_log(this, logfp, "\"%s.%s.fop.%s.per_sec\": %0.2lf,",
+                    key_prefix, str_prefix, lc_fop_name,
+                    (double)(fop_hits / interval_sec));
         }
-        return 0;
+    }
+
+    ret = syncop_getxattr(this, &unused_loc, &xattr, IO_THREADS_QUEUE_SIZE_KEY,
+                          NULL, NULL);
+    if (xattr) {
+        /*
+         * Iterate over the dictionary returned to us by io-threads and
+         * dump the results to the stats file.
+         */
+        data_pair_t *curr = NULL;
+
+        dict_foreach_inline(xattr, curr)
+        {
+            ios_log(this, logfp, "\"%s.%s.%s.queue_size\": %d,", key_prefix,
+                    str_prefix, curr->key, data_to_int32(curr->value));
+        }
+
+        /* Free the dictionary */
+        dict_unref(xattr);
+    } else {
+        gf_log(this->name, GF_LOG_WARNING,
+               "Unable to get queue size counts from "
+               "the io-threads translator!");
+    }
+
+    if (interval == -1) {
+        ios_log(this, logfp, "\"%s.%s.uptime\": %" PRIu64 ",", key_prefix,
+                str_prefix, (uint64_t)(now - stats->started_at));
+        ios_log(this, logfp,
+                "\"%s.%s.bytes_read\": "
+                "%" GF_PRI_ATOMIC ",",
+                key_prefix, str_prefix, GF_ATOMIC_GET(stats->data_read));
+        ios_log(this, logfp,
+                "\"%s.%s.bytes_written\": "
+                "%" GF_PRI_ATOMIC "",
+                key_prefix, str_prefix, GF_ATOMIC_GET(stats->data_written));
+    } else {
+        ios_log(this, logfp, "\"%s.%s.sample_interval_sec\": %0.2lf,",
+                key_prefix, str_prefix, interval_sec);
+        ios_log(this, logfp, "\"%s.%s.bytes_read_per_sec\": %0.2lf,",
+                key_prefix, str_prefix,
+                (double)(GF_ATOMIC_GET(stats->data_read) / interval_sec));
+        ios_log(this, logfp, "\"%s.%s.bytes_written_per_sec\": %0.2lf",
+                key_prefix, str_prefix,
+                (double)(GF_ATOMIC_GET(stats->data_written) / interval_sec));
+    }
+
+    ios_log(this, logfp, "}");
+    ret = 0;
+out:
+    GF_FREE(key_prefix);
+    return ret;
 }
 
+char *
+_resolve_username(xlator_t *this, uid_t uid)
+{
+    struct passwd pwd;
+    struct passwd *pwd_result = NULL;
+    size_t pwd_buf_len;
+    char *pwd_buf = NULL;
+    char *ret = NULL;
+
+    /* Prepare our buffer for the uid->username translation */
+#ifdef _SC_GETGR_R_SIZE_MAX
+    pwd_buf_len = sysconf(_SC_GETGR_R_SIZE_MAX);
+#else
+    pwd_buf_len = -1;
+#endif
+    if (pwd_buf_len == -1) {
+        pwd_buf_len = DEFAULT_PWD_BUF_SZ; /* per the man page */
+    }
+
+    pwd_buf = alloca(pwd_buf_len);
+    if (!pwd_buf)
+        goto err;
+
+    getpwuid_r(uid, &pwd, pwd_buf, pwd_buf_len, &pwd_result);
+    if (!pwd_result)
+        goto err;
+
+    ret = gf_strdup(pwd.pw_name);
+    if (ret)
+        return ret;
+    else
+        gf_log(this->name, GF_LOG_ERROR,
+               "gf_strdup failed, failing username "
+               "resolution.");
+err:
+    return ret;
+}
+
+char *
+_resolve_group_name(xlator_t *this, gid_t gid)
+{
+    struct group grp;
+    struct group *grp_result = NULL;
+    size_t grp_buf_len;
+    char *grp_buf = NULL;
+    char *ret = NULL;
+
+    /* Prepare our buffer for the gid->group name translation */
+#ifdef _SC_GETGR_R_SIZE_MAX
+    grp_buf_len = sysconf(_SC_GETGR_R_SIZE_MAX);
+#else
+    grp_buf_len = -1;
+#endif
+    if (grp_buf_len == -1) {
+        grp_buf_len = DEFAULT_GRP_BUF_SZ; /* per the man page */
+    }
+
+    grp_buf = alloca(grp_buf_len);
+    if (!grp_buf) {
+        goto err;
+    }
+
+    if (getgrgid_r(gid, &grp, grp_buf, grp_buf_len, &grp_result) != 0)
+        goto err;
+
+    if (!grp_result)
+        goto err;
+
+    ret = gf_strdup(grp.gr_name);
+    if (ret)
+        return ret;
+    else
+        gf_log(this->name, GF_LOG_ERROR,
+               "gf_strdup failed, failing username "
+               "resolution.");
+err:
+    return ret;
+}
+
+/*
+ * This function writes out a latency sample to a given file descriptor
+ * and beautifies the output in the process.
+ */
+void
+_io_stats_write_latency_sample(xlator_t *this, ios_sample_t *sample,
+                               FILE *logfp)
+{
+    double epoch_time = 0.00;
+    char *xlator_name = NULL;
+    char *instance_name = NULL;
+    char *hostname = NULL;
+    char *identifier = NULL;
+    char *port = NULL;
+    char *port_pos = NULL;
+    char *group_name = NULL;
+    char *username = NULL;
+    struct ios_conf *conf = NULL;
+
+    conf = this->private;
+
+    epoch_time = (sample->timestamp).tv_sec +
+                 ((sample->timestamp).tv_usec / 1000000.0);
+
+    if (strlen(sample->identifier) == 0) {
+        hostname = "Unknown";
+        port = "Unknown";
+    } else {
+        identifier = strdupa(sample->identifier);
+        port_pos = strrchr(identifier, ':');
+        if (!port_pos || strlen(port_pos) < 2)
+            goto err;
+        port = strdupa(port_pos + 1);
+        if (!port)
+            goto err;
+        *port_pos = '\0';
+        hostname = gf_rev_dns_lookup_cached(identifier, conf->dnscache);
+        if (!hostname)
+            hostname = "Unknown";
+    }
+
+    xlator_name = conf->unique_id;
+    if (!xlator_name || strlen(xlator_name) == 0)
+        xlator_name = "Unknown";
+
+    instance_name = this->instance_name;
+    if (!instance_name || strlen(instance_name) == 0)
+        instance_name = "N/A";
+
+    /* Resolve the UID to a string username */
+    username = _resolve_username(this, sample->uid);
+    if (!username) {
+        username = GF_MALLOC(30, gf_common_mt_char);
+        if (!username) {
+            goto out;
+        }
+        sprintf(username, "%d", (int32_t)sample->uid);
+    }
+
+    /* Resolve the GID to a string group name */
+    group_name = _resolve_group_name(this, sample->gid);
+    if (!group_name) {
+        group_name = GF_MALLOC(30, gf_common_mt_char);
+        if (!group_name) {
+            goto out;
+        }
+        sprintf(group_name, "%d", (int32_t)sample->gid);
+    }
+
+    ios_log(this, logfp, "%0.6lf,%s,%s,%0.4lf,%s,%s,%s,%s,%s,%s", epoch_time,
+            fop_enum_to_pri_string(sample->fop_type),
+            gf_fop_string(sample->fop_type), sample->elapsed, xlator_name,
+            instance_name, username, group_name, hostname, port);
+    goto out;
+err:
+    gf_log(this->name, GF_LOG_ERROR, "Error parsing socket identifier");
+out:
+    GF_FREE(group_name);
+    GF_FREE(username);
+}
+
+/*
+ * Takes our current sample buffer in conf->io_sample_buf, and saves
+ * a reference to this, init's a new buffer, and then dumps out the
+ * contents of the saved reference.
+ */
 int
-io_stats_dump_global_to_dict (xlator_t *this, struct ios_global_stats *stats,
-                              struct timeval *now, int interval, dict_t *dict)
+io_stats_dump_latency_samples_logfp(xlator_t *this, FILE *logfp)
 {
-        int             ret = 0;
-        char            key[256] = {0};
-        uint64_t        sec = 0;
-        int             i = 0;
-        uint64_t        count = 0;
+    uint64_t i = 0;
+    struct ios_conf *conf = NULL;
+    ios_sample_buf_t *sample_buf = NULL;
+    int ret = 1; /* Default to error */
+
+    conf = this->private;
+
+    /* Save pointer to old buffer; the CS equivalent of
+     * Indiana Jones: https://www.youtube.com/watch?v=Pr-8AP0To4k,
+     * though ours will end better I hope!
+     */
+    sample_buf = conf->ios_sample_buf;
+    if (!sample_buf) {
+        gf_log(this->name, GF_LOG_WARNING, "Sampling buffer is null, bailing!");
+        goto out;
+    }
+
+    /* Empty case, nothing to do, exit. */
+    if (sample_buf->collected == 0) {
+        gf_log(this->name, GF_LOG_DEBUG, "No samples, dump not required.");
+        ret = 0;
+        goto out;
+    }
+
+    /* Init a new buffer, so we are free to work on the one we saved a
+     * reference to above.
+     */
+    if (ios_init_sample_buf(conf) != 0) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "Failed to init new sampling buffer, out of memory?");
+        goto out;
+    }
+
+    /* Wrap-around case, dump from pos to sample_buf->size -1
+     * and then from 0 to sample_buf->pos (covered off by
+     * "simple case")
+     */
+    if (sample_buf->collected > sample_buf->pos + 1) {
+        for (i = sample_buf->pos; i < sample_buf->size; i++) {
+            _io_stats_write_latency_sample(this, &(sample_buf->ios_samples[i]),
+                                           logfp);
+        }
+    }
+
+    /* Simple case: Dump from 0 to sample_buf->pos */
+    for (i = 0; i < sample_buf->pos; i++) {
+        _io_stats_write_latency_sample(this, &(sample_buf->ios_samples[i]),
+                                       logfp);
+    }
+    ios_destroy_sample_buf(sample_buf);
 
-        GF_ASSERT (stats);
-        GF_ASSERT (now);
-        GF_ASSERT (dict);
-        GF_ASSERT (this);
+out:
+    return ret;
+}
 
-        if (interval == -1)
-                snprintf (key, sizeof (key), "cumulative");
+int
+io_stats_dump_global_to_logfp(xlator_t *this, struct ios_global_stats *stats,
+                              time_t now, int interval, FILE *logfp)
+{
+    int i = 0;
+    int per_line = 0;
+    int index = 0;
+    struct ios_stat_head *list_head = NULL;
+    struct ios_conf *conf = NULL;
+    char timestr[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    char str_header[128] = {0};
+    char str_read[128] = {0};
+    char str_write[128] = {0};
+    uint64_t fop_hits = 0;
+    uint64_t block_count_read = 0;
+    uint64_t block_count_write = 0;
+
+    conf = this->private;
+
+    if (interval == -1)
+        ios_log(this, logfp, "\n=== Cumulative stats ===");
+    else
+        ios_log(this, logfp, "\n=== Interval %d stats ===", interval);
+    ios_log(this, logfp, "      Duration : %" PRIu64 " secs",
+            (uint64_t)(now - stats->started_at));
+    ios_log(this, logfp, "     BytesRead : %" GF_PRI_ATOMIC,
+            GF_ATOMIC_GET(stats->data_read));
+    ios_log(this, logfp, "  BytesWritten : %" GF_PRI_ATOMIC "\n",
+            GF_ATOMIC_GET(stats->data_written));
+
+    snprintf(str_header, sizeof(str_header), "%-12s %c", "Block Size", ':');
+    snprintf(str_read, sizeof(str_read), "%-12s %c", "Read Count", ':');
+    snprintf(str_write, sizeof(str_write), "%-12s %c", "Write Count", ':');
+    index = 14;
+    for (i = 0; i < IOS_BLOCK_COUNT_SIZE; i++) {
+        block_count_read = GF_ATOMIC_GET(stats->block_count_read[i]);
+        block_count_write = GF_ATOMIC_GET(stats->block_count_write[i]);
+        if ((block_count_read == 0) && (block_count_write == 0))
+            continue;
+        per_line++;
+
+        snprintf(str_header + index, sizeof(str_header) - index, "%16dB+",
+                 (1 << i));
+        if (block_count_read)
+            snprintf(str_read + index, sizeof(str_read) - index, "%18" PRId64,
+                     block_count_read);
         else
-                snprintf (key, sizeof (key), "interval");
-        ret = dict_set_int32 (dict, key, interval);
-        if (ret)
-                gf_log (this->name, GF_LOG_ERROR, "failed to set "
-                        "interval %d", interval);
-
-        memset (key, 0, sizeof (key));
-        snprintf (key, sizeof (key), "%d-duration", interval);
-        sec = (uint64_t) (now->tv_sec - stats->started_at.tv_sec);
-        ret = dict_set_uint64 (dict, key, sec);
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "failed to set "
-                        "duration(%d) - %"PRId64, interval, sec);
-                goto out;
+            snprintf(str_read + index, sizeof(str_read) - index, "%18s", "0");
+        if (block_count_write)
+            snprintf(str_write + index, sizeof(str_write) - index,
+                     "%18" GF_PRI_ATOMIC, block_count_write);
+        else
+            snprintf(str_write + index, sizeof(str_write) - index, "%18s", "0");
+
+        index += 18;
+        if (per_line == 3) {
+            ios_log(this, logfp, "%s", str_header);
+            ios_log(this, logfp, "%s", str_read);
+            ios_log(this, logfp, "%s\n", str_write);
+
+            snprintf(str_header, sizeof(str_header), "%-12s %c", "Block Size",
+                     ':');
+            snprintf(str_read, sizeof(str_read), "%-12s %c", "Read Count", ':');
+            snprintf(str_write, sizeof(str_write), "%-12s %c", "Write Count",
+                     ':');
+
+            index = 14;
+            per_line = 0;
         }
+    }
+
+    if (per_line != 0) {
+        ios_log(this, logfp, "%s", str_header);
+        ios_log(this, logfp, "%s", str_read);
+        ios_log(this, logfp, "%s\n", str_write);
+    }
+
+    ios_log(this, logfp, "%-13s %10s %14s %14s %14s", "Fop", "Call Count",
+            "Avg-Latency", "Min-Latency", "Max-Latency");
+    ios_log(this, logfp, "%-13s %10s %14s %14s %14s", "---", "----------",
+            "-----------", "-----------", "-----------");
+
+    for (i = 0; i < GF_FOP_MAXVALUE; i++) {
+        fop_hits = GF_ATOMIC_GET(stats->fop_hits[i]);
+        if (fop_hits && !stats->latency[i].avg)
+            ios_log(this, logfp,
+                    "%-13s %10" GF_PRI_ATOMIC
+                    " %11s "
+                    "us %11s us %11s us",
+                    gf_fop_list[i], fop_hits, "0", "0", "0");
+        else if (fop_hits && stats->latency[i].avg)
+            ios_log(this, logfp,
+                    "%-13s %10" GF_PRI_ATOMIC
+                    " "
+                    "%11.2lf us %11.2lf us %11.2lf us",
+                    gf_fop_list[i], fop_hits, stats->latency[i].avg,
+                    stats->latency[i].min, stats->latency[i].max);
+    }
+
+    for (i = 0; i < GF_UPCALL_FLAGS_MAXVALUE; i++) {
+        fop_hits = GF_ATOMIC_GET(stats->upcall_hits[i]);
+        if (fop_hits)
+            ios_log(this, logfp,
+                    "%-13s %10" PRId64
+                    " %11s "
+                    "us %11s us %11s us",
+                    gf_upcall_list[i], fop_hits, "0", "0", "0");
+    }
+
+    ios_log(this, logfp,
+            "------ ----- ----- ----- ----- ----- ----- ----- "
+            " ----- ----- ----- -----\n");
+
+    if (interval == -1) {
+        LOCK(&conf->lock);
+        {
+            gf_time_fmt_tv(timestr, sizeof timestr,
+                           &conf->cumulative.max_openfd_time, gf_timefmt_FT);
+            ios_log(this, logfp,
+                    "Current open fd's: %" PRId64 " Max open fd's: %" PRId64
+                    " time %s",
+                    conf->cumulative.nr_opens, conf->cumulative.max_nr_opens,
+                    timestr);
+        }
+        UNLOCK(&conf->lock);
+        ios_log(this, logfp, "\n==========Open File Stats========");
+        ios_log(this, logfp, "\nCOUNT:  \t  FILE NAME");
+        list_head = &conf->list[IOS_STATS_TYPE_OPEN];
+        ios_dump_file_stats(list_head, this, logfp);
+
+        ios_log(this, logfp, "\n==========Read File Stats========");
+        ios_log(this, logfp, "\nCOUNT:  \t  FILE NAME");
+        list_head = &conf->list[IOS_STATS_TYPE_READ];
+        ios_dump_file_stats(list_head, this, logfp);
+
+        ios_log(this, logfp, "\n==========Write File Stats========");
+        ios_log(this, logfp, "\nCOUNT:  \t  FILE NAME");
+        list_head = &conf->list[IOS_STATS_TYPE_WRITE];
+        ios_dump_file_stats(list_head, this, logfp);
+
+        ios_log(this, logfp, "\n==========Directory open stats========");
+        ios_log(this, logfp, "\nCOUNT:  \t  DIRECTORY NAME");
+        list_head = &conf->list[IOS_STATS_TYPE_OPENDIR];
+        ios_dump_file_stats(list_head, this, logfp);
+
+        ios_log(this, logfp, "\n========Directory readdirp Stats=======");
+        ios_log(this, logfp, "\nCOUNT:  \t  DIRECTORY NAME");
+        list_head = &conf->list[IOS_STATS_TYPE_READDIRP];
+        ios_dump_file_stats(list_head, this, logfp);
+
+        ios_log(this, logfp, "\n========Read Throughput File Stats=====");
+        ios_log(this, logfp,
+                "\nTIMESTAMP \t\t\t THROUGHPUT(KBPS)"
+                "\tFILE NAME");
+        list_head = &conf->thru_list[IOS_STATS_THRU_READ];
+        ios_dump_throughput_stats(list_head, this, logfp, IOS_STATS_THRU_READ);
+
+        ios_log(this, logfp, "\n======Write Throughput File Stats======");
+        ios_log(this, logfp,
+                "\nTIMESTAMP \t\t\t THROUGHPUT(KBPS)"
+                "\tFILE NAME");
+        list_head = &conf->thru_list[IOS_STATS_THRU_WRITE];
+        ios_dump_throughput_stats(list_head, this, logfp, IOS_STATS_THRU_WRITE);
+    }
+    return 0;
+}
 
-        memset (key, 0, sizeof (key));
-        snprintf (key, sizeof (key), "%d-total-read", interval);
-        ret = dict_set_uint64 (dict, key, stats->data_read);
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "failed to set total "
-                       "read(%d) - %"PRId64, interval, stats->data_read);
+int
+io_stats_dump_global_to_dict(xlator_t *this, struct ios_global_stats *stats,
+                             time_t now, int interval, dict_t *dict)
+{
+    int ret = 0;
+    char key[64] = {0};
+    uint64_t sec = 0;
+    int i = 0;
+    uint64_t count = 0;
+    uint64_t fop_hits = 0;
+
+    GF_ASSERT(stats);
+    GF_ASSERT(now);
+    GF_ASSERT(dict);
+    GF_ASSERT(this);
+
+    if (interval == -1)
+        snprintf(key, sizeof(key), "cumulative");
+    else
+        snprintf(key, sizeof(key), "interval");
+    ret = dict_set_int32(dict, key, interval);
+    if (ret)
+        gf_log(this->name, GF_LOG_ERROR,
+               "failed to set "
+               "interval %d",
+               interval);
+
+    snprintf(key, sizeof(key), "%d-duration", interval);
+    sec = now - stats->started_at;
+    ret = dict_set_uint64(dict, key, sec);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "failed to set "
+               "duration(%d) - %" PRId64,
+               interval, sec);
+        goto out;
+    }
+
+    snprintf(key, sizeof(key), "%d-total-read", interval);
+    ret = dict_set_uint64(dict, key, GF_ATOMIC_GET(stats->data_read));
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "failed to set total "
+               "read(%d) - %" GF_PRI_ATOMIC,
+               interval, GF_ATOMIC_GET(stats->data_read));
+        goto out;
+    }
+
+    snprintf(key, sizeof(key), "%d-total-write", interval);
+    ret = dict_set_uint64(dict, key, GF_ATOMIC_GET(stats->data_written));
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "failed to set total "
+               "write(%d) - %" GF_PRI_ATOMIC,
+               interval, GF_ATOMIC_GET(stats->data_written));
+        goto out;
+    }
+    for (i = 0; i < 32; i++) {
+        count = GF_ATOMIC_GET(stats->block_count_read[i]);
+        if (count) {
+            snprintf(key, sizeof(key), "%d-read-%d", interval, (1 << i));
+            ret = dict_set_uint64(dict, key, count);
+            if (ret) {
+                gf_log(this->name, GF_LOG_ERROR,
+                       "failed to "
+                       "set read-%db+, with: %" PRId64,
+                       (1 << i), count);
                 goto out;
+            }
         }
-
-        memset (key, 0, sizeof (key));
-        snprintf (key, sizeof (key), "%d-total-write", interval);
-        ret = dict_set_uint64 (dict, key, stats->data_written);
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "failed to set total "
-                        "write(%d) - %"PRId64, interval, stats->data_written);
+    }
+
+    for (i = 0; i < IOS_BLOCK_COUNT_SIZE; i++) {
+        count = GF_ATOMIC_GET(stats->block_count_write[i]);
+        if (count) {
+            snprintf(key, sizeof(key), "%d-write-%d", interval, (1 << i));
+            ret = dict_set_uint64(dict, key, count);
+            if (ret) {
+                gf_log(this->name, GF_LOG_ERROR,
+                       "failed to "
+                       "set write-%db+, with: %" PRId64,
+                       (1 << i), count);
                 goto out;
+            }
         }
-        for (i = 0; i < 32; i++) {
-                if (stats->block_count_read[i]) {
-                        memset (key, 0, sizeof (key));
-                        snprintf (key, sizeof (key), "%d-read-%d", interval,
-                                  (1 << i));
-                        count = stats->block_count_read[i];
-                        ret = dict_set_uint64 (dict, key, count);
-                        if (ret) {
-                                gf_log (this->name, GF_LOG_ERROR, "failed to "
-                                        "set read-%db+, with: %"PRId64,
-                                        (1<<i), count);
-                                goto out;
-                        }
-                }
+    }
+
+    for (i = 0; i < GF_FOP_MAXVALUE; i++) {
+        fop_hits = GF_ATOMIC_GET(stats->fop_hits[i]);
+        if (fop_hits == 0)
+            continue;
+        snprintf(key, sizeof(key), "%d-%d-hits", interval, i);
+        ret = dict_set_uint64(dict, key, fop_hits);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "failed to set "
+                   "%s-fop-hits: %" GF_PRI_ATOMIC,
+                   gf_fop_list[i], fop_hits);
+            goto out;
         }
 
-        for (i = 0; i < 32; i++) {
-                if (stats->block_count_write[i]) {
-                        snprintf (key, sizeof (key), "%d-write-%d", interval,
-                                  (1<<i));
-                        count = stats->block_count_write[i];
-                        ret = dict_set_uint64 (dict, key, count);
-                        if (ret) {
-                                gf_log (this->name, GF_LOG_ERROR, "failed to "
-                                        "set write-%db+, with: %"PRId64,
-                                        (1<<i), count);
-                                goto out;
-                        }
-                }
+        if (stats->latency[i].avg == 0)
+            continue;
+        snprintf(key, sizeof(key), "%d-%d-avglatency", interval, i);
+        ret = dict_set_double(dict, key, stats->latency[i].avg);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "failed to set %s "
+                   "avglatency(%d) with %f",
+                   gf_fop_list[i], interval, stats->latency[i].avg);
+            goto out;
         }
-
-        for (i = 0; i < GF_FOP_MAXVALUE; i++) {
-                if (stats->fop_hits[i] == 0)
-                        continue;
-                snprintf (key, sizeof (key), "%d-%d-hits", interval, i);
-                ret = dict_set_uint64 (dict, key, stats->fop_hits[i]);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_ERROR, "failed to "
-                                "set %s-fop-hits: %"PRIu64, gf_fop_list[i],
-                                stats->fop_hits[i]);
-                        goto out;
-                }
-
-                if (stats->latency[i].avg == 0)
-                        continue;
-                snprintf (key, sizeof (key), "%d-%d-avglatency", interval, i);
-                ret = dict_set_double (dict, key, stats->latency[i].avg);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_ERROR, "failed to set %s "
-                                "avglatency(%d) with %f", gf_fop_list[i],
-                                interval, stats->latency[i].avg);
-                        goto out;
-                }
-                snprintf (key, sizeof (key), "%d-%d-minlatency", interval, i);
-                ret = dict_set_double (dict, key, stats->latency[i].min);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_ERROR, "failed to set %s "
-                                "minlatency(%d) with %f", gf_fop_list[i],
-                                interval, stats->latency[i].min);
-                        goto out;
-                }
-                snprintf (key, sizeof (key), "%d-%d-maxlatency", interval, i);
-                ret = dict_set_double (dict, key, stats->latency[i].max);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_ERROR, "failed to set %s "
-                                "maxlatency(%d) with %f", gf_fop_list[i],
-                                interval, stats->latency[i].max);
-                        goto out;
-                }
+        snprintf(key, sizeof(key), "%d-%d-minlatency", interval, i);
+        ret = dict_set_double(dict, key, stats->latency[i].min);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "failed to set %s "
+                   "minlatency(%d) with %f",
+                   gf_fop_list[i], interval, stats->latency[i].min);
+            goto out;
+        }
+        snprintf(key, sizeof(key), "%d-%d-maxlatency", interval, i);
+        ret = dict_set_double(dict, key, stats->latency[i].max);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "failed to set %s "
+                   "maxlatency(%d) with %f",
+                   gf_fop_list[i], interval, stats->latency[i].max);
+            goto out;
+        }
+    }
+    for (i = 0; i < GF_UPCALL_FLAGS_MAXVALUE; i++) {
+        fop_hits = GF_ATOMIC_GET(stats->upcall_hits[i]);
+        if (fop_hits == 0)
+            continue;
+        snprintf(key, sizeof(key), "%d-%d-upcall-hits", interval, i);
+        ret = dict_set_uint64(dict, key, fop_hits);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "failed to "
+                   "set %s-upcall-hits: %" PRIu64,
+                   gf_upcall_list[i], fop_hits);
+            goto out;
         }
+    }
 out:
-        gf_log (this->name, GF_LOG_DEBUG, "returning %d", ret);
-        return ret;
+    gf_log(this->name, GF_LOG_DEBUG, "returning %d", ret);
+    return ret;
 }
 
 int
-io_stats_dump_global (xlator_t *this, struct ios_global_stats *stats,
-                      struct timeval *now, int interval,
-                      struct ios_dump_args *args)
+io_stats_dump_global(xlator_t *this, struct ios_global_stats *stats, time_t now,
+                     int interval, struct ios_dump_args *args)
 {
-        int     ret = -1;
-
-        GF_ASSERT (args);
-        GF_ASSERT (now);
-        GF_ASSERT (stats);
-        GF_ASSERT (this);
-
-
-
-        switch (args->type) {
+    int ret = -1;
+
+    GF_ASSERT(args);
+    GF_ASSERT(now);
+    GF_ASSERT(stats);
+    GF_ASSERT(this);
+
+    switch (args->type) {
+        case IOS_DUMP_TYPE_JSON_FILE:
+            ret = io_stats_dump_global_to_json_logfp(this, stats, now, interval,
+                                                     args->u.logfp);
+            break;
         case IOS_DUMP_TYPE_FILE:
-                ret = io_stats_dump_global_to_logfp (this, stats, now,
-                                                     interval, args->u.logfp);
-        break;
+            ret = io_stats_dump_global_to_logfp(this, stats, now, interval,
+                                                args->u.logfp);
+            break;
         case IOS_DUMP_TYPE_DICT:
-                ret = io_stats_dump_global_to_dict (this, stats, now,
-                                                    interval, args->u.dict);
-        break;
+            ret = io_stats_dump_global_to_dict(this, stats, now, interval,
+                                               args->u.dict);
+            break;
         default:
-                GF_ASSERT (0);
-                ret = -1;
-        break;
-        }
-        return ret;
+            GF_ASSERT(0);
+            ret = -1;
+            break;
+    }
+    return ret;
 }
 
 int
-ios_dump_args_init (struct ios_dump_args *args, ios_dump_type_t type,
-                    void *output)
+ios_dump_args_init(struct ios_dump_args *args, ios_dump_type_t type,
+                   void *output)
 {
-        int             ret = 0;
+    int ret = 0;
 
-        GF_ASSERT (args);
-        GF_ASSERT (type > IOS_DUMP_TYPE_NONE && type < IOS_DUMP_TYPE_MAX);
-        GF_ASSERT (output);
+    GF_ASSERT(args);
+    GF_ASSERT(type > IOS_DUMP_TYPE_NONE && type < IOS_DUMP_TYPE_MAX);
+    GF_ASSERT(output);
 
-        args->type = type;
-        switch (args->type) {
+    args->type = type;
+    switch (args->type) {
+        case IOS_DUMP_TYPE_JSON_FILE:
         case IOS_DUMP_TYPE_FILE:
-                args->u.logfp = output;
-                break;
+            args->u.logfp = output;
+            break;
         case IOS_DUMP_TYPE_DICT:
-                args->u.dict = output;
-                break;
+            args->u.dict = output;
+            break;
         default:
-                GF_ASSERT (0);
-                ret = -1;
-        }
+            GF_ASSERT(0);
+            ret = -1;
+    }
 
-        return ret;
+    return ret;
 }
 
-int
-io_stats_dump (xlator_t *this, struct ios_dump_args *args)
+static void
+ios_global_stats_clear(struct ios_global_stats *stats, time_t now)
 {
-        struct ios_conf         *conf = NULL;
-        struct ios_global_stats  cumulative = {0, };
-        struct ios_global_stats  incremental = {0, };
-        int                      increment = 0;
-        struct timeval           now;
-
-        GF_ASSERT (this);
-        GF_ASSERT (args);
-        GF_ASSERT (args->type > IOS_DUMP_TYPE_NONE);
-        GF_ASSERT (args->type < IOS_DUMP_TYPE_MAX);
-
-        conf = this->private;
+    GF_ASSERT(stats);
+    GF_ASSERT(now);
 
-        gettimeofday (&now, NULL);
-        LOCK (&conf->lock);
-        {
-                cumulative  = conf->cumulative;
-                incremental = conf->incremental;
+    memset(stats, 0, sizeof(*stats));
+    stats->started_at = now;
+}
 
+int
+io_stats_dump(xlator_t *this, struct ios_dump_args *args, ios_info_op_t op,
+              gf_boolean_t is_peek)
+{
+    struct ios_conf *conf = NULL;
+    struct ios_global_stats cumulative = {};
+    struct ios_global_stats incremental = {};
+    int increment = 0;
+    time_t now = 0;
+
+    GF_ASSERT(this);
+    GF_ASSERT(args);
+    GF_ASSERT(args->type > IOS_DUMP_TYPE_NONE);
+    GF_ASSERT(args->type < IOS_DUMP_TYPE_MAX);
+
+    conf = this->private;
+    now = gf_time();
+
+    LOCK(&conf->lock);
+    {
+        if (op == GF_IOS_INFO_ALL || op == GF_IOS_INFO_CUMULATIVE)
+            cumulative = conf->cumulative;
+
+        if (op == GF_IOS_INFO_ALL || op == GF_IOS_INFO_INCREMENTAL) {
+            incremental = conf->incremental;
+            increment = conf->increment;
+
+            if (!is_peek) {
                 increment = conf->increment++;
 
-                memset (&conf->incremental, 0, sizeof (conf->incremental));
-                conf->incremental.started_at = now;
+                ios_global_stats_clear(&conf->incremental, now);
+            }
         }
-        UNLOCK (&conf->lock);
+    }
+    UNLOCK(&conf->lock);
 
-        io_stats_dump_global (this, &cumulative, &now, -1, args);
-        io_stats_dump_global (this, &incremental, &now, increment, args);
+    if (op == GF_IOS_INFO_ALL || op == GF_IOS_INFO_CUMULATIVE)
+        io_stats_dump_global(this, &cumulative, now, -1, args);
 
-        return 0;
-}
+    if (op == GF_IOS_INFO_ALL || op == GF_IOS_INFO_INCREMENTAL)
+        io_stats_dump_global(this, &incremental, now, increment, args);
 
+    return 0;
+}
 
 int
-io_stats_dump_fd (xlator_t *this, struct ios_fd *iosfd)
+io_stats_dump_fd(xlator_t *this, struct ios_fd *iosfd)
 {
-        struct ios_conf         *conf = NULL;
-        struct timeval           now;
-        uint64_t                 sec = 0;
-        uint64_t                 usec = 0;
-        int                      i = 0;
-
-        conf = this->private;
-
-        if (!conf->dump_fd_stats)
-                return 0;
-
-        if (!iosfd)
-                return 0;
+    struct ios_conf *conf = NULL;
+    struct timeval now;
+    int i = 0;
+    double usecs = 0;
+    uint64_t data_read = 0;
+    uint64_t data_written = 0;
+    uint64_t block_count_read = 0;
+    uint64_t block_count_write = 0;
+
+    conf = this->private;
+
+    if (!conf->dump_fd_stats)
+        return 0;
 
-        gettimeofday (&now, NULL);
+    if (!iosfd)
+        return 0;
 
-        if (iosfd->opened_at.tv_usec > now.tv_usec) {
-                now.tv_usec += 1000000;
-                now.tv_usec--;
-        }
+    gettimeofday(&now, NULL);
+    usecs = gf_tvdiff(&iosfd->opened_at, &now);
+
+    gf_log(this->name, GF_LOG_INFO, "--- fd stats ---");
+
+    if (iosfd->filename)
+        gf_log(this->name, GF_LOG_INFO, "      Filename : %s", iosfd->filename);
+
+    if (usecs)
+        gf_log(this->name, GF_LOG_INFO, "      Lifetime : %lf secs", usecs);
+
+    data_read = GF_ATOMIC_GET(iosfd->data_read);
+    if (data_read)
+        gf_log(this->name, GF_LOG_INFO, "     BytesRead : %" PRId64 " bytes",
+               data_read);
+
+    data_written = GF_ATOMIC_GET(iosfd->data_written);
+    if (data_written)
+        gf_log(this->name, GF_LOG_INFO, "  BytesWritten : %" PRId64 " bytes",
+               data_written);
+
+    for (i = 0; i < 32; i++) {
+        block_count_read = GF_ATOMIC_GET(iosfd->block_count_read[i]);
+        if (block_count_read)
+            gf_log(this->name, GF_LOG_INFO,
+                   " Read %06db+ :"
+                   "%" PRId64,
+                   (1 << i), block_count_read);
+    }
+    for (i = 0; i < IOS_BLOCK_COUNT_SIZE; i++) {
+        block_count_write = GF_ATOMIC_GET(iosfd->block_count_write[i]);
+        if (block_count_write)
+            gf_log(this->name, GF_LOG_INFO, "Write %06db+ : %" PRId64, (1 << i),
+                   block_count_write);
+    }
+    return 0;
+}
 
-        sec = now.tv_sec - iosfd->opened_at.tv_sec;
-        usec = now.tv_usec - iosfd->opened_at.tv_usec;
-
-        gf_log (this->name, GF_LOG_INFO,
-                "--- fd stats ---");
-
-        if (iosfd->filename)
-                gf_log (this->name, GF_LOG_INFO,
-                        "      Filename : %s",
-                        iosfd->filename);
-
-        if (sec)
-                gf_log (this->name, GF_LOG_INFO,
-                        "      Lifetime : %"PRId64"secs, %"PRId64"usecs",
-                        sec, usec);
-
-        if (iosfd->data_read)
-                gf_log (this->name, GF_LOG_INFO,
-                        "     BytesRead : %"PRId64" bytes",
-                        iosfd->data_read);
-
-        if (iosfd->data_written)
-                gf_log (this->name, GF_LOG_INFO,
-                        "  BytesWritten : %"PRId64" bytes",
-                        iosfd->data_written);
-
-        for (i = 0; i < 32; i++) {
-                if (iosfd->block_count_read[i])
-                        gf_log (this->name, GF_LOG_INFO,
-                                " Read %06db+ : %"PRId64,
-                                (1 << i), iosfd->block_count_read[i]);
-        }
-        for (i = 0; i < 32; i++) {
-                if (iosfd->block_count_write[i])
-                        gf_log (this->name, GF_LOG_INFO,
-                                "Write %06db+ : %"PRId64,
-                                (1 << i), iosfd->block_count_write[i]);
-        }
-        return 0;
+void
+collect_ios_latency_sample(struct ios_conf *conf, glusterfs_fop_t fop_type,
+                           double elapsed, call_frame_t *frame)
+{
+    ios_sample_buf_t *ios_sample_buf = NULL;
+    ios_sample_t *ios_sample = NULL;
+    struct timespec *timestamp = NULL;
+    call_stack_t *root = NULL;
+
+    ios_sample_buf = conf->ios_sample_buf;
+    LOCK(&conf->ios_sampling_lock);
+    if (conf->ios_sample_interval == 0 ||
+        ios_sample_buf->observed % conf->ios_sample_interval != 0)
+        goto out;
+
+    timestamp = &frame->begin;
+    root = frame->root;
+
+    ios_sample = &(ios_sample_buf->ios_samples[ios_sample_buf->pos]);
+    ios_sample->elapsed = elapsed;
+    ios_sample->fop_type = fop_type;
+    ios_sample->uid = root->uid;
+    ios_sample->gid = root->gid;
+    (ios_sample->timestamp).tv_sec = timestamp->tv_sec;
+    (ios_sample->timestamp).tv_usec = timestamp->tv_nsec / 1000;
+    memcpy(&ios_sample->identifier, &root->identifier,
+           sizeof(root->identifier));
+
+    /* We've reached the end of the circular buffer, start from the
+     * beginning. */
+    if (ios_sample_buf->pos == (ios_sample_buf->size - 1))
+        ios_sample_buf->pos = 0;
+    else
+        ios_sample_buf->pos++;
+    ios_sample_buf->collected++;
+out:
+    ios_sample_buf->observed++;
+    UNLOCK(&conf->ios_sampling_lock);
+    return;
 }
 
 static void
-update_ios_latency_stats (struct ios_global_stats   *stats, double elapsed,
-                          glusterfs_fop_t op)
+update_ios_latency_stats(struct ios_global_stats *stats, double elapsed,
+                         glusterfs_fop_t op)
 {
-        double avg;
+    double avg;
 
-        GF_ASSERT (stats);
+    GF_ASSERT(stats);
 
-        if (!stats->latency[op].min)
-                stats->latency[op].min = elapsed;
-        if (stats->latency[op].min > elapsed)
-                stats->latency[op].min = elapsed;
-        if (stats->latency[op].max < elapsed)
-                stats->latency[op].max = elapsed;
+    stats->latency[op].total += elapsed;
 
-        avg = stats->latency[op].avg;
+    if (!stats->latency[op].min)
+        stats->latency[op].min = elapsed;
+    if (stats->latency[op].min > elapsed)
+        stats->latency[op].min = elapsed;
+    if (stats->latency[op].max < elapsed)
+        stats->latency[op].max = elapsed;
 
-        stats->latency[op].avg = avg + (elapsed - avg) / stats->fop_hits[op];
+    avg = stats->latency[op].avg;
+
+    stats->latency[op].avg = avg + (elapsed - avg) /
+                                       GF_ATOMIC_GET(stats->fop_hits[op]);
 }
 
 int
-update_ios_latency (struct ios_conf *conf, call_frame_t *frame,
-                    glusterfs_fop_t op)
+update_ios_latency(struct ios_conf *conf, call_frame_t *frame,
+                   glusterfs_fop_t op)
 {
-        double elapsed;
-        struct timeval *begin, *end;
+    double elapsed;
+    struct timespec *begin, *end;
 
-        begin = &frame->begin;
-        end   = &frame->end;
+    begin = &frame->begin;
+    end = &frame->end;
 
-        elapsed = (end->tv_sec - begin->tv_sec) * 1e6
-                + (end->tv_usec - begin->tv_usec);
+    elapsed = gf_tsdiff(begin, end) / 1000.0;
 
-        update_ios_latency_stats (&conf->cumulative, elapsed, op);
-        update_ios_latency_stats (&conf->incremental, elapsed, op);
+    update_ios_latency_stats(&conf->cumulative, elapsed, op);
+    update_ios_latency_stats(&conf->incremental, elapsed, op);
+    collect_ios_latency_sample(conf, op, elapsed, frame);
 
-        return 0;
+    return 0;
 }
 
 int32_t
-io_stats_dump_stats_to_dict (xlator_t *this, dict_t *resp,
-                             ios_stats_type_t flags, int32_t list_cnt)
-{
-        struct ios_conf         *conf = NULL;
-        int                      cnt  = 0;
-        char                     key[256];
-        struct ios_stat_head    *list_head = NULL;
-        struct ios_stat_list    *entry = NULL;
-        int                      ret = -1;
-        ios_stats_thru_t         index = IOS_STATS_THRU_MAX;
-        char                     timestr[256] = {0, };
-        char                    *dict_timestr = NULL;
-
-        conf = this->private;
-
-        switch (flags) {
-                case IOS_STATS_TYPE_OPEN:
-                        list_head = &conf->list[IOS_STATS_TYPE_OPEN];
-                        LOCK (&conf->lock);
-                        {
-                                ret = dict_set_uint64 (resp, "current-open",
-                                                     conf->cumulative.nr_opens);
-                                if (ret)
-                                        goto unlock;
-                                ret = dict_set_uint64 (resp, "max-open",
-                                                       conf->cumulative.max_nr_opens);
-
-                                gf_time_fmt (timestr, sizeof timestr,
-                                             conf->cumulative.max_openfd_time.tv_sec,
-                                             gf_timefmt_FT);
-                                if (conf->cumulative.max_openfd_time.tv_sec)
-                                        snprintf (timestr + strlen (timestr), sizeof timestr - strlen (timestr),
-                                                  ".%"GF_PRI_SUSECONDS,
-                                                  conf->cumulative.max_openfd_time.tv_usec);
-
-                                dict_timestr = gf_strdup (timestr);
-                                if (!dict_timestr)
-                                        goto unlock;
-                                ret = dict_set_dynstr (resp, "max-openfd-time",
-                                                       dict_timestr);
-                                if (ret)
-                                        goto unlock;
-                        }
+io_stats_dump_stats_to_dict(xlator_t *this, dict_t *resp,
+                            ios_stats_type_t flags, int32_t list_cnt)
+{
+    struct ios_conf *conf = NULL;
+    int cnt = 0;
+    char key[32];
+    int keylen;
+    struct ios_stat_head *list_head = NULL;
+    struct ios_stat_list *entry = NULL;
+    int ret = -1;
+    ios_stats_thru_t index = IOS_STATS_THRU_MAX;
+    char timestr[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    char *dict_timestr = NULL;
+
+    conf = this->private;
+
+    switch (flags) {
+        case IOS_STATS_TYPE_OPEN:
+            list_head = &conf->list[IOS_STATS_TYPE_OPEN];
+            LOCK(&conf->lock);
+            {
+                ret = dict_set_uint64(resp, "current-open",
+                                      conf->cumulative.nr_opens);
+                if (ret)
+                    goto unlock;
+                ret = dict_set_uint64(resp, "max-open",
+                                      conf->cumulative.max_nr_opens);
+
+                gf_time_fmt_tv(timestr, sizeof timestr,
+                               &conf->cumulative.max_openfd_time,
+                               gf_timefmt_FT);
+
+                dict_timestr = gf_strdup(timestr);
+                if (!dict_timestr)
+                    goto unlock;
+                ret = dict_set_dynstr(resp, "max-openfd-time", dict_timestr);
+                if (ret)
+                    goto unlock;
+            }
         unlock:
-                        UNLOCK (&conf->lock);
-                        /* Do not proceed if we came here because of some error
-                         * during the dict operation */
-                        if (ret)
-                                goto out;
-                        break;
-                case IOS_STATS_TYPE_READ:
-                        list_head = &conf->list[IOS_STATS_TYPE_READ];
-                        break;
-                case IOS_STATS_TYPE_WRITE:
-                        list_head = &conf->list[IOS_STATS_TYPE_WRITE];
-                        break;
-                case IOS_STATS_TYPE_OPENDIR:
-                        list_head = &conf->list[IOS_STATS_TYPE_OPENDIR];
-                        break;
-                case IOS_STATS_TYPE_READDIRP:
-                        list_head = &conf->list[IOS_STATS_TYPE_READDIRP];
-                        break;
-                case IOS_STATS_TYPE_READ_THROUGHPUT:
-                        list_head = &conf->thru_list[IOS_STATS_THRU_READ];
-                        index = IOS_STATS_THRU_READ;
-                        break;
-                case IOS_STATS_TYPE_WRITE_THROUGHPUT:
-                        list_head = &conf->thru_list[IOS_STATS_THRU_WRITE];
-                        index = IOS_STATS_THRU_WRITE;
-                        break;
-
-                default:
-                       goto out;
-        }
-        ret = dict_set_int32 (resp, "top-op", flags);
-        if (!list_cnt)
+            UNLOCK(&conf->lock);
+            /* Do not proceed if we came here because of some error
+             * during the dict operation */
+            if (ret)
                 goto out;
-        LOCK (&list_head->lock);
-        {
-                list_for_each_entry (entry, &list_head->iosstats->list, list) {
-
-                        cnt++;
-                        snprintf (key, 256, "%s-%d", "filename", cnt);
-                        ret = dict_set_str (resp, key, entry->iosstat->filename);
-                        if (ret)
-                                goto unlock_list_head;
-                         snprintf (key, 256, "%s-%d", "value",cnt);
-                         ret = dict_set_uint64 (resp, key, entry->value);
-                         if (ret)
-                                 goto unlock_list_head;
-                         if (index != IOS_STATS_THRU_MAX) {
-                                 snprintf (key, 256, "%s-%d", "time-sec", cnt);
-                                 ret = dict_set_int32 (resp, key,
-                                         entry->iosstat->thru_counters[index].time.tv_sec);
-                                 if (ret)
-                                         goto unlock_list_head;
-                                 snprintf (key, 256, "%s-%d", "time-usec", cnt);
-                                 ret = dict_set_int32 (resp, key,
-                                         entry->iosstat->thru_counters[index].time.tv_usec);
-                                 if (ret)
-                                         goto unlock_list_head;
-                         }
-                         if (cnt == list_cnt)
-                                 break;
+            break;
+        case IOS_STATS_TYPE_READ:
+            list_head = &conf->list[IOS_STATS_TYPE_READ];
+            break;
+        case IOS_STATS_TYPE_WRITE:
+            list_head = &conf->list[IOS_STATS_TYPE_WRITE];
+            break;
+        case IOS_STATS_TYPE_OPENDIR:
+            list_head = &conf->list[IOS_STATS_TYPE_OPENDIR];
+            break;
+        case IOS_STATS_TYPE_READDIRP:
+            list_head = &conf->list[IOS_STATS_TYPE_READDIRP];
+            break;
+        case IOS_STATS_TYPE_READ_THROUGHPUT:
+            list_head = &conf->thru_list[IOS_STATS_THRU_READ];
+            index = IOS_STATS_THRU_READ;
+            break;
+        case IOS_STATS_TYPE_WRITE_THROUGHPUT:
+            list_head = &conf->thru_list[IOS_STATS_THRU_WRITE];
+            index = IOS_STATS_THRU_WRITE;
+            break;
 
-                }
+        default:
+            goto out;
+    }
+    ret = dict_set_int32_sizen(resp, "top-op", flags);
+    if (!list_cnt)
+        goto out;
+    LOCK(&list_head->lock);
+    {
+        list_for_each_entry(entry, &list_head->iosstats->list, list)
+        {
+            cnt++;
+            keylen = snprintf(key, sizeof(key), "filename-%d", cnt);
+            ret = dict_set_strn(resp, key, keylen, entry->iosstat->filename);
+            if (ret)
+                goto unlock_list_head;
+            snprintf(key, sizeof(key), "value-%d", cnt);
+            ret = dict_set_uint64(resp, key, entry->value);
+            if (ret)
+                goto unlock_list_head;
+            if (index != IOS_STATS_THRU_MAX) {
+                keylen = snprintf(key, sizeof(key), "time-sec-%d", cnt);
+                ret = dict_set_int32n(
+                    resp, key, keylen,
+                    entry->iosstat->thru_counters[index].time.tv_sec);
+                if (ret)
+                    goto unlock_list_head;
+                keylen = snprintf(key, sizeof(key), "time-usec-%d", cnt);
+                ret = dict_set_int32n(
+                    resp, key, keylen,
+                    entry->iosstat->thru_counters[index].time.tv_usec);
+                if (ret)
+                    goto unlock_list_head;
+            }
+            if (cnt == list_cnt)
+                break;
         }
+    }
 unlock_list_head:
-        UNLOCK (&list_head->lock);
-        /* ret is !=0 if some dict operation in the above critical region
-         * failed. */
-        if (ret)
-                goto out;
-        ret = dict_set_int32 (resp, "members", cnt);
- out:
-        return ret;
+    UNLOCK(&list_head->lock);
+    /* ret is !=0 if some dict operation in the above critical region
+     * failed. */
+    if (ret)
+        goto out;
+    ret = dict_set_int32_sizen(resp, "members", cnt);
+out:
+    return ret;
 }
 
-int
-io_stats_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, fd_t *fd,
-                     inode_t *inode, struct iatt *buf,
-                     struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+static struct ios_stat *
+ios_init_iosstat(xlator_t *this, char *path, uuid_t gfid, inode_t *inode)
 {
-        struct ios_fd *iosfd = NULL;
-        char          *path = NULL;
-        struct ios_stat *iosstat = NULL;
-        struct ios_conf   *conf = NULL;
+    struct ios_stat *iosstat = NULL;
+    int i = 0;
 
-        conf = this->private;
+    iosstat = GF_CALLOC(1, sizeof(*iosstat), gf_io_stats_mt_ios_stat);
+    if (!iosstat)
+        goto out;
 
-        path = frame->local;
-        frame->local = NULL;
+    iosstat->filename = gf_strdup(path);
+    gf_uuid_copy(iosstat->gfid, gfid);
+    LOCK_INIT(&iosstat->lock);
 
-        if (!path)
-                goto unwind;
+    for (i = 0; i < IOS_STATS_TYPE_MAX; i++)
+        GF_ATOMIC_INIT(iosstat->counters[i], 0);
 
-        if (op_ret < 0) {
-                GF_FREE (path);
-                goto unwind;
-        }
+    ios_inode_ctx_set(inode, this, iosstat);
 
-        iosfd = GF_CALLOC (1, sizeof (*iosfd), gf_io_stats_mt_ios_fd);
-        if (!iosfd) {
-                GF_FREE (path);
-                goto unwind;
-        }
-
-        iosfd->filename = path;
-        gettimeofday (&iosfd->opened_at, NULL);
+out:
+    return iosstat;
+}
 
-        ios_fd_ctx_set (fd, this, iosfd);
-        LOCK (&conf->lock);
-        {
-                conf->cumulative.nr_opens++;
-                if (conf->cumulative.nr_opens > conf->cumulative.max_nr_opens) {
-                        conf->cumulative.max_nr_opens = conf->cumulative.nr_opens;
-                        conf->cumulative.max_openfd_time = iosfd->opened_at;
-                }
+int
+io_stats_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, fd_t *fd, inode_t *inode,
+                    struct iatt *buf, struct iatt *preparent,
+                    struct iatt *postparent, dict_t *xdata)
+{
+    struct ios_fd *iosfd = NULL;
+    char *path = NULL;
+    struct ios_stat *iosstat = NULL;
+    struct ios_conf *conf = NULL;
+
+    conf = this->private;
+
+    path = frame->local;
+    frame->local = NULL;
+
+    if (!path)
+        goto unwind;
+
+    if (op_ret < 0) {
+        GF_FREE(path);
+        goto unwind;
+    }
+
+    iosfd = GF_CALLOC(1, sizeof(*iosfd), gf_io_stats_mt_ios_fd);
+    if (!iosfd) {
+        GF_FREE(path);
+        goto unwind;
+    }
+
+    iosfd->filename = path;
+    gettimeofday(&iosfd->opened_at, NULL);
+
+    ios_fd_ctx_set(fd, this, iosfd);
+    LOCK(&conf->lock);
+    {
+        conf->cumulative.nr_opens++;
+        if (conf->cumulative.nr_opens > conf->cumulative.max_nr_opens) {
+            conf->cumulative.max_nr_opens = conf->cumulative.nr_opens;
+            conf->cumulative.max_openfd_time = iosfd->opened_at;
         }
-        UNLOCK (&conf->lock);
+    }
+    UNLOCK(&conf->lock);
 
-        iosstat = GF_CALLOC (1, sizeof (*iosstat), gf_io_stats_mt_ios_stat);
-        if (!iosstat) {
-                GF_FREE (path);
-                goto unwind;
-        }
-        iosstat->filename = gf_strdup (path);
-        uuid_copy (iosstat->gfid, buf->ia_gfid);
-        LOCK_INIT (&iosstat->lock);
-        ios_inode_ctx_set (fd->inode, this, iosstat);
+    iosstat = ios_init_iosstat(this, path, buf->ia_gfid, inode);
+    if (!iosstat)
+        GF_FREE(path);
 
 unwind:
-        UPDATE_PROFILE_STATS (frame, CREATE);
-        STACK_UNWIND_STRICT (create, frame, op_ret, op_errno, fd, inode, buf,
-                             preparent, postparent, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, CREATE);
+    STACK_UNWIND_STRICT(create, frame, op_ret, op_errno, fd, inode, buf,
+                        preparent, postparent, xdata);
+    return 0;
 }
 
-
 int
-io_stats_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
+io_stats_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
 {
-        struct ios_fd *iosfd = NULL;
-        char          *path = NULL;
-        struct   ios_stat *iosstat = NULL;
-        struct ios_conf   *conf = NULL;
-
-        conf = this->private;
-        path = frame->local;
-        frame->local = NULL;
-
-        if (!path)
-                goto unwind;
-
-        if (op_ret < 0) {
-                GF_FREE (path);
-                goto unwind;
-        }
-
-        iosfd = GF_CALLOC (1, sizeof (*iosfd), gf_io_stats_mt_ios_fd);
-        if (!iosfd) {
-                GF_FREE (path);
-                goto unwind;
-        }
-
-        iosfd->filename = path;
-        gettimeofday (&iosfd->opened_at, NULL);
-
-        ios_fd_ctx_set (fd, this, iosfd);
-
-        ios_inode_ctx_get (fd->inode, this, &iosstat);
-        if (!iosstat) {
-                iosstat = GF_CALLOC (1, sizeof (*iosstat),
-                                     gf_io_stats_mt_ios_stat);
-                if (iosstat) {
-                        iosstat->filename = gf_strdup (path);
-                        uuid_copy (iosstat->gfid, fd->inode->gfid);
-                        LOCK_INIT (&iosstat->lock);
-                        ios_inode_ctx_set (fd->inode, this, iosstat);
-                }
-        }
-
-        LOCK (&conf->lock);
-        {
-                conf->cumulative.nr_opens++;
-                if (conf->cumulative.nr_opens > conf->cumulative.max_nr_opens) {
-                        conf->cumulative.max_nr_opens = conf->cumulative.nr_opens;
-                        conf->cumulative.max_openfd_time = iosfd->opened_at;
-                }
-        }
-        UNLOCK (&conf->lock);
-        if (iosstat) {
-              BUMP_STATS (iosstat, IOS_STATS_TYPE_OPEN);
-              iosstat = NULL;
+    struct ios_fd *iosfd = NULL;
+    char *path = NULL;
+    struct ios_stat *iosstat = NULL;
+    struct ios_conf *conf = NULL;
+    int i = 0;
+
+    conf = this->private;
+    path = frame->local;
+    frame->local = NULL;
+
+    if (!path)
+        goto unwind;
+
+    if (op_ret < 0) {
+        GF_FREE(path);
+        goto unwind;
+    }
+
+    iosfd = GF_CALLOC(1, sizeof(*iosfd), gf_io_stats_mt_ios_fd);
+    if (!iosfd) {
+        GF_FREE(path);
+        goto unwind;
+    }
+
+    iosfd->filename = path;
+    GF_ATOMIC_INIT(iosfd->data_read, 0);
+    GF_ATOMIC_INIT(iosfd->data_written, 0);
+    for (i = 0; i < IOS_BLOCK_COUNT_SIZE; i++) {
+        GF_ATOMIC_INIT(iosfd->block_count_write[i], 0);
+        GF_ATOMIC_INIT(iosfd->block_count_read[i], 0);
+    }
+    gettimeofday(&iosfd->opened_at, NULL);
+
+    ios_fd_ctx_set(fd, this, iosfd);
+
+    ios_inode_ctx_get(fd->inode, this, &iosstat);
+    if (!iosstat) {
+        iosstat = ios_init_iosstat(this, path, fd->inode->gfid, fd->inode);
+    }
+
+    LOCK(&conf->lock);
+    {
+        conf->cumulative.nr_opens++;
+        if (conf->cumulative.nr_opens > conf->cumulative.max_nr_opens) {
+            conf->cumulative.max_nr_opens = conf->cumulative.nr_opens;
+            conf->cumulative.max_openfd_time = iosfd->opened_at;
         }
+    }
+    UNLOCK(&conf->lock);
+    if (iosstat) {
+        ios_bump_stats(this, iosstat, IOS_STATS_TYPE_OPEN);
+        iosstat = NULL;
+    }
 unwind:
-        UPDATE_PROFILE_STATS (frame, OPEN);
-
-        STACK_UNWIND_STRICT (open, frame, op_ret, op_errno, fd, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, OPEN);
 
+    STACK_UNWIND_STRICT(open, frame, op_ret, op_errno, fd, xdata);
+    return 0;
 }
 
-
 int
-io_stats_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct iatt *buf, dict_t *xdata)
+io_stats_stat_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                  dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, STAT);
-        STACK_UNWIND_STRICT (stat, frame, op_ret, op_errno, buf, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, STAT);
+    STACK_UNWIND_STRICT(stat, frame, op_ret, op_errno, buf, xdata);
+    return 0;
 }
 
+int
+io_stats_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iovec *vector,
+                   int32_t count, struct iatt *buf, struct iobref *iobref,
+                   dict_t *xdata)
+{
+    int len = 0;
+    fd_t *fd = NULL;
+    struct ios_stat *iosstat = NULL;
+
+    fd = frame->local;
+    frame->local = NULL;
+
+    if (op_ret > 0) {
+        len = iov_length(vector, count);
+        ios_bump_read(this, fd, len);
+    }
+
+    UPDATE_PROFILE_STATS(frame, READ);
+    ios_inode_ctx_get(fd->inode, this, &iosstat);
+
+    if (iosstat) {
+        ios_bump_stats(this, iosstat, IOS_STATS_TYPE_READ);
+        BUMP_THROUGHPUT(iosstat, IOS_STATS_THRU_READ);
+        iosstat = NULL;
+    }
+
+    STACK_UNWIND_STRICT(readv, frame, op_ret, op_errno, vector, count, buf,
+                        iobref, xdata);
+    return 0;
+}
 
 int
-io_stats_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno,
-                    struct iovec *vector, int32_t count,
-                    struct iatt *buf, struct iobref *iobref, dict_t *xdata)
+io_stats_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                    struct iatt *postbuf, dict_t *xdata)
 {
-        int              len = 0;
-        fd_t            *fd = NULL;
-        struct ios_stat *iosstat = NULL;
+    struct ios_stat *iosstat = NULL;
+    inode_t *inode = NULL;
 
-        fd = frame->local;
+    UPDATE_PROFILE_STATS(frame, WRITE);
+    if (frame->local) {
+        inode = frame->local;
         frame->local = NULL;
-
-        if (op_ret > 0) {
-                len = iov_length (vector, count);
-                BUMP_READ (fd, len);
-        }
-
-        UPDATE_PROFILE_STATS (frame, READ);
-        ios_inode_ctx_get (fd->inode, this, &iosstat);
-
+        ios_inode_ctx_get(inode, this, &iosstat);
         if (iosstat) {
-              BUMP_STATS (iosstat, IOS_STATS_TYPE_READ);
-              BUMP_THROUGHPUT (iosstat, IOS_STATS_THRU_READ);
-              iosstat = NULL;
+            ios_bump_stats(this, iosstat, IOS_STATS_TYPE_WRITE);
+            BUMP_THROUGHPUT(iosstat, IOS_STATS_THRU_WRITE);
+            inode = NULL;
+            iosstat = NULL;
         }
+    }
 
-        STACK_UNWIND_STRICT (readv, frame, op_ret, op_errno,
-                             vector, count, buf, iobref, xdata);
-        return 0;
-
+    STACK_UNWIND_STRICT(writev, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+    return 0;
 }
 
-
 int
-io_stats_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno,
-                     struct iatt *prebuf, struct iatt *postbuf, dict_t *xdata)
+io_stats_copy_file_range_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                             int32_t op_ret, int32_t op_errno,
+                             struct iatt *stbuf, struct iatt *prebuf_dst,
+                             struct iatt *postbuf_dst, dict_t *xdata)
 {
-        struct ios_stat *iosstat = NULL;
-        inode_t         *inode   = NULL;
-
-        UPDATE_PROFILE_STATS (frame, WRITE);
-        if (frame->local){
-                inode = frame->local;
-                frame->local = NULL;
-                ios_inode_ctx_get (inode, this, &iosstat);
-                if (iosstat) {
-                        BUMP_STATS (iosstat, IOS_STATS_TYPE_WRITE);
-                        BUMP_THROUGHPUT (iosstat, IOS_STATS_THRU_WRITE);
-                        inode = NULL;
-                        iosstat = NULL;
-                }
-        }
-
-        STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno, prebuf, postbuf, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, COPY_FILE_RANGE);
 
+    STACK_UNWIND_STRICT(copy_file_range, frame, op_ret, op_errno, stbuf,
+                        prebuf_dst, postbuf_dst, xdata);
+    return 0;
 }
 
-
-
-
 int
-io_stats_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, gf_dirent_t *buf, dict_t *xdata)
+io_stats_readdirp_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, gf_dirent_t *buf,
+                      dict_t *xdata)
 {
-        struct ios_stat *iosstat = NULL;
-        inode_t         *inode   = frame->local;
+    struct ios_stat *iosstat = NULL;
+    inode_t *inode = frame->local;
 
-        frame->local = NULL;
+    frame->local = NULL;
 
-        UPDATE_PROFILE_STATS (frame, READDIRP);
+    UPDATE_PROFILE_STATS(frame, READDIRP);
 
-        ios_inode_ctx_get (inode, this, &iosstat);
+    ios_inode_ctx_get(inode, this, &iosstat);
 
-        if (iosstat) {
-              BUMP_STATS (iosstat, IOS_STATS_TYPE_READDIRP);
-              iosstat = NULL;
-        }
+    if (iosstat) {
+        ios_bump_stats(this, iosstat, IOS_STATS_TYPE_READDIRP);
+        iosstat = NULL;
+    }
 
-        STACK_UNWIND_STRICT (readdirp, frame, op_ret, op_errno, buf, xdata);
-        return 0;
+    STACK_UNWIND_STRICT(readdirp, frame, op_ret, op_errno, buf, xdata);
+    return 0;
 }
 
-
 int
-io_stats_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno, gf_dirent_t *buf, dict_t *xdata)
+io_stats_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, gf_dirent_t *buf,
+                     dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, READDIR);
-        STACK_UNWIND_STRICT (readdir, frame, op_ret, op_errno, buf, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, READDIR);
+    STACK_UNWIND_STRICT(readdir, frame, op_ret, op_errno, buf, xdata);
+    return 0;
 }
 
-
 int
-io_stats_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno,
-                    struct iatt *prebuf, struct iatt *postbuf, dict_t *xdata)
+io_stats_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                   struct iatt *postbuf, dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, FSYNC);
-        STACK_UNWIND_STRICT (fsync, frame, op_ret, op_errno, prebuf, postbuf, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, FSYNC);
+    STACK_UNWIND_STRICT(fsync, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+    return 0;
 }
 
-
 int
-io_stats_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno,
-                      struct iatt *preop, struct iatt *postop, dict_t *xdata)
+io_stats_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *preop,
+                     struct iatt *postop, dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, SETATTR);
-        STACK_UNWIND_STRICT (setattr, frame, op_ret, op_errno, preop, postop, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, SETATTR);
+    STACK_UNWIND_STRICT(setattr, frame, op_ret, op_errno, preop, postop, xdata);
+    return 0;
 }
 
-
 int
-io_stats_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno,
-                     struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+io_stats_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                    struct iatt *postparent, dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, UNLINK);
-        STACK_UNWIND_STRICT (unlink, frame, op_ret, op_errno,
-                             preparent, postparent, xdata);
-        return 0;
-
+    UPDATE_PROFILE_STATS(frame, UNLINK);
+    STACK_UNWIND_STRICT(unlink, frame, op_ret, op_errno, preparent, postparent,
+                        xdata);
+    return 0;
 }
 
-
 int
-io_stats_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, struct iatt *buf,
-                     struct iatt *preoldparent, struct iatt *postoldparent,
-                     struct iatt *prenewparent, struct iatt *postnewparent, dict_t *xdata)
+io_stats_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                    struct iatt *preoldparent, struct iatt *postoldparent,
+                    struct iatt *prenewparent, struct iatt *postnewparent,
+                    dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, RENAME);
-        STACK_UNWIND_STRICT (rename, frame, op_ret, op_errno, buf,
-                             preoldparent, postoldparent,
-                             prenewparent, postnewparent, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, RENAME);
+    STACK_UNWIND_STRICT(rename, frame, op_ret, op_errno, buf, preoldparent,
+                        postoldparent, prenewparent, postnewparent, xdata);
+    return 0;
 }
 
-
 int
-io_stats_readlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, const char *buf,
-                       struct iatt *sbuf, dict_t *xdata)
+io_stats_readlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, const char *buf,
+                      struct iatt *sbuf, dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, READLINK);
-        STACK_UNWIND_STRICT (readlink, frame, op_ret, op_errno, buf, sbuf, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, READLINK);
+    STACK_UNWIND_STRICT(readlink, frame, op_ret, op_errno, buf, sbuf, xdata);
+    return 0;
 }
 
-
 int
-io_stats_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno,
-                     inode_t *inode, struct iatt *buf,
-                     dict_t *xdata, struct iatt *postparent)
+io_stats_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, inode_t *inode,
+                    struct iatt *buf, dict_t *xdata, struct iatt *postparent)
 {
-        UPDATE_PROFILE_STATS (frame, LOOKUP);
-        STACK_UNWIND_STRICT (lookup, frame, op_ret, op_errno, inode, buf, xdata,
-                             postparent);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, LOOKUP);
+    STACK_UNWIND_STRICT(lookup, frame, op_ret, op_errno, inode, buf, xdata,
+                        postparent);
+    return 0;
 }
 
-
 int
-io_stats_symlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno,
-                      inode_t *inode, struct iatt *buf,
-                      struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+io_stats_symlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, inode_t *inode,
+                     struct iatt *buf, struct iatt *preparent,
+                     struct iatt *postparent, dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, SYMLINK);
-        STACK_UNWIND_STRICT (symlink, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, SYMLINK);
+    STACK_UNWIND_STRICT(symlink, frame, op_ret, op_errno, inode, buf, preparent,
+                        postparent, xdata);
+    return 0;
 }
 
-
 int
-io_stats_mknod_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno,
-                    inode_t *inode, struct iatt *buf,
-                    struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+io_stats_mknod_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, inode_t *inode,
+                   struct iatt *buf, struct iatt *preparent,
+                   struct iatt *postparent, dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, MKNOD);
-        STACK_UNWIND_STRICT (mknod, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, MKNOD);
+    STACK_UNWIND_STRICT(mknod, frame, op_ret, op_errno, inode, buf, preparent,
+                        postparent, xdata);
+    return 0;
 }
 
-
 int
-io_stats_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno,
-                    inode_t *inode, struct iatt *buf,
-                    struct iatt *preparent, struct iatt *postparent,
-                    dict_t *xdata)
+io_stats_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, inode_t *inode,
+                   struct iatt *buf, struct iatt *preparent,
+                   struct iatt *postparent, dict_t *xdata)
 {
-        struct ios_stat *iosstat = NULL;
-        char   *path = frame->local;
+    char *path = frame->local;
 
-        UPDATE_PROFILE_STATS (frame, MKDIR);
-        if (op_ret < 0)
-                goto unwind;
+    if (!path)
+        goto unwind;
 
-        iosstat = GF_CALLOC (1, sizeof (*iosstat), gf_io_stats_mt_ios_stat);
-        if (iosstat) {
-                LOCK_INIT (&iosstat->lock);
-                iosstat->filename = gf_strdup(path);
-                uuid_copy (iosstat->gfid, buf->ia_gfid);
-                ios_inode_ctx_set (inode, this, iosstat);
-        }
+    UPDATE_PROFILE_STATS(frame, MKDIR);
+    if (op_ret < 0)
+        goto unwind;
+
+    /* allocate a struct ios_stat and set the inode ctx */
+    ios_init_iosstat(this, path, buf->ia_gfid, inode);
 
 unwind:
-        /* local is assigned with path */
-        GF_FREE (frame->local);
-        frame->local = NULL;
-        STACK_UNWIND_STRICT (mkdir, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent, xdata);
-        return 0;
+    /* local is assigned with path */
+    GF_FREE(frame->local);
+    frame->local = NULL;
+    STACK_UNWIND_STRICT(mkdir, frame, op_ret, op_errno, inode, buf, preparent,
+                        postparent, xdata);
+    return 0;
 }
 
-
 int
-io_stats_link_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno,
-                   inode_t *inode, struct iatt *buf,
-                   struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+io_stats_link_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, inode_t *inode,
+                  struct iatt *buf, struct iatt *preparent,
+                  struct iatt *postparent, dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, LINK);
-        STACK_UNWIND_STRICT (link, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, LINK);
+    STACK_UNWIND_STRICT(link, frame, op_ret, op_errno, inode, buf, preparent,
+                        postparent, xdata);
+    return 0;
 }
 
-
 int
-io_stats_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
+io_stats_flush_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, FLUSH);
-        STACK_UNWIND_STRICT (flush, frame, op_ret, op_errno, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, FLUSH);
+    STACK_UNWIND_STRICT(flush, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
-
 int
-io_stats_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
+io_stats_opendir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
 {
-        struct ios_stat *iosstat = NULL;
-        int              ret     = -1;
+    struct ios_stat *iosstat = NULL;
+    int ret = -1;
 
-        UPDATE_PROFILE_STATS (frame, OPENDIR);
-        if (op_ret < 0)
-                goto unwind;
+    UPDATE_PROFILE_STATS(frame, OPENDIR);
+    if (op_ret < 0)
+        goto unwind;
 
-        ios_fd_ctx_set (fd, this, 0);
+    ios_fd_ctx_set(fd, this, 0);
 
-        ret = ios_inode_ctx_get (fd->inode, this, &iosstat);
-        if (!ret)
-                BUMP_STATS (iosstat, IOS_STATS_TYPE_OPENDIR);
+    ret = ios_inode_ctx_get(fd->inode, this, &iosstat);
+    if (!ret)
+        ios_bump_stats(this, iosstat, IOS_STATS_TYPE_OPENDIR);
 
 unwind:
-        STACK_UNWIND_STRICT (opendir, frame, op_ret, op_errno, fd, xdata);
-        return 0;
+    STACK_UNWIND_STRICT(opendir, frame, op_ret, op_errno, fd, xdata);
+    return 0;
 }
 
-
 int
-io_stats_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno,
-                    struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+io_stats_rmdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                   struct iatt *postparent, dict_t *xdata)
 {
+    UPDATE_PROFILE_STATS(frame, RMDIR);
 
-        UPDATE_PROFILE_STATS (frame, RMDIR);
-
-        STACK_UNWIND_STRICT (rmdir, frame, op_ret, op_errno,
-                             preparent, postparent, xdata);
-        return 0;
+    STACK_UNWIND_STRICT(rmdir, frame, op_ret, op_errno, preparent, postparent,
+                        xdata);
+    return 0;
 }
 
+int
+io_stats_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                      struct iatt *postbuf, dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, TRUNCATE);
+    STACK_UNWIND_STRICT(truncate, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+    return 0;
+}
 
 int
-io_stats_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno,
-                       struct iatt *prebuf, struct iatt *postbuf, dict_t *xdata)
+io_stats_statfs_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct statvfs *buf,
+                    dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, TRUNCATE);
-        STACK_UNWIND_STRICT (truncate, frame, op_ret, op_errno,
-                             prebuf, postbuf, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, STATFS);
+    STACK_UNWIND_STRICT(statfs, frame, op_ret, op_errno, buf, xdata);
+    return 0;
 }
 
+int
+io_stats_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, SETXATTR);
+    STACK_UNWIND_STRICT(setxattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
 int
-io_stats_statfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, struct statvfs *buf, dict_t *xdata)
+io_stats_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *dict,
+                      dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, STATFS);
-        STACK_UNWIND_STRICT (statfs, frame, op_ret, op_errno, buf, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, GETXATTR);
+    STACK_UNWIND_STRICT(getxattr, frame, op_ret, op_errno, dict, xdata);
+    return 0;
 }
 
+int
+io_stats_removexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, REMOVEXATTR);
+    STACK_UNWIND_STRICT(removexattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
 int
-io_stats_setxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+io_stats_fsetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, SETXATTR);
-        STACK_UNWIND_STRICT (setxattr, frame, op_ret, op_errno, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, FSETXATTR);
+    STACK_UNWIND_STRICT(fsetxattr, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
-
 int
-io_stats_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+io_stats_fgetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *dict,
+                       dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, GETXATTR);
-        STACK_UNWIND_STRICT (getxattr, frame, op_ret, op_errno, dict, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, FGETXATTR);
+    STACK_UNWIND_STRICT(fgetxattr, frame, op_ret, op_errno, dict, xdata);
+    return 0;
 }
 
-
 int
-io_stats_removexattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+io_stats_fremovexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
                           int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, REMOVEXATTR);
-        STACK_UNWIND_STRICT (removexattr, frame, op_ret, op_errno, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, FREMOVEXATTR);
+    STACK_UNWIND_STRICT(fremovexattr, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
 int
-io_stats_fsetxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+io_stats_fsyncdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, FSETXATTR);
-        STACK_UNWIND_STRICT (fsetxattr, frame, op_ret, op_errno, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, FSYNCDIR);
+    STACK_UNWIND_STRICT(fsyncdir, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
-
 int
-io_stats_fgetxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+io_stats_access_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, FGETXATTR);
-        STACK_UNWIND_STRICT (fgetxattr, frame, op_ret, op_errno, dict, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, ACCESS);
+    STACK_UNWIND_STRICT(access, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
-
 int
-io_stats_fremovexattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                           int32_t op_ret, int32_t op_errno, dict_t *xdata)
+io_stats_ftruncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                       struct iatt *postbuf, dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, FREMOVEXATTR);
-        STACK_UNWIND_STRICT (fremovexattr, frame, op_ret, op_errno, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, FTRUNCATE);
+    STACK_UNWIND_STRICT(ftruncate, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+    return 0;
 }
 
-
 int
-io_stats_fsyncdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
+io_stats_fstat_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                   dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, FSYNCDIR);
-        STACK_UNWIND_STRICT (fsyncdir, frame, op_ret, op_errno, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, FSTAT);
+    STACK_UNWIND_STRICT(fstat, frame, op_ret, op_errno, buf, xdata);
+    return 0;
 }
 
-
 int
-io_stats_access_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, dict_t *xdata)
+io_stats_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                       struct iatt *postbuf, dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, ACCESS);
-        STACK_UNWIND_STRICT (access, frame, op_ret, op_errno, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, FALLOCATE);
+    STACK_UNWIND_STRICT(fallocate, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+    return 0;
 }
 
-
 int
-io_stats_ftruncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno,
-                        struct iatt *prebuf, struct iatt *postbuf, dict_t *xdata)
+io_stats_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, FTRUNCATE);
-        STACK_UNWIND_STRICT (ftruncate, frame, op_ret, op_errno,
-                             prebuf, postbuf, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, DISCARD);
+    STACK_UNWIND_STRICT(discard, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+    return 0;
 }
 
-
 int
-io_stats_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, struct iatt *buf, dict_t *xdata)
+io_stats_zerofill_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                      struct iatt *postbuf, dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, FSTAT);
-        STACK_UNWIND_STRICT (fstat, frame, op_ret, op_errno, buf, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, ZEROFILL);
+    STACK_UNWIND_STRICT(zerofill, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+    return 0;
 }
 
+int32_t
+io_stats_ipc_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, IPC);
+    STACK_UNWIND_STRICT(ipc, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
 int
-io_stats_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
-		       int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-		       struct iatt *postbuf, dict_t *xdata)
+io_stats_lk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct gf_flock *lock,
+                dict_t *xdata)
 {
-	UPDATE_PROFILE_STATS(frame, FALLOCATE);
-	STACK_UNWIND_STRICT(fallocate, frame, op_ret, op_errno, prebuf, postbuf,
-			    xdata);
-	return 0;
+    UPDATE_PROFILE_STATS(frame, LK);
+    STACK_UNWIND_STRICT(lk, frame, op_ret, op_errno, lock, xdata);
+    return 0;
 }
 
-
 int
-io_stats_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
-		     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-		     struct iatt *postbuf, dict_t *xdata)
+io_stats_entrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	UPDATE_PROFILE_STATS(frame, DISCARD);
-	STACK_UNWIND_STRICT(discard, frame, op_ret, op_errno, prebuf, postbuf,
-			    xdata);
-	return 0;
+    UPDATE_PROFILE_STATS(frame, ENTRYLK);
+    STACK_UNWIND_STRICT(entrylk, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
+int
+io_stats_fentrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, FENTRYLK);
+    STACK_UNWIND_STRICT(fentrylk, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
 int
-io_stats_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno, struct gf_flock *lock, dict_t *xdata)
+io_stats_rchecksum_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, uint32_t weak_checksum,
+                       uint8_t *strong_checksum, dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, LK);
-        STACK_UNWIND_STRICT (lk, frame, op_ret, op_errno, lock, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, RCHECKSUM);
+    STACK_UNWIND_STRICT(rchecksum, frame, op_ret, op_errno, weak_checksum,
+                        strong_checksum, xdata);
+    return 0;
 }
 
+int
+io_stats_seek_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, off_t offset, dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, SEEK);
+    STACK_UNWIND_STRICT(seek, frame, op_ret, op_errno, offset, xdata);
+    return 0;
+}
 
 int
-io_stats_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
+io_stats_lease_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct gf_lease *lease,
+                   dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, ENTRYLK);
-        STACK_UNWIND_STRICT (entrylk, frame, op_ret, op_errno, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, LEASE);
+    STACK_UNWIND_STRICT(lease, frame, op_ret, op_errno, lease, xdata);
+    return 0;
 }
 
+int
+io_stats_getactivelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno,
+                         lock_migration_info_t *locklist, dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, GETACTIVELK);
+    STACK_UNWIND_STRICT(getactivelk, frame, op_ret, op_errno, locklist, xdata);
+    return 0;
+}
 
 int
-io_stats_xattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+io_stats_setactivelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, XATTROP);
-        STACK_UNWIND_STRICT (xattrop, frame, op_ret, op_errno, dict, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, SETACTIVELK);
+    STACK_UNWIND_STRICT(setactivelk, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
+int
+io_stats_compound_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, void *data,
+                      dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, COMPOUND);
+    STACK_UNWIND_STRICT(compound, frame, op_ret, op_errno, data, xdata);
+    return 0;
+}
 
 int
-io_stats_fxattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+io_stats_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, dict_t *dict,
+                     dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, FXATTROP);
-        STACK_UNWIND_STRICT (fxattrop, frame, op_ret, op_errno, dict, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, XATTROP);
+    STACK_UNWIND_STRICT(xattrop, frame, op_ret, op_errno, dict, xdata);
+    return 0;
 }
 
+int
+io_stats_fxattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *dict,
+                      dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, FXATTROP);
+    STACK_UNWIND_STRICT(fxattrop, frame, op_ret, op_errno, dict, xdata);
+    return 0;
+}
 
 int
-io_stats_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
+io_stats_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        UPDATE_PROFILE_STATS (frame, INODELK);
-        STACK_UNWIND_STRICT (inodelk, frame, op_ret, op_errno, xdata);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, INODELK);
+    STACK_UNWIND_STRICT(inodelk, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
 int
-io_stats_entrylk (call_frame_t *frame, xlator_t *this,
-                  const char *volume, loc_t *loc, const char *basename,
-                  entrylk_cmd cmd, entrylk_type type, dict_t *xdata)
+io_stats_entrylk(call_frame_t *frame, xlator_t *this, const char *volume,
+                 loc_t *loc, const char *basename, entrylk_cmd cmd,
+                 entrylk_type type, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_entrylk_cbk,
-                    FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->entrylk,
-                    volume, loc, basename, cmd, type, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, volume, loc, basename, cmd,
+               type, xdata);
+    return 0;
 }
 
-
 int
-io_stats_inodelk (call_frame_t *frame, xlator_t *this,
-                  const char *volume, loc_t *loc, int32_t cmd, struct gf_flock *flock, dict_t *xdata)
+io_stats_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume,
+                  fd_t *fd, const char *basename, entrylk_cmd cmd,
+                  entrylk_type type, dict_t *xdata)
 {
+    START_FOP_LATENCY(frame);
 
-        START_FOP_LATENCY (frame);
-
-        STACK_WIND (frame, io_stats_inodelk_cbk,
-                    FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->inodelk,
-                    volume, loc, cmd, flock, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_fentrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fentrylk, volume, fd, basename, cmd,
+               type, xdata);
+    return 0;
 }
 
-
 int
-io_stats_finodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
+io_stats_inodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+                 loc_t *loc, int32_t cmd, struct gf_flock *flock, dict_t *xdata)
 {
+    START_FOP_LATENCY(frame);
 
-        UPDATE_PROFILE_STATS (frame, FINODELK);
-        STACK_UNWIND_STRICT (finodelk, frame, op_ret, op_errno, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_inodelk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->inodelk, volume, loc, cmd, flock,
+               xdata);
+    return 0;
 }
 
+int
+io_stats_finodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, FINODELK);
+    STACK_UNWIND_STRICT(finodelk, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
 int
-io_stats_finodelk (call_frame_t *frame, xlator_t *this, const char *volume,
-                   fd_t *fd, int32_t cmd, struct gf_flock *flock, dict_t *xdata)
+io_stats_finodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+                  fd_t *fd, int32_t cmd, struct gf_flock *flock, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_finodelk_cbk,
-                    FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->finodelk,
-                    volume, fd, cmd, flock, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_finodelk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->finodelk, volume, fd, cmd, flock,
+               xdata);
+    return 0;
 }
 
-
 int
-io_stats_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                  gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+io_stats_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                 gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_xattrop_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->xattrop,
-                    loc, flags, dict, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_xattrop_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->xattrop, loc, flags, dict, xdata);
+    return 0;
 }
 
-
 int
-io_stats_fxattrop (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                   gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+io_stats_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                  gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_fxattrop_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fxattrop,
-                    fd, flags, dict, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_fxattrop_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fxattrop, fd, flags, dict, xdata);
+    return 0;
 }
 
-
 int
-io_stats_lookup (call_frame_t *frame, xlator_t *this,
-                 loc_t *loc, dict_t *xdata)
+io_stats_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_lookup_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->lookup,
-                    loc, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_lookup_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lookup, loc, xdata);
+    return 0;
 }
 
-
 int
-io_stats_stat (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+io_stats_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_stat_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->stat,
-                    loc, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_stat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->stat, loc, xdata);
+    return 0;
 }
 
-
 int
-io_stats_readlink (call_frame_t *frame, xlator_t *this,
-                   loc_t *loc, size_t size, dict_t *xdata)
+io_stats_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+                  dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_readlink_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->readlink,
-                    loc, size, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_readlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readlink, loc, size, xdata);
+    return 0;
 }
 
-
 int
-io_stats_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                mode_t mode, dev_t dev, mode_t umask, dict_t *xdata)
+io_stats_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+               dev_t dev, mode_t umask, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_mknod_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->mknod,
-                    loc, mode, dev, umask, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_mknod_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mknod, loc, mode, dev, umask, xdata);
+    return 0;
 }
 
-
 int
-io_stats_mkdir (call_frame_t *frame, xlator_t *this,
-                loc_t *loc, mode_t mode, mode_t umask, dict_t *xdata)
+io_stats_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+               mode_t umask, dict_t *xdata)
 {
-        frame->local = gf_strdup (loc->path);
+    if (loc->path)
+        frame->local = gf_strdup(loc->path);
 
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_mkdir_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->mkdir,
-                    loc, mode, umask, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_mkdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mkdir, loc, mode, umask, xdata);
+    return 0;
 }
 
-
 int
-io_stats_unlink (call_frame_t *frame, xlator_t *this,
-                 loc_t *loc, int xflag, dict_t *xdata)
+io_stats_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+                dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_unlink_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->unlink,
-                    loc, xflag, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_unlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->unlink, loc, xflag, xdata);
+    return 0;
 }
 
-
 int
-io_stats_rmdir (call_frame_t *frame, xlator_t *this,
-                loc_t *loc, int flags, dict_t *xdata)
+io_stats_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+               dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_rmdir_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->rmdir,
-                    loc, flags, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_rmdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rmdir, loc, flags, xdata);
+    return 0;
 }
 
-
 int
-io_stats_symlink (call_frame_t *frame, xlator_t *this, const char *linkpath,
-                  loc_t *loc, mode_t umask, dict_t *xdata)
+io_stats_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath,
+                 loc_t *loc, mode_t umask, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_symlink_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->symlink,
-                    linkpath, loc, umask, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_symlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->symlink, linkpath, loc, umask, xdata);
+    return 0;
 }
 
-
 int
-io_stats_rename (call_frame_t *frame, xlator_t *this,
-                 loc_t *oldloc, loc_t *newloc, dict_t *xdata)
+io_stats_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc,
+                loc_t *newloc, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_rename_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->rename,
-                    oldloc, newloc, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_rename_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rename, oldloc, newloc, xdata);
+    return 0;
 }
 
-
 int
-io_stats_link (call_frame_t *frame, xlator_t *this,
-               loc_t *oldloc, loc_t *newloc, dict_t *xdata)
+io_stats_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+              dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_link_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->link,
-                    oldloc, newloc, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_link_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->link, oldloc, newloc, xdata);
+    return 0;
 }
 
-
 int
-io_stats_setattr (call_frame_t *frame, xlator_t *this,
-                  loc_t *loc, struct iatt *stbuf, int32_t valid, dict_t *xdata)
+io_stats_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                 struct iatt *stbuf, int32_t valid, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_setattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->setattr,
-                    loc, stbuf, valid, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_setattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setattr, loc, stbuf, valid, xdata);
+    return 0;
 }
 
-
 int
-io_stats_truncate (call_frame_t *frame, xlator_t *this,
-                   loc_t *loc, off_t offset, dict_t *xdata)
+io_stats_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+                  dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_truncate_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->truncate,
-                    loc, offset, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_truncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
+    return 0;
 }
 
-
 int
-io_stats_open (call_frame_t *frame, xlator_t *this, loc_t *loc,
-               int32_t flags, fd_t *fd, dict_t *xdata)
+io_stats_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+              fd_t *fd, dict_t *xdata)
 {
-        frame->local = gf_strdup (loc->path);
+    if (loc->path)
+        frame->local = gf_strdup(loc->path);
 
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_open_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->open,
-                    loc, flags, fd, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_open_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->open, loc, flags, fd, xdata);
+    return 0;
 }
 
-
 int
-io_stats_create (call_frame_t *frame, xlator_t *this,
-                 loc_t *loc, int32_t flags, mode_t mode,
-                 mode_t umask, fd_t *fd, dict_t *xdata)
+io_stats_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+                mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
 {
-        frame->local = gf_strdup (loc->path);
+    if (loc->path)
+        frame->local = gf_strdup(loc->path);
 
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_create_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->create,
-                    loc, flags, mode, umask, fd, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_create_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->create, loc, flags, mode, umask, fd,
+               xdata);
+    return 0;
 }
 
-
 int
-io_stats_readv (call_frame_t *frame, xlator_t *this,
-                fd_t *fd, size_t size, off_t offset, uint32_t flags, dict_t *xdata)
+io_stats_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+               off_t offset, uint32_t flags, dict_t *xdata)
 {
-        frame->local = fd;
+    frame->local = fd;
 
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_readv_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->readv,
-                    fd, size, offset, flags, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_readv_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readv, fd, size, offset, flags, xdata);
+    return 0;
 }
 
-
 int
-io_stats_writev (call_frame_t *frame, xlator_t *this,
-                 fd_t *fd, struct iovec *vector,
-                 int32_t count, off_t offset,
-                 uint32_t flags, struct iobref *iobref, dict_t *xdata)
+io_stats_writev(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                struct iovec *vector, int32_t count, off_t offset,
+                uint32_t flags, struct iobref *iobref, dict_t *xdata)
 {
-        int                 len = 0;
+    int len = 0;
 
-        if (fd->inode)
-                frame->local = fd->inode;
-        len = iov_length (vector, count);
-
-        BUMP_WRITE (fd, len);
-        START_FOP_LATENCY (frame);
+    if (fd->inode)
+        frame->local = fd->inode;
+    len = iov_length(vector, count);
 
-        STACK_WIND (frame, io_stats_writev_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->writev,
-                    fd, vector, count, offset, flags, iobref, xdata);
-        return 0;
+    ios_bump_write(this, fd, len);
+    START_FOP_LATENCY(frame);
 
+    STACK_WIND(frame, io_stats_writev_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->writev, fd, vector, count, offset,
+               flags, iobref, xdata);
+    return 0;
 }
 
-
 int
-io_stats_statfs (call_frame_t *frame, xlator_t *this,
-                 loc_t *loc, dict_t *xdata)
+io_stats_copy_file_range(call_frame_t *frame, xlator_t *this, fd_t *fd_in,
+                         off_t off_in, fd_t *fd_out, off_t off_out, size_t len,
+                         uint32_t flags, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_statfs_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->statfs,
-                    loc, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_copy_file_range_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->copy_file_range, fd_in, off_in, fd_out,
+               off_out, len, flags, xdata);
+    return 0;
 }
 
-
 int
-io_stats_flush (call_frame_t *frame, xlator_t *this,
-                fd_t *fd, dict_t *xdata)
+io_stats_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_flush_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->flush,
-                    fd, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_statfs_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->statfs, loc, xdata);
+    return 0;
 }
 
-
 int
-io_stats_fsync (call_frame_t *frame, xlator_t *this,
-                fd_t *fd, int32_t flags, dict_t *xdata)
+io_stats_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_fsync_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fsync,
-                    fd, flags, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_flush_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->flush, fd, xdata);
+    return 0;
 }
 
-
 int
-conditional_dump (dict_t *dict, char *key, data_t *value, void *data)
+io_stats_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+               dict_t *xdata)
 {
-        struct {
-                xlator_t       *this;
-                inode_t        *inode;
-                const char     *path;
-        } *stub;
-        xlator_t             *this = NULL;
-        char                 *filename = NULL;
-        FILE                 *logfp = NULL;
-        struct ios_dump_args args = {0};
+    START_FOP_LATENCY(frame);
 
-        stub  = data;
-        this  = stub->this;
+    STACK_WIND(frame, io_stats_fsync_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsync, fd, flags, xdata);
+    return 0;
+}
 
-        filename = alloca (value->len + 1);
-        memset (filename, 0, value->len + 1);
-        memcpy (filename, data_to_str (value), value->len);
+int
+conditional_dump(dict_t *dict, char *key, data_t *value, void *data)
+{
+    struct {
+        xlator_t *this;
+        inode_t *inode;
+        const char *path;
+    } * stub;
+    xlator_t *this = NULL;
+    char *filename = NULL;
+    FILE *logfp = NULL;
+    struct ios_dump_args args = {0};
+    int pid, namelen, dirlen;
+    char dump_key[100];
+    char *slash_ptr = NULL;
+    char *path_in_value = NULL;
+    char *identifier = NULL;
+    struct ios_conf *conf = NULL;
+
+    stub = data;
+    this = stub->this;
+    conf = this->private;
+
+    /* Don't do this on 'brick-side', only do this on client side */
+    /* Addresses CVE-2018-14659 */
+    if (this->ctx->process_mode != GF_CLIENT_PROCESS) {
+        gf_log(this->name, GF_LOG_DEBUG,
+               "taking io-stats dump using setxattr not permitted on brick."
+               " Use 'gluster profile' instead");
+        return -1;
+    }
+
+    /* Create a file name that is appended with the io-stats instance
+    name as well. This helps when there is more than a single io-stats
+    instance in the graph, or the client and server processes are running
+    on the same node */
+    /* For the sanity of where the file should be located, we should make
+       sure file is written only inside RUNDIR (ie, /var/run/gluster) */
+    /* TODO: provide an option to dump it to different directory of
+       choice, based on options */
+    /* name format: /var/run/gluster/<passed in path/filename>.<xlator name
+     * slashes to -> */
+
+    path_in_value = alloca0(value->len + 1);
+
+    /* We need a memcpy here because of the way dict_unserialize works */
+
+    memcpy(path_in_value, data_to_str(value), value->len);
+    path_in_value[value->len] = '\0';
+
+    if (strstr(path_in_value, "../")) {
+        gf_log(this->name, GF_LOG_ERROR, "%s: no \"../\" allowed in path",
+               path_in_value);
+        return -1;
+    }
+
+    if (path_in_value[0] == '/') {
+        path_in_value = path_in_value + 1;
+    }
+
+    dirlen = strlen(IOS_STATS_DUMP_DIR);
+    if (conf->unique_id) {
+        /* this->name will be the same for all bricks of the volume */
+        identifier = conf->unique_id;
+    } else {
+        identifier = this->name;
+    }
+
+    namelen = (dirlen + value->len + strlen(identifier) + 3);
+    /* +3 for '/', '.' and '\0' added in snprintf below*/
+
+    filename = alloca0(namelen);
+    snprintf(filename, namelen, "%s/%s.%s", IOS_STATS_DUMP_DIR, path_in_value,
+             identifier);
+
+    /* convert any slashes to '-' so that fopen works correctly */
+    slash_ptr = strchr(filename + dirlen + 1, '/');
+    while (slash_ptr) {
+        *slash_ptr = '-';
+        slash_ptr = strchr(slash_ptr, '/');
+    }
+
+    pid = getpid();
+
+    if (!strncmp(filename, "", 1)) {
+        gf_log(this->name, GF_LOG_ERROR, "No filename given");
+        return -1;
+    }
+    logfp = fopen(filename, "w+");
+    if (!logfp) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "failed to open %s "
+               "for writing",
+               filename);
+        return -1;
+    }
+    sprintf(dump_key, "*io*stat*%d_json_dump", pid);
+    if (fnmatch(dump_key, key, 0) == 0) {
+        (void)ios_dump_args_init(&args, IOS_DUMP_TYPE_JSON_FILE, logfp);
+    } else {
+        (void)ios_dump_args_init(&args, IOS_DUMP_TYPE_FILE, logfp);
+    }
+    io_stats_dump(this, &args, GF_IOS_INFO_ALL, _gf_false);
+    fclose(logfp);
+    return 0;
+}
 
-        if (fnmatch ("*io*stat*dump", key, 0) == 0) {
+int
+_ios_destroy_dump_thread(struct ios_conf *conf)
+{
+    conf->dump_thread_should_die = _gf_true;
+    if (conf->dump_thread_running) {
+        (void)pthread_cancel(conf->dump_thread);
+        (void)pthread_join(conf->dump_thread, NULL);
+    }
+    return 0;
+}
 
-                if (!strncmp (filename, "", 1)) {
-                        gf_log (this->name, GF_LOG_ERROR, "No filename given");
-                        return -1;
-                }
-                logfp = fopen (filename, "w+");
-                if (!logfp) {
-                        gf_log (this->name, GF_LOG_ERROR, "failed to open %s "
-                                "for writing", filename);
-                        return -1;
-                }
-                (void) ios_dump_args_init (&args, IOS_DUMP_TYPE_FILE,
-                                           logfp);
-                io_stats_dump (this, &args);
-                fclose (logfp);
+void *
+_ios_dump_thread(xlator_t *this)
+{
+    struct ios_conf *conf = NULL;
+    FILE *stats_logfp = NULL;
+    FILE *samples_logfp = NULL;
+    struct ios_dump_args args = {0};
+    int i;
+    int stats_bytes_written = 0;
+    int samples_bytes_written = 0;
+    char stats_filename[PATH_MAX];
+    char samples_filename[PATH_MAX];
+    char *xlator_name;
+    char *instance_name;
+    gf_boolean_t log_stats_fopen_failure = _gf_true;
+    gf_boolean_t log_samples_fopen_failure = _gf_true;
+    int old_cancel_type;
+
+    conf = this->private;
+    gf_log(this->name, GF_LOG_INFO,
+           "IO stats dump thread started, "
+           "polling IO stats every %d seconds",
+           conf->ios_dump_interval);
+    xlator_name = strdupa(conf->unique_id);
+    for (i = 0; i < strlen(xlator_name); i++) {
+        if (xlator_name[i] == '/')
+            xlator_name[i] = '_';
+    }
+    instance_name = this->instance_name;
+    if (this->name && strcmp(this->name, "glustershd") == 0) {
+        xlator_name = "shd";
+    } else if (this->prev && strcmp(this->prev->name, "nfs-server") == 0) {
+        xlator_name = "nfsd";
+        instance_name = this->prev->instance_name;
+    }
+    if (sys_mkdir(_IOS_DUMP_DIR, S_IRWXU | S_IRWXO | S_IRWXG) == (-1)) {
+        if (errno != EEXIST) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "could not create stats-dump directory %s", _IOS_DUMP_DIR);
+            goto out;
         }
-        return 0;
+    }
+    if (sys_mkdir(_IOS_SAMP_DIR, S_IRWXU | S_IRWXO | S_IRWXG) == (-1)) {
+        if (errno != EEXIST) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "could not create stats-sample directory %s", _IOS_SAMP_DIR);
+            goto out;
+        }
+    }
+    if (instance_name) {
+        stats_bytes_written = snprintf(stats_filename, PATH_MAX,
+                                       "%s/%s_%s_%s.dump", _IOS_DUMP_DIR,
+                                       __progname, xlator_name, instance_name);
+        samples_bytes_written = snprintf(
+            samples_filename, PATH_MAX, "%s/%s_%s_%s.samp", _IOS_SAMP_DIR,
+            __progname, xlator_name, instance_name);
+    } else {
+        stats_bytes_written = snprintf(stats_filename, PATH_MAX,
+                                       "%s/%s_%s.dump", _IOS_DUMP_DIR,
+                                       __progname, xlator_name);
+        samples_bytes_written = snprintf(samples_filename, PATH_MAX,
+                                         "%s/%s_%s.samp", _IOS_SAMP_DIR,
+                                         __progname, xlator_name);
+    }
+    if ((stats_bytes_written >= PATH_MAX) ||
+        (samples_bytes_written >= PATH_MAX)) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Invalid path for stats dump (%s) and/or latency "
+               "samples (%s)",
+               stats_filename, samples_filename);
+        goto out;
+    }
+    while (1) {
+        if (conf->dump_thread_should_die)
+            break;
+        (void)pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS,
+                                    &old_cancel_type);
+        sleep(conf->ios_dump_interval);
+        (void)pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, &old_cancel_type);
+        /*
+         * It's not clear whether we should reopen this each time, or
+         * just hold it open and rewind/truncate on each iteration.
+         * Leaving it alone for now.
+         */
+        stats_logfp = fopen(stats_filename, "w+");
+        if (stats_logfp) {
+            (void)ios_dump_args_init(&args, conf->dump_format, stats_logfp);
+            io_stats_dump(this, &args, GF_IOS_INFO_ALL, _gf_false);
+            fclose(stats_logfp);
+            log_stats_fopen_failure = _gf_true;
+        } else if (log_stats_fopen_failure) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "could not open stats-dump file %s (%s)", stats_filename,
+                   strerror(errno));
+            log_stats_fopen_failure = _gf_false;
+        }
+        samples_logfp = fopen(samples_filename, "w+");
+        if (samples_logfp) {
+            io_stats_dump_latency_samples_logfp(this, samples_logfp);
+            fclose(samples_logfp);
+            log_samples_fopen_failure = _gf_true;
+        } else if (log_samples_fopen_failure) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "could not open samples-dump file %s (%s)", samples_filename,
+                   strerror(errno));
+            log_samples_fopen_failure = _gf_false;
+        }
+    }
+out:
+    conf->dump_thread_running = _gf_false;
+    gf_log(this->name, GF_LOG_INFO, "IO stats dump thread terminated");
+    return NULL;
 }
 
+static gf_boolean_t
+match_special_xattr(dict_t *d, char *k, data_t *val, void *mdata)
+{
+    gf_boolean_t ret = _gf_false;
+    if (fnmatch("*io*stat*dump", k, 0) == 0) {
+        ret = _gf_true;
+    }
+
+    return ret;
+}
 
 int
-io_stats_setxattr (call_frame_t *frame, xlator_t *this,
-                   loc_t *loc, dict_t *dict,
-                   int32_t flags, dict_t *xdata)
+io_stats_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+                  int32_t flags, dict_t *xdata)
 {
-        struct {
-                xlator_t     *this;
-                inode_t      *inode;
-                const char   *path;
-        } stub;
+    struct {
+        xlator_t *this;
+        inode_t *inode;
+        const char *path;
+    } stub;
 
-        stub.this  = this;
-        stub.inode = loc->inode;
-        stub.path  = loc->path;
+    stub.this = this;
+    stub.inode = loc->inode;
+    stub.path = loc->path;
 
-        dict_foreach (dict, conditional_dump, &stub);
+    (void)dict_foreach_match(dict, match_special_xattr, NULL, conditional_dump,
+                             &stub);
 
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_setxattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->setxattr,
-                    loc, dict, flags, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_setxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setxattr, loc, dict, flags, xdata);
+    return 0;
 }
 
-
 int
-io_stats_getxattr (call_frame_t *frame, xlator_t *this,
-                   loc_t *loc, const char *name, dict_t *xdata)
+io_stats_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                  const char *name, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_getxattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->getxattr,
-                    loc, name, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_getxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->getxattr, loc, name, xdata);
+    return 0;
 }
 
-
 int
-io_stats_removexattr (call_frame_t *frame, xlator_t *this,
-                      loc_t *loc, const char *name, dict_t *xdata)
+io_stats_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                     const char *name, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_removexattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->removexattr,
-                    loc, name, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_removexattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->removexattr, loc, name, xdata);
+    return 0;
 }
 
-
 int
-io_stats_fsetxattr (call_frame_t *frame, xlator_t *this,
-                    fd_t *fd, dict_t *dict,
-                    int32_t flags, dict_t *xdata)
+io_stats_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+                   int32_t flags, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_fsetxattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fsetxattr,
-                    fd, dict, flags, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_fsetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags, xdata);
+    return 0;
 }
 
-
 int
-io_stats_fgetxattr (call_frame_t *frame, xlator_t *this,
-                    fd_t *fd, const char *name, dict_t *xdata)
+io_stats_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                   const char *name, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_fgetxattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fgetxattr,
-                    fd, name, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_fgetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fgetxattr, fd, name, xdata);
+    return 0;
 }
 
-
 int
-io_stats_fremovexattr (call_frame_t *frame, xlator_t *this,
-                       fd_t *fd, const char *name, dict_t *xdata)
+io_stats_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                      const char *name, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_fremovexattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fremovexattr,
-                    fd, name, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_fremovexattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fremovexattr, fd, name, xdata);
+    return 0;
 }
 
-
 int
-io_stats_opendir (call_frame_t *frame, xlator_t *this,
-                  loc_t *loc, fd_t *fd, dict_t *xdata)
+io_stats_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+                 dict_t *xdata)
 {
+    START_FOP_LATENCY(frame);
 
-        START_FOP_LATENCY (frame);
-
-        STACK_WIND (frame, io_stats_opendir_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->opendir,
-                    loc, fd, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_opendir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->opendir, loc, fd, xdata);
+    return 0;
 }
 
 int
-io_stats_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-                   off_t offset, dict_t *dict)
+io_stats_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                  off_t offset, dict_t *dict)
 {
-        frame->local = fd->inode;
-        START_FOP_LATENCY (frame);
+    frame->local = fd->inode;
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_readdirp_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->readdirp,
-                    fd, size, offset, dict);
-        return 0;
+    STACK_WIND(frame, io_stats_readdirp_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readdirp, fd, size, offset, dict);
+    return 0;
 }
 
-
 int
-io_stats_readdir (call_frame_t *frame, xlator_t *this,
-                  fd_t *fd, size_t size, off_t offset, dict_t *xdata)
+io_stats_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                 off_t offset, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_readdir_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->readdir,
-                    fd, size, offset, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_readdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readdir, fd, size, offset, xdata);
+    return 0;
 }
 
-
 int
-io_stats_fsyncdir (call_frame_t *frame, xlator_t *this,
-                   fd_t *fd, int32_t datasync, dict_t *xdata)
+io_stats_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                  int32_t datasync, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_fsyncdir_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fsyncdir,
-                    fd, datasync, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_fsyncdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsyncdir, fd, datasync, xdata);
+    return 0;
 }
 
-
 int
-io_stats_access (call_frame_t *frame, xlator_t *this,
-                 loc_t *loc, int32_t mask, dict_t *xdata)
+io_stats_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+                dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_access_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->access,
-                    loc, mask, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_access_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->access, loc, mask, xdata);
+    return 0;
 }
 
-
 int
-io_stats_ftruncate (call_frame_t *frame, xlator_t *this,
-                    fd_t *fd, off_t offset, dict_t *xdata)
+io_stats_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                   dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_ftruncate_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->ftruncate,
-                    fd, offset, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_ftruncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+    return 0;
 }
 
-
 int
-io_stats_fsetattr (call_frame_t *frame, xlator_t *this,
-                   fd_t *fd, struct iatt *stbuf, int32_t valid, dict_t *xdata)
+io_stats_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                  struct iatt *stbuf, int32_t valid, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_setattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fsetattr,
-                    fd, stbuf, valid, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_setattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetattr, fd, stbuf, valid, xdata);
+    return 0;
 }
 
+int
+io_stats_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
+
+    STACK_WIND(frame, io_stats_fstat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fstat, fd, xdata);
+    return 0;
+}
 
 int
-io_stats_fstat (call_frame_t *frame, xlator_t *this,
-                fd_t *fd, dict_t *xdata)
+io_stats_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+                   off_t offset, size_t len, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_fstat_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fstat,
-                    fd, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_fallocate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fallocate, fd, mode, offset, len,
+               xdata);
+
+    return 0;
 }
 
+int
+io_stats_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                 size_t len, dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
+
+    STACK_WIND(frame, io_stats_discard_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->discard, fd, offset, len, xdata);
+
+    return 0;
+}
 
 int
-io_stats_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
-		   off_t offset, size_t len, dict_t *xdata)
+io_stats_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                  off_t len, dict_t *xdata)
 {
-	START_FOP_LATENCY(frame);
+    START_FOP_LATENCY(frame);
 
-	STACK_WIND(frame, io_stats_fallocate_cbk, FIRST_CHILD(this),
-		   FIRST_CHILD(this)->fops->fallocate, fd, mode, offset, len,
-		   xdata);
+    STACK_WIND(frame, io_stats_zerofill_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->zerofill, fd, offset, len, xdata);
 
-	return 0;
+    return 0;
 }
 
+int32_t
+io_stats_ipc(call_frame_t *frame, xlator_t *this, int32_t op, dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
+
+    STACK_WIND(frame, io_stats_ipc_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ipc, op, xdata);
+    return 0;
+}
 
 int
-io_stats_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
-		 size_t len, dict_t *xdata)
+io_stats_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
+            struct gf_flock *lock, dict_t *xdata)
 {
-	START_FOP_LATENCY(frame);
+    START_FOP_LATENCY(frame);
 
-	STACK_WIND(frame, io_stats_discard_cbk, FIRST_CHILD(this),
-		   FIRST_CHILD(this)->fops->discard, fd, offset, len, xdata);
+    STACK_WIND(frame, io_stats_lk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lk, fd, cmd, lock, xdata);
+    return 0;
+}
+
+int
+io_stats_rchecksum(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                   int32_t len, dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
 
-	return 0;
+    STACK_WIND(frame, io_stats_rchecksum_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rchecksum, fd, offset, len, xdata);
+    return 0;
 }
 
+int
+io_stats_seek(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+              gf_seek_what_t what, dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
+
+    STACK_WIND(frame, io_stats_seek_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->seek, fd, offset, what, xdata);
+    return 0;
+}
 
 int
-io_stats_lk (call_frame_t *frame, xlator_t *this,
-             fd_t *fd, int32_t cmd, struct gf_flock *lock, dict_t *xdata)
+io_stats_lease(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               struct gf_lease *lease, dict_t *xdata)
 {
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_lk_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->lk,
-                    fd, cmd, lock, xdata);
-        return 0;
+    STACK_WIND(frame, io_stats_lease_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lease, loc, lease, xdata);
+    return 0;
 }
 
+int
+io_stats_getactivelk(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                     dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
+
+    STACK_WIND(frame, io_stats_getactivelk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->getactivelk, loc, xdata);
+    return 0;
+}
 
 int
-io_stats_release (xlator_t *this, fd_t *fd)
+io_stats_setactivelk(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                     lock_migration_info_t *locklist, dict_t *xdata)
 {
-        struct ios_fd  *iosfd = NULL;
-        struct ios_conf *conf = NULL;
+    START_FOP_LATENCY(frame);
 
-        BUMP_FOP (RELEASE);
+    STACK_WIND(frame, io_stats_setactivelk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setactivelk, loc, locklist, xdata);
+    return 0;
+}
+
+int
+io_stats_compound(call_frame_t *frame, xlator_t *this, void *args,
+                  dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
+
+    STACK_WIND(frame, io_stats_compound_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->compound, args, xdata);
+    return 0;
+}
 
-        conf = this->private;
+int
+io_stats_release(xlator_t *this, fd_t *fd)
+{
+    struct ios_fd *iosfd = NULL;
+    struct ios_conf *conf = NULL;
 
-        LOCK (&conf->lock);
+    BUMP_FOP(RELEASE);
+
+    conf = this->private;
+    if (conf) {
+        LOCK(&conf->lock);
         {
-                conf->cumulative.nr_opens--;
+            conf->cumulative.nr_opens--;
         }
-        UNLOCK (&conf->lock);
+        UNLOCK(&conf->lock);
+    }
 
-        ios_fd_ctx_get (fd, this, &iosfd);
-        if (iosfd) {
-                io_stats_dump_fd (this, iosfd);
+    ios_fd_ctx_get(fd, this, &iosfd);
+    if (iosfd) {
+        io_stats_dump_fd(this, iosfd);
 
-                GF_FREE (iosfd->filename);
-                GF_FREE (iosfd);
-        }
+        GF_FREE(iosfd->filename);
+        GF_FREE(iosfd);
+    }
 
-        return 0;
+    return 0;
 }
 
-
 int
-io_stats_releasedir (xlator_t *this, fd_t *fd)
+io_stats_releasedir(xlator_t *this, fd_t *fd)
 {
-        BUMP_FOP (RELEASEDIR);
+    BUMP_FOP(RELEASEDIR);
 
-        return 0;
+    return 0;
 }
 
-
 int
-io_stats_forget (xlator_t *this, inode_t *inode)
+io_stats_forget(xlator_t *this, inode_t *inode)
 {
-        BUMP_FOP (FORGET);
-        ios_stats_cleanup (this, inode);
-        return 0;
+    BUMP_FOP(FORGET);
+    ios_stats_cleanup(this, inode);
+    return 0;
 }
 
 static int
-ios_init_top_stats (struct ios_conf *conf)
+ios_init_top_stats(struct ios_conf *conf)
 {
-        int     i = 0;
+    int i = 0;
 
-        GF_ASSERT (conf);
+    GF_ASSERT(conf);
 
-        for (i = 0; i <IOS_STATS_TYPE_MAX; i++) {
-                conf->list[i].iosstats = GF_CALLOC (1,
-                                         sizeof(*conf->list[i].iosstats),
-                                         gf_io_stats_mt_ios_stat);
+    for (i = 0; i < IOS_STATS_TYPE_MAX; i++) {
+        conf->list[i].iosstats = GF_CALLOC(1, sizeof(*conf->list[i].iosstats),
+                                           gf_io_stats_mt_ios_stat);
 
-                if (!conf->list[i].iosstats)
-                        return -1;
+        if (!conf->list[i].iosstats)
+            return -1;
 
-                INIT_LIST_HEAD(&conf->list[i].iosstats->list);
-                LOCK_INIT (&conf->list[i].lock);
-        }
+        INIT_LIST_HEAD(&conf->list[i].iosstats->list);
+        LOCK_INIT(&conf->list[i].lock);
+    }
 
-        for (i = 0; i < IOS_STATS_THRU_MAX; i ++) {
-                conf->thru_list[i].iosstats = GF_CALLOC (1,
-                                 sizeof (*conf->thru_list[i].iosstats),
-                                 gf_io_stats_mt_ios_stat);
+    for (i = 0; i < IOS_STATS_THRU_MAX; i++) {
+        conf->thru_list[i].iosstats = GF_CALLOC(
+            1, sizeof(*conf->thru_list[i].iosstats), gf_io_stats_mt_ios_stat);
 
-                if (!conf->thru_list[i].iosstats)
-                        return -1;
+        if (!conf->thru_list[i].iosstats)
+            return -1;
 
-                INIT_LIST_HEAD(&conf->thru_list[i].iosstats->list);
-                LOCK_INIT (&conf->thru_list[i].lock);
-        }
+        INIT_LIST_HEAD(&conf->thru_list[i].iosstats->list);
+        LOCK_INIT(&conf->thru_list[i].lock);
+    }
 
-        return 0;
+    return 0;
 }
 
 static void
-ios_destroy_top_stats (struct ios_conf *conf)
-{
-        int                     i = 0;
-        struct ios_stat_head    *list_head = NULL;
-        struct ios_stat_list    *entry     = NULL;
-        struct ios_stat_list    *tmp       = NULL;
-        struct ios_stat_list    *list      = NULL;
-        struct ios_stat         *stat      = NULL;
-
-        GF_ASSERT (conf);
-
-        LOCK (&conf->lock);
-
-        conf->cumulative.nr_opens = 0;
-        conf->cumulative.max_nr_opens = 0;
-        conf->cumulative.max_openfd_time.tv_sec = 0;
-        conf->cumulative.max_openfd_time.tv_usec = 0;
-
-        for (i = 0; i < IOS_STATS_TYPE_MAX; i++) {
-                list_head = &conf->list[i];
-                if (!list_head)
-                        continue;
-                list_for_each_entry_safe (entry, tmp,
-                                          &list_head->iosstats->list, list) {
-                        list = entry;
-                        stat = list->iosstat;
-                        ios_stat_unref (stat);
-                        list_del (&list->list);
-                        GF_FREE (list);
-                        list_head->members--;
-                }
+ios_destroy_top_stats(struct ios_conf *conf)
+{
+    int i = 0;
+    struct ios_stat_head *list_head = NULL;
+    struct ios_stat_list *entry = NULL;
+    struct ios_stat_list *tmp = NULL;
+    struct ios_stat_list *list = NULL;
+    struct ios_stat *stat = NULL;
+
+    GF_ASSERT(conf);
+
+    LOCK(&conf->lock);
+
+    conf->cumulative.nr_opens = 0;
+    conf->cumulative.max_nr_opens = 0;
+    conf->cumulative.max_openfd_time.tv_sec = 0;
+    conf->cumulative.max_openfd_time.tv_usec = 0;
+
+    for (i = 0; i < IOS_STATS_TYPE_MAX; i++) {
+        list_head = &conf->list[i];
+        if (!list_head)
+            continue;
+        list_for_each_entry_safe(entry, tmp, &list_head->iosstats->list, list)
+        {
+            list = entry;
+            stat = list->iosstat;
+            ios_stat_unref(stat);
+            list_del(&list->list);
+            GF_FREE(list);
+            list_head->members--;
         }
-
-        for (i = 0; i < IOS_STATS_THRU_MAX; i++) {
-                list_head = &conf->thru_list[i];
-                if (!list_head)
-                        continue;
-                list_for_each_entry_safe (entry, tmp,
-                                          &list_head->iosstats->list, list) {
-                        list = entry;
-                        stat = list->iosstat;
-                        ios_stat_unref (stat);
-                        list_del (&list->list);
-                        GF_FREE (list);
-                        list_head->members--;
-                }
+        GF_FREE(list_head->iosstats);
+    }
+
+    for (i = 0; i < IOS_STATS_THRU_MAX; i++) {
+        list_head = &conf->thru_list[i];
+        if (!list_head)
+            continue;
+        list_for_each_entry_safe(entry, tmp, &list_head->iosstats->list, list)
+        {
+            list = entry;
+            stat = list->iosstat;
+            ios_stat_unref(stat);
+            list_del(&list->list);
+            GF_FREE(list);
+            list_head->members--;
         }
+        GF_FREE(list_head->iosstats);
+    }
 
-        UNLOCK (&conf->lock);
+    UNLOCK(&conf->lock);
 
-        return;
+    return;
 }
 
-int
-reconfigure (xlator_t *this, dict_t *options)
+static void
+io_stats_clear(struct ios_conf *conf)
 {
-        struct ios_conf    *conf = NULL;
-        int                 ret = -1;
-        char               *sys_log_str = NULL;
-        int                 sys_log_level = -1;
-        char               *log_str = NULL;
-        int                 log_level = -1;
+    time_t now = 0;
+
+    GF_ASSERT(conf);
+    now = gf_time();
+
+    LOCK(&conf->lock);
+    {
+        ios_global_stats_clear(&conf->cumulative, now);
+        ios_global_stats_clear(&conf->incremental, now);
+        conf->increment = 0;
+    }
+    UNLOCK(&conf->lock);
+}
 
-        if (!this || !this->private)
-                goto out;
+int32_t
+io_priv(xlator_t *this)
+{
+    int i;
+    char key[GF_DUMP_MAX_BUF_LEN];
+    char key_prefix_cumulative[GF_DUMP_MAX_BUF_LEN];
+    char key_prefix_incremental[GF_DUMP_MAX_BUF_LEN];
+    double min, max, avg;
+    uint64_t count, total;
+    struct ios_conf *conf = NULL;
+
+    conf = this->private;
+    if (!conf)
+        return -1;
+
+    if (!conf->count_fop_hits || !conf->measure_latency)
+        return -1;
+
+    gf_proc_dump_write("cumulative.data_read", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(conf->cumulative.data_read));
+    gf_proc_dump_write("cumulative.data_written", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(conf->cumulative.data_written));
+
+    gf_proc_dump_write("incremental.data_read", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(conf->incremental.data_read));
+    gf_proc_dump_write("incremental.data_written", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(conf->incremental.data_written));
+
+    snprintf(key_prefix_cumulative, GF_DUMP_MAX_BUF_LEN, "%s.cumulative",
+             this->name);
+    snprintf(key_prefix_incremental, GF_DUMP_MAX_BUF_LEN, "%s.incremental",
+             this->name);
+
+    for (i = 0; i < GF_FOP_MAXVALUE; i++) {
+        count = GF_ATOMIC_GET(conf->cumulative.fop_hits[i]);
+        total = conf->cumulative.latency[i].total;
+        min = conf->cumulative.latency[i].min;
+        max = conf->cumulative.latency[i].max;
+        avg = conf->cumulative.latency[i].avg;
+
+        gf_proc_dump_build_key(key, key_prefix_cumulative, "%s",
+                               (char *)gf_fop_list[i]);
+
+        gf_proc_dump_write(key, "%" PRId64 ",%" PRId64 ",%.03f,%.03f,%.03f",
+                           count, total, min, max, avg);
+
+        count = GF_ATOMIC_GET(conf->incremental.fop_hits[i]);
+        total = conf->incremental.latency[i].total;
+        min = conf->incremental.latency[i].min;
+        max = conf->incremental.latency[i].max;
+        avg = conf->incremental.latency[i].avg;
+
+        gf_proc_dump_build_key(key, key_prefix_incremental, "%s",
+                               (char *)gf_fop_list[i]);
+
+        gf_proc_dump_write(key, "%" PRId64 ",%" PRId64 ",%.03f,%.03f,%.03f",
+                           count, total, min, max, avg);
+    }
+
+    return 0;
+}
 
-        conf = this->private;
+static void
+ios_set_log_format_code(struct ios_conf *conf, char *dump_format_str)
+{
+    if (strcmp(dump_format_str, "json") == 0)
+        conf->dump_format = IOS_DUMP_TYPE_JSON_FILE;
+    else if (strcmp(dump_format_str, "text") == 0)
+        conf->dump_format = IOS_DUMP_TYPE_FILE;
+    else if (strcmp(dump_format_str, "dict") == 0)
+        conf->dump_format = IOS_DUMP_TYPE_DICT;
+    else if (strcmp(dump_format_str, "samples") == 0)
+        conf->dump_format = IOS_DUMP_TYPE_SAMPLES;
+}
 
-        GF_OPTION_RECONF ("dump-fd-stats", conf->dump_fd_stats, options, bool,
-                          out);
+void
+xlator_set_loglevel(xlator_t *this, int log_level)
+{
+    glusterfs_ctx_t *ctx = NULL;
+    glusterfs_graph_t *active = NULL;
+    xlator_t *top = NULL;
+    xlator_t *trav = this;
 
-        GF_OPTION_RECONF ("count-fop-hits", conf->count_fop_hits, options, bool,
-                          out);
+    ctx = this->ctx;
+    GF_ASSERT(ctx);
+    active = ctx->active;
+    top = active->first;
 
-        GF_OPTION_RECONF ("latency-measurement", conf->measure_latency,
-                          options, bool, out);
+    if (log_level == -1)
+        return;
 
-        GF_OPTION_RECONF ("sys-log-level", sys_log_str, options, str, out);
-        if (sys_log_str) {
-                sys_log_level = glusterd_check_log_level (sys_log_str);
-                set_sys_log_level (sys_log_level);
-        }
+    if (ctx->cmd_args.brick_mux) {
+        /* Set log-level for all brick xlators */
+        top->loglevel = log_level;
 
-        GF_OPTION_RECONF ("log-level", log_str, options, str, out);
-        if (log_str) {
-                log_level = glusterd_check_log_level (log_str);
-                gf_log_set_loglevel (log_level);
+        /* Set log-level for parent xlator */
+        if (this->parents)
+            this->parents->xlator->loglevel = log_level;
+
+        while (trav) {
+            trav->loglevel = log_level;
+            trav = trav->next;
         }
+    } else {
+        gf_log_set_loglevel(this->ctx, log_level);
+    }
+}
 
-        ret = 0;
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+    struct ios_conf *conf = NULL;
+    int ret = -1;
+    char *sys_log_str = NULL;
+    char *log_format_str = NULL;
+    char *logger_str = NULL;
+    char *dump_format_str = NULL;
+    int sys_log_level = -1;
+    char *log_str = NULL;
+    int log_level = -1;
+    int log_format = -1;
+    int logger = -1;
+    uint32_t log_buf_size = 0;
+    uint32_t log_flush_timeout = 0;
+    int32_t old_dump_interval;
+    int32_t threads;
+
+    if (!this || !this->private)
+        goto out;
+
+    conf = this->private;
+
+    GF_OPTION_RECONF("dump-fd-stats", conf->dump_fd_stats, options, bool, out);
+
+    GF_OPTION_RECONF("count-fop-hits", conf->count_fop_hits, options, bool,
+                     out);
+
+    GF_OPTION_RECONF("latency-measurement", conf->measure_latency, options,
+                     bool, out);
+
+    old_dump_interval = conf->ios_dump_interval;
+    GF_OPTION_RECONF("ios-dump-interval", conf->ios_dump_interval, options,
+                     int32, out);
+    if ((old_dump_interval <= 0) && (conf->ios_dump_interval > 0)) {
+        conf->dump_thread_running = _gf_true;
+        conf->dump_thread_should_die = _gf_false;
+        ret = gf_thread_create(&conf->dump_thread, NULL,
+                               (void *)&_ios_dump_thread, this, "iosdump");
+        if (ret) {
+            conf->dump_thread_running = _gf_false;
+            gf_log(this ? this->name : "io-stats", GF_LOG_ERROR,
+                   "Failed to start thread"
+                   "while reconfigure. Returning %d",
+                   ret);
+            goto out;
+        }
+    } else if ((old_dump_interval > 0) && (conf->ios_dump_interval == 0)) {
+        _ios_destroy_dump_thread(conf);
+    }
+
+    GF_OPTION_RECONF("ios-sample-interval", conf->ios_sample_interval, options,
+                     int32, out);
+    GF_OPTION_RECONF("ios-dump-format", dump_format_str, options, str, out);
+    ios_set_log_format_code(conf, dump_format_str);
+    GF_OPTION_RECONF("ios-sample-buf-size", conf->ios_sample_buf_size, options,
+                     int32, out);
+    GF_OPTION_RECONF("sys-log-level", sys_log_str, options, str, out);
+    if (sys_log_str) {
+        sys_log_level = glusterd_check_log_level(sys_log_str);
+        set_sys_log_level(sys_log_level);
+    }
+
+    GF_OPTION_RECONF("log-level", log_str, options, str, out);
+    if (log_str) {
+        log_level = glusterd_check_log_level(log_str);
+        /* Set loglevel for all children and server xlators */
+        xlator_set_loglevel(this, log_level);
+    }
+
+    GF_OPTION_RECONF("logger", logger_str, options, str, out);
+    if (logger_str) {
+        logger = gf_check_logger(logger_str);
+        gf_log_set_logger(logger);
+    }
+
+    GF_OPTION_RECONF("log-format", log_format_str, options, str, out);
+    if (log_format_str) {
+        log_format = gf_check_log_format(log_format_str);
+        gf_log_set_logformat(log_format);
+    }
+
+    GF_OPTION_RECONF("log-buf-size", log_buf_size, options, uint32, out);
+    gf_log_set_log_buf_size(log_buf_size);
+
+    GF_OPTION_RECONF("log-flush-timeout", log_flush_timeout, options, time,
+                     out);
+    gf_log_set_log_flush_timeout(log_flush_timeout);
+
+    GF_OPTION_RECONF("threads", threads, options, int32, out);
+    gf_async_adjust_threads(threads);
+
+    ret = 0;
 out:
-        gf_log (this->name, GF_LOG_DEBUG, "reconfigure returning %d", ret);
-        return ret;
+    gf_log(this ? this->name : "io-stats", GF_LOG_DEBUG,
+           "reconfigure returning %d", ret);
+    return ret;
 }
 
-
 int32_t
-mem_acct_init (xlator_t *this)
+mem_acct_init(xlator_t *this)
 {
-        int     ret = -1;
-
-        if (!this)
-                return ret;
+    int ret = -1;
 
-        ret = xlator_mem_acct_init (this, gf_io_stats_mt_end + 1);
+    if (!this)
+        return ret;
 
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
-                        " failed");
-                return ret;
-        }
+    ret = xlator_mem_acct_init(this, gf_io_stats_mt_end + 1);
 
+    if (ret != 0) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Memory accounting init"
+               " failed");
         return ret;
+    }
+
+    return ret;
 }
 
-int
-init (xlator_t *this)
+void
+ios_conf_destroy(struct ios_conf *conf)
 {
-        struct ios_conf    *conf = NULL;
-        char               *sys_log_str = NULL;
-        int                 sys_log_level = -1;
-        char               *log_str = NULL;
-        int                 log_level = -1;
-        int                 ret = -1;
-
-        if (!this)
-                return -1;
-
-        if (!this->children) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "io_stats translator requires atleast one subvolume");
-                return -1;
-        }
-
-        if (!this->parents) {
-                /* This is very much valid as io-stats currently is loaded
-                 * on top of volumes on both client and server, hence this is
-                 * not an warning message */
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "dangling volume. check volfile ");
-        }
-
-        conf = GF_CALLOC (1, sizeof(*conf), gf_io_stats_mt_ios_conf);
-
-        if (!conf) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory.");
-                return -1;
-        }
+    if (!conf)
+        return;
 
-        LOCK_INIT (&conf->lock);
+    ios_destroy_top_stats(conf);
+    _ios_destroy_dump_thread(conf);
+    ios_destroy_sample_buf(conf->ios_sample_buf);
+    LOCK_DESTROY(&conf->lock);
+    gf_dnscache_deinit(conf->dnscache);
+    GF_FREE(conf);
+}
 
-        gettimeofday (&conf->cumulative.started_at, NULL);
-        gettimeofday (&conf->incremental.started_at, NULL);
+static void
+ios_init_stats(struct ios_global_stats *stats)
+{
+    int i = 0;
 
-        ret = ios_init_top_stats (conf);
-        if (ret)
-                return -1;
+    GF_ATOMIC_INIT(stats->data_read, 0);
+    GF_ATOMIC_INIT(stats->data_written, 0);
 
-        GF_OPTION_INIT ("dump-fd-stats", conf->dump_fd_stats, bool, out);
+    for (i = 0; i < IOS_BLOCK_COUNT_SIZE; i++) {
+        GF_ATOMIC_INIT(stats->block_count_write[i], 0);
+        GF_ATOMIC_INIT(stats->block_count_read[i], 0);
+    }
 
-        GF_OPTION_INIT ("count-fop-hits", conf->count_fop_hits, bool, out);
+    for (i = 0; i < GF_FOP_MAXVALUE; i++)
+        GF_ATOMIC_INIT(stats->fop_hits[i], 0);
 
-        GF_OPTION_INIT ("latency-measurement", conf->measure_latency,
-                          bool, out);
+    for (i = 0; i < GF_UPCALL_FLAGS_MAXVALUE; i++)
+        GF_ATOMIC_INIT(stats->upcall_hits[i], 0);
 
-        GF_OPTION_INIT ("sys-log-level", sys_log_str, str, out);
-        if (sys_log_str) {
-                sys_log_level = glusterd_check_log_level (sys_log_str);
-                set_sys_log_level (sys_log_level);
-        }
+    stats->started_at = gf_time();
+}
 
-        GF_OPTION_INIT ("log-level", log_str, str, out);
-        if (log_str) {
-                log_level = glusterd_check_log_level (log_str);
-                gf_log_set_loglevel (log_level);
+int
+init(xlator_t *this)
+{
+    struct ios_conf *conf = NULL;
+    char *volume_id = NULL;
+    char *sys_log_str = NULL;
+    char *logger_str = NULL;
+    char *log_format_str = NULL;
+    char *dump_format_str = NULL;
+    int logger = -1;
+    int log_format = -1;
+    int sys_log_level = -1;
+    char *log_str = NULL;
+    int log_level = -1;
+    int ret = -1;
+    uint32_t log_buf_size = 0;
+    uint32_t log_flush_timeout = 0;
+    int32_t threads;
+
+    if (!this)
+        return -1;
+
+    if (!this->children) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "io_stats translator requires at least one subvolume");
+        return -1;
+    }
+
+    if (!this->parents) {
+        /* This is very much valid as io-stats currently is loaded
+         * on top of volumes on both client and server, hence this is
+         * not an warning message */
+        gf_log(this->name, GF_LOG_DEBUG, "dangling volume. check volfile ");
+    }
+
+    conf = GF_CALLOC(1, sizeof(*conf), gf_io_stats_mt_ios_conf);
+
+    if (!conf)
+        goto out;
+
+    if (dict_get_str(this->options, "unique-id", &conf->unique_id) != 0) {
+        /* This is always set on servers, so we must be a client. */
+        conf->unique_id = this->name;
+    }
+
+    ret = dict_get_strn(this->options, "volume-id", SLEN("volume-id"),
+                        &volume_id);
+    if (!ret) {
+        strncpy(this->graph->volume_id, volume_id, GF_UUID_BUF_SIZE);
+    }
+    /*
+     * Init it just after calloc, so that we are sure the lock is inited
+     * in case of error paths.
+     */
+    LOCK_INIT(&conf->lock);
+    LOCK_INIT(&conf->ios_sampling_lock);
+
+    ios_init_stats(&conf->cumulative);
+    ios_init_stats(&conf->incremental);
+
+    ret = ios_init_top_stats(conf);
+    if (ret)
+        goto out;
+
+    GF_OPTION_INIT("dump-fd-stats", conf->dump_fd_stats, bool, out);
+
+    GF_OPTION_INIT("count-fop-hits", conf->count_fop_hits, bool, out);
+
+    GF_OPTION_INIT("latency-measurement", conf->measure_latency, bool, out);
+
+    GF_OPTION_INIT("ios-dump-interval", conf->ios_dump_interval, int32, out);
+
+    GF_OPTION_INIT("ios-sample-interval", conf->ios_sample_interval, int32,
+                   out);
+
+    GF_OPTION_INIT("ios-dump-format", dump_format_str, str, out);
+    ios_set_log_format_code(conf, dump_format_str);
+
+    GF_OPTION_INIT("ios-sample-buf-size", conf->ios_sample_buf_size, int32,
+                   out);
+
+    ret = ios_init_sample_buf(conf);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR, "Out of memory.");
+        goto out;
+    }
+
+    GF_OPTION_INIT("ios-dnscache-ttl-sec", conf->ios_dnscache_ttl_sec, int32,
+                   out);
+    conf->dnscache = gf_dnscache_init(conf->ios_dnscache_ttl_sec);
+    if (!conf->dnscache) {
+        ret = -1;
+        goto out;
+    }
+
+    GF_OPTION_INIT("sys-log-level", sys_log_str, str, out);
+    if (sys_log_str) {
+        sys_log_level = glusterd_check_log_level(sys_log_str);
+        set_sys_log_level(sys_log_level);
+    }
+
+    GF_OPTION_INIT("log-level", log_str, str, out);
+    if (log_str) {
+        log_level = glusterd_check_log_level(log_str);
+        if (DEFAULT_LOG_LEVEL != log_level)
+            gf_log_set_loglevel(this->ctx, log_level);
+    }
+
+    GF_OPTION_INIT("logger", logger_str, str, out);
+    if (logger_str) {
+        logger = gf_check_logger(logger_str);
+        gf_log_set_logger(logger);
+    }
+
+    GF_OPTION_INIT("log-format", log_format_str, str, out);
+    if (log_format_str) {
+        log_format = gf_check_log_format(log_format_str);
+        gf_log_set_logformat(log_format);
+    }
+
+    GF_OPTION_INIT("log-buf-size", log_buf_size, uint32, out);
+    gf_log_set_log_buf_size(log_buf_size);
+
+    GF_OPTION_INIT("log-flush-timeout", log_flush_timeout, time, out);
+    gf_log_set_log_flush_timeout(log_flush_timeout);
+
+    GF_OPTION_INIT("threads", threads, int32, out);
+    gf_async_adjust_threads(threads);
+
+    this->private = conf;
+    if (conf->ios_dump_interval > 0) {
+        conf->dump_thread_running = _gf_true;
+        conf->dump_thread_should_die = _gf_false;
+        ret = gf_thread_create(&conf->dump_thread, NULL,
+                               (void *)&_ios_dump_thread, this, "iosdump");
+        if (ret) {
+            conf->dump_thread_running = _gf_false;
+            gf_log(this ? this->name : "io-stats", GF_LOG_ERROR,
+                   "Failed to start thread"
+                   "in init. Returning %d",
+                   ret);
+            goto out;
         }
-
-        this->private = conf;
-        ret = 0;
+    }
+    return 0;
 out:
-        return ret;
+    ios_conf_destroy(conf);
+    return ret;
 }
 
-
 void
-fini (xlator_t *this)
+fini(xlator_t *this)
 {
-        struct ios_conf *conf = NULL;
-
-        if (!this)
-                return;
-
-        conf = this->private;
-
-        if (!conf)
-                return;
-        this->private = NULL;
+    struct ios_conf *conf = NULL;
 
-        ios_destroy_top_stats (conf);
+    if (!this)
+        return;
 
-        GF_FREE(conf);
+    conf = this->private;
 
-        gf_log (this->name, GF_LOG_INFO,
-                "io-stats translator unloaded");
-        return;
+    ios_conf_destroy(conf);
+    this->private = NULL;
+    gf_log(this->name, GF_LOG_INFO, "io-stats translator unloaded");
+    return;
 }
 
 int
-notify (xlator_t *this, int32_t event, void *data, ...)
+notify(xlator_t *this, int32_t event, void *data, ...)
 {
-        int          ret = 0;
-        struct ios_dump_args args = {0};
-        dict_t       *output = NULL;
-        dict_t       *dict = NULL;
-        int32_t       top_op = 0;
-        int32_t       list_cnt = 0;
-        double        throughput = 0;
-        double        time = 0;
-        va_list ap;
-
-        dict = data;
-        va_start (ap, data);
-        output = va_arg (ap, dict_t*);
-        va_end (ap);
-        switch (event) {
+    int ret = 0;
+    struct ios_dump_args args = {0};
+    dict_t *output = NULL;
+    dict_t *dict = NULL;
+    int32_t op = 0;
+    int32_t list_cnt = 0;
+    double throughput = 0;
+    double time = 0;
+    gf_boolean_t is_peek = _gf_false;
+    va_list ap;
+    struct gf_upcall *up_data = NULL;
+    struct gf_upcall_cache_invalidation *up_ci = NULL;
+
+    dict = data;
+    va_start(ap, data);
+    output = va_arg(ap, dict_t *);
+    va_end(ap);
+    switch (event) {
         case GF_EVENT_TRANSLATOR_INFO:
-                ret = dict_get_str_boolean (dict, "clear-stats", _gf_false);
+            ret = dict_get_str_boolean(dict, "clear-stats", _gf_false);
+            if (ret) {
+                ret = dict_set_int32(output, "top-op", op);
                 if (ret) {
-                         ret = dict_set_int32 (output, "top-op", top_op);
-                        if (ret) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "Failed to set top-op in dict");
-                                goto out;
-                        }
-                        ios_destroy_top_stats (this->private);
-                        ret = ios_init_top_stats (this->private);
+                    gf_log(this->name, GF_LOG_ERROR,
+                           "Failed to set top-op in dict");
+                    goto out;
+                }
+                ios_destroy_top_stats(this->private);
+                ret = ios_init_top_stats(this->private);
+                if (ret)
+                    gf_log(this->name, GF_LOG_ERROR,
+                           "Failed to reset top stats");
+                ret = dict_set_int32(output, "stats-cleared", ret ? 0 : 1);
+                if (ret)
+                    gf_log(this->name, GF_LOG_ERROR,
+                           "Failed to set stats-cleared"
+                           " in dict");
+                goto out;
+            }
+
+            ret = dict_get_int32(dict, "top-op", &op);
+            if (!ret) {
+                ret = dict_get_int32(dict, "list-cnt", &list_cnt);
+                if (op > IOS_STATS_TYPE_NONE && op < IOS_STATS_TYPE_MAX)
+                    ret = io_stats_dump_stats_to_dict(this, output, op,
+                                                      list_cnt);
+                if (op == IOS_STATS_TYPE_READ_THROUGHPUT ||
+                    op == IOS_STATS_TYPE_WRITE_THROUGHPUT) {
+                    ret = dict_get_double(dict, "throughput", &throughput);
+                    if (!ret) {
+                        ret = dict_get_double(dict, "time", &time);
                         if (ret)
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "Failed to reset top stats");
-                        ret = dict_set_int32 (output, "stats-cleared",
-                                              ret ? 0 : 1);
+                            goto out;
+                        ret = dict_set_double(output, "throughput", throughput);
                         if (ret)
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "Failed to set stats-cleared"
-                                        " in dict");
-                        goto out;
+                            goto out;
+                        ret = dict_set_double(output, "time", time);
+                        if (ret)
+                            goto out;
+                    }
+                    ret = 0;
                 }
+            } else {
+                ret = dict_get_int32(dict, "info-op", &op);
+                if (ret || op < GF_IOS_INFO_ALL || GF_IOS_INFO_CLEAR < op)
+                    op = GF_IOS_INFO_ALL;
+
+                ret = dict_set_int32(output, "info-op", op);
+                if (ret) {
+                    gf_log(this->name, GF_LOG_ERROR,
+                           "Failed to set info-op in dict");
+                    goto out;
+                }
+
+                if (GF_IOS_INFO_CLEAR == op) {
+                    io_stats_clear(this->private);
 
-                ret = dict_get_int32 (dict, "top-op", &top_op);
-                if (!ret) {
-                        ret = dict_get_int32 (dict, "list-cnt", &list_cnt);
-                        if (top_op > IOS_STATS_TYPE_NONE &&
-                            top_op < IOS_STATS_TYPE_MAX)
-                                ret = io_stats_dump_stats_to_dict (this, output,
-                                                             top_op, list_cnt);
-                        if (top_op == IOS_STATS_TYPE_READ_THROUGHPUT ||
-                                top_op == IOS_STATS_TYPE_WRITE_THROUGHPUT) {
-                                ret = dict_get_double (dict, "throughput",
-                                                        &throughput);
-                                if (!ret) {
-                                        ret = dict_get_double (dict, "time",
-                                                                &time);
-                                        if (ret)
-                                                goto out;
-                                        ret = dict_set_double (output,
-                                                "throughput", throughput);
-                                        if (ret)
-                                                goto out;
-                                        ret = dict_set_double (output, "time",
-                                                        time);
-                                        if (ret)
-                                                goto out;
-                                }
-                                ret = 0;
-
-                        }
+                    ret = dict_set_int32(output, "stats-cleared", 1);
+                    if (ret)
+                        gf_log(this->name, GF_LOG_ERROR,
+                               "Failed to set stats-cleared"
+                               " in dict");
                 } else {
-                        (void) ios_dump_args_init (&args, IOS_DUMP_TYPE_DICT,
-                                           output);
-                        ret = io_stats_dump (this, &args);
+                    ret = dict_get_str_boolean(dict, "peek", _gf_false);
+                    if (-1 != ret)
+                        is_peek = ret;
+
+                    (void)ios_dump_args_init(&args, IOS_DUMP_TYPE_DICT, output);
+                    ret = io_stats_dump(this, &args, op, is_peek);
                 }
-                break;
+            }
+            break;
+        case GF_EVENT_UPCALL:
+            up_data = (struct gf_upcall *)data;
+            ios_bump_upcall(this, GF_UPCALL);
+
+            switch (up_data->event_type) {
+                case GF_UPCALL_RECALL_LEASE:
+                    ios_bump_upcall(this, GF_UPCALL_LEASE_RECALL);
+                    break;
+                case GF_UPCALL_CACHE_INVALIDATION:
+                    up_ci = (struct gf_upcall_cache_invalidation *)
+                                up_data->data;
+                    if (up_ci->flags & (UP_XATTR | UP_XATTR_RM))
+                        ios_bump_upcall(this, GF_UPCALL_CI_XATTR);
+                    if (up_ci->flags & IATT_UPDATE_FLAGS)
+                        ios_bump_upcall(this, GF_UPCALL_CI_STAT);
+                    if (up_ci->flags & UP_RENAME_FLAGS)
+                        ios_bump_upcall(this, GF_UPCALL_CI_RENAME);
+                    if (up_ci->flags & UP_FORGET)
+                        ios_bump_upcall(this, GF_UPCALL_CI_FORGET);
+                    if (up_ci->flags & UP_NLINK)
+                        ios_bump_upcall(this, GF_UPCALL_CI_NLINK);
+                    break;
+                default:
+                    gf_msg_debug(this->name, 0,
+                                 "Unknown upcall event "
+                                 "type :%d",
+                                 up_data->event_type);
+                    break;
+            }
+
+            default_notify(this, event, data);
+            break;
         default:
-                default_notify (this, event, data);
-                break;
-
-        }
+            default_notify(this, event, data);
+            break;
+    }
 out:
-        return ret;
+    return ret;
 }
 
+struct xlator_dumpops dumpops = {.priv = io_priv};
+
 struct xlator_fops fops = {
-        .stat        = io_stats_stat,
-        .readlink    = io_stats_readlink,
-        .mknod       = io_stats_mknod,
-        .mkdir       = io_stats_mkdir,
-        .unlink      = io_stats_unlink,
-        .rmdir       = io_stats_rmdir,
-        .symlink     = io_stats_symlink,
-        .rename      = io_stats_rename,
-        .link        = io_stats_link,
-        .truncate    = io_stats_truncate,
-        .open        = io_stats_open,
-        .readv       = io_stats_readv,
-        .writev      = io_stats_writev,
-        .statfs      = io_stats_statfs,
-        .flush       = io_stats_flush,
-        .fsync       = io_stats_fsync,
-        .setxattr    = io_stats_setxattr,
-        .getxattr    = io_stats_getxattr,
-        .removexattr = io_stats_removexattr,
-        .fsetxattr    = io_stats_fsetxattr,
-        .fgetxattr    = io_stats_fgetxattr,
-        .fremovexattr = io_stats_fremovexattr,
-        .opendir     = io_stats_opendir,
-        .readdir     = io_stats_readdir,
-        .readdirp    = io_stats_readdirp,
-        .fsyncdir    = io_stats_fsyncdir,
-        .access      = io_stats_access,
-        .ftruncate   = io_stats_ftruncate,
-        .fstat       = io_stats_fstat,
-        .create      = io_stats_create,
-        .lk          = io_stats_lk,
-        .inodelk     = io_stats_inodelk,
-        .finodelk    = io_stats_finodelk,
-        .entrylk     = io_stats_entrylk,
-        .lookup      = io_stats_lookup,
-        .xattrop     = io_stats_xattrop,
-        .fxattrop    = io_stats_fxattrop,
-        .setattr     = io_stats_setattr,
-        .fsetattr    = io_stats_fsetattr,
-	.fallocate   = io_stats_fallocate,
-	.discard     = io_stats_discard,
+    .stat = io_stats_stat,
+    .readlink = io_stats_readlink,
+    .mknod = io_stats_mknod,
+    .mkdir = io_stats_mkdir,
+    .unlink = io_stats_unlink,
+    .rmdir = io_stats_rmdir,
+    .symlink = io_stats_symlink,
+    .rename = io_stats_rename,
+    .link = io_stats_link,
+    .truncate = io_stats_truncate,
+    .open = io_stats_open,
+    .readv = io_stats_readv,
+    .writev = io_stats_writev,
+    .statfs = io_stats_statfs,
+    .flush = io_stats_flush,
+    .fsync = io_stats_fsync,
+    .setxattr = io_stats_setxattr,
+    .getxattr = io_stats_getxattr,
+    .removexattr = io_stats_removexattr,
+    .fsetxattr = io_stats_fsetxattr,
+    .fgetxattr = io_stats_fgetxattr,
+    .fremovexattr = io_stats_fremovexattr,
+    .opendir = io_stats_opendir,
+    .readdir = io_stats_readdir,
+    .readdirp = io_stats_readdirp,
+    .fsyncdir = io_stats_fsyncdir,
+    .access = io_stats_access,
+    .ftruncate = io_stats_ftruncate,
+    .fstat = io_stats_fstat,
+    .create = io_stats_create,
+    .lk = io_stats_lk,
+    .inodelk = io_stats_inodelk,
+    .finodelk = io_stats_finodelk,
+    .entrylk = io_stats_entrylk,
+    .fentrylk = io_stats_fentrylk,
+    .lookup = io_stats_lookup,
+    .xattrop = io_stats_xattrop,
+    .fxattrop = io_stats_fxattrop,
+    .setattr = io_stats_setattr,
+    .fsetattr = io_stats_fsetattr,
+    .fallocate = io_stats_fallocate,
+    .discard = io_stats_discard,
+    .zerofill = io_stats_zerofill,
+    .ipc = io_stats_ipc,
+    .rchecksum = io_stats_rchecksum,
+    .seek = io_stats_seek,
+    .lease = io_stats_lease,
+    .getactivelk = io_stats_getactivelk,
+    .setactivelk = io_stats_setactivelk,
+    .compound = io_stats_compound,
+    .copy_file_range = io_stats_copy_file_range,
 };
 
 struct xlator_cbks cbks = {
-        .release     = io_stats_release,
-        .releasedir  = io_stats_releasedir,
-        .forget      = io_stats_forget,
+    .release = io_stats_release,
+    .releasedir = io_stats_releasedir,
+    .forget = io_stats_forget,
 };
 
 struct volume_options options[] = {
-        { .key  = {"dump-fd-stats"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "off",
-          .description = "If on stats related to file-operations would be "
-                         "tracked inside GlusterFS data-structures."
-        },
-        { .key  = { "latency-measurement" },
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "off",
-          .description = "If on stats related to the latency of each operation "
-                         "would be tracked inside GlusterFS data-structures. "
-        },
-        { .key  = {"count-fop-hits"},
-          .type = GF_OPTION_TYPE_BOOL,
-        },
-        { .key = {"log-level"},
-          .type = GF_OPTION_TYPE_STR,
-          .value = { "DEBUG", "WARNING", "ERROR", "INFO",
-                     "CRITICAL", "NONE", "TRACE"}
-        },
-
-        /* These are synthetic entries to assist validation of CLI's  *
-         *  volume set  command                                       */
-        { .key = {"client-log-level"},
-          .type = GF_OPTION_TYPE_STR,
-          .default_value = "INFO",
-          .description = "Changes the log-level of the clients",
-          .value = { "DEBUG", "WARNING", "ERROR", "INFO",
-                     "CRITICAL", "NONE", "TRACE"}
-        },
-        { .key = {"sys-log-level"},
-          .type = GF_OPTION_TYPE_STR,
-          .default_value = "CRITICAL",
-          .description = "Gluster's syslog log-level",
-          .value = { "WARNING", "ERROR", "INFO", "CRITICAL"}
-        },
-        { .key = {"brick-log-level"},
-          .type = GF_OPTION_TYPE_STR,
-          .default_value = "INFO",
-          .description = "Changes the log-level of the bricks",
-          .value = { "DEBUG", "WARNING", "ERROR", "INFO",
-                     "CRITICAL", "NONE", "TRACE"}
-        },
-        { .key  = {NULL} },
+    {.key = {"dump-fd-stats"},
+     .op_version = {1},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description = "If on stats related to file-operations would be "
+                    "tracked inside GlusterFS data-structures."},
+    {.key = {"ios-dump-interval"},
+     .type = GF_OPTION_TYPE_INT,
+     .op_version = {1},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .min = 0,
+     .max = 3600,
+     .default_value = "0",
+     .description = "Interval (in seconds) at which to auto-dump "
+                    "statistics. Zero disables automatic dumping."},
+    {.key = {"ios-sample-interval"},
+     .type = GF_OPTION_TYPE_INT,
+     .op_version = {1},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .min = 0,
+     .max = 65535,
+     .default_value = "0",
+     .description = "Interval in which we want to collect FOP latency "
+                    "samples.  2 means collect a sample every 2nd FOP."},
+    {.key = {"ios-dump-format"},
+     .type = GF_OPTION_TYPE_STR,
+     .op_version = {GD_OP_VERSION_3_12_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .default_value = "json",
+     .description = " The dump-format option specifies the format in which"
+                    " to dump the statistics. Select between \"text\", "
+                    "\"json\", \"dict\" and \"samples\". Default is "
+                    "\"json\".",
+     .value = {"text", "json", "dict", "samples"}},
+    {.key = {"ios-sample-buf-size"},
+     .type = GF_OPTION_TYPE_INT,
+     .op_version = {1},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .min = 1024,
+     .max = 1024 * 1024,
+     .default_value = "65535",
+     .description = "The maximum size of our FOP sampling ring buffer."},
+    {.key = {"ios-dnscache-ttl-sec"},
+     .type = GF_OPTION_TYPE_INT,
+     .op_version = {1},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .min = 1,
+     .max = 3600 * 72,
+     .default_value = "86400",
+     .description = "The interval after wish a cached DNS entry will be "
+                    "re-validated.  Default: 24 hrs"},
+    {.key = {"latency-measurement"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .op_version = {1},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .default_value = "off",
+     .description = "If on stats related to the latency of each operation "
+                    "would be tracked inside GlusterFS data-structures. "},
+    {
+        .key = {"count-fop-hits"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .op_version = {1},
+        .flags = OPT_FLAG_SETTABLE,
+        .tags = {"io-stats"},
+    },
+    {.key = {"log-level"},
+     .type = GF_OPTION_TYPE_STR,
+     .value = {"DEBUG", "WARNING", "ERROR", "INFO", "CRITICAL", "NONE",
+               "TRACE"}},
+
+    /* These are synthetic entries to assist validation of CLI's  *
+     *  volume set  command                                       */
+    {.key = {"client-log-level"},
+     .type = GF_OPTION_TYPE_STR,
+     .op_version = {1},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .default_value = "INFO",
+     .description = "Changes the log-level of the clients",
+     .value = {"DEBUG", "WARNING", "ERROR", "INFO", "CRITICAL", "NONE",
+               "TRACE"}},
+    {.key = {"sys-log-level"},
+     .type = GF_OPTION_TYPE_STR,
+     .default_value = "CRITICAL",
+     .description = "Gluster's syslog log-level",
+     .value = {"WARNING", "ERROR", "INFO", "CRITICAL"}},
+    {.key = {"brick-log-level"},
+     .type = GF_OPTION_TYPE_STR,
+     .op_version = {1},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .default_value = "INFO",
+     .description = "Changes the log-level of the bricks",
+     .value = {"DEBUG", "WARNING", "ERROR", "INFO", "CRITICAL", "NONE",
+               "TRACE"}},
+    {.key = {"logger"},
+     .type = GF_OPTION_TYPE_STR,
+     .value = {GF_LOGGER_GLUSTER_LOG, GF_LOGGER_SYSLOG}},
+    {.key = {"client-logger"},
+     .type = GF_OPTION_TYPE_STR,
+     .op_version = {GD_OP_VERSION_3_6_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .default_value = GF_LOGGER_GLUSTER_LOG,
+     .description = "Changes the logging sub-system to log to, for the "
+                    "clients",
+     .value = {GF_LOGGER_GLUSTER_LOG, GF_LOGGER_SYSLOG}},
+    {.key = {"brick-logger"},
+     .type = GF_OPTION_TYPE_STR,
+     .op_version = {GD_OP_VERSION_3_6_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .default_value = GF_LOGGER_GLUSTER_LOG,
+     .description = "Changes the logging sub-system to log to, for the "
+                    "bricks",
+     .value = {GF_LOGGER_GLUSTER_LOG, GF_LOGGER_SYSLOG}},
+    {.key = {"log-format"},
+     .type = GF_OPTION_TYPE_STR,
+     .value = {GF_LOG_FORMAT_NO_MSG_ID, GF_LOG_FORMAT_WITH_MSG_ID}},
+    {.key = {"client-log-format"},
+     .type = GF_OPTION_TYPE_STR,
+     .op_version = {GD_OP_VERSION_3_6_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .default_value = GF_LOG_FORMAT_WITH_MSG_ID,
+     .description = "Changes log format for the clients",
+     .value = {GF_LOG_FORMAT_NO_MSG_ID, GF_LOG_FORMAT_WITH_MSG_ID}},
+    {.key = {"brick-log-format"},
+     .type = GF_OPTION_TYPE_STR,
+     .op_version = {GD_OP_VERSION_3_6_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .default_value = GF_LOG_FORMAT_WITH_MSG_ID,
+     .description = "Changes the log format for the bricks",
+     .value = {GF_LOG_FORMAT_NO_MSG_ID, GF_LOG_FORMAT_WITH_MSG_ID}},
+    {
+        .key = {"log-buf-size"},
+        .type = GF_OPTION_TYPE_INT,
+        .min = GF_LOG_LRU_BUFSIZE_MIN,
+        .max = GF_LOG_LRU_BUFSIZE_MAX,
+        .default_value = "5",
+    },
+    {.key = {"client-log-buf-size"},
+     .type = GF_OPTION_TYPE_INT,
+     .op_version = {GD_OP_VERSION_3_6_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .min = GF_LOG_LRU_BUFSIZE_MIN,
+     .max = GF_LOG_LRU_BUFSIZE_MAX,
+     .default_value = "5",
+     .description = "This option determines the maximum number of unique "
+                    "log messages that can be buffered for a time equal to"
+                    " the value of the option client-log-flush-timeout."},
+    {.key = {"brick-log-buf-size"},
+     .type = GF_OPTION_TYPE_INT,
+     .op_version = {GD_OP_VERSION_3_6_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .min = GF_LOG_LRU_BUFSIZE_MIN,
+     .max = GF_LOG_LRU_BUFSIZE_MAX,
+     .default_value = "5",
+     .description = "This option determines the maximum number of unique "
+                    "log messages that can be buffered for a time equal to"
+                    " the value of the option brick-log-flush-timeout."},
+    {
+        .key = {"log-flush-timeout"},
+        .type = GF_OPTION_TYPE_TIME,
+        .min = GF_LOG_FLUSH_TIMEOUT_MIN,
+        .max = GF_LOG_FLUSH_TIMEOUT_MAX,
+        .default_value = "120",
+    },
+    {.key = {"client-log-flush-timeout"},
+     .type = GF_OPTION_TYPE_TIME,
+     .op_version = {GD_OP_VERSION_3_6_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .min = GF_LOG_FLUSH_TIMEOUT_MIN,
+     .max = GF_LOG_FLUSH_TIMEOUT_MAX,
+     .default_value = "120",
+     .description = "This option determines the maximum number of unique "
+                    "log messages that can be buffered for a time equal to"
+                    " the value of the option client-log-flush-timeout."},
+    {.key = {"brick-log-flush-timeout"},
+     .type = GF_OPTION_TYPE_TIME,
+     .op_version = {GD_OP_VERSION_3_6_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .min = GF_LOG_FLUSH_TIMEOUT_MIN,
+     .max = GF_LOG_FLUSH_TIMEOUT_MAX,
+     .default_value = "120",
+     .description = "This option determines the maximum number of unique "
+                    "log messages that can be buffered for a time equal to"
+                    " the value of the option brick-log-flush-timeout."},
+    {.key = {"unique-id"},
+     .type = GF_OPTION_TYPE_STR,
+     .default_value = "/no/such/path",
+     .description = "Unique ID for our files."},
+    {.key = {"global-threading"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .op_version = {GD_OP_VERSION_6_0},
+     .flags = OPT_FLAG_SETTABLE,
+     .tags = {"io-stats", "threading"},
+     .description = "This option enables the global threading support for "
+                    "bricks. If enabled, it's recommended to also enable "
+                    "'performance.iot-pass-through'"},
+    {.key = {"threads"}, .type = GF_OPTION_TYPE_INT},
+    {.key = {"brick-threads"},
+     .type = GF_OPTION_TYPE_INT,
+     .default_value = "16",
+     .min = 0,
+     .max = GF_ASYNC_MAX_THREADS,
+     .op_version = {GD_OP_VERSION_6_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats", "threading"},
+     .description = "When global threading is used, this value determines the "
+                    "maximum amount of threads that can be created on bricks"},
+    {.key = {"client-threads"},
+     .type = GF_OPTION_TYPE_INT,
+     .default_value = "16",
+     .min = 0,
+     .max = GF_ASYNC_MAX_THREADS,
+     .op_version = {GD_OP_VERSION_6_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC | OPT_FLAG_CLIENT_OPT,
+     .tags = {"io-stats", "threading"},
+     .description = "When global threading is used, this value determines the "
+                    "maximum amount of threads that can be created on clients"},
+    {.key = {"volume-id"},
+     .type = GF_OPTION_TYPE_STR,
+     .op_version = {GD_OP_VERSION_7_1},
+     .tags = {"global", "volume-id"},
+     .description =
+         "This option points to the 'unique' UUID particular to this "
+         "volume, which would be set in 'graph->volume_id'"},
+    {.key = {NULL}},
+};
 
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .dumpops = &dumpops,
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "io-stats",
+    .category = GF_MAINTAINED,
 };
diff --git a/xlators/cluster/stripe/Makefile.am b/xlators/debug/sink/Makefile.am
index d471a3f9243..f2689244371 100644
--- a/xlators/cluster/stripe/Makefile.am
+++ b/xlators/debug/sink/Makefile.am
@@ -1,3 +1,2 @@
 SUBDIRS = src
 
-CLEANFILES = 
diff --git a/xlators/debug/sink/src/Makefile.am b/xlators/debug/sink/src/Makefile.am
new file mode 100644
index 00000000000..f952c2ce6bc
--- /dev/null
+++ b/xlators/debug/sink/src/Makefile.am
@@ -0,0 +1,14 @@
+xlator_LTLIBRARIES = sink.la
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/debug
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+        -I$(top_builddir)/rpc/xdr/src
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+sink_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+sink_la_SOURCES = sink.c
+sink_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+CLEANFILES =
+
diff --git a/xlators/debug/sink/src/sink.c b/xlators/debug/sink/src/sink.c
new file mode 100644
index 00000000000..9822bbb732e
--- /dev/null
+++ b/xlators/debug/sink/src/sink.c
@@ -0,0 +1,94 @@
+/*
+   Copyright (c) 2017 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+int32_t
+init(xlator_t *this)
+{
+    return 0;
+}
+
+void
+fini(xlator_t *this)
+{
+    return;
+}
+
+/*
+ * notify - when parent sends PARENT_UP, send CHILD_UP event from here
+ */
+int32_t
+notify(xlator_t *this, int32_t event, void *data, ...)
+{
+    switch (event) {
+        case GF_EVENT_PARENT_UP:
+            /* Tell the parent that this xlator is up */
+            default_notify(this, GF_EVENT_CHILD_UP, data);
+            break;
+        case GF_EVENT_PARENT_DOWN:
+            /* Tell the parent that this xlator is down */
+            default_notify(this, GF_EVENT_CHILD_DOWN, data);
+            break;
+        default:
+            break;
+    }
+
+    return 0;
+}
+
+/*
+ * A lookup on "/" is done while mounting or glfs_init() is performed. This
+ * needs to return a valid directory for the root of the mountpoint.
+ *
+ * In case this xlator is used for more advanced debugging, it will need to be
+ * extended to support different LOOKUPs too.
+ */
+static int32_t
+sink_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    struct iatt stbuf = {
+        0,
+    };
+    struct iatt postparent = {
+        0,
+    };
+
+    /* the root of the volume always need to be a directory */
+    stbuf.ia_type = IA_IFDIR;
+
+    STACK_UNWIND_STRICT(lookup, frame, 0, 0, loc ? loc->inode : NULL, &stbuf,
+                        xdata, &postparent);
+
+    return 0;
+}
+
+struct xlator_fops fops = {
+    .lookup = sink_lookup,
+};
+
+struct xlator_cbks cbks = {};
+
+struct volume_options options[] = {
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .op_version = {GD_OP_VERSION_3_12_0},
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "sink",
+    .category = GF_TECH_PREVIEW,
+};
diff --git a/xlators/debug/trace/src/Makefile.am b/xlators/debug/trace/src/Makefile.am
index 7b2597b4db6..a37ea63af04 100644
--- a/xlators/debug/trace/src/Makefile.am
+++ b/xlators/debug/trace/src/Makefile.am
@@ -2,13 +2,14 @@
 xlator_LTLIBRARIES = trace.la
 xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/debug
 
-trace_la_LDFLAGS = -module -avoid-version
+trace_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
 
 trace_la_SOURCES = trace.c
 trace_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
 noinst_HEADERS = trace.h trace-mem-types.h
-AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
 
 AM_CFLAGS = -Wall $(GF_CFLAGS)
 
diff --git a/xlators/debug/trace/src/trace-mem-types.h b/xlators/debug/trace/src/trace-mem-types.h
index 9fa7d97c2ca..18a7e0414a6 100644
--- a/xlators/debug/trace/src/trace-mem-types.h
+++ b/xlators/debug/trace/src/trace-mem-types.h
@@ -8,14 +8,13 @@
    cases as published by the Free Software Foundation.
 */
 
-
 #ifndef __TRACE_MEM_TYPES_H__
 #define __TRACE_MEM_TYPES_H__
 
-#include "mem-types.h"
+#include <glusterfs/mem-types.h>
 
 enum gf_trace_mem_types_ {
-        gf_trace_mt_trace_conf_t = gf_common_mt_end + 1,
-        gf_trace_mt_end
+    gf_trace_mt_trace_conf_t = gf_common_mt_end + 1,
+    gf_trace_mt_end
 };
 #endif
diff --git a/xlators/debug/trace/src/trace.c b/xlators/debug/trace/src/trace.c
index a9c11babec2..6ed0ca00342 100644
--- a/xlators/debug/trace/src/trace.c
+++ b/xlators/debug/trace/src/trace.c
@@ -17,3258 +17,3518 @@
  *    their _cbk functions, which later passes the call to next layer.
  *    Very helpful translator for debugging.
  */
+#define TRACE_STAT_TO_STR(buf, str) trace_stat_to_str(buf, str, sizeof(str))
+
+static void
+trace_stat_to_str(struct iatt *buf, char *str, size_t len)
+{
+    char atime_buf[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    char mtime_buf[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    char ctime_buf[GF_TIMESTR_SIZE] = {
+        0,
+    };
+
+    if (!buf)
+        return;
 
-int
-dump_history_trace (circular_buffer_t *cb, void *data)
-{
-        char    *string      = NULL;
-        struct  tm      *tm  = NULL;
-        char    timestr[256] = {0,};
-
-        string = (char *)cb->data;
-        tm    = localtime (&cb->tv.tv_sec);
+    gf_time_fmt(atime_buf, sizeof atime_buf, buf->ia_atime, gf_timefmt_dirent);
 
-        /* Since we are continuing with adding entries to the buffer even when
-           gettimeofday () fails, it's safe to check tm and then dump the time
-           at which the entry was added to the buffer */
-        if (tm) {
-                strftime (timestr, 256, "%Y-%m-%d %H:%M:%S", tm);
-                snprintf (timestr + strlen (timestr), 256 - strlen (timestr),
-                          ".%"GF_PRI_SUSECONDS, cb->tv.tv_usec);
-                gf_proc_dump_write ("TIME", "%s", timestr);
-        }
+    gf_time_fmt(mtime_buf, sizeof mtime_buf, buf->ia_mtime, gf_timefmt_dirent);
 
-        gf_proc_dump_write ("FOP", "%s\n", string);
+    gf_time_fmt(ctime_buf, sizeof ctime_buf, buf->ia_ctime, gf_timefmt_dirent);
 
-        return 0;
+    snprintf(str, len,
+             "gfid=%s ino=%" PRIu64
+             ", mode=%o, "
+             "nlink=%" GF_PRI_NLINK ", uid=%u, gid=%u, size=%" PRIu64
+             ", "
+             "blocks=%" PRIu64
+             ", atime=%s mtime=%s ctime=%s "
+             "atime_sec=%" PRId64 ", atime_nsec=%" PRIu32
+             ","
+             " mtime_sec=%" PRId64 ", mtime_nsec=%" PRIu32
+             ", "
+             "ctime_sec=%" PRId64 ", ctime_nsec=%" PRIu32 "",
+             uuid_utoa(buf->ia_gfid), buf->ia_ino,
+             st_mode_from_ia(buf->ia_prot, buf->ia_type), buf->ia_nlink,
+             buf->ia_uid, buf->ia_gid, buf->ia_size, buf->ia_blocks, atime_buf,
+             mtime_buf, ctime_buf, buf->ia_atime, buf->ia_atime_nsec,
+             buf->ia_mtime, buf->ia_mtime_nsec, buf->ia_ctime,
+             buf->ia_ctime_nsec);
 }
 
 int
-trace_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno, fd_t *fd,
-                  inode_t *inode, struct iatt *buf,
-                  struct iatt *preparent, struct iatt *postparent,
-                  dict_t *xdata)
+dump_history_trace(circular_buffer_t *cb, void *data)
 {
-        char          statstr[4096]       = {0, };
-        char          preparentstr[4096]  = {0, };
-        char          postparentstr[4096] = {0, };
-        trace_conf_t  *conf               = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_CREATE].enabled) {
-                char  string[4096] = {0,};
-                if (op_ret >= 0) {
-                        trace_stat_to_str (buf, statstr);
-                        trace_stat_to_str (preparent, preparentstr);
-                        trace_stat_to_str (postparent, postparentstr);
-
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s (op_ret=%d, fd=%p"
-                                  "*stbuf {%s}, *preparent {%s}, "
-                                  "*postparent = {%s})",
-                                  frame->root->unique,
-                                  uuid_utoa (inode->gfid), op_ret, fd,
-                                  statstr, preparentstr, postparentstr);
-
-                        /* for 'release' log */
-                        fd_ctx_set (fd, this, 0);
-                } else {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                  frame->root->unique, op_ret,
-                                  op_errno);
-                }
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (create, frame, op_ret, op_errno, fd, inode, buf,
-                            preparent, postparent, xdata);
-        return 0;
-}
+    char timestr[GF_TIMESTR_SIZE] = {
+        0,
+    };
 
-int
-trace_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
-{
-        trace_conf_t      *conf = NULL;
+    /* Since we are continuing with adding entries to the buffer even when
+       gettimeofday () fails, it's safe to check tm and then dump the time
+       at which the entry was added to the buffer */
 
-        conf = this->private;
+    gf_time_fmt_tv(timestr, sizeof timestr, &cb->tv, gf_timefmt_Ymd_T);
+    gf_proc_dump_write("TIME", "%s", timestr);
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_OPEN].enabled) {
-                char     string[4096] = {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s op_ret=%d, op_errno=%d, "
-                          "*fd=%p", frame->root->unique,
-                          uuid_utoa (frame->local), op_ret, op_errno,
-                          fd);
+    gf_proc_dump_write("FOP", "%s\n", (char *)cb->data);
 
-                LOG_ELEMENT (conf, string);
-        }
-
-out:
-        /* for 'release' log */
-        if (op_ret >= 0)
-                fd_ctx_set (fd, this, 0);
-
-        TRACE_STACK_UNWIND (open, frame, op_ret, op_errno, fd, xdata);
-        return 0;
+    return 0;
 }
 
 int
-trace_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                int32_t op_ret, int32_t op_errno, struct iatt *buf,
-                dict_t *xdata)
+trace_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, fd_t *fd, inode_t *inode,
+                 struct iatt *buf, struct iatt *preparent,
+                 struct iatt *postparent, dict_t *xdata)
 {
-        char          statstr[4096] = {0, };
-        trace_conf_t  *conf         = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_STAT].enabled) {
-                char string[4096] = {0,};
-                if (op_ret == 0) {
-                        trace_stat_to_str (buf, statstr);
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s op_ret=%d buf=%s",
-                                  frame->root->unique,
-                                  uuid_utoa (frame->local), op_ret,
-                                  statstr);
-                } else {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s op_ret=%d, "
-                                  "op_errno=%d)",
-                                  frame->root->unique,
-                                  uuid_utoa (frame->local), op_ret,
-                                  op_errno);
-                }
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (stat, frame, op_ret, op_errno, buf, xdata);
-        return 0;
-}
-
-int
-trace_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno, struct iovec *vector,
-                 int32_t count, struct iatt *buf, struct iobref *iobref,
-                 dict_t *xdata)
-{
-        char          statstr[4096] = {0, };
-        trace_conf_t  *conf         = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_READ].enabled) {
-                char  string[4096] = {0,};
-                if (op_ret >= 0) {
-                        trace_stat_to_str (buf, statstr);
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s op_ret=%d buf=%s",
-                                  frame->root->unique,
-                                  uuid_utoa (frame->local), op_ret,
-                                  statstr);
-                } else {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s op_ret=%d, "
-                                  "op_errno=%d)",
-                                  frame->root->unique,
-                                  uuid_utoa (frame->local), op_ret,
-                                  op_errno);
-                }
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (readv, frame, op_ret, op_errno, vector, count,
-                            buf, iobref, xdata);
-        return 0;
-}
-
-int
-trace_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno,
-                  struct iatt *prebuf, struct iatt *postbuf, dict_t *xdata)
-{
-        char         preopstr[4096]  = {0, };
-        char         postopstr[4096] = {0, };
-        trace_conf_t *conf           = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_WRITE].enabled) {
-                char  string[4096] = {0,};
-                if (op_ret >= 0) {
-                        trace_stat_to_str (prebuf, preopstr);
-                        trace_stat_to_str (postbuf, postopstr);
-
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": (op_ret=%d, "
-                                  "*prebuf = {%s}, *postbuf = {%s})",
-                                  frame->root->unique, op_ret,
-                                  preopstr, postopstr);
-                } else {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s op_ret=%d, "
-                                  "op_errno=%d", frame->root->unique,
-                                  uuid_utoa (frame->local), op_ret,
-                                  op_errno);
-                }
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (writev, frame, op_ret, op_errno, prebuf, postbuf,
-                            xdata);
-        return 0;
-}
-
-int
-trace_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, gf_dirent_t *buf,
-                   dict_t *xdata)
-{
-        trace_conf_t  *conf = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_READDIR].enabled) {
-                char    string[4096]  = {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64" : gfid=%s op_ret=%d, op_errno=%d",
-                          frame->root->unique, uuid_utoa (frame->local),
-                          op_ret, op_errno);
-
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (readdir, frame, op_ret, op_errno, buf, xdata);
-
-        return 0;
-}
+    char statstr[1024] = {
+        0,
+    };
+    char preparentstr[1024] = {
+        0,
+    };
+    char postparentstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
 
-int
-trace_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, gf_dirent_t *buf,
-                    dict_t *xdata)
-{
-        trace_conf_t   *conf = NULL;
+    conf = this->private;
 
-        conf = this->private;
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_CREATE].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret >= 0) {
+            TRACE_STAT_TO_STR(buf, statstr);
+            TRACE_STAT_TO_STR(preparent, preparentstr);
+            TRACE_STAT_TO_STR(postparent, postparentstr);
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_READDIRP].enabled) {
-                char        string[4096] = {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64" : gfid=%s op_ret=%d, op_errno=%d",
-                          frame->root->unique, uuid_utoa (frame->local),
-                          op_ret, op_errno);
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s (op_ret=%d, fd=%p"
+                     "*stbuf {%s}, *preparent {%s}, "
+                     "*postparent = {%s})",
+                     frame->root->unique, uuid_utoa(inode->gfid), op_ret, fd,
+                     statstr, preparentstr, postparentstr);
 
-                LOG_ELEMENT (conf, string);
+            /* for 'release' log */
+            fd_ctx_set(fd, this, 0);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64 ": (op_ret=%d, op_errno=%d)",
+                     frame->root->unique, op_ret, op_errno);
         }
-
+        LOG_ELEMENT(conf, string);
+    }
 out:
-        TRACE_STACK_UNWIND (readdirp, frame, op_ret, op_errno, buf, xdata);
-        return 0;
+    TRACE_STACK_UNWIND(create, frame, op_ret, op_errno, fd, inode, buf,
+                       preparent, postparent, xdata);
+    return 0;
 }
 
 int
-trace_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno,
-                 struct iatt *prebuf, struct iatt *postbuf, dict_t *xdata)
+trace_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
 {
-        char          preopstr[4096]  = {0, };
-        char          postopstr[4096] = {0, };
-        trace_conf_t  *conf           = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_FSYNC].enabled) {
-                char  string[4096] = {0,};
-                if (op_ret == 0) {
-                        trace_stat_to_str (prebuf, preopstr);
-                        trace_stat_to_str (postbuf, postopstr);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_OPEN].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s op_ret=%d, op_errno=%d, "
+                 "*fd=%p",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret, op_errno,
+                 fd);
 
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": (op_ret=%d, "
-                                  "*prebuf = {%s}, *postbuf = {%s}",
-                                  frame->root->unique, op_ret,
-                                  preopstr, postopstr);
-                } else {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s op_ret=%d, "
-                                  "op_errno=%d", frame->root->unique,
-                                  uuid_utoa (frame->local), op_ret,
-                                  op_errno);
+        LOG_ELEMENT(conf, string);
+    }
 
-                }
-                LOG_ELEMENT (conf, string);
-        }
 out:
-        TRACE_STACK_UNWIND (fsync, frame, op_ret, op_errno, prebuf, postbuf,
-                            xdata);
+    /* for 'release' log */
+    if (op_ret >= 0)
+        fd_ctx_set(fd, this, 0);
 
-        return 0;
+    TRACE_STACK_UNWIND(open, frame, op_ret, op_errno, fd, xdata);
+    return 0;
 }
 
 int
-trace_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno,
-                   struct iatt *statpre, struct iatt *statpost, dict_t *xdata)
+trace_stat_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, struct iatt *buf,
+               dict_t *xdata)
 {
-        char          preopstr[4096]  = {0, };
-        char          postopstr[4096] = {0, };
-        trace_conf_t  *conf           = NULL;
+    char statstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_STAT].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(buf, statstr);
+            (void)snprintf(
+                string, sizeof(string), "%" PRId64 ": gfid=%s op_ret=%d buf=%s",
+                frame->root->unique, uuid_utoa(frame->local), op_ret, statstr);
+        } else {
+            (void)snprintf(string, sizeof(string),
+                           "%" PRId64
+                           ": gfid=%s op_ret=%d, "
+                           "op_errno=%d)",
+                           frame->root->unique, uuid_utoa(frame->local), op_ret,
+                           op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(stat, frame, op_ret, op_errno, buf, xdata);
+    return 0;
+}
+
+int
+trace_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iovec *vector,
+                int32_t count, struct iatt *buf, struct iobref *iobref,
+                dict_t *xdata)
+{
+    char statstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_READ].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret >= 0) {
+            TRACE_STAT_TO_STR(buf, statstr);
+            snprintf(
+                string, sizeof(string), "%" PRId64 ": gfid=%s op_ret=%d buf=%s",
+                frame->root->unique, uuid_utoa(frame->local), op_ret, statstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d)",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(readv, frame, op_ret, op_errno, vector, count, buf,
+                       iobref, xdata);
+    return 0;
+}
+
+int
+trace_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                 struct iatt *postbuf, dict_t *xdata)
+{
+    char preopstr[1024] = {
+        0,
+    };
+    char postopstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_WRITE].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret >= 0) {
+            TRACE_STAT_TO_STR(prebuf, preopstr);
+            TRACE_STAT_TO_STR(postbuf, postopstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": (op_ret=%d, "
+                     "*prebuf = {%s}, *postbuf = {%s})",
+                     frame->root->unique, op_ret, preopstr, postopstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(writev, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+    return 0;
+}
+
+int
+trace_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, gf_dirent_t *buf,
+                  dict_t *xdata)
+{
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_SETATTR].enabled) {
-                char  string[4096]  = {0,};
-                if (op_ret == 0) {
-                        trace_stat_to_str (statpre, preopstr);
-                        trace_stat_to_str (statpost, postopstr);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_READDIR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 " : gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
 
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": (op_ret=%d, "
-                                  "*prebuf = {%s}, *postbuf = {%s})",
-                                  frame->root->unique, op_ret,
-                                  preopstr, postopstr);
-                } else {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s op_ret=%d, "
-                                  "op_errno=%d)", frame->root->unique,
-                                  uuid_utoa (frame->local), op_ret,
-                                  op_errno);
-                }
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 out:
-        TRACE_STACK_UNWIND (setattr, frame, op_ret, op_errno, statpre,
-                            statpost, xdata);
-        return 0;
-}
-
-int
-trace_fsetattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno,
-                    struct iatt *statpre, struct iatt *statpost, dict_t *xdata)
-{
-        char          preopstr[4096]  = {0, };
-        char          postopstr[4096] = {0, };
-        trace_conf_t  *conf           = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_FSETATTR].enabled) {
-                char  string[4096] = {0,};
-                if (op_ret == 0) {
-                        trace_stat_to_str (statpre, preopstr);
-                        trace_stat_to_str (statpost, postopstr);
-
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": (op_ret=%d, "
-                                  "*prebuf = {%s}, *postbuf = {%s})",
-                                  frame->root->unique, op_ret,
-                                  preopstr, postopstr);
-                } else {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s op_ret=%d, op_errno=%d)",
-                                  frame->root->unique, uuid_utoa (frame->local),
-                                  op_ret, op_errno);
-                }
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (fsetattr, frame, op_ret, op_errno,
-                            statpre, statpost, xdata);
-        return 0;
-}
+    TRACE_STACK_UNWIND(readdir, frame, op_ret, op_errno, buf, xdata);
 
-int
-trace_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno,
-                  struct iatt *preparent, struct iatt *postparent,
-                  dict_t *xdata)
-{
-        char          preparentstr[4096]  = {0, };
-        char          postparentstr[4096] = {0, };
-        trace_conf_t  *conf               = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_UNLINK].enabled) {
-                char string[4096] = {0,};
-                if (op_ret == 0) {
-                        trace_stat_to_str (preparent, preparentstr);
-                        trace_stat_to_str (postparent, postparentstr);
-
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s op_ret=%d, "
-                                  " *preparent = {%s}, "
-                                  "*postparent = {%s})",
-                                  frame->root->unique,
-                                  uuid_utoa (frame->local),
-                                  op_ret, preparentstr,
-                                  postparentstr);
-                } else {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s op_ret=%d, "
-                                  "op_errno=%d)",
-                                  frame->root->unique,
-                                  uuid_utoa (frame->local), op_ret,
-                                  op_errno);
-                }
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (unlink, frame, op_ret, op_errno,
-                            preparent, postparent, xdata);
-        return 0;
+    return 0;
 }
 
 int
-trace_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno, struct iatt *buf,
-                  struct iatt *preoldparent, struct iatt *postoldparent,
-                  struct iatt *prenewparent, struct iatt *postnewparent,
-                  dict_t *xdata)
-{
-        char           statstr[4096]          = {0, };
-        char           preoldparentstr[4096]  = {0, };
-        char           postoldparentstr[4096] = {0, };
-        char           prenewparentstr[4096]  = {0, };
-        char           postnewparentstr[4096] = {0, };
-        trace_conf_t   *conf                  = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_RENAME].enabled) {
-                char  string[4096] = {0,};
-                if (op_ret == 0) {
-                        trace_stat_to_str (buf, statstr);
-                        trace_stat_to_str (preoldparent, preoldparentstr);
-                        trace_stat_to_str (postoldparent, postoldparentstr);
-                        trace_stat_to_str (prenewparent, prenewparentstr);
-                        trace_stat_to_str (postnewparent, postnewparentstr);
-
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": (op_ret=%d, "
-                                  "*stbuf = {%s}, *preoldparent = {%s},"
-                                  " *postoldparent = {%s}"
-                                  " *prenewparent = {%s}, "
-                                  "*postnewparent = {%s})",
-                                  frame->root->unique, op_ret, statstr,
-                                  preoldparentstr, postoldparentstr,
-                                  prenewparentstr, postnewparentstr);
-                } else {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s op_ret=%d, "
-                                  "op_errno=%d", frame->root->unique,
-                                  uuid_utoa (frame->local),
-                                  op_ret, op_errno);
-
-                }
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (rename, frame, op_ret, op_errno, buf,
-                            preoldparent, postoldparent,
-                            prenewparent, postnewparent, xdata);
-        return 0;
-}
-
-int
-trace_readlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno,
-                    const char *buf, struct iatt *stbuf, dict_t *xdata)
-{
-        char          statstr[4096] = {0, };
-        trace_conf_t  *conf         = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_READLINK].enabled) {
-                char string[4096] = {0,};
-                if (op_ret == 0) {
-                        trace_stat_to_str (stbuf, statstr);
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": (op_ret=%d, op_errno=%d,"
-                                  "buf=%s, stbuf = { %s })",
-                                  frame->root->unique, op_ret, op_errno,
-                                  buf, statstr);
-                } else {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s op_ret=%d, "
-                                  "op_errno=%d",
-                                  frame->root->unique,
-                                  uuid_utoa (frame->local), op_ret,
-                                  op_errno);
-                }
-
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (readlink, frame, op_ret, op_errno, buf, stbuf,
-                            xdata);
-        return 0;
-}
-
-int
-trace_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno,
-                  inode_t *inode, struct iatt *buf,
-                  dict_t *xdata, struct iatt *postparent)
-{
-        char          statstr[4096]       = {0, };
-        char          postparentstr[4096] = {0, };
-        trace_conf_t  *conf               = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_LOOKUP].enabled) {
-                char  string[4096] = {0,};
-                if (op_ret == 0) {
-                        trace_stat_to_str (buf, statstr);
-                        trace_stat_to_str (postparent, postparentstr);
-                        /* print buf->ia_gfid instead of inode->gfid,
-                         * since if the inode is not yet linked to the
-                         * inode table (fresh lookup) then null gfid
-                         * will be printed.
-                         */
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s (op_ret=%d "
-                                  "*buf {%s}, *postparent {%s}",
-                                  frame->root->unique,
-                                  uuid_utoa (buf->ia_gfid),
-                                  op_ret, statstr, postparentstr);
-
-                        /* For 'forget' */
-                        inode_ctx_put (inode, this, 0);
-                } else {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s op_ret=%d, "
-                                  "op_errno=%d)",
-                                  frame->root->unique,
-                                  uuid_utoa (frame->local), op_ret,
-                                  op_errno);
-                }
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (lookup, frame, op_ret, op_errno, inode, buf,
-                            xdata, postparent);
-        return 0;
-}
-
-int
-trace_symlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno,
-                   inode_t *inode, struct iatt *buf,
-                   struct iatt *preparent, struct iatt *postparent,
+trace_readdirp_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, gf_dirent_t *buf,
                    dict_t *xdata)
 {
-        char          statstr[4096]       = {0, };
-        char          preparentstr[4096]  = {0, };
-        char          postparentstr[4096] = {0, };
-        trace_conf_t  *conf               = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_SYMLINK].enabled) {
-                char  string[4096] = {0,};
-                if (op_ret == 0) {
-                        trace_stat_to_str (buf, statstr);
-                        trace_stat_to_str (preparent, preparentstr);
-                        trace_stat_to_str (postparent, postparentstr);
-
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s (op_ret=%d "
-                                  "*stbuf = {%s}, *preparent = {%s}, "
-                                  "*postparent = {%s})",
-                                  frame->root->unique,
-                                  uuid_utoa (inode->gfid),
-                                  op_ret, statstr, preparentstr,
-                                  postparentstr);
-                } else {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": op_ret=%d, op_errno=%d",
-                                  frame->root->unique, op_ret,
-                                  op_errno);
-                }
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (symlink, frame, op_ret, op_errno, inode, buf,
-                            preparent, postparent, xdata);
-        return 0;
-}
-
-int
-trace_mknod_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno,
-                 inode_t *inode, struct iatt *buf,
-                 struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
-{
-        char          statstr[4096]       = {0, };
-        char          preparentstr[4096]  = {0, };
-        char          postparentstr[4096] = {0, };
-        trace_conf_t  *conf               = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        char string[4096]  = {0,};
-        if (trace_fop_names[GF_FOP_MKNOD].enabled) {
-                if (op_ret == 0) {
-                        trace_stat_to_str (buf, statstr);
-                        trace_stat_to_str (preparent, preparentstr);
-                        trace_stat_to_str (postparent, postparentstr);
-
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s (op_ret=%d "
-                                  "*stbuf = {%s}, *preparent = {%s}, "
-                                  "*postparent = {%s})",
-                                  frame->root->unique,
-                                  uuid_utoa (inode->gfid),
-                                  op_ret, statstr, preparentstr,
-                                  postparentstr);
-                } else {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                  frame->root->unique, op_ret,
-                                  op_errno);
-                }
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (mknod, frame, op_ret, op_errno, inode, buf,
-                            preparent, postparent, xdata);
-        return 0;
-}
-
-int
-trace_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno,
-                 inode_t *inode, struct iatt *buf,
-                 struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
-{
-        char          statstr[4096]       = {0, };
-        char          preparentstr[4096]  = {0, };
-        char          postparentstr[4096] = {0, };
-        trace_conf_t  *conf               = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_MKDIR].enabled) {
-                char  string[4096]  = {0,};
-                if (op_ret == 0) {
-                        trace_stat_to_str (buf, statstr);
-                        trace_stat_to_str (preparent, preparentstr);
-                        trace_stat_to_str (postparent, postparentstr);
-
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s (op_ret=%d "
-                                  ", *stbuf = {%s}, *prebuf = {%s}, "
-                                  "*postbuf = {%s} )",
-                                  frame->root->unique,
-                                  uuid_utoa (inode->gfid),
-                                  op_ret, statstr, preparentstr,
-                                  postparentstr);
-                } else {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                  frame->root->unique, op_ret,
-                                  op_errno);
-                }
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (mkdir, frame, op_ret, op_errno, inode, buf,
-                            preparent, postparent, xdata);
-        return 0;
-}
-
-int
-trace_link_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                int32_t op_ret, int32_t op_errno,
-                inode_t *inode, struct iatt *buf,
-                struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
-{
-        char          statstr[4096]       = {0, };
-        char          preparentstr[4096]  = {0, };
-        char          postparentstr[4096] = {0, };
-        trace_conf_t  *conf               = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        char  string[4096]  = {0,};
-        if (trace_fop_names[GF_FOP_LINK].enabled) {
-                if (op_ret == 0) {
-                        trace_stat_to_str (buf, statstr);
-                        trace_stat_to_str (preparent, preparentstr);
-                        trace_stat_to_str (postparent, postparentstr);
-
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": (op_ret=%d, "
-                                  "*stbuf = {%s},  *prebuf = {%s},"
-                                  " *postbuf = {%s})",
-                                  frame->root->unique, op_ret,
-                                  statstr, preparentstr, postparentstr);
-                } else {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s op_ret=%d, "
-                                  "op_errno=%d",
-                                  frame->root->unique,
-                                  uuid_utoa (frame->local),
-                                  op_ret, op_errno);
-                }
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (link, frame, op_ret, op_errno, inode, buf,
-                            preparent, postparent, xdata);
-        return 0;
-}
-
-int
-trace_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
+    int count = 0;
+    char statstr[1024] = {
+        0,
+    };
+    char string[4096] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+    gf_dirent_t *entry = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_READDIRP].enabled) {
+        snprintf(string, sizeof(string),
+                 "%" PRId64 " : gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
+
+        LOG_ELEMENT(conf, string);
+    }
+    if (op_ret < 0)
+        goto out;
+
+    list_for_each_entry(entry, &buf->list, list)
+    {
+        count++;
+        TRACE_STAT_TO_STR(&entry->d_stat, statstr);
+        snprintf(string, sizeof(string),
+                 "entry no. %d, pargfid=%s, "
+                 "bname=%s *buf {%s}",
+                 count, uuid_utoa(frame->local), entry->d_name, statstr);
+        LOG_ELEMENT(conf, string);
+    }
+
+out:
+    TRACE_STACK_UNWIND(readdirp, frame, op_ret, op_errno, buf, xdata);
+    return 0;
+}
+
+int
+trace_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                struct iatt *postbuf, dict_t *xdata)
+{
+    char preopstr[1024] = {
+        0,
+    };
+    char postopstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FSYNC].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(prebuf, preopstr);
+            TRACE_STAT_TO_STR(postbuf, postopstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": (op_ret=%d, "
+                     "*prebuf = {%s}, *postbuf = {%s}",
+                     frame->root->unique, op_ret, preopstr, postopstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(fsync, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+
+    return 0;
+}
+
+int
+trace_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct iatt *statpre,
+                  struct iatt *statpost, dict_t *xdata)
+{
+    char preopstr[1024] = {
+        0,
+    };
+    char postopstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_SETATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(statpre, preopstr);
+            TRACE_STAT_TO_STR(statpost, postopstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": (op_ret=%d, "
+                     "*prebuf = {%s}, *postbuf = {%s})",
+                     frame->root->unique, op_ret, preopstr, postopstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d)",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(setattr, frame, op_ret, op_errno, statpre, statpost,
+                       xdata);
+    return 0;
+}
+
+int
+trace_fsetattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *statpre,
+                   struct iatt *statpost, dict_t *xdata)
+{
+    char preopstr[1024] = {
+        0,
+    };
+    char postopstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FSETATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(statpre, preopstr);
+            TRACE_STAT_TO_STR(statpost, postopstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": (op_ret=%d, "
+                     "*prebuf = {%s}, *postbuf = {%s})",
+                     frame->root->unique, op_ret, preopstr, postopstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d)",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(fsetattr, frame, op_ret, op_errno, statpre, statpost,
+                       xdata);
+    return 0;
+}
+
+int
+trace_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                 struct iatt *postparent, dict_t *xdata)
+{
+    char preparentstr[1024] = {
+        0,
+    };
+    char postparentstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_UNLINK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(preparent, preparentstr);
+            TRACE_STAT_TO_STR(postparent, postparentstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     " *preparent = {%s}, "
+                     "*postparent = {%s})",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     preparentstr, postparentstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d)",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(unlink, frame, op_ret, op_errno, preparent, postparent,
+                       xdata);
+    return 0;
+}
+
+int
+trace_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                 struct iatt *preoldparent, struct iatt *postoldparent,
+                 struct iatt *prenewparent, struct iatt *postnewparent,
+                 dict_t *xdata)
 {
-        trace_conf_t   *conf = NULL;
+    char statstr[1024] = {
+        0,
+    };
+    char preoldparentstr[1024] = {
+        0,
+    };
+    char postoldparentstr[1024] = {
+        0,
+    };
+    char prenewparentstr[1024] = {
+        0,
+    };
+    char postnewparentstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_RENAME].enabled) {
+        char string[6044] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(buf, statstr);
+            TRACE_STAT_TO_STR(preoldparent, preoldparentstr);
+            TRACE_STAT_TO_STR(postoldparent, postoldparentstr);
+            TRACE_STAT_TO_STR(prenewparent, prenewparentstr);
+            TRACE_STAT_TO_STR(postnewparent, postnewparentstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": (op_ret=%d, "
+                     "*stbuf = {%s}, *preoldparent = {%s},"
+                     " *postoldparent = {%s}"
+                     " *prenewparent = {%s}, "
+                     "*postnewparent = {%s})",
+                     frame->root->unique, op_ret, statstr, preoldparentstr,
+                     postoldparentstr, prenewparentstr, postnewparentstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(rename, frame, op_ret, op_errno, buf, preoldparent,
+                       postoldparent, prenewparent, postnewparent, xdata);
+    return 0;
+}
+
+int
+trace_readlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, const char *buf,
+                   struct iatt *stbuf, dict_t *xdata)
+{
+    char statstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_READLINK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(stbuf, statstr);
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": (op_ret=%d, op_errno=%d,"
+                     "buf=%s, stbuf = { %s })",
+                     frame->root->unique, op_ret, op_errno, buf, statstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(readlink, frame, op_ret, op_errno, buf, stbuf, xdata);
+    return 0;
+}
+
+int
+trace_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, inode_t *inode,
+                 struct iatt *buf, dict_t *xdata, struct iatt *postparent)
+{
+    char statstr[1024] = {
+        0,
+    };
+    char postparentstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_LOOKUP].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(buf, statstr);
+            TRACE_STAT_TO_STR(postparent, postparentstr);
+            /* print buf->ia_gfid instead of inode->gfid,
+             * since if the inode is not yet linked to the
+             * inode table (fresh lookup) then null gfid
+             * will be printed.
+             */
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s (op_ret=%d "
+                     "*buf {%s}, *postparent {%s}",
+                     frame->root->unique, uuid_utoa(buf->ia_gfid), op_ret,
+                     statstr, postparentstr);
+
+            /* For 'forget' */
+            inode_ctx_put(inode, this, 0);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d)",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(lookup, frame, op_ret, op_errno, inode, buf, xdata,
+                       postparent);
+    return 0;
+}
+
+int
+trace_symlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, inode_t *inode,
+                  struct iatt *buf, struct iatt *preparent,
+                  struct iatt *postparent, dict_t *xdata)
+{
+    char statstr[1024] = {
+        0,
+    };
+    char preparentstr[1024] = {
+        0,
+    };
+    char postparentstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_SYMLINK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(buf, statstr);
+            TRACE_STAT_TO_STR(preparent, preparentstr);
+            TRACE_STAT_TO_STR(postparent, postparentstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s (op_ret=%d "
+                     "*stbuf = {%s}, *preparent = {%s}, "
+                     "*postparent = {%s})",
+                     frame->root->unique, uuid_utoa(inode->gfid), op_ret,
+                     statstr, preparentstr, postparentstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64 ": op_ret=%d, op_errno=%d", frame->root->unique,
+                     op_ret, op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(symlink, frame, op_ret, op_errno, inode, buf, preparent,
+                       postparent, xdata);
+    return 0;
+}
+
+int
+trace_mknod_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, inode_t *inode,
+                struct iatt *buf, struct iatt *preparent,
+                struct iatt *postparent, dict_t *xdata)
+{
+    char statstr[1024] = {
+        0,
+    };
+    char preparentstr[1024] = {
+        0,
+    };
+    char postparentstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    char string[4096] = {
+        0,
+    };
+    if (trace_fop_names[GF_FOP_MKNOD].enabled) {
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(buf, statstr);
+            TRACE_STAT_TO_STR(preparent, preparentstr);
+            TRACE_STAT_TO_STR(postparent, postparentstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s (op_ret=%d "
+                     "*stbuf = {%s}, *preparent = {%s}, "
+                     "*postparent = {%s})",
+                     frame->root->unique, uuid_utoa(inode->gfid), op_ret,
+                     statstr, preparentstr, postparentstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64 ": (op_ret=%d, op_errno=%d)",
+                     frame->root->unique, op_ret, op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(mknod, frame, op_ret, op_errno, inode, buf, preparent,
+                       postparent, xdata);
+    return 0;
+}
+
+int
+trace_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, inode_t *inode,
+                struct iatt *buf, struct iatt *preparent,
+                struct iatt *postparent, dict_t *xdata)
+{
+    char statstr[1024] = {
+        0,
+    };
+    char preparentstr[1024] = {
+        0,
+    };
+    char postparentstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_MKDIR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(buf, statstr);
+            TRACE_STAT_TO_STR(preparent, preparentstr);
+            TRACE_STAT_TO_STR(postparent, postparentstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s (op_ret=%d "
+                     ", *stbuf = {%s}, *prebuf = {%s}, "
+                     "*postbuf = {%s} )",
+                     frame->root->unique, uuid_utoa(inode->gfid), op_ret,
+                     statstr, preparentstr, postparentstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64 ": (op_ret=%d, op_errno=%d)",
+                     frame->root->unique, op_ret, op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(mkdir, frame, op_ret, op_errno, inode, buf, preparent,
+                       postparent, xdata);
+    return 0;
+}
+
+int
+trace_link_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, inode_t *inode,
+               struct iatt *buf, struct iatt *preparent,
+               struct iatt *postparent, dict_t *xdata)
+{
+    char statstr[1024] = {
+        0,
+    };
+    char preparentstr[1024] = {
+        0,
+    };
+    char postparentstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    char string[4096] = {
+        0,
+    };
+    if (trace_fop_names[GF_FOP_LINK].enabled) {
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(buf, statstr);
+            TRACE_STAT_TO_STR(preparent, preparentstr);
+            TRACE_STAT_TO_STR(postparent, postparentstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": (op_ret=%d, "
+                     "*stbuf = {%s},  *prebuf = {%s},"
+                     " *postbuf = {%s})",
+                     frame->root->unique, op_ret, statstr, preparentstr,
+                     postparentstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(link, frame, op_ret, op_errno, inode, buf, preparent,
+                       postparent, xdata);
+    return 0;
+}
+
+int
+trace_flush_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    char string[4096] = {
+        0,
+    };
+    if (trace_fop_names[GF_FOP_FLUSH].enabled) {
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
+
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(flush, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+int
+trace_opendir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
+{
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    char string[4096] = {
+        0,
+    };
+    if (trace_fop_names[GF_FOP_OPENDIR].enabled) {
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s op_ret=%d, op_errno=%d,"
+                 " fd=%p",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret, op_errno,
+                 fd);
+
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    /* for 'releasedir' log */
+    if (op_ret >= 0)
+        fd_ctx_set(fd, this, 0);
+
+    TRACE_STACK_UNWIND(opendir, frame, op_ret, op_errno, fd, xdata);
+    return 0;
+}
+
+int
+trace_rmdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                struct iatt *postparent, dict_t *xdata)
+{
+    char preparentstr[1024] = {
+        0,
+    };
+    char postparentstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_RMDIR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(preparent, preparentstr);
+            TRACE_STAT_TO_STR(postparent, postparentstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "*prebuf={%s},  *postbuf={%s}",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     preparentstr, postparentstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(rmdir, frame, op_ret, op_errno, preparent, postparent,
+                       xdata);
+    return 0;
+}
+
+int
+trace_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                   struct iatt *postbuf, dict_t *xdata)
+{
+    char preopstr[1024] = {
+        0,
+    };
+    char postopstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_TRUNCATE].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(prebuf, preopstr);
+            TRACE_STAT_TO_STR(postbuf, postopstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": (op_ret=%d, "
+                     "*prebuf = {%s}, *postbuf = {%s} )",
+                     frame->root->unique, op_ret, preopstr, postopstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(truncate, frame, op_ret, op_errno, prebuf, postbuf,
+                       xdata);
+    return 0;
+}
+
+int
+trace_statfs_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct statvfs *buf,
+                 dict_t *xdata)
+{
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_STATFS].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": ({f_bsize=%lu, "
+                     "f_frsize=%lu, "
+                     "f_blocks=%" GF_PRI_FSBLK ", f_bfree=%" GF_PRI_FSBLK
+                     ", "
+                     "f_bavail=%" GF_PRI_FSBLK
+                     ", "
+                     "f_files=%" GF_PRI_FSBLK
+                     ", "
+                     "f_ffree=%" GF_PRI_FSBLK
+                     ", "
+                     "f_favail=%" GF_PRI_FSBLK
+                     ", "
+                     "f_fsid=%lu, f_flag=%lu, "
+                     "f_namemax=%lu}) => ret=%d",
+                     frame->root->unique, buf->f_bsize, buf->f_frsize,
+                     buf->f_blocks, buf->f_bfree, buf->f_bavail, buf->f_files,
+                     buf->f_ffree, buf->f_favail, buf->f_fsid, buf->f_flag,
+                     buf->f_namemax, op_ret);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": (op_ret=%d, "
+                     "op_errno=%d)",
+                     frame->root->unique, op_ret, op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(statfs, frame, op_ret, op_errno, buf, xdata);
+    return 0;
+}
+
+int
+trace_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        char      string[4096] = {0,};
-        if (trace_fop_names[GF_FOP_FLUSH].enabled) {
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s op_ret=%d, op_errno=%d",
-                          frame->root->unique, uuid_utoa (frame->local),
-                          op_ret, op_errno);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_SETXATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 out:
-        TRACE_STACK_UNWIND (flush, frame, op_ret, op_errno, xdata);
-        return 0;
+    TRACE_STACK_UNWIND(setxattr, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
 int
-trace_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
+trace_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *dict,
+                   dict_t *xdata)
 {
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        char    string[4096] = {0,};
-        if (trace_fop_names[GF_FOP_OPENDIR].enabled) {
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s op_ret=%d, op_errno=%d,"
-                          " fd=%p",
-                          frame->root->unique, uuid_utoa (frame->local),
-                          op_ret, op_errno, fd);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_GETXATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s op_ret=%d, op_errno=%d,"
+                 " dict=%p",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret, op_errno,
+                 dict);
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 out:
-        /* for 'releasedir' log */
-        if (op_ret >= 0)
-                fd_ctx_set (fd, this, 0);
-
-        TRACE_STACK_UNWIND (opendir, frame, op_ret, op_errno, fd, xdata);
-        return 0;
-}
-
-int
-trace_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno,
-                 struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
-{
-        char           preparentstr[4096]  = {0, };
-        char           postparentstr[4096] = {0, };
-        trace_conf_t   *conf               = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_RMDIR].enabled) {
-                char  string[4096] = {0,};
-                if (op_ret == 0) {
-                        trace_stat_to_str (preparent, preparentstr);
-                        trace_stat_to_str (postparent, postparentstr);
-
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s op_ret=%d, "
-                                  "*prebuf={%s},  *postbuf={%s}",
-                                  frame->root->unique,
-                                  uuid_utoa (frame->local),
-                                  op_ret, preparentstr, postparentstr);
-                } else {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s op_ret=%d, "
-                                  "op_errno=%d", frame->root->unique,
-                                  uuid_utoa (frame->local),
-                                  op_ret, op_errno);
-                }
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (rmdir, frame, op_ret, op_errno,
-                            preparent, postparent, xdata);
-        return 0;
-}
-
-int
-trace_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno,
-                    struct iatt *prebuf, struct iatt *postbuf, dict_t *xdata)
-{
-        char           preopstr[4096]  = {0, };
-        char           postopstr[4096] = {0, };
-        trace_conf_t   *conf           = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_TRUNCATE].enabled) {
-                char   string[4096] = {0,};
-                if (op_ret == 0) {
-                        trace_stat_to_str (prebuf, preopstr);
-                        trace_stat_to_str (postbuf, postopstr);
-
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": (op_ret=%d, "
-                                  "*prebuf = {%s}, *postbuf = {%s} )",
-                                  frame->root->unique, op_ret,
-                                  preopstr, postopstr);
-                } else {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s op_ret=%d, "
-                                  "op_errno=%d", frame->root->unique,
-                                  uuid_utoa (frame->local), op_ret,
-                                  op_errno);
-                }
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (truncate, frame, op_ret, op_errno, prebuf,
-                            postbuf, xdata);
-        return 0;
-}
+    TRACE_STACK_UNWIND(getxattr, frame, op_ret, op_errno, dict, xdata);
 
-int
-trace_statfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno, struct statvfs *buf,
-                  dict_t *xdata)
-{
-        trace_conf_t  *conf = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_STATFS].enabled) {
-                char   string[4096] = {0,};
-                if (op_ret == 0) {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": ({f_bsize=%lu, "
-                                  "f_frsize=%lu, "
-                                  "f_blocks=%"GF_PRI_FSBLK
-                                  ", f_bfree=%"GF_PRI_FSBLK", "
-                                  "f_bavail=%"GF_PRI_FSBLK", "
-                                  "f_files=%"GF_PRI_FSBLK", "
-                                  "f_ffree=%"GF_PRI_FSBLK", "
-                                  "f_favail=%"GF_PRI_FSBLK", "
-                                  "f_fsid=%lu, f_flag=%lu, "
-                                  "f_namemax=%lu}) => ret=%d",
-                                  frame->root->unique, buf->f_bsize,
-                                  buf->f_frsize, buf->f_blocks,
-                                  buf->f_bfree, buf->f_bavail,
-                                  buf->f_files, buf->f_ffree,
-                                  buf->f_favail, buf->f_fsid,
-                                  buf->f_flag, buf->f_namemax, op_ret);
-                } else {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": (op_ret=%d, "
-                                  "op_errno=%d)",
-                                  frame->root->unique, op_ret,
-                                  op_errno);
-                }
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (statfs, frame, op_ret, op_errno, buf, xdata);
-        return 0;
+    return 0;
 }
 
 int
-trace_setxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+trace_fsetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
                     int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        trace_conf_t     *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_SETXATTR].enabled) {
-                char      string[4096] = {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s op_ret=%d, op_errno=%d",
-                          frame->root->unique,
-                          uuid_utoa (frame->local), op_ret,
-                          op_errno);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FSETXATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 out:
-        TRACE_STACK_UNWIND (setxattr, frame, op_ret, op_errno, xdata);
-        return 0;
+    TRACE_STACK_UNWIND(fsetxattr, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
 int
-trace_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+trace_fgetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
                     int32_t op_ret, int32_t op_errno, dict_t *dict,
                     dict_t *xdata)
 {
-        trace_conf_t    *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_GETXATTR].enabled) {
-                char       string[4096] = {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s op_ret=%d, op_errno=%d,"
-                          " dict=%p", frame->root->unique,
-                          uuid_utoa (frame->local), op_ret, op_errno,
-                          dict);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FGETXATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s op_ret=%d, op_errno=%d,"
+                 " dict=%p",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret, op_errno,
+                 dict);
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 out:
-        TRACE_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict, xdata);
-
-        return 0;
-}
-
-int
-trace_fsetxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-        trace_conf_t   *conf = NULL;
-
-        conf = this->private;
+    TRACE_STACK_UNWIND(fgetxattr, frame, op_ret, op_errno, dict, xdata);
 
-        if (!conf->log_file && !conf->log_history)
-                goto out;
-        if (trace_fop_names[GF_FOP_FSETXATTR].enabled) {
-                char    string[4096] = {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s op_ret=%d, op_errno=%d",
-                          frame->root->unique,
-                          uuid_utoa (frame->local), op_ret, op_errno);
-
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (fsetxattr, frame, op_ret, op_errno, xdata);
-        return 0;
+    return 0;
 }
 
 int
-trace_fgetxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, dict_t *dict,
-                     dict_t *xdata)
+trace_removexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        trace_conf_t    *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_FGETXATTR].enabled) {
-                char      string[4096]  = {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s op_ret=%d, op_errno=%d,"
-                          " dict=%p", frame->root->unique,
-                          uuid_utoa (frame->local), op_ret, op_errno,
-                          dict);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_REMOVEXATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 out:
-        TRACE_STACK_UNWIND (fgetxattr, frame, op_ret, op_errno, dict, xdata);
+    TRACE_STACK_UNWIND(removexattr, frame, op_ret, op_errno, xdata);
 
-        return 0;
+    return 0;
 }
 
 int
-trace_removexattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
+trace_fsyncdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        trace_conf_t  *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_REMOVEXATTR].enabled) {
-                char       string[4096] = {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s op_ret=%d, op_errno=%d",
-                          frame->root->unique,
-                          uuid_utoa (frame->local), op_ret, op_errno);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FSYNCDIR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 out:
-        TRACE_STACK_UNWIND (removexattr, frame, op_ret, op_errno, xdata);
-
-        return 0;
+    TRACE_STACK_UNWIND(fsyncdir, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
 int
-trace_fsyncdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
+trace_access_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        trace_conf_t     *conf = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_FSYNCDIR].enabled) {
-                char       string[4096]   =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s op_ret=%d, op_errno=%d",
-                          frame->root->unique,
-                          uuid_utoa (frame->local), op_ret, op_errno);
-
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (fsyncdir, frame, op_ret, op_errno, xdata);
-        return 0;
-}
-
-int
-trace_access_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_ACCESS].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s op_ret=%d, "
+                 "op_errno=%d)",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
+
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(access, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+int
+trace_ftruncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                    struct iatt *postbuf, dict_t *xdata)
+{
+    char prebufstr[1024] = {
+        0,
+    };
+    char postbufstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FTRUNCATE].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(prebuf, prebufstr);
+            TRACE_STAT_TO_STR(postbuf, postbufstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": op_ret=%d, "
+                     "*prebuf = {%s}, *postbuf = {%s} )",
+                     frame->root->unique, op_ret, prebufstr, postbufstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(ftruncate, frame, op_ret, op_errno, prebuf, postbuf,
+                       xdata);
+    return 0;
+}
+
+int
+trace_fstat_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                dict_t *xdata)
+{
+    char statstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FSTAT].enabled) {
+        char string[4096] = {0.};
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(buf, statstr);
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d "
+                     "buf=%s",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     statstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(fstat, frame, op_ret, op_errno, buf, xdata);
+    return 0;
+}
+
+int
+trace_lk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, struct gf_flock *lock, dict_t *xdata)
+{
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_LK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "{l_type=%d, l_whence=%d, "
+                     "l_start=%" PRId64
+                     ", "
+                     "l_len=%" PRId64 ", l_pid=%u})",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     lock->l_type, lock->l_whence, lock->l_start, lock->l_len,
+                     lock->l_pid);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d)",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(lk, frame, op_ret, op_errno, lock, xdata);
+    return 0;
+}
+
+int
+trace_entrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
                   int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_ACCESS].enabled) {
-                char      string[4096] = {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s op_ret=%d, "
-                          "op_errno=%d)", frame->root->unique,
-                          uuid_utoa (frame->local), op_ret, op_errno);
-
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (access, frame, op_ret, op_errno, xdata);
-        return 0;
-}
-
-int
-trace_ftruncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno,
-                     struct iatt *prebuf, struct iatt *postbuf, dict_t *xdata)
-{
-        char          prebufstr[4096]  = {0, };
-        char          postbufstr[4096] = {0, };
-        trace_conf_t  *conf            = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_FTRUNCATE].enabled) {
-                char  string[4096]  = {0,};
-                if (op_ret == 0) {
-                        trace_stat_to_str (prebuf, prebufstr);
-                        trace_stat_to_str (postbuf, postbufstr);
-
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": op_ret=%d, "
-                                  "*prebuf = {%s}, *postbuf = {%s} )",
-                                  frame->root->unique, op_ret,
-                                  prebufstr, postbufstr);
-                } else {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s op_ret=%d, "
-                                  "op_errno=%d", frame->root->unique,
-                                  uuid_utoa (frame->local), op_ret,
-                                  op_errno);
-                }
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (ftruncate, frame, op_ret, op_errno, prebuf, postbuf,
-                            xdata);
-        return 0;
-}
-
-int
-trace_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno, struct iatt *buf, dict_t *xdata)
-{
-        char          statstr[4096] = {0, };
-        trace_conf_t  *conf         = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_FSTAT].enabled) {
-                char string[4096]  = {0.};
-                if (op_ret == 0) {
-                        trace_stat_to_str (buf, statstr);
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s op_ret=%d "
-                                  "buf=%s", frame->root->unique,
-                                  uuid_utoa (frame->local), op_ret,
-                                  statstr);
-                } else {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s op_ret=%d, "
-                                  "op_errno=%d", frame->root->unique,
-                                  uuid_utoa (frame->local), op_ret,
-                                  op_errno);
-                }
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (fstat, frame, op_ret, op_errno, buf, xdata);
-        return 0;
-}
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_ENTRYLK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
 
-int
-trace_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-              int32_t op_ret, int32_t op_errno, struct gf_flock *lock,
-              dict_t *xdata)
-{
-        trace_conf_t   *conf = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_LK].enabled) {
-                char      string[4096] = {0,};
-                if (op_ret == 0) {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s op_ret=%d, "
-                                  "{l_type=%d, l_whence=%d, "
-                                  "l_start=%"PRId64", "
-                                  "l_len=%"PRId64", l_pid=%u})",
-                                  frame->root->unique,
-                                  uuid_utoa (frame->local),
-                                  op_ret, lock->l_type, lock->l_whence,
-                                  lock->l_start, lock->l_len,
-                                  lock->l_pid);
-                } else {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s op_ret=%d, "
-                                  "op_errno=%d)", frame->root->unique,
-                                  uuid_utoa (frame->local), op_ret,
-                                  op_errno);
-                }
-
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 out:
-        TRACE_STACK_UNWIND (lk, frame, op_ret, op_errno, lock, xdata);
-        return 0;
+    TRACE_STACK_UNWIND(entrylk, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
 int
-trace_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+trace_fentrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        trace_conf_t  *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_ENTRYLK].enabled) {
-                char   string[4096] = {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s op_ret=%d, op_errno=%d",
-                          frame->root->unique,
-                          uuid_utoa (frame->local), op_ret, op_errno);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FENTRYLK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 out:
-        TRACE_STACK_UNWIND (entrylk, frame, op_ret, op_errno, xdata);
-        return 0;
+    TRACE_STACK_UNWIND(fentrylk, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
 int
-trace_fentrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
+trace_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
 {
-        trace_conf_t    *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_FENTRYLK].enabled) {
-                char      string[4096] = {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s op_ret=%d, op_errno=%d",
-                          frame->root->unique,
-                          uuid_utoa (frame->local), op_ret, op_errno);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_XATTROP].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 out:
-        TRACE_STACK_UNWIND (fentrylk, frame, op_ret, op_errno, xdata);
-        return 0;
+    TRACE_STACK_UNWIND(xattrop, frame, op_ret, op_errno, dict, xdata);
+    return 0;
 }
 
 int
-trace_xattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+trace_fxattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
                    int32_t op_ret, int32_t op_errno, dict_t *dict,
                    dict_t *xdata)
 {
-        trace_conf_t  *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_XATTROP].enabled) {
-                char    string[4096] = {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s op_ret=%d, op_errno=%d",
-                          frame->root->unique,
-                          uuid_utoa (frame->local), op_ret, op_errno);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FXATTROP].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 out:
-        TRACE_STACK_UNWIND (xattrop, frame, op_ret, op_errno, dict, xdata);
-        return 0;
+    TRACE_STACK_UNWIND(fxattrop, frame, op_ret, op_errno, dict, xdata);
+    return 0;
 }
 
 int
-trace_fxattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, dict_t *dict,
-                    dict_t *xdata)
+trace_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        trace_conf_t    *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_FXATTROP].enabled) {
-                char      string[4096] = {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s op_ret=%d, op_errno=%d",
-                          frame->root->unique,
-                          uuid_utoa (frame->local), op_ret, op_errno);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_INODELK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 out:
-        TRACE_STACK_UNWIND (fxattrop, frame, op_ret, op_errno, dict, xdata);
-        return 0;
+    TRACE_STACK_UNWIND(inodelk, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
 int
-trace_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+trace_finodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        trace_conf_t   *conf = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_INODELK].enabled) {
-                char       string[4096] = {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s op_ret=%d, op_errno=%d",
-                          frame->root->unique,
-                          uuid_utoa (frame->local),op_ret, op_errno);
-
-                LOG_ELEMENT (conf, string);
-        }
-out:
-        TRACE_STACK_UNWIND (inodelk, frame, op_ret, op_errno, xdata);
-        return 0;
-}
-
-int
-trace_finodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
-{
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_FINODELK].enabled) {
-                char      string[4096] = {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s op_ret=%d, op_errno=%d",
-                          frame->root->unique,
-                          uuid_utoa (frame->local), op_ret, op_errno);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FINODELK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 out:
-        TRACE_STACK_UNWIND (finodelk, frame, op_ret, op_errno, xdata);
-        return 0;
+    TRACE_STACK_UNWIND(finodelk, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
 int
-trace_rchecksum_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno,
-                     uint32_t weak_checksum, uint8_t *strong_checksum,
-                     dict_t *xdata)
+trace_rchecksum_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, uint32_t weak_checksum,
+                    uint8_t *strong_checksum, dict_t *xdata)
 {
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-                goto out;
-        if (trace_fop_names[GF_FOP_RCHECKSUM].enabled) {
-                char      string[4096] = {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s op_ret=%d op_errno=%d",
-                          frame->root->unique,
-                          uuid_utoa (frame->local), op_ret, op_errno);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_RCHECKSUM].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        TRACE_STACK_UNWIND (rchecksum, frame, op_ret, op_errno, weak_checksum,
-                            strong_checksum, xdata);
+    TRACE_STACK_UNWIND(rchecksum, frame, op_ret, op_errno, weak_checksum,
+                       strong_checksum, xdata);
 
-        return 0;
+    return 0;
 }
 
 /* *_cbk section over <----------> fop section start */
 
 int
-trace_entrylk (call_frame_t *frame, xlator_t *this,
-               const char *volume, loc_t *loc, const char *basename,
-               entrylk_cmd cmd, entrylk_type type, dict_t *xdata)
+trace_entrylk(call_frame_t *frame, xlator_t *this, const char *volume,
+              loc_t *loc, const char *basename, entrylk_cmd cmd,
+              entrylk_type type, dict_t *xdata)
 {
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_ENTRYLK].enabled) {
-                char     string[4096] = {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s volume=%s, (path=%s "
-                          "basename=%s, cmd=%s, type=%s)",
-                          frame->root->unique,
-                          uuid_utoa (loc->inode->gfid),
-                          volume, loc->path, basename,
-                          ((cmd == ENTRYLK_LOCK) ? "ENTRYLK_LOCK" :
-                           "ENTRYLK_UNLOCK"),
-                          ((type == ENTRYLK_RDLCK) ? "ENTRYLK_RDLCK" :
-                           "ENTRYLK_WRLCK"));
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_ENTRYLK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s volume=%s, (path=%s "
+                 "basename=%s, cmd=%s, type=%s)",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), volume,
+                 loc->path, basename,
+                 ((cmd == ENTRYLK_LOCK) ? "ENTRYLK_LOCK" : "ENTRYLK_UNLOCK"),
+                 ((type == ENTRYLK_RDLCK) ? "ENTRYLK_RDLCK" : "ENTRYLK_WRLCK"));
 
-                frame->local = loc->inode->gfid;
+        frame->local = loc->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_entrylk_cbk,
-                    FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->entrylk,
-                    volume, loc, basename, cmd, type, xdata);
-        return 0;
+    STACK_WIND(frame, trace_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, volume, loc, basename, cmd,
+               type, xdata);
+    return 0;
 }
 
 int
-trace_inodelk (call_frame_t *frame, xlator_t *this, const char *volume,
-               loc_t *loc, int32_t cmd, struct gf_flock *flock, dict_t *xdata)
+trace_inodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+              loc_t *loc, int32_t cmd, struct gf_flock *flock, dict_t *xdata)
 {
-        char         *cmd_str  = NULL;
-        char         *type_str = NULL;
-        trace_conf_t *conf     = NULL;
+    char *cmd_str = NULL;
+    char *type_str = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_INODELK].enabled) {
-                char string[4096]  = {0,};
-                switch (cmd) {
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_INODELK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        switch (cmd) {
 #if F_GETLK != F_GETLK64
-                case F_GETLK64:
+            case F_GETLK64:
 #endif
-                case F_GETLK:
-                        cmd_str = "GETLK";
-                        break;
+            case F_GETLK:
+                cmd_str = "GETLK";
+                break;
 
 #if F_SETLK != F_SETLK64
-                case F_SETLK64:
+            case F_SETLK64:
 #endif
-                case F_SETLK:
-                        cmd_str = "SETLK";
-                        break;
+            case F_SETLK:
+                cmd_str = "SETLK";
+                break;
 
 #if F_SETLKW != F_SETLKW64
-                case F_SETLKW64:
+            case F_SETLKW64:
 #endif
-                case F_SETLKW:
-                        cmd_str = "SETLKW";
-                        break;
-
-                default:
-                        cmd_str = "UNKNOWN";
-                        break;
-                }
-
-                switch (flock->l_type) {
-                case F_RDLCK:
-                        type_str = "READ";
-                        break;
-                case F_WRLCK:
-                        type_str = "WRITE";
-                        break;
-                case F_UNLCK:
-                        type_str = "UNLOCK";
-                        break;
-                default:
-                        type_str = "UNKNOWN";
-                        break;
-                }
-
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s volume=%s, (path=%s "
-                          "cmd=%s, type=%s, start=%llu, len=%llu, "
-                          "pid=%llu)", frame->root->unique,
-                          uuid_utoa (loc->inode->gfid), volume,
-                          loc->path, cmd_str, type_str,
-                          (unsigned long long)flock->l_start,
-                          (unsigned long long) flock->l_len,
-                          (unsigned long long) flock->l_pid);
-
-                frame->local = loc->inode->gfid;
-
-                LOG_ELEMENT (conf, string);
+            case F_SETLKW:
+                cmd_str = "SETLKW";
+                break;
+
+            default:
+                cmd_str = "UNKNOWN";
+                break;
+        }
+
+        switch (flock->l_type) {
+            case F_RDLCK:
+                type_str = "READ";
+                break;
+            case F_WRLCK:
+                type_str = "WRITE";
+                break;
+            case F_UNLCK:
+                type_str = "UNLOCK";
+                break;
+            default:
+                type_str = "UNKNOWN";
+                break;
         }
 
+        snprintf(
+            string, sizeof(string),
+            "%" PRId64
+            ": gfid=%s volume=%s, (path=%s "
+            "cmd=%s, type=%s, start=%llu, len=%llu, "
+            "pid=%llu)",
+            frame->root->unique, uuid_utoa(loc->inode->gfid), volume, loc->path,
+            cmd_str, type_str, (unsigned long long)flock->l_start,
+            (unsigned long long)flock->l_len, (unsigned long long)flock->l_pid);
+
+        frame->local = loc->inode->gfid;
+
+        LOG_ELEMENT(conf, string);
+    }
+
 out:
-        STACK_WIND (frame, trace_inodelk_cbk,
-                    FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->inodelk,
-                    volume, loc, cmd, flock, xdata);
-        return 0;
+    STACK_WIND(frame, trace_inodelk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->inodelk, volume, loc, cmd, flock,
+               xdata);
+    return 0;
 }
 
 int
-trace_finodelk (call_frame_t *frame, xlator_t *this, const char *volume,
-                fd_t *fd, int32_t cmd, struct gf_flock *flock, dict_t *xdata)
+trace_finodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+               fd_t *fd, int32_t cmd, struct gf_flock *flock, dict_t *xdata)
 {
-        char          *cmd_str  = NULL;
-        char          *type_str = NULL;
-        trace_conf_t  *conf     = NULL;
+    char *cmd_str = NULL;
+    char *type_str = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_FINODELK].enabled) {
-                char  string[4096] = {0,};
-                switch (cmd) {
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FINODELK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        switch (cmd) {
 #if F_GETLK != F_GETLK64
-                case F_GETLK64:
+            case F_GETLK64:
 #endif
-                case F_GETLK:
-                        cmd_str = "GETLK";
-                        break;
+            case F_GETLK:
+                cmd_str = "GETLK";
+                break;
 
 #if F_SETLK != F_SETLK64
-                case F_SETLK64:
+            case F_SETLK64:
 #endif
-                case F_SETLK:
-                        cmd_str = "SETLK";
-                        break;
+            case F_SETLK:
+                cmd_str = "SETLK";
+                break;
 
 #if F_SETLKW != F_SETLKW64
-                case F_SETLKW64:
+            case F_SETLKW64:
 #endif
-                case F_SETLKW:
-                        cmd_str = "SETLKW";
-                        break;
-
-                default:
-                        cmd_str = "UNKNOWN";
-                        break;
-                }
-
-                switch (flock->l_type) {
-                case F_RDLCK:
-                        type_str = "READ";
-                        break;
-                case F_WRLCK:
-                        type_str = "WRITE";
-                        break;
-                case F_UNLCK:
-                        type_str = "UNLOCK";
-                        break;
-                default:
-                        type_str = "UNKNOWN";
-                        break;
-                }
-
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s volume=%s, (fd =%p "
-                          "cmd=%s, type=%s, start=%llu, len=%llu, "
-                          "pid=%llu)", frame->root->unique,
-                          uuid_utoa (fd->inode->gfid), volume, fd,
-                          cmd_str, type_str,
-                          (unsigned long long) flock->l_start,
-                          (unsigned long long) flock->l_len,
-                          (unsigned long long) flock->l_pid);
-
-                frame->local = fd->inode->gfid;
-
-                LOG_ELEMENT (conf, string);
+            case F_SETLKW:
+                cmd_str = "SETLKW";
+                break;
+
+            default:
+                cmd_str = "UNKNOWN";
+                break;
         }
+
+        switch (flock->l_type) {
+            case F_RDLCK:
+                type_str = "READ";
+                break;
+            case F_WRLCK:
+                type_str = "WRITE";
+                break;
+            case F_UNLCK:
+                type_str = "UNLOCK";
+                break;
+            default:
+                type_str = "UNKNOWN";
+                break;
+        }
+
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s volume=%s, (fd =%p "
+                 "cmd=%s, type=%s, start=%llu, len=%llu, "
+                 "pid=%llu)",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), volume, fd,
+                 cmd_str, type_str, (unsigned long long)flock->l_start,
+                 (unsigned long long)flock->l_len,
+                 (unsigned long long)flock->l_pid);
+
+        frame->local = fd->inode->gfid;
+
+        LOG_ELEMENT(conf, string);
+    }
 out:
-        STACK_WIND (frame, trace_finodelk_cbk,
-                    FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->finodelk,
-                    volume, fd, cmd, flock, xdata);
-        return 0;
+    STACK_WIND(frame, trace_finodelk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->finodelk, volume, fd, cmd, flock,
+               xdata);
+    return 0;
 }
 
 int
-trace_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc,
-               gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+trace_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+              gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
 {
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_XATTROP].enabled) {
-                char    string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s (path=%s flags=%d)",
-                          frame->root->unique,
-                          uuid_utoa (loc->inode->gfid), loc->path,
-                          flags);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_XATTROP].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s (path=%s flags=%d)", frame->root->unique,
+                 uuid_utoa(loc->inode->gfid), loc->path, flags);
 
-                frame->local = loc->inode->gfid;
+        frame->local = loc->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_xattrop_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->xattrop,
-                    loc, flags, dict, xdata);
+    STACK_WIND(frame, trace_xattrop_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->xattrop, loc, flags, dict, xdata);
 
-        return 0;
+    return 0;
 }
 
 int
-trace_fxattrop (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+trace_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+               gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
 {
-        trace_conf_t    *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_FXATTROP].enabled) {
-                char    string[4096]  = {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s fd=%p, flags=%d",
-                          frame->root->unique,
-                          uuid_utoa (fd->inode->gfid), fd, flags);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FXATTROP].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s fd=%p, flags=%d",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), fd, flags);
 
-                frame->local = fd->inode->gfid;
+        frame->local = fd->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_fxattrop_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fxattrop,
-                    fd, flags, dict, xdata);
+    STACK_WIND(frame, trace_fxattrop_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fxattrop, fd, flags, dict, xdata);
 
-        return 0;
+    return 0;
 }
 
 int
-trace_lookup (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, dict_t *xdata)
+trace_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-        trace_conf_t *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_LOOKUP].enabled) {
-                char string[4096] = {0,};
-                /* TODO: print all the keys mentioned in xattr_req */
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s path=%s",
-                          frame->root->unique,
-                          uuid_utoa (loc->inode->gfid), loc->path);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_LOOKUP].enabled) {
+        char string[4096] = {
+            0,
+        };
+        /* TODO: print all the keys mentioned in xattr_req */
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s path=%s",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path);
 
-                frame->local = loc->inode->gfid;
+        frame->local = loc->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_lookup_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->lookup,
-                    loc, xdata);
+    STACK_WIND(frame, trace_lookup_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lookup, loc, xdata);
 
-        return 0;
+    return 0;
 }
 
 int
-trace_stat (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+trace_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-        trace_conf_t  *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-                goto out;
-        if (trace_fop_names[GF_FOP_STAT].enabled) {
-                char  string[4096]  = {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s path=%s",
-                          frame->root->unique,
-                          uuid_utoa (loc->inode->gfid), loc->path);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_STAT].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s path=%s",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path);
 
-                frame->local = loc->inode->gfid;
+        frame->local = loc->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_stat_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->stat,
-                    loc, xdata);
+    STACK_WIND(frame, trace_stat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->stat, loc, xdata);
 
-        return 0;
+    return 0;
 }
 
 int
-trace_readlink (call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
-                dict_t *xdata)
+trace_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+               dict_t *xdata)
 {
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_READLINK].enabled) {
-                char      string[4096] = {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s path=%s, "
-                          "size=%"GF_PRI_SIZET")", frame->root->unique,
-                          uuid_utoa (loc->inode->gfid), loc->path,
-                          size);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_READLINK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s path=%s, "
+                 "size=%" GF_PRI_SIZET ")",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path,
+                 size);
 
-                frame->local = loc->inode->gfid;
+        frame->local = loc->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_readlink_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->readlink,
-                    loc, size, xdata);
+    STACK_WIND(frame, trace_readlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readlink, loc, size, xdata);
 
-        return 0;
+    return 0;
 }
 
 int
-trace_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc,
-             mode_t mode, dev_t dev, mode_t umask, dict_t *xdata)
+trace_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+            dev_t dev, mode_t umask, dict_t *xdata)
 {
-        trace_conf_t  *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-                goto out;
-        if (trace_fop_names[GF_FOP_MKNOD].enabled) {
-                char     string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s path=%s mode=%d "
-                          "umask=0%o, dev=%"GF_PRI_DEV")",
-                          frame->root->unique,
-                          uuid_utoa (loc->inode->gfid), loc->path,
-                          mode, umask, dev);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_MKNOD].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s path=%s mode=%d "
+                 "umask=0%o, dev=%" GF_PRI_DEV ")",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path,
+                 mode, umask, dev);
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_mknod_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->mknod,
-                    loc, mode, dev, umask, xdata);
+    STACK_WIND(frame, trace_mknod_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mknod, loc, mode, dev, umask, xdata);
 
-        return 0;
+    return 0;
 }
 
 int
-trace_mkdir (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
-             mode_t umask, dict_t *xdata)
+trace_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+            mode_t umask, dict_t *xdata)
 {
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_MKDIR].enabled) {
-                char     string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s path=%s mode=%d"
-                          " umask=0%o", frame->root->unique,
-                          uuid_utoa (loc->inode->gfid), loc->path,
-                          mode, umask);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_MKDIR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s path=%s mode=%d"
+                 " umask=0%o",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path,
+                 mode, umask);
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_mkdir_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->mkdir,
-                    loc, mode, umask, xdata);
-        return 0;
+    STACK_WIND(frame, trace_mkdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mkdir, loc, mode, umask, xdata);
+    return 0;
 }
 
 int
-trace_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
-              dict_t *xdata)
+trace_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+             dict_t *xdata)
 {
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_UNLINK].enabled) {
-                char     string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s path=%s flag=%d",
-                          frame->root->unique,
-                          uuid_utoa (loc->inode->gfid), loc->path,
-                          xflag);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_UNLINK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s path=%s flag=%d",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path,
+                 xflag);
 
-                frame->local = loc->inode->gfid;
+        frame->local = loc->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 out:
-        STACK_WIND (frame, trace_unlink_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->unlink,
-                    loc, xflag, xdata);
-        return 0;
+    STACK_WIND(frame, trace_unlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->unlink, loc, xflag, xdata);
+    return 0;
 }
 
 int
-trace_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
-             dict_t *xdata)
+trace_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+            dict_t *xdata)
 {
-        trace_conf_t    *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_RMDIR].enabled) {
-                char     string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s path=%s flags=%d",
-                          frame->root->unique,
-                          uuid_utoa (loc->inode->gfid), loc->path,
-                          flags);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_RMDIR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s path=%s flags=%d", frame->root->unique,
+                 uuid_utoa(loc->inode->gfid), loc->path, flags);
 
-                frame->local = loc->inode->gfid;
+        frame->local = loc->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_rmdir_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->rmdir,
-                    loc, flags, xdata);
+    STACK_WIND(frame, trace_rmdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rmdir, loc, flags, xdata);
 
-        return 0;
+    return 0;
 }
 
 int
-trace_symlink (call_frame_t *frame, xlator_t *this, const char *linkpath,
-               loc_t *loc, mode_t umask, dict_t *xdata)
+trace_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath,
+              loc_t *loc, mode_t umask, dict_t *xdata)
 {
-        trace_conf_t     *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_SYMLINK].enabled) {
-                char     string[4096]   =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s linkpath=%s, path=%s"
-                          " umask=0%o", frame->root->unique,
-                          uuid_utoa (loc->inode->gfid), linkpath,
-                          loc->path, umask);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_SYMLINK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s linkpath=%s, path=%s"
+                 " umask=0%o",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), linkpath,
+                 loc->path, umask);
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_symlink_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->symlink,
-                    linkpath, loc, umask, xdata);
+    STACK_WIND(frame, trace_symlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->symlink, linkpath, loc, umask, xdata);
 
-        return 0;
+    return 0;
 }
 
 int
-trace_rename (call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
-              dict_t *xdata)
+trace_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+             dict_t *xdata)
 {
-        char         oldgfid[50] = {0,};
-        char         newgfid[50] = {0,};
-        trace_conf_t *conf       = NULL;
+    char oldgfid[50] = {
+        0,
+    };
+    char newgfid[50] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_RENAME].enabled) {
-                char string[4096] = {0,};
-                if (newloc->inode)
-                        uuid_utoa_r (newloc->inode->gfid, newgfid);
-                else
-                        strcpy (newgfid, "0");
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_RENAME].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (newloc->inode)
+            uuid_utoa_r(newloc->inode->gfid, newgfid);
+        else
+            strcpy(newgfid, "0");
 
-                uuid_utoa_r (oldloc->inode->gfid, oldgfid);
+        uuid_utoa_r(oldloc->inode->gfid, oldgfid);
 
-                snprintf (string, sizeof (string),
-                          "%"PRId64": oldgfid=%s oldpath=%s --> "
-                          "newgfid=%s newpath=%s",
-                          frame->root->unique, oldgfid,
-                          oldloc->path, newgfid, newloc->path);
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": oldgfid=%s oldpath=%s --> "
+                 "newgfid=%s newpath=%s",
+                 frame->root->unique, oldgfid, oldloc->path, newgfid,
+                 newloc->path);
 
-                frame->local = oldloc->inode->gfid;
+        frame->local = oldloc->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_rename_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->rename,
-                    oldloc, newloc, xdata);
+    STACK_WIND(frame, trace_rename_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rename, oldloc, newloc, xdata);
 
-        return 0;
+    return 0;
 }
 
 int
-trace_link (call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
-            dict_t *xdata)
+trace_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+           dict_t *xdata)
 {
-        char         oldgfid[50] = {0,};
-        char         newgfid[50] = {0,};
-        trace_conf_t *conf       = NULL;
+    char oldgfid[50] = {
+        0,
+    };
+    char newgfid[50] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
 
-        conf = this->private;
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_LINK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (newloc->inode)
+            uuid_utoa_r(newloc->inode->gfid, newgfid);
+        else
+            strcpy(newgfid, "0");
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_LINK].enabled) {
-                char string[4096]  = {0,};
-                if (newloc->inode)
-                        uuid_utoa_r (newloc->inode->gfid, newgfid);
-                else
-                        strcpy (newgfid, "0");
+        uuid_utoa_r(oldloc->inode->gfid, oldgfid);
 
-                uuid_utoa_r (oldloc->inode->gfid, oldgfid);
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": oldgfid=%s oldpath=%s --> "
+                 "newgfid=%s newpath=%s",
+                 frame->root->unique, oldgfid, oldloc->path, newgfid,
+                 newloc->path);
 
-                snprintf (string, sizeof (string),
-                          "%"PRId64": oldgfid=%s oldpath=%s --> "
-                          "newgfid=%s newpath=%s", frame->root->unique,
-                          oldgfid, oldloc->path, newgfid,
-                          newloc->path);
+        frame->local = oldloc->inode->gfid;
 
-                frame->local = oldloc->inode->gfid;
+        LOG_ELEMENT(conf, string);
+    }
 
-                LOG_ELEMENT (conf, string);
+out:
+    STACK_WIND(frame, trace_link_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->link, oldloc, newloc, xdata);
+    return 0;
+}
+
+int
+trace_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+              struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    char actime_str[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    char modtime_str[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_SETATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (valid & GF_SET_ATTR_MODE) {
+            snprintf(
+                string, sizeof(string), "%" PRId64 ": gfid=%s path=%s mode=%o)",
+                frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path,
+                st_mode_from_ia(stbuf->ia_prot, stbuf->ia_type));
+
+            LOG_ELEMENT(conf, string);
+            memset(string, 0, sizeof(string));
+        }
+
+        if (valid & (GF_SET_ATTR_UID | GF_SET_ATTR_GID)) {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s path=%s uid=%o,"
+                     " gid=%o",
+                     frame->root->unique, uuid_utoa(loc->inode->gfid),
+                     loc->path, stbuf->ia_uid, stbuf->ia_gid);
+
+            LOG_ELEMENT(conf, string);
+            memset(string, 0, sizeof(string));
+        }
+
+        if (valid & (GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME)) {
+            gf_time_fmt(actime_str, sizeof actime_str, stbuf->ia_atime,
+                        gf_timefmt_bdT);
+
+            gf_time_fmt(modtime_str, sizeof modtime_str, stbuf->ia_mtime,
+                        gf_timefmt_bdT);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s path=%s "
+                     "ia_atime=%s, ia_mtime=%s",
+                     frame->root->unique, uuid_utoa(loc->inode->gfid),
+                     loc->path, actime_str, modtime_str);
+
+            LOG_ELEMENT(conf, string);
+            memset(string, 0, sizeof(string));
         }
+        frame->local = loc->inode->gfid;
+    }
 
 out:
-        STACK_WIND (frame, trace_link_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->link,
-                    oldloc, newloc, xdata);
-        return 0;
+    STACK_WIND(frame, trace_setattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setattr, loc, stbuf, valid, xdata);
+
+    return 0;
 }
 
 int
-trace_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
+trace_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
                struct iatt *stbuf, int32_t valid, dict_t *xdata)
 {
-        uint64_t     ia_time          = 0;
-        char         actime_str[256]  = {0,};
-        char         modtime_str[256] = {0,};
-        trace_conf_t *conf            = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_SETATTR].enabled) {
-                char     string[4096]  =  {0,};
-                if (valid & GF_SET_ATTR_MODE) {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s path=%s mode=%o)",
-                                  frame->root->unique,
-                                  uuid_utoa (loc->inode->gfid),
-                                  loc->path,
-                                  st_mode_from_ia (stbuf->ia_prot,
-                                                   stbuf->ia_type));
-
-                        LOG_ELEMENT (conf, string);
-                        memset (string, 0 , sizeof (string));
-                }
-
-                if (valid & (GF_SET_ATTR_UID | GF_SET_ATTR_GID)) {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s path=%s uid=%o,"
-                                  " gid=%o", frame->root->unique,
-                                  uuid_utoa (loc->inode->gfid),
-                                  loc->path, stbuf->ia_uid,
-                                  stbuf->ia_gid);
-
-                        LOG_ELEMENT (conf, string);
-                        memset (string, 0 , sizeof (string));
-                }
-
-                if (valid & (GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME)) {
-                        ia_time = stbuf->ia_atime;
-                        strftime (actime_str, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        ia_time = stbuf->ia_mtime;
-                        strftime (modtime_str, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s path=%s "
-                                  "ia_atime=%s, ia_mtime=%s",
-                                  frame->root->unique,
-                                  uuid_utoa (loc->inode->gfid),
-                                  loc->path, actime_str, modtime_str);
-
-                        LOG_ELEMENT (conf, string);
-                        memset (string, 0 , sizeof (string));
-                }
-                frame->local = loc->inode->gfid;
+    char actime_str[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    char modtime_str[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FSETATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (valid & GF_SET_ATTR_MODE) {
+            snprintf(string, sizeof(string),
+                     "%" PRId64 ": gfid=%s fd=%p, mode=%o", frame->root->unique,
+                     uuid_utoa(fd->inode->gfid), fd,
+                     st_mode_from_ia(stbuf->ia_prot, stbuf->ia_type));
+
+            LOG_ELEMENT(conf, string);
+            memset(string, 0, sizeof(string));
         }
 
-out:
-        STACK_WIND (frame, trace_setattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->setattr,
-                    loc, stbuf, valid, xdata);
-
-        return 0;
-}
-
-int
-trace_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                struct iatt *stbuf, int32_t valid, dict_t *xdata)
-{
-        uint64_t       ia_time          = 0;
-        char           actime_str[256]  = {0,};
-        char           modtime_str[256] = {0,};
-        trace_conf_t   *conf            = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_FSETATTR].enabled) {
-                char     string[4096]  =  {0,};
-                if (valid & GF_SET_ATTR_MODE) {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s fd=%p, mode=%o",
-                                  frame->root->unique,
-                                  uuid_utoa (fd->inode->gfid), fd,
-                                  st_mode_from_ia (stbuf->ia_prot,
-                                                   stbuf->ia_type));
-
-                        LOG_ELEMENT (conf, string);
-                        memset (string, 0, sizeof (string));
-                }
-
-                if (valid & (GF_SET_ATTR_UID | GF_SET_ATTR_GID)) {
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s fd=%p, uid=%o, "
-                                  "gid=%o", frame->root->unique,
-                                  uuid_utoa (fd->inode->gfid),
-                                  fd, stbuf->ia_uid, stbuf->ia_gid);
-
-                        LOG_ELEMENT (conf, string);
-                        memset (string, 0, sizeof (string));
-                }
-
-                if (valid & (GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME)) {
-                        ia_time = stbuf->ia_atime;
-                        strftime (actime_str, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        ia_time = stbuf->ia_mtime;
-                        strftime (modtime_str, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        snprintf (string, sizeof (string),
-                                  "%"PRId64": gfid=%s fd=%p "
-                                  "ia_atime=%s, ia_mtime=%s",
-                                  frame->root->unique,
-                                  uuid_utoa (fd->inode->gfid),
-                                  fd, actime_str, modtime_str);
-
-                        LOG_ELEMENT (conf, string);
-                        memset (string, 0, sizeof (string));
-                }
-                frame->local = fd->inode->gfid;
+        if (valid & (GF_SET_ATTR_UID | GF_SET_ATTR_GID)) {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s fd=%p, uid=%o, "
+                     "gid=%o",
+                     frame->root->unique, uuid_utoa(fd->inode->gfid), fd,
+                     stbuf->ia_uid, stbuf->ia_gid);
+
+            LOG_ELEMENT(conf, string);
+            memset(string, 0, sizeof(string));
+        }
+
+        if (valid & (GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME)) {
+            gf_time_fmt(actime_str, sizeof actime_str, stbuf->ia_atime,
+                        gf_timefmt_bdT);
+
+            gf_time_fmt(modtime_str, sizeof modtime_str, stbuf->ia_mtime,
+                        gf_timefmt_bdT);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s fd=%p "
+                     "ia_atime=%s, ia_mtime=%s",
+                     frame->root->unique, uuid_utoa(fd->inode->gfid), fd,
+                     actime_str, modtime_str);
+
+            LOG_ELEMENT(conf, string);
+            memset(string, 0, sizeof(string));
         }
+        frame->local = fd->inode->gfid;
+    }
 
 out:
-        STACK_WIND (frame, trace_fsetattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fsetattr,
-                    fd, stbuf, valid, xdata);
+    STACK_WIND(frame, trace_fsetattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetattr, fd, stbuf, valid, xdata);
 
-        return 0;
+    return 0;
+}
+
+static int
+trace_seek_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, off_t offset, dict_t *xdata)
+{
+    trace_conf_t *conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_SEEK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s op_ret=%d op_errno=%d, "
+                 "offset=%" PRId64 "",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret, op_errno,
+                 offset);
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(seek, frame, op_ret, op_errno, offset, xdata);
+    return 0;
+}
+
+static int
+trace_seek(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+           gf_seek_what_t what, dict_t *xdata)
+{
+    trace_conf_t *conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_SEEK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s fd=%p "
+                 "offset=%" PRId64 " what=%d",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), fd, offset,
+                 what);
+
+        frame->local = fd->inode->gfid;
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    STACK_WIND(frame, trace_seek_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->seek, fd, offset, what, xdata);
+
+    return 0;
 }
 
 int
-trace_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                off_t offset, dict_t *xdata)
+trace_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+               dict_t *xdata)
 {
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_TRUNCATE].enabled) {
-                char     string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s path=%s, "
-                          "offset=%"PRId64"", frame->root->unique,
-                          uuid_utoa (loc->inode->gfid), loc->path,
-                          offset);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_TRUNCATE].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s path=%s, "
+                 "offset=%" PRId64 "",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path,
+                 offset);
 
-                frame->local = loc->inode->gfid;
+        frame->local = loc->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_truncate_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->truncate,
-                    loc, offset, xdata);
+    STACK_WIND(frame, trace_truncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
 
-        return 0;
+    return 0;
 }
 
 int
-trace_open (call_frame_t *frame, xlator_t *this, loc_t *loc,
-            int32_t flags, fd_t *fd, dict_t *xdata)
+trace_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+           fd_t *fd, dict_t *xdata)
 {
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_OPEN].enabled) {
-                char      string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s path=%s flags=%d fd=%p",
-                          frame->root->unique,
-                          uuid_utoa (loc->inode->gfid), loc->path,
-                          flags, fd);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_OPEN].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s path=%s flags=%d fd=%p",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path,
+                 flags, fd);
 
-                frame->local = loc->inode->gfid;
+        frame->local = loc->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_open_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->open,
-                    loc, flags, fd, xdata);
-        return 0;
+    STACK_WIND(frame, trace_open_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->open, loc, flags, fd, xdata);
+    return 0;
 }
 
 int
-trace_create (call_frame_t *frame, xlator_t *this, loc_t *loc,
-              int32_t flags, mode_t mode, mode_t umask, fd_t *fd,
-              dict_t *xdata)
+trace_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+             mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
 {
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_CREATE].enabled) {
-                char    string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s path=%s, fd=%p, "
-                          "flags=0%o mode=0%o umask=0%o",
-                          frame->root->unique,
-                          uuid_utoa (loc->inode->gfid), loc->path,
-                          fd, flags, mode, umask);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_CREATE].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s path=%s, fd=%p, "
+                 "flags=0%o mode=0%o umask=0%o",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path,
+                 fd, flags, mode, umask);
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_create_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->create,
-                    loc, flags, mode, umask, fd, xdata);
-        return 0;
+    STACK_WIND(frame, trace_create_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->create, loc, flags, mode, umask, fd,
+               xdata);
+    return 0;
 }
 
 int
-trace_readv (call_frame_t *frame, xlator_t *this, fd_t *fd,
-             size_t size, off_t offset, uint32_t flags, dict_t *xdata)
+trace_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t offset, uint32_t flags, dict_t *xdata)
 {
-        trace_conf_t  *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_READ].enabled) {
-                char     string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s fd=%p, size=%"
-                          GF_PRI_SIZET"offset=%"PRId64" flags=0%x)",
-                          frame->root->unique,
-                          uuid_utoa (fd->inode->gfid), fd, size,
-                          offset, flags);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_READ].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s fd=%p, size=%" GF_PRI_SIZET
+                 "offset=%" PRId64 " flags=0%x)",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), fd, size,
+                 offset, flags);
 
-                frame->local = fd->inode->gfid;
+        frame->local = fd->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_readv_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->readv,
-                    fd, size, offset, flags, xdata);
-        return 0;
+    STACK_WIND(frame, trace_readv_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readv, fd, size, offset, flags, xdata);
+    return 0;
 }
 
 int
-trace_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
-              struct iovec *vector, int32_t count,
-              off_t offset, uint32_t flags, struct iobref *iobref, dict_t *xdata)
+trace_writev(call_frame_t *frame, xlator_t *this, fd_t *fd,
+             struct iovec *vector, int32_t count, off_t offset, uint32_t flags,
+             struct iobref *iobref, dict_t *xdata)
 {
-        trace_conf_t    *conf = NULL;
+    trace_conf_t *conf = NULL;
+    int i = 0;
+    size_t total_size = 0;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_WRITE].enabled) {
-                char     string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s fd=%p, count=%d, "
-                          " offset=%"PRId64" flags=0%x)",
-                          frame->root->unique,
-                          uuid_utoa (fd->inode->gfid), fd, count,
-                          offset, flags);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_WRITE].enabled) {
+        char string[4096] = {
+            0,
+        };
+        for (i = 0; i < count; i++)
+            total_size += vector[i].iov_len;
 
-                frame->local = fd->inode->gfid;
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s fd=%p, count=%d, "
+                 " offset=%" PRId64 " flags=0%x write_size=%zu",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), fd, count,
+                 offset, flags, total_size);
 
-                LOG_ELEMENT (conf, string);
-        }
+        frame->local = fd->inode->gfid;
+
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_writev_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->writev,
-                    fd, vector, count, offset, flags, iobref, xdata);
-        return 0;
+    STACK_WIND(frame, trace_writev_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->writev, fd, vector, count, offset,
+               flags, iobref, xdata);
+    return 0;
 }
 
 int
-trace_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+trace_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-        trace_conf_t  *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_STATFS].enabled) {
-                char  string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s path=%s",
-                          frame->root->unique, (loc->inode)?
-                          uuid_utoa (loc->inode->gfid):"0", loc->path);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_STATFS].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s path=%s",
+                 frame->root->unique,
+                 (loc->inode) ? uuid_utoa(loc->inode->gfid) : "0", loc->path);
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_statfs_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->statfs,
-                    loc, xdata);
-        return 0;
+    STACK_WIND(frame, trace_statfs_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->statfs, loc, xdata);
+    return 0;
 }
 
 int
-trace_flush (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+trace_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
-        trace_conf_t  *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_FLUSH].enabled) {
-                char    string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s fd=%p",
-                          frame->root->unique,
-                          uuid_utoa (fd->inode->gfid), fd);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FLUSH].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s fd=%p",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), fd);
 
-                frame->local = fd->inode->gfid;
+        frame->local = fd->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_flush_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->flush,
-                    fd, xdata);
-        return 0;
+    STACK_WIND(frame, trace_flush_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->flush, fd, xdata);
+    return 0;
 }
 
 int
-trace_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
-             dict_t *xdata)
+trace_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+            dict_t *xdata)
 {
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_FSYNC].enabled) {
-                char     string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s flags=%d fd=%p",
-                          frame->root->unique,
-                          uuid_utoa (fd->inode->gfid), flags, fd);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FSYNC].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s flags=%d fd=%p",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), flags, fd);
 
-                frame->local = fd->inode->gfid;
+        frame->local = fd->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_fsync_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fsync,
-                    fd, flags, xdata);
-        return 0;
+    STACK_WIND(frame, trace_fsync_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsync, fd, flags, xdata);
+    return 0;
 }
 
 int
-trace_setxattr (call_frame_t *frame, xlator_t *this,
-                loc_t *loc, dict_t *dict, int32_t flags, dict_t *xdata)
+trace_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+               int32_t flags, dict_t *xdata)
 {
-        trace_conf_t    *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_SETXATTR].enabled) {
-                char     string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s path=%s flags=%d",
-                          frame->root->unique,
-                          uuid_utoa (loc->inode->gfid), loc->path,
-                          flags);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_SETXATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s path=%s flags=%d", frame->root->unique,
+                 uuid_utoa(loc->inode->gfid), loc->path, flags);
 
-                frame->local = loc->inode->gfid;
+        frame->local = loc->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_setxattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->setxattr,
-                    loc, dict, flags, xdata);
-        return 0;
+    STACK_WIND(frame, trace_setxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setxattr, loc, dict, flags, xdata);
+    return 0;
 }
 
 int
-trace_getxattr (call_frame_t *frame, xlator_t *this,
-                loc_t *loc, const char *name, dict_t *xdata)
+trace_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               const char *name, dict_t *xdata)
 {
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_GETXATTR].enabled) {
-                char    string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s path=%s name=%s",
-                          frame->root->unique,
-                          uuid_utoa (loc->inode->gfid), loc->path,
-                          name);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_GETXATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s path=%s name=%s",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path,
+                 name);
 
-                frame->local = loc->inode->gfid;
+        frame->local = loc->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_getxattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->getxattr,
-                    loc, name, xdata);
-        return 0;
+    STACK_WIND(frame, trace_getxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->getxattr, loc, name, xdata);
+    return 0;
 }
 
 int
-trace_removexattr (call_frame_t *frame, xlator_t *this,
-                   loc_t *loc, const char *name, dict_t *xdata)
+trace_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                  const char *name, dict_t *xdata)
 {
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_REMOVEXATTR].enabled) {
-                char   string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s path=%s name=%s",
-                          frame->root->unique,
-                          uuid_utoa (loc->inode->gfid), loc->path,
-                          name);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_REMOVEXATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s path=%s name=%s",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path,
+                 name);
 
-                frame->local = loc->inode->gfid;
+        frame->local = loc->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_removexattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->removexattr,
-                    loc, name, xdata);
+    STACK_WIND(frame, trace_removexattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->removexattr, loc, name, xdata);
 
-        return 0;
+    return 0;
 }
 
 int
-trace_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
-               dict_t *xdata)
+trace_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+              dict_t *xdata)
 {
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_OPENDIR].enabled) {
-                char   string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s path=%s fd=%p",
-                          frame->root->unique,
-                          uuid_utoa (loc->inode->gfid), loc->path, fd);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_OPENDIR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s path=%s fd=%p",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path,
+                 fd);
 
-                frame->local = loc->inode->gfid;
+        frame->local = loc->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_opendir_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->opendir,
-                    loc, fd, xdata);
-        return 0;
+    STACK_WIND(frame, trace_opendir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->opendir, loc, fd, xdata);
+    return 0;
 }
 
 int
-trace_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-                off_t offset, dict_t *dict)
+trace_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+               off_t offset, dict_t *dict)
 {
-        trace_conf_t *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_READDIRP].enabled) {
-                char   string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s fd=%p, size=%"GF_PRI_SIZET
-                          ", offset=%"PRId64" dict=%p",
-                          frame->root->unique,
-                          uuid_utoa (fd->inode->gfid), fd, size,
-                          offset, dict);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_READDIRP].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s fd=%p, size=%" GF_PRI_SIZET
+                 ", offset=%" PRId64 " dict=%p",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), fd, size,
+                 offset, dict);
 
-                frame->local = fd->inode->gfid;
+        frame->local = fd->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_readdirp_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->readdirp,
-                    fd, size, offset, dict);
+    STACK_WIND(frame, trace_readdirp_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readdirp, fd, size, offset, dict);
 
-        return 0;
+    return 0;
 }
 
 int
-trace_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd,
-               size_t size, off_t offset, dict_t *xdata)
+trace_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+              off_t offset, dict_t *xdata)
 {
-        trace_conf_t  *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_READDIR].enabled) {
-                char   string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s fd=%p, size=%"GF_PRI_SIZET
-                          ", offset=%"PRId64,
-                          frame->root->unique,
-                          uuid_utoa (fd->inode->gfid), fd, size,
-                          offset);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_READDIR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s fd=%p, size=%" GF_PRI_SIZET
+                 ", offset=%" PRId64,
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), fd, size,
+                 offset);
 
-                frame->local = fd->inode->gfid;
+        frame->local = fd->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_readdir_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->readdir,
-                    fd, size, offset, xdata);
+    STACK_WIND(frame, trace_readdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readdir, fd, size, offset, xdata);
 
-        return 0;
+    return 0;
 }
 
 int
-trace_fsyncdir (call_frame_t *frame, xlator_t *this,
-                fd_t *fd, int32_t datasync, dict_t *xdata)
+trace_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+               dict_t *xdata)
 {
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_FSYNCDIR].enabled) {
-                char     string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s datasync=%d fd=%p",
-                          frame->root->unique,
-                          uuid_utoa (fd->inode->gfid), datasync, fd);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FSYNCDIR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s datasync=%d fd=%p", frame->root->unique,
+                 uuid_utoa(fd->inode->gfid), datasync, fd);
 
-                frame->local = fd->inode->gfid;
+        frame->local = fd->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_fsyncdir_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fsyncdir,
-                    fd, datasync, xdata);
-        return 0;
+    STACK_WIND(frame, trace_fsyncdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsyncdir, fd, datasync, xdata);
+    return 0;
 }
 
 int
-trace_access (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
-              dict_t *xdata)
+trace_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+             dict_t *xdata)
 {
-        trace_conf_t  *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_ACCESS].enabled) {
-                char     string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s path=%s mask=0%o",
-                          frame->root->unique,
-                          uuid_utoa (loc->inode->gfid),
-                          loc->path, mask);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_ACCESS].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s path=%s mask=0%o", frame->root->unique,
+                 uuid_utoa(loc->inode->gfid), loc->path, mask);
 
-                frame->local = loc->inode->gfid;
+        frame->local = loc->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_access_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->access,
-                    loc, mask, xdata);
-        return 0;
+    STACK_WIND(frame, trace_access_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->access, loc, mask, xdata);
+    return 0;
 }
 
 int32_t
-trace_rchecksum (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
-                 int32_t len, dict_t *xdata)
+trace_rchecksum(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                int32_t len, dict_t *xdata)
 {
+    trace_conf_t *conf = NULL;
 
-        trace_conf_t *conf = NULL;
+    conf = this->private;
 
-        conf = this->private;
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_RCHECKSUM].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s offset=%" PRId64 "len=%u fd=%p",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), offset, len,
+                 fd);
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_RCHECKSUM].enabled) {
-                char    string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s offset=%"PRId64
-                          "len=%u fd=%p", frame->root->unique,
-                          uuid_utoa (fd->inode->gfid), offset, len, fd);
+        frame->local = fd->inode->gfid;
 
-                frame->local = fd->inode->gfid;
-
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_rchecksum_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->rchecksum,
-                    fd, offset, len, xdata);
-
-        return 0;
+    STACK_WIND(frame, trace_rchecksum_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rchecksum, fd, offset, len, xdata);
 
+    return 0;
 }
 
 int32_t
-trace_fentrylk (call_frame_t *frame, xlator_t *this, const char *volume,
-                fd_t *fd, const char *basename, entrylk_cmd cmd,
-                entrylk_type type, dict_t *xdata)
-{
-        trace_conf_t   *conf = NULL;
-
-        conf = this->private;
-
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_FENTRYLK].enabled) {
-                char      string[4096]   =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s volume=%s, (fd=%p "
-                          "basename=%s, cmd=%s, type=%s)",
-                          frame->root->unique,
-                          uuid_utoa (fd->inode->gfid), volume, fd,
-                          basename,
-                          ((cmd == ENTRYLK_LOCK) ? "ENTRYLK_LOCK" :
-                           "ENTRYLK_UNLOCK"),
-                          ((type == ENTRYLK_RDLCK) ? "ENTRYLK_RDLCK" :
-                           "ENTRYLK_WRLCK"));
-
-                frame->local = fd->inode->gfid;
-
-                LOG_ELEMENT (conf, string);
-        }
+trace_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume,
+               fd_t *fd, const char *basename, entrylk_cmd cmd,
+               entrylk_type type, dict_t *xdata)
+{
+    trace_conf_t *conf = NULL;
 
-out:
-        STACK_WIND (frame, trace_fentrylk_cbk,
-                    FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->fentrylk,
-                    volume, fd, basename, cmd, type, xdata);
-        return 0;
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FENTRYLK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s volume=%s, (fd=%p "
+                 "basename=%s, cmd=%s, type=%s)",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), volume, fd,
+                 basename,
+                 ((cmd == ENTRYLK_LOCK) ? "ENTRYLK_LOCK" : "ENTRYLK_UNLOCK"),
+                 ((type == ENTRYLK_RDLCK) ? "ENTRYLK_RDLCK" : "ENTRYLK_WRLCK"));
+
+        frame->local = fd->inode->gfid;
 
+        LOG_ELEMENT(conf, string);
+    }
+
+out:
+    STACK_WIND(frame, trace_fentrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fentrylk, volume, fd, basename, cmd,
+               type, xdata);
+    return 0;
 }
 
 int32_t
-trace_fgetxattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                 const char *name, dict_t *xdata)
+trace_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+                dict_t *xdata)
 {
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-                goto out;
-        if (trace_fop_names[GF_FOP_FGETXATTR].enabled) {
-                char     string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s fd=%p name=%s",
-                          frame->root->unique,
-                          uuid_utoa (fd->inode->gfid), fd, name);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FGETXATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s fd=%p name=%s",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), fd, name);
 
-                frame->local = fd->inode->gfid;
+        frame->local = fd->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_fgetxattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fgetxattr,
-                    fd, name, xdata);
-        return 0;
+    STACK_WIND(frame, trace_fgetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fgetxattr, fd, name, xdata);
+    return 0;
 }
 
 int32_t
-trace_fsetxattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                 dict_t *dict, int32_t flags, dict_t *xdata)
+trace_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+                int32_t flags, dict_t *xdata)
 {
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_FSETXATTR].enabled) {
-                char   string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s fd=%p flags=%d",
-                          frame->root->unique,
-                          uuid_utoa (fd->inode->gfid), fd, flags);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FSETXATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s fd=%p flags=%d",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), fd, flags);
 
-                frame->local = fd->inode->gfid;
+        frame->local = fd->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_fsetxattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fsetxattr,
-                    fd, dict, flags, xdata);
-        return 0;
+    STACK_WIND(frame, trace_fsetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags, xdata);
+    return 0;
 }
 
 int
-trace_ftruncate (call_frame_t *frame, xlator_t *this,
-                 fd_t *fd, off_t offset, dict_t *xdata)
+trace_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                dict_t *xdata)
 {
-        trace_conf_t  *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_FTRUNCATE].enabled) {
-                char    string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s offset=%"PRId64" fd=%p",
-                          frame->root->unique,
-                          uuid_utoa (fd->inode->gfid), offset, fd);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FTRUNCATE].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s offset=%" PRId64 " fd=%p",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), offset, fd);
 
-                frame->local = fd->inode->gfid;
+        frame->local = fd->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_ftruncate_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->ftruncate,
-                    fd, offset, xdata);
+    STACK_WIND(frame, trace_ftruncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
 
-        return 0;
+    return 0;
 }
 
 int
-trace_fstat (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+trace_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_FSTAT].enabled) {
-                char   string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s fd=%p",
-                          frame->root->unique,
-                          uuid_utoa (fd->inode->gfid), fd);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FSTAT].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s fd=%p",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), fd);
 
-                frame->local = fd->inode->gfid;
+        frame->local = fd->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_fstat_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fstat,
-                    fd, xdata);
-        return 0;
+    STACK_WIND(frame, trace_fstat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fstat, fd, xdata);
+    return 0;
 }
 
 int
-trace_lk (call_frame_t *frame, xlator_t *this, fd_t *fd,
-          int32_t cmd, struct gf_flock *lock, dict_t *xdata)
+trace_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
+         struct gf_flock *lock, dict_t *xdata)
 {
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_LK].enabled) {
-                char     string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "%"PRId64": gfid=%s fd=%p, cmd=%d, "
-                          "lock {l_type=%d, "
-                          "l_whence=%d, l_start=%"PRId64", "
-                          "l_len=%"PRId64", l_pid=%u})",
-                          frame->root->unique,
-                          uuid_utoa (fd->inode->gfid), fd, cmd,
-                          lock->l_type, lock->l_whence,
-                          lock->l_start, lock->l_len, lock->l_pid);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_LK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s fd=%p, cmd=%d, "
+                 "lock {l_type=%d, "
+                 "l_whence=%d, l_start=%" PRId64
+                 ", "
+                 "l_len=%" PRId64 ", l_pid=%u})",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), fd, cmd,
+                 lock->l_type, lock->l_whence, lock->l_start, lock->l_len,
+                 lock->l_pid);
 
-                frame->local = fd->inode->gfid;
+        frame->local = fd->inode->gfid;
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        STACK_WIND (frame, trace_lk_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->lk,
-                    fd, cmd, lock, xdata);
-        return 0;
+    STACK_WIND(frame, trace_lk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lk, fd, cmd, lock, xdata);
+    return 0;
 }
 
 int32_t
-trace_forget (xlator_t *this, inode_t *inode)
+trace_forget(xlator_t *this, inode_t *inode)
 {
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
-        /* If user want to understand when a lookup happens,
-           he should know about 'forget' too */
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_LOOKUP].enabled) {
-                char   string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "gfid=%s", uuid_utoa (inode->gfid));
+    conf = this->private;
+    /* If user want to understand when a lookup happens,
+       he should know about 'forget' too */
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_LOOKUP].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "gfid=%s", uuid_utoa(inode->gfid));
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        return 0;
+    return 0;
 }
 
 int32_t
-trace_releasedir (xlator_t *this, fd_t *fd)
+trace_releasedir(xlator_t *this, fd_t *fd)
 {
-        trace_conf_t  *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_OPENDIR].enabled) {
-                char   string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "gfid=%s fd=%p",
-                          uuid_utoa (fd->inode->gfid), fd);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_OPENDIR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "gfid=%s fd=%p",
+                 uuid_utoa(fd->inode->gfid), fd);
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        return 0;
+    return 0;
 }
 
 int32_t
-trace_release (xlator_t *this, fd_t *fd)
+trace_release(xlator_t *this, fd_t *fd)
 {
-        trace_conf_t   *conf = NULL;
+    trace_conf_t *conf = NULL;
 
-        conf = this->private;
+    conf = this->private;
 
-        if (!conf->log_file && !conf->log_history)
-		goto out;
-        if (trace_fop_names[GF_FOP_OPEN].enabled ||
-            trace_fop_names[GF_FOP_CREATE].enabled) {
-                char   string[4096]  =  {0,};
-                snprintf (string, sizeof (string),
-                          "gfid=%s fd=%p",
-                          uuid_utoa (fd->inode->gfid), fd);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_OPEN].enabled ||
+        trace_fop_names[GF_FOP_CREATE].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "gfid=%s fd=%p",
+                 uuid_utoa(fd->inode->gfid), fd);
 
-                LOG_ELEMENT (conf, string);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
 out:
-        return 0;
+    return 0;
 }
 
-
 void
-enable_all_calls (int enabled)
+enable_all_calls(int enabled)
 {
-        int i;
+    int i;
 
-        for (i = 0; i < GF_FOP_MAXVALUE; i++)
-                trace_fop_names[i].enabled = enabled;
+    for (i = 0; i < GF_FOP_MAXVALUE; i++)
+        trace_fop_names[i].enabled = enabled;
 }
 
 void
-enable_call (const char *name, int enabled)
+enable_call(const char *name, int enabled)
 {
-        int i;
-        for (i = 0; i < GF_FOP_MAXVALUE; i++)
-                if (!strcasecmp(trace_fop_names[i].name, name))
-                        trace_fop_names[i].enabled = enabled;
+    int i;
+    for (i = 0; i < GF_FOP_MAXVALUE; i++)
+        if (!strcasecmp(trace_fop_names[i].name, name))
+            trace_fop_names[i].enabled = enabled;
 }
 
-
 /*
   include = 1 for "include-ops"
   = 0 for "exclude-ops"
 */
 void
-process_call_list (const char *list, int include)
+process_call_list(const char *list, int include)
 {
-        enable_all_calls (include ? 0 : 1);
+    enable_all_calls(include ? 0 : 1);
 
-        char *call = strsep ((char **)&list, ",");
+    char *call = strsep((char **)&list, ",");
 
-        while (call) {
-                enable_call (call, include);
-                call = strsep ((char **)&list, ",");
-        }
+    while (call) {
+        enable_call(call, include);
+        call = strsep((char **)&list, ",");
+    }
 }
 
 int32_t
-trace_dump_history (xlator_t *this)
-{
-        int ret = -1;
-        char key_prefix[GF_DUMP_MAX_BUF_LEN] = {0,};
-        trace_conf_t *conf = NULL;
-
-        GF_VALIDATE_OR_GOTO ("trace", this, out);
-        GF_VALIDATE_OR_GOTO (this->name, this->history, out);
-
-        conf = this->private;
-        // Is it ok to return silently if log-history option his off?
-        if (conf && conf->log_history == _gf_true) {
-                gf_proc_dump_build_key (key_prefix, "xlator.debug.trace",
-                                        "history");
-                gf_proc_dump_add_section (key_prefix);
-                eh_dump (this->history, NULL, dump_history_trace);
-        }
-        ret = 0;
+trace_dump_history(xlator_t *this)
+{
+    int ret = -1;
+    char key_prefix[GF_DUMP_MAX_BUF_LEN] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    GF_VALIDATE_OR_GOTO("trace", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->history, out);
+
+    conf = this->private;
+    // Is it ok to return silently if log-history option his off?
+    if (conf && conf->log_history == _gf_true) {
+        gf_proc_dump_build_key(key_prefix, "xlator.debug.trace", "history");
+        gf_proc_dump_add_section("%s", key_prefix);
+        eh_dump(this->history, NULL, dump_history_trace);
+    }
+    ret = 0;
 
 out:
-        return ret;
+    return ret;
 }
 
 int32_t
-mem_acct_init (xlator_t *this)
+mem_acct_init(xlator_t *this)
 {
-        int     ret = -1;
-
-        if (!this)
-                return ret;
+    int ret = -1;
 
-        ret = xlator_mem_acct_init (this, gf_trace_mt_end + 1);
+    if (!this)
+        return ret;
 
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
-                        " failed");
-                return ret;
-        }
+    ret = xlator_mem_acct_init(this, gf_trace_mt_end + 1);
 
+    if (ret != 0) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Memory accounting init"
+               " failed");
         return ret;
+    }
+
+    return ret;
 }
 
 int
-reconfigure (xlator_t *this, dict_t *options)
+reconfigure(xlator_t *this, dict_t *options)
 {
-        int32_t       ret            = -1;
-        trace_conf_t    *conf           = NULL;
-        char            *includes = NULL, *excludes = NULL;
+    int32_t ret = -1;
+    trace_conf_t *conf = NULL;
+    char *includes = NULL, *excludes = NULL;
 
-        GF_VALIDATE_OR_GOTO ("quick-read", this, out);
-        GF_VALIDATE_OR_GOTO (this->name, this->private, out);
-        GF_VALIDATE_OR_GOTO (this->name, options, out);
+    GF_VALIDATE_OR_GOTO("quick-read", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+    GF_VALIDATE_OR_GOTO(this->name, options, out);
 
-        conf = this->private;
+    conf = this->private;
 
-        includes = data_to_str (dict_get (options, "include-ops"));
-        excludes = data_to_str (dict_get (options, "exclude-ops"));
+    includes = data_to_str(dict_get(options, "include-ops"));
+    excludes = data_to_str(dict_get(options, "exclude-ops"));
 
-        {
-                int i;
-                for (i = 0; i < GF_FOP_MAXVALUE; i++) {
-                        if (gf_fop_list[i])
-                                strncpy (trace_fop_names[i].name,
-                                         gf_fop_list[i],
-                                         strlen (gf_fop_list[i]));
-                        else
-                                strncpy (trace_fop_names[i].name, ":O",
-                                         strlen (":O"));
-                        trace_fop_names[i].enabled = 1;
-                }
+    {
+        int i;
+        for (i = 0; i < GF_FOP_MAXVALUE; i++) {
+            if (gf_fop_list[i])
+                strncpy(trace_fop_names[i].name, gf_fop_list[i],
+                        sizeof(trace_fop_names[i].name));
+            else
+                strncpy(trace_fop_names[i].name, ":0",
+                        sizeof(trace_fop_names[i].name));
+            trace_fop_names[i].enabled = 1;
+            trace_fop_names[i].name[sizeof(trace_fop_names[i].name) - 1] = 0;
         }
+    }
 
-        if (includes && excludes) {
-                gf_log (this->name,
-                        GF_LOG_ERROR,
-                        "must specify only one of 'include-ops' and "
-                        "'exclude-ops'");
-                goto out;
-        }
+    if (includes && excludes) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "must specify only one of 'include-ops' and "
+               "'exclude-ops'");
+        goto out;
+    }
 
-        if (includes)
-                process_call_list (includes, 1);
-        if (excludes)
-                process_call_list (excludes, 0);
+    if (includes)
+        process_call_list(includes, 1);
+    if (excludes)
+        process_call_list(excludes, 0);
 
-        /* Should resizing of the event-history be allowed in reconfigure?
-         * for which a new event_history might have to be allocated and the
-         * older history has to be freed.
-         */
-        GF_OPTION_RECONF ("log-file", conf->log_file, options, bool, out);
+    /* Should resizing of the event-history be allowed in reconfigure?
+     * for which a new event_history might have to be allocated and the
+     * older history has to be freed.
+     */
+    GF_OPTION_RECONF("log-file", conf->log_file, options, bool, out);
 
-        GF_OPTION_RECONF ("log-history", conf->log_history, options, bool, out);
+    GF_OPTION_RECONF("log-history", conf->log_history, options, bool, out);
 
-        ret = 0;
+    ret = 0;
 
 out:
-        return ret;
+    return ret;
 }
 
 int32_t
-init (xlator_t *this)
-{
-        dict_t *options = NULL;
-        char *includes = NULL, *excludes = NULL;
-        char *forced_loglevel = NULL;
-        eh_t *history = NULL;
-        int  ret = -1;
-        size_t  history_size = TRACE_DEFAULT_HISTORY_SIZE;
-        trace_conf_t    *conf = NULL;
-
-        if (!this)
-                return -1;
-
-        if (!this->children || this->children->next) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "trace translator requires one subvolume");
-                return -1;
-        }
-        if (!this->parents) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "dangling volume. check volfile ");
-        }
-
-        conf = GF_CALLOC (1, sizeof (trace_conf_t), gf_trace_mt_trace_conf_t);
-        if (!conf) {
-                gf_log (this->name, GF_LOG_ERROR, "cannot allocate "
-                        "xl->private");
-                return -1;
-        }
-
-        options = this->options;
-        includes = data_to_str (dict_get (options, "include-ops"));
-        excludes = data_to_str (dict_get (options, "exclude-ops"));
-
-        {
-                int i;
-                for (i = 0; i < GF_FOP_MAXVALUE; i++) {
-                        if (gf_fop_list[i])
-                                strncpy (trace_fop_names[i].name,
-                                         gf_fop_list[i],
-                                         strlen (gf_fop_list[i]));
-                        else
-                                strncpy (trace_fop_names[i].name, ":O",
-                                         strlen (":O"));
-                        trace_fop_names[i].enabled = 1;
-                }
-        }
-
-        if (includes && excludes) {
-                gf_log (this->name,
-                        GF_LOG_ERROR,
-                        "must specify only one of 'include-ops' and "
-                        "'exclude-ops'");
-                return -1;
-        }
-
-        if (includes)
-                process_call_list (includes, 1);
-        if (excludes)
-                process_call_list (excludes, 0);
-
-
-        GF_OPTION_INIT ("history-size", conf->history_size, size, out);
-
-        gf_log (this->name, GF_LOG_INFO, "history size %"GF_PRI_SIZET,
-                history_size);
-
-        GF_OPTION_INIT ("log-file", conf->log_file, bool, out);
-
-        gf_log (this->name, GF_LOG_INFO, "logging to file %s",
-                (conf->log_file == _gf_true)?"enabled":"disabled");
-
-        GF_OPTION_INIT ("log-history", conf->log_history, bool, out);
-
-        gf_log (this->name, GF_LOG_DEBUG, "logging to history %s",
-                (conf->log_history == _gf_true)?"enabled":"disabled");
-
-        history = eh_new (history_size, _gf_false);
-        if (!history) {
-                gf_log (this->name, GF_LOG_ERROR, "event history cannot be "
-                        "initialized");
-                return -1;
-        }
-
-        this->history = history;
-
-        conf->trace_log_level = GF_LOG_INFO;
-
-        if (dict_get (options, "force-log-level")) {
-                forced_loglevel = data_to_str (dict_get (options,
-                                                         "force-log-level"));
-                if (!forced_loglevel)
-                        goto setloglevel;
-
-                if (strcmp (forced_loglevel, "INFO") == 0)
-                        conf->trace_log_level = GF_LOG_INFO;
-                else if (strcmp (forced_loglevel, "TRACE") == 0)
-                        conf->trace_log_level = GF_LOG_TRACE;
-                else if (strcmp (forced_loglevel, "ERROR") == 0)
-                        conf->trace_log_level = GF_LOG_ERROR;
-                else if (strcmp (forced_loglevel, "DEBUG") == 0)
-                        conf->trace_log_level = GF_LOG_DEBUG;
-                else if (strcmp (forced_loglevel, "WARNING") == 0)
-                        conf->trace_log_level = GF_LOG_WARNING;
-                else if (strcmp (forced_loglevel, "CRITICAL") == 0)
-                        conf->trace_log_level = GF_LOG_CRITICAL;
-                else if (strcmp (forced_loglevel, "NONE") == 0)
-                        conf->trace_log_level = GF_LOG_NONE;
-        }
+init(xlator_t *this)
+{
+    dict_t *options = NULL;
+    char *includes = NULL, *excludes = NULL;
+    char *forced_loglevel = NULL;
+    eh_t *history = NULL;
+    int ret = -1;
+    uint64_t history_size = TRACE_DEFAULT_HISTORY_SIZE;
+    trace_conf_t *conf = NULL;
+
+    if (!this)
+        return -1;
+
+    if (!this->children || this->children->next) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "trace translator requires one subvolume");
+        return -1;
+    }
+    if (!this->parents) {
+        gf_log(this->name, GF_LOG_WARNING, "dangling volume. check volfile ");
+    }
+
+    conf = GF_CALLOC(1, sizeof(trace_conf_t), gf_trace_mt_trace_conf_t);
+    if (!conf) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "cannot allocate "
+               "xl->private");
+        return -1;
+    }
+
+    options = this->options;
+    includes = data_to_str(dict_get(options, "include-ops"));
+    excludes = data_to_str(dict_get(options, "exclude-ops"));
+
+    {
+        int i;
+        for (i = 0; i < GF_FOP_MAXVALUE; i++) {
+            if (gf_fop_list[i])
+                strncpy(trace_fop_names[i].name, gf_fop_list[i],
+                        sizeof(trace_fop_names[i].name));
+            else
+                strncpy(trace_fop_names[i].name, ":O",
+                        sizeof(trace_fop_names[i].name));
+            trace_fop_names[i].enabled = 1;
+            trace_fop_names[i].name[sizeof(trace_fop_names[i].name) - 1] = 0;
+        }
+    }
+
+    if (includes && excludes) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "must specify only one of 'include-ops' and "
+               "'exclude-ops'");
+        return -1;
+    }
+
+    if (includes)
+        process_call_list(includes, 1);
+    if (excludes)
+        process_call_list(excludes, 0);
+
+    GF_OPTION_INIT("history-size", history_size, size, out);
+    conf->history_size = history_size;
+
+    gf_log(this->name, GF_LOG_INFO, "history size %" PRIu64, history_size);
+
+    GF_OPTION_INIT("log-file", conf->log_file, bool, out);
+
+    gf_log(this->name, GF_LOG_INFO, "logging to file %s",
+           (conf->log_file == _gf_true) ? "enabled" : "disabled");
+
+    GF_OPTION_INIT("log-history", conf->log_history, bool, out);
+
+    gf_log(this->name, GF_LOG_DEBUG, "logging to history %s",
+           (conf->log_history == _gf_true) ? "enabled" : "disabled");
+
+    history = eh_new(history_size, _gf_false, NULL);
+    if (!history) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "event history cannot be "
+               "initialized");
+        return -1;
+    }
+
+    this->history = history;
+
+    conf->trace_log_level = GF_LOG_INFO;
+
+    if (dict_get(options, "force-log-level")) {
+        forced_loglevel = data_to_str(dict_get(options, "force-log-level"));
+        if (!forced_loglevel)
+            goto setloglevel;
+
+        if (strcmp(forced_loglevel, "INFO") == 0)
+            conf->trace_log_level = GF_LOG_INFO;
+        else if (strcmp(forced_loglevel, "TRACE") == 0)
+            conf->trace_log_level = GF_LOG_TRACE;
+        else if (strcmp(forced_loglevel, "ERROR") == 0)
+            conf->trace_log_level = GF_LOG_ERROR;
+        else if (strcmp(forced_loglevel, "DEBUG") == 0)
+            conf->trace_log_level = GF_LOG_DEBUG;
+        else if (strcmp(forced_loglevel, "WARNING") == 0)
+            conf->trace_log_level = GF_LOG_WARNING;
+        else if (strcmp(forced_loglevel, "CRITICAL") == 0)
+            conf->trace_log_level = GF_LOG_CRITICAL;
+        else if (strcmp(forced_loglevel, "NONE") == 0)
+            conf->trace_log_level = GF_LOG_NONE;
+    }
 
 setloglevel:
-        gf_log_set_loglevel (conf->trace_log_level);
-        this->private = conf;
-        ret = 0;
-out:
-        if (ret == -1) {
-                if (history)
-                        GF_FREE (history);
-                if (conf)
-                        GF_FREE (conf);
-        }
+    gf_log_set_loglevel(this->ctx, conf->trace_log_level);
+    this->private = conf;
+    ret = 0;
+out:
+    if (ret == -1) {
+        if (history)
+            GF_FREE(history);
+        if (conf)
+            GF_FREE(conf);
+    }
 
-        return ret;
+    return ret;
 }
 
 void
-fini (xlator_t *this)
+fini(xlator_t *this)
 {
-        if (!this)
-                return;
+    if (!this)
+        return;
 
-        if (this->history)
-                eh_destroy (this->history);
+    if (this->history)
+        eh_destroy(this->history);
 
-        gf_log (this->name, GF_LOG_INFO,
-                "trace translator unloaded");
-        return;
+    gf_log(this->name, GF_LOG_INFO, "trace translator unloaded");
+    return;
 }
 
 struct xlator_fops fops = {
-        .stat        = trace_stat,
-        .readlink    = trace_readlink,
-        .mknod       = trace_mknod,
-        .mkdir       = trace_mkdir,
-        .unlink      = trace_unlink,
-        .rmdir       = trace_rmdir,
-        .symlink     = trace_symlink,
-        .rename      = trace_rename,
-        .link        = trace_link,
-        .truncate    = trace_truncate,
-        .open        = trace_open,
-        .readv       = trace_readv,
-        .writev      = trace_writev,
-        .statfs      = trace_statfs,
-        .flush       = trace_flush,
-        .fsync       = trace_fsync,
-        .setxattr    = trace_setxattr,
-        .getxattr    = trace_getxattr,
-        .fsetxattr   = trace_fsetxattr,
-        .fgetxattr   = trace_fgetxattr,
-        .removexattr = trace_removexattr,
-        .opendir     = trace_opendir,
-        .readdir     = trace_readdir,
-        .readdirp    = trace_readdirp,
-        .fsyncdir    = trace_fsyncdir,
-        .access      = trace_access,
-        .ftruncate   = trace_ftruncate,
-        .fstat       = trace_fstat,
-        .create      = trace_create,
-        .lk          = trace_lk,
-        .inodelk     = trace_inodelk,
-        .finodelk    = trace_finodelk,
-        .entrylk     = trace_entrylk,
-        .fentrylk    = trace_fentrylk,
-        .lookup      = trace_lookup,
-        .rchecksum   = trace_rchecksum,
-        .xattrop     = trace_xattrop,
-        .fxattrop    = trace_fxattrop,
-        .setattr     = trace_setattr,
-        .fsetattr    = trace_fsetattr,
+    .stat = trace_stat,
+    .readlink = trace_readlink,
+    .mknod = trace_mknod,
+    .mkdir = trace_mkdir,
+    .unlink = trace_unlink,
+    .rmdir = trace_rmdir,
+    .symlink = trace_symlink,
+    .rename = trace_rename,
+    .link = trace_link,
+    .truncate = trace_truncate,
+    .open = trace_open,
+    .readv = trace_readv,
+    .writev = trace_writev,
+    .statfs = trace_statfs,
+    .flush = trace_flush,
+    .fsync = trace_fsync,
+    .setxattr = trace_setxattr,
+    .getxattr = trace_getxattr,
+    .fsetxattr = trace_fsetxattr,
+    .fgetxattr = trace_fgetxattr,
+    .removexattr = trace_removexattr,
+    .opendir = trace_opendir,
+    .readdir = trace_readdir,
+    .readdirp = trace_readdirp,
+    .fsyncdir = trace_fsyncdir,
+    .access = trace_access,
+    .ftruncate = trace_ftruncate,
+    .fstat = trace_fstat,
+    .create = trace_create,
+    .lk = trace_lk,
+    .inodelk = trace_inodelk,
+    .finodelk = trace_finodelk,
+    .entrylk = trace_entrylk,
+    .fentrylk = trace_fentrylk,
+    .lookup = trace_lookup,
+    .rchecksum = trace_rchecksum,
+    .xattrop = trace_xattrop,
+    .fxattrop = trace_fxattrop,
+    .setattr = trace_setattr,
+    .fsetattr = trace_fsetattr,
+    .seek = trace_seek,
 };
 
 struct xlator_cbks cbks = {
-        .release     = trace_release,
-        .releasedir  = trace_releasedir,
-        .forget      = trace_forget,
+    .release = trace_release,
+    .releasedir = trace_releasedir,
+    .forget = trace_forget,
 };
 
 struct volume_options options[] = {
-        { .key  = {"include-ops", "include"},
-          .type = GF_OPTION_TYPE_STR,
-          /*.value = { ""} */
-        },
-        { .key  = {"exclude-ops", "exclude"},
-          .type = GF_OPTION_TYPE_STR
-          /*.value = { ""} */
-        },
-        { .key  = {"history-size"},
-          .type = GF_OPTION_TYPE_SIZET,
-          .default_value = "1024",
-        },
-        { .key  = {"log-file"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "no",
-        },
-        { .key  = {"log-history"},
-          .type = GF_OPTION_TYPE_BOOL,
-          .default_value = "no",
-        },
-        { .key  = {NULL} },
+    {
+        .key = {"include-ops", "include"},
+        .type = GF_OPTION_TYPE_STR,
+        /*.value = { ""} */
+    },
+    {
+        .key = {"exclude-ops", "exclude"},
+        .type = GF_OPTION_TYPE_STR
+        /*.value = { ""} */
+    },
+    {
+        .key = {"history-size"},
+        .type = GF_OPTION_TYPE_SIZET,
+        .default_value = "1024",
+    },
+    {
+        .key = {"log-file"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "no",
+    },
+    {
+        .key = {"log-history"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "no",
+    },
+    {.key = {NULL}},
 };
 
-struct xlator_dumpops dumpops = {
-        .history = trace_dump_history
+struct xlator_dumpops dumpops = {.history = trace_dump_history};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1},
+    .dumpops = &dumpops,
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "trace",
+    .category = GF_TECH_PREVIEW,
 };
diff --git a/xlators/debug/trace/src/trace.h b/xlators/debug/trace/src/trace.h
index 045eefb3682..b16304799da 100644
--- a/xlators/debug/trace/src/trace.h
+++ b/xlators/debug/trace/src/trace.h
@@ -8,91 +8,48 @@
    cases as published by the Free Software Foundation.
 */
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
 #include <time.h>
 #include <errno.h>
-#include "glusterfs.h"
-#include "xlator.h"
-#include "common-utils.h"
-#include "event-history.h"
-#include "logging.h"
-#include "circ-buff.h"
-#include "statedump.h"
-#include "options.h"
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/event-history.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/circ-buff.h>
+#include <glusterfs/statedump.h>
+#include <glusterfs/options.h>
 
 #define TRACE_DEFAULT_HISTORY_SIZE 1024
 
 typedef struct {
-        /* Since the longest fop name is fremovexattr i.e 12 characters, array size
-         * is kept 24, i.e double of the maximum.
-         */
-        char name[24];
-        int enabled;
+    /* Since the longest fop name is fremovexattr i.e 12 characters, array size
+     * is kept 24, i.e double of the maximum.
+     */
+    char name[24];
+    int enabled;
 } trace_fop_name_t;
 
 trace_fop_name_t trace_fop_names[GF_FOP_MAXVALUE];
 
 typedef struct {
-        gf_boolean_t log_file;
-        gf_boolean_t log_history;
-        size_t       history_size;
-        int trace_log_level;
+    gf_boolean_t log_file;
+    gf_boolean_t log_history;
+    uint64_t history_size;
+    int trace_log_level;
 } trace_conf_t;
 
-#define TRACE_STACK_UNWIND(op, frame, params ...)                       \
-        do {                                                            \
-                frame->local = NULL;                                    \
-                STACK_UNWIND_STRICT (op, frame, params);                \
-        } while (0);
-
-#define LOG_ELEMENT(_conf, _string)                                     \
-        do {                                                            \
-                if (_conf) {                                            \
-                        if ((_conf->log_history) == _gf_true)           \
-                                gf_log_eh ("%s", _string);              \
-                        if ((_conf->log_file) == _gf_true)              \
-                                gf_log (THIS->name, _conf->trace_log_level, \
-                                        "%s", _string);                 \
-                }                                                       \
-        } while (0);
-
-#define trace_stat_to_str(buf, statstr)                                 \
-        do {                                                            \
-                char     atime_buf[256]    = {0,};                      \
-                char     mtime_buf[256]    = {0,};                      \
-                char     ctime_buf[256]    = {0,};                      \
-                uint64_t ia_time           = 0;                         \
-                                                                        \
-                if (!buf)                                               \
-                        break;                                          \
-                                                                        \
-                ia_time = buf->ia_atime;                                \
-                strftime (atime_buf, 256, "[%b %d %H:%M:%S]",           \
-                          localtime ((time_t *)&ia_time));              \
-                                                                        \
-                ia_time = buf->ia_mtime;                                \
-                strftime (mtime_buf, 256, "[%b %d %H:%M:%S]",           \
-                          localtime ((time_t *)&ia_time));              \
-                                                                        \
-                ia_time = buf->ia_ctime;                                \
-                strftime (ctime_buf, 256, "[%b %d %H:%M:%S]",           \
-                          localtime ((time_t *)&ia_time));              \
-                                                                        \
-                snprintf (statstr, sizeof (statstr),                    \
-                          "gfid=%s ino=%"PRIu64", mode=%o, "            \
-                          "nlink=%"GF_PRI_NLINK", uid=%u, "             \
-                          "gid=%u, size=%"PRIu64", "                    \
-                          "blocks=%"PRIu64", atime=%s, "                \
-                          "mtime=%s, ctime=%s",                         \
-                          uuid_utoa (buf->ia_gfid), buf->ia_ino,        \
-                          st_mode_from_ia (buf->ia_prot,                \
-                                           buf->ia_type),               \
-                          buf->ia_nlink, buf->ia_uid,                   \
-                          buf->ia_gid, buf->ia_size,                    \
-                          buf->ia_blocks, atime_buf,                    \
-                          mtime_buf, ctime_buf);                        \
-        } while (0);
+#define TRACE_STACK_UNWIND(op, frame, params...)                               \
+    do {                                                                       \
+        frame->local = NULL;                                                   \
+        STACK_UNWIND_STRICT(op, frame, params);                                \
+    } while (0);
+
+#define LOG_ELEMENT(_conf, _string)                                            \
+    do {                                                                       \
+        if (_conf) {                                                           \
+            if ((_conf->log_history) == _gf_true)                              \
+                gf_log_eh("%s", _string);                                      \
+            if ((_conf->log_file) == _gf_true)                                 \
+                gf_log(THIS->name, _conf->trace_log_level, "%s", _string);     \
+        }                                                                      \
+    } while (0);
diff --git a/xlators/encryption/Makefile.am b/xlators/encryption/Makefile.am
deleted file mode 100644
index 2cbde680fac..00000000000
--- a/xlators/encryption/Makefile.am
+++ /dev/null
@@ -1,3 +0,0 @@
-SUBDIRS = rot-13
-
-CLEANFILES = 
diff --git a/xlators/encryption/rot-13/src/rot-13.c b/xlators/encryption/rot-13/src/rot-13.c
deleted file mode 100644
index b9ac29a72e1..00000000000
--- a/xlators/encryption/rot-13/src/rot-13.c
+++ /dev/null
@@ -1,199 +0,0 @@
-/*
-   Copyright (c) 2006-2012 Red Hat, Inc. <http://www.redhat.com>
-   This file is part of GlusterFS.
-
-   This file is licensed to you under your choice of the GNU Lesser
-   General Public License, version 3 or any later version (LGPLv3 or
-   later), or the GNU General Public License, version 2 (GPLv2), in all
-   cases as published by the Free Software Foundation.
-*/
-#include <ctype.h>
-#include <sys/uio.h>
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "xlator.h"
-#include "logging.h"
-
-#include "rot-13.h"
-
-/*
- * This is a rot13 ``encryption'' xlator. It rot13's data when
- * writing to disk and rot13's it back when reading it.
- * This xlator is meant as an example, NOT FOR PRODUCTION
- * USE ;) (hence no error-checking)
- */
-
-void
-rot13 (char *buf, int len)
-{
-	int i;
-	for (i = 0; i < len; i++) {
-		if (buf[i] >= 'a' && buf[i] <= 'z')
-			buf[i] = 'a' + ((buf[i] - 'a' + 13) % 26);
-		else if (buf[i] >= 'A' && buf[i] <= 'Z')
-			buf[i] = 'A' + ((buf[i] - 'A' + 13) % 26);
-	}
-}
-
-void
-rot13_iovec (struct iovec *vector, int count)
-{
-	int i;
-	for (i = 0; i < count; i++) {
-		rot13 (vector[i].iov_base, vector[i].iov_len);
-	}
-}
-
-int32_t
-rot13_readv_cbk (call_frame_t *frame,
-                 void *cookie,
-                 xlator_t *this,
-                 int32_t op_ret,
-                 int32_t op_errno,
-                 struct iovec *vector,
-                 int32_t count,
-		 struct iatt *stbuf,
-                 struct iobref *iobref, dict_t *xdata)
-{
-	rot_13_private_t *priv = (rot_13_private_t *)this->private;
-
-	if (priv->decrypt_read)
-		rot13_iovec (vector, count);
-
-	STACK_UNWIND_STRICT (readv, frame, op_ret, op_errno, vector, count,
-                             stbuf, iobref, xdata);
-	return 0;
-}
-
-int32_t
-rot13_readv (call_frame_t *frame,
-             xlator_t *this,
-             fd_t *fd,
-             size_t size,
-             off_t offset, uint32_t flags, dict_t *xdata)
-{
-	STACK_WIND (frame,
-		    rot13_readv_cbk,
-		    FIRST_CHILD (this),
-		    FIRST_CHILD (this)->fops->readv,
-		    fd, size, offset, flags, xdata);
-	return 0;
-}
-
-int32_t
-rot13_writev_cbk (call_frame_t *frame,
-                  void *cookie,
-                  xlator_t *this,
-                  int32_t op_ret,
-                  int32_t op_errno,
-                  struct iatt *prebuf,
-		  struct iatt *postbuf, dict_t *xdata)
-{
-	STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno, prebuf, postbuf,
-                             xdata);
-	return 0;
-}
-
-int32_t
-rot13_writev (call_frame_t *frame,
-              xlator_t *this,
-              fd_t *fd,
-              struct iovec *vector,
-              int32_t count,
-              off_t offset, uint32_t flags,
-              struct iobref *iobref, dict_t *xdata)
-{
-	rot_13_private_t *priv = (rot_13_private_t *)this->private;
-	if (priv->encrypt_write)
-		rot13_iovec (vector, count);
-
-	STACK_WIND (frame,
-		    rot13_writev_cbk,
-		    FIRST_CHILD (this),
-		    FIRST_CHILD (this)->fops->writev,
-		    fd, vector, count, offset, flags,
-                    iobref, xdata);
-	return 0;
-}
-
-int32_t
-init (xlator_t *this)
-{
-	data_t *data = NULL;
-	rot_13_private_t *priv = NULL;
-
-	if (!this->children || this->children->next) {
-		gf_log ("rot13", GF_LOG_ERROR,
-			"FATAL: rot13 should have exactly one child");
-		return -1;
-	}
-
-	if (!this->parents) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"dangling volume. check volfile ");
-	}
-
-	priv = GF_CALLOC (sizeof (rot_13_private_t), 1, 0);
-        if (!priv)
-                return -1;
-
-	priv->decrypt_read = 1;
-	priv->encrypt_write = 1;
-
-	data = dict_get (this->options, "encrypt-write");
-	if (data) {
-		if (gf_string2boolean (data->data, &priv->encrypt_write) == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"encrypt-write takes only boolean options");
-			return -1;
-		}
-	}
-
-	data = dict_get (this->options, "decrypt-read");
-	if (data) {
-		if (gf_string2boolean (data->data, &priv->decrypt_read) == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"decrypt-read takes only boolean options");
-			return -1;
-		}
-	}
-
-	this->private = priv;
-	gf_log ("rot13", GF_LOG_DEBUG, "rot13 xlator loaded");
-	return 0;
-}
-
-void
-fini (xlator_t *this)
-{
-	rot_13_private_t *priv = this->private;
-
-        if (!priv)
-                return;
-        this->private = NULL;
-	GF_FREE (priv);
-
-	return;
-}
-
-struct xlator_fops fops = {
-	.readv        = rot13_readv,
-	.writev       = rot13_writev
-};
-
-struct xlator_cbks cbks;
-
-struct volume_options options[] = {
-	{ .key  = {"encrypt-write"},
-	  .type = GF_OPTION_TYPE_BOOL
-	},
-	{ .key  = {"decrypt-read"},
-	  .type = GF_OPTION_TYPE_BOOL
-	},
-	{ .key  = {NULL} },
-};
diff --git a/xlators/features/Makefile.am b/xlators/features/Makefile.am
index 0e37fc0b7d6..c57897f11ea 100644
--- a/xlators/features/Makefile.am
+++ b/xlators/features/Makefile.am
@@ -1,4 +1,14 @@
-SUBDIRS = locks quota read-only mac-compat quiesce marker index \
-	  protect changelog gfid-access $(GLUPY_SUBDIR) qemu-block # trash path-converter # filter
+if BUILD_CLOUDSYNC
+  CLOUDSYNC_DIR = cloudsync
+endif
+
+if BUILD_METADISP
+  METADISP_DIR = metadisp
+endif
+
+SUBDIRS = locks quota read-only quiesce marker index barrier arbiter upcall \
+	compress changelog gfid-access snapview-client snapview-server trash \
+	shard bit-rot leases selinux sdfs namespace $(CLOUDSYNC_DIR) thin-arbiter \
+	utime $(METADISP_DIR)
 
 CLEANFILES =
diff --git a/xlators/storage/bd_map/Makefile.am b/xlators/features/arbiter/Makefile.am
index a985f42a877..a985f42a877 100644
--- a/xlators/storage/bd_map/Makefile.am
+++ b/xlators/features/arbiter/Makefile.am
diff --git a/xlators/features/arbiter/src/Makefile.am b/xlators/features/arbiter/src/Makefile.am
new file mode 100644
index 00000000000..badc42f37be
--- /dev/null
+++ b/xlators/features/arbiter/src/Makefile.am
@@ -0,0 +1,19 @@
+if WITH_SERVER
+xlator_LTLIBRARIES = arbiter.la
+endif
+
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+arbiter_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+arbiter_la_SOURCES = arbiter.c
+arbiter_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+noinst_HEADERS = arbiter.h arbiter-mem-types.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/arbiter/src/arbiter-mem-types.h b/xlators/features/arbiter/src/arbiter-mem-types.h
new file mode 100644
index 00000000000..05d18374c46
--- /dev/null
+++ b/xlators/features/arbiter/src/arbiter-mem-types.h
@@ -0,0 +1,18 @@
+/*
+  Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __ARBITER_MEM_TYPES_H__
+#define __ARBITER_MEM_TYPES_H__
+#include <glusterfs/mem-types.h>
+
+typedef enum gf_arbiter_mem_types_ {
+    gf_arbiter_mt_inode_ctx_t = gf_common_mt_end + 1,
+    gf_arbiter_mt_end
+} gf_arbiter_mem_types_t;
+#endif
diff --git a/xlators/features/arbiter/src/arbiter.c b/xlators/features/arbiter/src/arbiter.c
new file mode 100644
index 00000000000..83a97e3354b
--- /dev/null
+++ b/xlators/features/arbiter/src/arbiter.c
@@ -0,0 +1,380 @@
+/*
+  Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "arbiter.h"
+#include "arbiter-mem-types.h"
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/logging.h>
+
+static arbiter_inode_ctx_t *
+__arbiter_inode_ctx_get(inode_t *inode, xlator_t *this)
+{
+    arbiter_inode_ctx_t *ctx = NULL;
+    int ret = 0;
+    uint64_t ctx_addr = 0;
+
+    ret = __inode_ctx_get(inode, this, &ctx_addr);
+    if (ret == 0) {
+        ctx = (arbiter_inode_ctx_t *)(long)ctx_addr;
+        goto out;
+    }
+
+    ctx = GF_CALLOC(1, sizeof(*ctx), gf_arbiter_mt_inode_ctx_t);
+    if (!ctx)
+        goto out;
+
+    ret = __inode_ctx_put(inode, this, (uint64_t)(uintptr_t)ctx);
+    if (ret) {
+        GF_FREE(ctx);
+        ctx = NULL;
+        gf_log_callingfn(this->name, GF_LOG_ERROR,
+                         "failed to "
+                         "set the inode ctx (%s)",
+                         uuid_utoa(inode->gfid));
+    }
+out:
+    return ctx;
+}
+
+static arbiter_inode_ctx_t *
+arbiter_inode_ctx_get(inode_t *inode, xlator_t *this)
+{
+    arbiter_inode_ctx_t *ctx = NULL;
+
+    LOCK(&inode->lock);
+    {
+        ctx = __arbiter_inode_ctx_get(inode, this);
+    }
+    UNLOCK(&inode->lock);
+    return ctx;
+}
+
+int32_t
+arbiter_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, inode_t *inode,
+                   struct iatt *buf, dict_t *xdata, struct iatt *postparent)
+{
+    arbiter_inode_ctx_t *ctx = NULL;
+
+    if (op_ret != 0)
+        goto unwind;
+    ctx = arbiter_inode_ctx_get(inode, this);
+    if (!ctx) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+    memcpy(&ctx->iattbuf, buf, sizeof(ctx->iattbuf));
+
+unwind:
+    STACK_UNWIND_STRICT(lookup, frame, op_ret, op_errno, inode, buf, xdata,
+                        postparent);
+    return 0;
+}
+
+int32_t
+arbiter_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    STACK_WIND(frame, arbiter_lookup_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lookup, loc, xdata);
+    return 0;
+}
+
+int32_t
+arbiter_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+                 dict_t *xdata)
+{
+    arbiter_inode_ctx_t *ctx = NULL;
+    struct iatt *buf = NULL;
+    int32_t op_ret = 0;
+    int32_t op_errno = 0;
+
+    ctx = arbiter_inode_ctx_get(loc->inode, this);
+    if (!ctx) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+    buf = &ctx->iattbuf;
+unwind:
+    STACK_UNWIND_STRICT(truncate, frame, op_ret, op_errno, buf, buf, NULL);
+    return 0;
+}
+
+int32_t
+arbiter_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                  dict_t *xdata)
+
+{
+    arbiter_inode_ctx_t *ctx = NULL;
+    struct iatt *buf = NULL;
+    int32_t op_ret = 0;
+    int32_t op_errno = 0;
+
+    ctx = arbiter_inode_ctx_get(fd->inode, this);
+    if (!ctx) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+    buf = &ctx->iattbuf;
+unwind:
+    STACK_UNWIND_STRICT(ftruncate, frame, op_ret, op_errno, buf, buf, NULL);
+    return 0;
+}
+
+dict_t *
+arbiter_fill_writev_xdata(fd_t *fd, dict_t *xdata, xlator_t *this)
+{
+    dict_t *rsp_xdata = NULL;
+    int32_t ret = 0;
+    int is_append = 1;
+
+    if (!fd || !fd->inode || gf_uuid_is_null(fd->inode->gfid)) {
+        goto out;
+    }
+
+    if (!xdata)
+        goto out;
+
+    rsp_xdata = dict_new();
+    if (!rsp_xdata)
+        goto out;
+
+    if (dict_get(xdata, GLUSTERFS_OPEN_FD_COUNT)) {
+        ret = dict_set_uint32(rsp_xdata, GLUSTERFS_OPEN_FD_COUNT,
+                              fd->inode->fd_count);
+        if (ret < 0) {
+            gf_msg_debug(this->name, 0,
+                         "Failed to set dict value"
+                         " for GLUSTERFS_OPEN_FD_COUNT");
+        }
+    }
+    if (dict_get(xdata, GLUSTERFS_WRITE_IS_APPEND)) {
+        ret = dict_set_uint32(rsp_xdata, GLUSTERFS_WRITE_IS_APPEND, is_append);
+        if (ret < 0) {
+            gf_msg_debug(this->name, 0,
+                         "Failed to set dict value"
+                         " for GLUSTERFS_WRITE_IS_APPEND");
+        }
+    }
+out:
+    return rsp_xdata;
+}
+
+int32_t
+arbiter_writev(call_frame_t *frame, xlator_t *this, fd_t *fd,
+               struct iovec *vector, int32_t count, off_t off, uint32_t flags,
+               struct iobref *iobref, dict_t *xdata)
+{
+    arbiter_inode_ctx_t *ctx = NULL;
+    struct iatt *buf = NULL;
+    dict_t *rsp_xdata = NULL;
+    int op_ret = 0;
+    int op_errno = 0;
+
+    ctx = arbiter_inode_ctx_get(fd->inode, this);
+    if (!ctx) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+    buf = &ctx->iattbuf;
+    op_ret = iov_length(vector, count);
+    rsp_xdata = arbiter_fill_writev_xdata(fd, xdata, this);
+unwind:
+    STACK_UNWIND_STRICT(writev, frame, op_ret, op_errno, buf, buf, rsp_xdata);
+    if (rsp_xdata)
+        dict_unref(rsp_xdata);
+    return 0;
+}
+
+int32_t
+arbiter_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                  int32_t keep_size, off_t offset, size_t len, dict_t *xdata)
+{
+    arbiter_inode_ctx_t *ctx = NULL;
+    struct iatt *buf = NULL;
+    int op_ret = 0;
+    int op_errno = 0;
+
+    ctx = arbiter_inode_ctx_get(fd->inode, this);
+    if (!ctx) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+    buf = &ctx->iattbuf;
+unwind:
+    STACK_UNWIND_STRICT(fallocate, frame, op_ret, op_errno, buf, buf, NULL);
+    return 0;
+}
+
+int32_t
+arbiter_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                size_t len, dict_t *xdata)
+{
+    arbiter_inode_ctx_t *ctx = NULL;
+    struct iatt *buf = NULL;
+    int op_ret = 0;
+    int op_errno = 0;
+
+    ctx = arbiter_inode_ctx_get(fd->inode, this);
+    if (!ctx) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+    buf = &ctx->iattbuf;
+unwind:
+    STACK_UNWIND_STRICT(discard, frame, op_ret, op_errno, buf, buf, NULL);
+    return 0;
+}
+
+int32_t
+arbiter_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                 off_t len, dict_t *xdata)
+{
+    arbiter_inode_ctx_t *ctx = NULL;
+    struct iatt *buf = NULL;
+    int op_ret = 0;
+    int op_errno = 0;
+
+    ctx = arbiter_inode_ctx_get(fd->inode, this);
+    if (!ctx) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+    buf = &ctx->iattbuf;
+unwind:
+    STACK_UNWIND_STRICT(zerofill, frame, op_ret, op_errno, buf, buf, NULL);
+    return 0;
+}
+
+static int32_t
+arbiter_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+              off_t offset, uint32_t flags, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(readv, frame, -1, ENOSYS, NULL, 0, NULL, NULL, NULL);
+    return 0;
+}
+
+static int32_t
+arbiter_seek(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             gf_seek_what_t what, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(seek, frame, -1, ENOSYS, 0, xdata);
+    return 0;
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    ret = xlator_mem_acct_init(this, gf_arbiter_mt_end + 1);
+    if (ret)
+        gf_log(this->name, GF_LOG_ERROR,
+               "Memory accounting "
+               "initialization failed.");
+    return ret;
+}
+
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+    return 0;
+}
+
+int
+arbiter_forget(xlator_t *this, inode_t *inode)
+{
+    arbiter_inode_ctx_t *ctx = NULL;
+    uint64_t ctx_addr = 0;
+
+    inode_ctx_del(inode, this, &ctx_addr);
+    if (!ctx_addr)
+        return 0;
+    ctx = (arbiter_inode_ctx_t *)(long)ctx_addr;
+    GF_FREE(ctx);
+    return 0;
+}
+
+int32_t
+init(xlator_t *this)
+{
+    if (!this->children || this->children->next) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "'arbiter' not configured with exactly one child");
+        return -1;
+    }
+
+    if (!this->parents)
+        gf_log(this->name, GF_LOG_ERROR, "dangling volume. check volfile ");
+
+    return 0;
+}
+
+void
+fini(xlator_t *this)
+{
+    return;
+}
+
+struct xlator_fops fops = {
+    .lookup = arbiter_lookup,
+
+    /* Return success for these inode write FOPS without winding it down to
+     * posix; this is needed for AFR write transaction logic to work.*/
+    .truncate = arbiter_truncate,
+    .writev = arbiter_writev,
+    .ftruncate = arbiter_ftruncate,
+    .fallocate = arbiter_fallocate,
+    .discard = arbiter_discard,
+    .zerofill = arbiter_zerofill,
+
+    /* AFR is not expected to wind these inode read FOPS initiated by the
+     * application to the arbiter brick. But in case a bug causes them
+     * to be called, we return ENOSYS. */
+    .readv = arbiter_readv,
+    .seek = arbiter_seek,
+
+    /* The following inode read FOPS initiated by the application are not
+     * wound by AFR either but internal logic like  shd, glfsheal and
+     * client side healing in AFR will send them for selfheal/ inode refresh
+     * operations etc.,so we need to wind them down to posix:
+     *
+     * (f)stat, readdir(p), readlink, (f)getxattr.*/
+
+    /* All other FOPs not listed here are safe to be wound down to posix.*/
+};
+
+struct xlator_cbks cbks = {
+    .forget = arbiter_forget,
+};
+
+struct volume_options options[] = {
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "arbiter",
+    .category = GF_MAINTAINED,
+};
diff --git a/xlators/features/arbiter/src/arbiter.h b/xlators/features/arbiter/src/arbiter.h
new file mode 100644
index 00000000000..546db7b751a
--- /dev/null
+++ b/xlators/features/arbiter/src/arbiter.h
@@ -0,0 +1,21 @@
+/*
+  Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _ARBITER_H
+#define _ARBITER_H
+
+#include <glusterfs/locking.h>
+#include <glusterfs/common-utils.h>
+
+typedef struct arbiter_inode_ctx_ {
+    struct iatt iattbuf;
+} arbiter_inode_ctx_t;
+
+#endif /* _ARBITER_H */
diff --git a/xlators/features/mac-compat/Makefile.am b/xlators/features/barrier/Makefile.am
index d471a3f9243..a985f42a877 100644
--- a/xlators/features/mac-compat/Makefile.am
+++ b/xlators/features/barrier/Makefile.am
@@ -1,3 +1,3 @@
 SUBDIRS = src
 
-CLEANFILES = 
+CLEANFILES =
diff --git a/xlators/features/barrier/src/Makefile.am b/xlators/features/barrier/src/Makefile.am
new file mode 100644
index 00000000000..25099bc56e5
--- /dev/null
+++ b/xlators/features/barrier/src/Makefile.am
@@ -0,0 +1,17 @@
+xlator_LTLIBRARIES = barrier.la
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+barrier_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+barrier_la_SOURCES = barrier.c
+
+barrier_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+noinst_HEADERS = barrier.h barrier-mem-types.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/filter/src/filter-mem-types.h b/xlators/features/barrier/src/barrier-mem-types.h
index 47a17249b8d..71ed7898d9c 100644
--- a/xlators/features/filter/src/filter-mem-types.h
+++ b/xlators/features/barrier/src/barrier-mem-types.h
@@ -1,5 +1,5 @@
 /*
-   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
    This file is part of GlusterFS.
 
    This file is licensed to you under your choice of the GNU Lesser
@@ -7,14 +7,14 @@
    later), or the GNU General Public License, version 2 (GPLv2), in all
    cases as published by the Free Software Foundation.
 */
-#ifndef __FILTER_MEM_TYPES_H__
-#define __FILTER_MEM_TYPES_H__
 
-#include "mem-types.h"
+#ifndef __BARRIER_MEM_TYPES_H__
+#define __BARRIER_MEM_TYPES_H__
 
-enum gf_filter_mem_types_ {
-        gf_filter_mt_gf_filter = gf_common_mt_end + 1,
-        gf_filter_mt_end
+#include <glusterfs/mem-types.h>
+
+enum gf_barrier_mem_types_ {
+    gf_barrier_mt_priv_t = gf_common_mt_end + 1,
+    gf_barrier_mt_end
 };
 #endif
-
diff --git a/xlators/features/barrier/src/barrier.c b/xlators/features/barrier/src/barrier.c
new file mode 100644
index 00000000000..852bbacb99d
--- /dev/null
+++ b/xlators/features/barrier/src/barrier.c
@@ -0,0 +1,809 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include "barrier.h"
+#include <glusterfs/defaults.h>
+#include <glusterfs/call-stub.h>
+
+#include <glusterfs/statedump.h>
+
+void
+barrier_local_set_gfid(call_frame_t *frame, uuid_t gfid, xlator_t *this)
+{
+    if (gfid) {
+        uuid_t *id = GF_MALLOC(sizeof(uuid_t), gf_common_mt_uuid_t);
+        if (!id) {
+            gf_log(this->name, GF_LOG_WARNING,
+                   "Could not set gfid"
+                   ". gfid will not be dumped in statedump file.");
+            return;
+        }
+        gf_uuid_copy(*id, gfid);
+        frame->local = id;
+    }
+}
+
+void
+barrier_local_free_gfid(call_frame_t *frame)
+{
+    if (frame->local) {
+        GF_FREE(frame->local);
+        frame->local = NULL;
+    }
+}
+
+int32_t
+barrier_truncate_cbk_resume(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int32_t op_ret, int32_t op_errno,
+                            struct iatt *prebuf, struct iatt *postbuf,
+                            dict_t *xdata)
+{
+    barrier_local_free_gfid(frame);
+    STACK_UNWIND_STRICT(truncate, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+    return 0;
+}
+
+int32_t
+barrier_ftruncate_cbk_resume(call_frame_t *frame, void *cookie, xlator_t *this,
+                             int32_t op_ret, int32_t op_errno,
+                             struct iatt *prebuf, struct iatt *postbuf,
+                             dict_t *xdata)
+{
+    barrier_local_free_gfid(frame);
+    STACK_UNWIND_STRICT(ftruncate, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+    return 0;
+}
+
+int32_t
+barrier_unlink_cbk_resume(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno,
+                          struct iatt *preparent, struct iatt *postparent,
+                          dict_t *xdata)
+{
+    barrier_local_free_gfid(frame);
+    STACK_UNWIND_STRICT(unlink, frame, op_ret, op_errno, preparent, postparent,
+                        xdata);
+    return 0;
+}
+
+int32_t
+barrier_rmdir_cbk_resume(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno,
+                         struct iatt *preparent, struct iatt *postparent,
+                         dict_t *xdata)
+{
+    barrier_local_free_gfid(frame);
+    STACK_UNWIND_STRICT(rmdir, frame, op_ret, op_errno, preparent, postparent,
+                        xdata);
+    return 0;
+}
+
+int32_t
+barrier_rename_cbk_resume(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                          struct iatt *preoldparent, struct iatt *postoldparent,
+                          struct iatt *prenewparent, struct iatt *postnewparent,
+                          dict_t *xdata)
+{
+    barrier_local_free_gfid(frame);
+    STACK_UNWIND_STRICT(rename, frame, op_ret, op_errno, buf, preoldparent,
+                        postoldparent, prenewparent, postnewparent, xdata);
+    return 0;
+}
+
+int32_t
+barrier_writev_cbk_resume(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                          struct iatt *postbuf, dict_t *xdata)
+{
+    barrier_local_free_gfid(frame);
+    STACK_UNWIND_STRICT(writev, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+    return 0;
+}
+
+int32_t
+barrier_fsync_cbk_resume(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                         struct iatt *postbuf, dict_t *xdata)
+{
+    barrier_local_free_gfid(frame);
+    STACK_UNWIND_STRICT(fsync, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+    return 0;
+}
+
+int32_t
+barrier_removexattr_cbk_resume(call_frame_t *frame, void *cookie,
+                               xlator_t *this, int32_t op_ret, int32_t op_errno,
+                               dict_t *xdata)
+{
+    barrier_local_free_gfid(frame);
+    STACK_UNWIND_STRICT(removexattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+int32_t
+barrier_fremovexattr_cbk_resume(call_frame_t *frame, void *cookie,
+                                xlator_t *this, int32_t op_ret,
+                                int32_t op_errno, dict_t *xdata)
+{
+    barrier_local_free_gfid(frame);
+    STACK_UNWIND_STRICT(fremovexattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+int32_t
+barrier_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                   struct iatt *postbuf, dict_t *xdata)
+{
+    BARRIER_FOP_CBK(writev, out, frame, this, op_ret, op_errno, prebuf, postbuf,
+                    xdata);
+out:
+    return 0;
+}
+
+int32_t
+barrier_fremovexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    BARRIER_FOP_CBK(fremovexattr, out, frame, this, op_ret, op_errno, xdata);
+out:
+    return 0;
+}
+
+int32_t
+barrier_removexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    BARRIER_FOP_CBK(removexattr, out, frame, this, op_ret, op_errno, xdata);
+out:
+    return 0;
+}
+
+int32_t
+barrier_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
+{
+    BARRIER_FOP_CBK(truncate, out, frame, this, op_ret, op_errno, prebuf,
+                    postbuf, xdata);
+out:
+    return 0;
+}
+
+int32_t
+barrier_ftruncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                      struct iatt *postbuf, dict_t *xdata)
+{
+    BARRIER_FOP_CBK(ftruncate, out, frame, this, op_ret, op_errno, prebuf,
+                    postbuf, xdata);
+out:
+    return 0;
+}
+
+int32_t
+barrier_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                   struct iatt *preoldparent, struct iatt *postoldparent,
+                   struct iatt *prenewparent, struct iatt *postnewparent,
+                   dict_t *xdata)
+{
+    BARRIER_FOP_CBK(rename, out, frame, this, op_ret, op_errno, buf,
+                    preoldparent, postoldparent, prenewparent, postnewparent,
+                    xdata);
+out:
+    return 0;
+}
+
+int32_t
+barrier_rmdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                  struct iatt *postparent, dict_t *xdata)
+{
+    BARRIER_FOP_CBK(rmdir, out, frame, this, op_ret, op_errno, preparent,
+                    postparent, xdata);
+out:
+    return 0;
+}
+
+int32_t
+barrier_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                   struct iatt *postparent, dict_t *xdata)
+{
+    BARRIER_FOP_CBK(unlink, out, frame, this, op_ret, op_errno, preparent,
+                    postparent, xdata);
+out:
+    return 0;
+}
+
+int32_t
+barrier_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                  struct iatt *postbuf, dict_t *xdata)
+{
+    BARRIER_FOP_CBK(fsync, out, frame, this, op_ret, op_errno, prebuf, postbuf,
+                    xdata);
+out:
+    return 0;
+}
+
+int32_t
+barrier_writev(call_frame_t *frame, xlator_t *this, fd_t *fd,
+               struct iovec *vector, int32_t count, off_t off, uint32_t flags,
+               struct iobref *iobref, dict_t *xdata)
+{
+    if (!((flags | fd->flags) & (O_SYNC | O_DSYNC))) {
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->writev, fd, vector, count, off,
+                        flags, iobref, xdata);
+
+        return 0;
+    }
+
+    barrier_local_set_gfid(frame, fd->inode->gfid, this);
+    STACK_WIND(frame, barrier_writev_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->writev, fd, vector, count, off, flags,
+               iobref, xdata);
+    return 0;
+}
+
+int32_t
+barrier_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                     const char *name, dict_t *xdata)
+{
+    barrier_local_set_gfid(frame, fd->inode->gfid, this);
+    STACK_WIND(frame, barrier_fremovexattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fremovexattr, fd, name, xdata);
+    return 0;
+}
+
+int32_t
+barrier_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                    const char *name, dict_t *xdata)
+{
+    barrier_local_set_gfid(frame, loc->inode->gfid, this);
+    STACK_WIND(frame, barrier_removexattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->removexattr, loc, name, xdata);
+    return 0;
+}
+
+int32_t
+barrier_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+                 dict_t *xdata)
+{
+    barrier_local_set_gfid(frame, loc->inode->gfid, this);
+    STACK_WIND(frame, barrier_truncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
+    return 0;
+}
+
+int32_t
+barrier_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc,
+               loc_t *newloc, dict_t *xdata)
+{
+    barrier_local_set_gfid(frame, oldloc->inode->gfid, this);
+    STACK_WIND(frame, barrier_rename_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rename, oldloc, newloc, xdata);
+    return 0;
+}
+
+int
+barrier_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+              dict_t *xdata)
+{
+    barrier_local_set_gfid(frame, loc->inode->gfid, this);
+    STACK_WIND(frame, barrier_rmdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rmdir, loc, flags, xdata);
+    return 0;
+}
+
+int32_t
+barrier_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+               dict_t *xdata)
+{
+    barrier_local_set_gfid(frame, loc->inode->gfid, this);
+    STACK_WIND(frame, barrier_unlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->unlink, loc, xflag, xdata);
+    return 0;
+}
+
+int32_t
+barrier_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                  dict_t *xdata)
+{
+    barrier_local_set_gfid(frame, fd->inode->gfid, this);
+    STACK_WIND(frame, barrier_ftruncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+    return 0;
+}
+
+int32_t
+barrier_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+              dict_t *xdata)
+{
+    barrier_local_set_gfid(frame, fd->inode->gfid, this);
+    STACK_WIND(frame, barrier_fsync_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsync, fd, flags, xdata);
+    return 0;
+}
+
+call_stub_t *
+__barrier_dequeue(xlator_t *this, struct list_head *queue)
+{
+    call_stub_t *stub = NULL;
+    barrier_priv_t *priv = NULL;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    if (list_empty(queue))
+        goto out;
+
+    stub = list_entry(queue->next, call_stub_t, list);
+    list_del_init(&stub->list);
+
+out:
+    return stub;
+}
+
+void
+barrier_dequeue_all(xlator_t *this, struct list_head *queue)
+{
+    call_stub_t *stub = NULL;
+
+    gf_log(this->name, GF_LOG_INFO, "Dequeuing all the barriered fops");
+
+    /* TODO: Start the below task in a new thread */
+    while ((stub = __barrier_dequeue(this, queue)))
+        call_resume(stub);
+
+    gf_log(this->name, GF_LOG_INFO,
+           "Dequeuing the barriered fops is "
+           "finished");
+    return;
+}
+
+void
+barrier_timeout(void *data)
+{
+    xlator_t *this = NULL;
+    barrier_priv_t *priv = NULL;
+    struct list_head queue = {
+        0,
+    };
+
+    this = data;
+    THIS = this;
+    priv = this->private;
+
+    INIT_LIST_HEAD(&queue);
+
+    gf_log(this->name, GF_LOG_CRITICAL,
+           "Disabling barrier because of "
+           "the barrier timeout.");
+
+    LOCK(&priv->lock);
+    {
+        __barrier_disable(this, &queue);
+    }
+    UNLOCK(&priv->lock);
+
+    barrier_dequeue_all(this, &queue);
+
+    return;
+}
+
+void
+__barrier_enqueue(xlator_t *this, call_stub_t *stub)
+{
+    barrier_priv_t *priv = NULL;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    list_add_tail(&stub->list, &priv->queue);
+    priv->queue_size++;
+
+    return;
+}
+
+void
+__barrier_disable(xlator_t *this, struct list_head *queue)
+{
+    GF_UNUSED int ret = 0;
+    barrier_priv_t *priv = NULL;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    if (priv->timer) {
+        ret = gf_timer_call_cancel(this->ctx, priv->timer);
+        priv->timer = NULL;
+    }
+
+    list_splice_init(&priv->queue, queue);
+    priv->queue_size = 0;
+    priv->barrier_enabled = _gf_false;
+}
+
+int
+__barrier_enable(xlator_t *this, barrier_priv_t *priv)
+{
+    int ret = -1;
+
+    priv->timer = gf_timer_call_after(this->ctx, priv->timeout, barrier_timeout,
+                                      (void *)this);
+    if (!priv->timer) {
+        gf_log(this->name, GF_LOG_CRITICAL,
+               "Couldn't add barrier "
+               "timeout event.");
+        goto out;
+    }
+
+    priv->barrier_enabled = _gf_true;
+    ret = 0;
+out:
+    return ret;
+}
+
+int
+notify(xlator_t *this, int event, void *data, ...)
+{
+    barrier_priv_t *priv = this->private;
+    dict_t *dict = NULL;
+    int ret = -1;
+    int barrier_enabled = _gf_false;
+    struct list_head queue = {
+        0,
+    };
+
+    GF_ASSERT(priv);
+    INIT_LIST_HEAD(&queue);
+
+    switch (event) {
+        case GF_EVENT_TRANSLATOR_OP: {
+            dict = data;
+            barrier_enabled = dict_get_str_boolean(dict, "barrier", -1);
+
+            if (barrier_enabled == -1) {
+                gf_log(this->name, GF_LOG_ERROR,
+                       "Could not fetch "
+                       " barrier key from the dictionary.");
+                goto out;
+            }
+
+            LOCK(&priv->lock);
+            {
+                if (!priv->barrier_enabled) {
+                    if (barrier_enabled) {
+                        ret = __barrier_enable(this, priv);
+                    } else {
+                        UNLOCK(&priv->lock);
+                        gf_log(this->name, GF_LOG_ERROR, "Already disabled.");
+                        goto post_unlock;
+                    }
+                } else {
+                    if (!barrier_enabled) {
+                        __barrier_disable(this, &queue);
+                        ret = 0;
+                    } else {
+                        UNLOCK(&priv->lock);
+                        gf_log(this->name, GF_LOG_ERROR, "Already enabled");
+                        goto post_unlock;
+                    }
+                }
+            }
+            UNLOCK(&priv->lock);
+        post_unlock:
+            if (!list_empty(&queue))
+                barrier_dequeue_all(this, &queue);
+
+            break;
+        }
+        default: {
+            default_notify(this, event, data);
+            ret = 0;
+            goto out;
+        }
+    }
+out:
+    return ret;
+}
+
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+    barrier_priv_t *priv = NULL;
+    int ret = -1;
+    gf_boolean_t barrier_enabled = _gf_false;
+    uint32_t timeout = {
+        0,
+    };
+    struct list_head queue = {
+        0,
+    };
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    GF_OPTION_RECONF("barrier", barrier_enabled, options, bool, out);
+    GF_OPTION_RECONF("barrier-timeout", timeout, options, time, out);
+
+    INIT_LIST_HEAD(&queue);
+
+    LOCK(&priv->lock);
+    {
+        if (!priv->barrier_enabled) {
+            if (barrier_enabled) {
+                ret = __barrier_enable(this, priv);
+                if (ret) {
+                    goto unlock;
+                }
+            }
+        } else {
+            if (!barrier_enabled) {
+                __barrier_disable(this, &queue);
+            }
+        }
+        priv->timeout.tv_sec = timeout;
+        ret = 0;
+    }
+unlock:
+    UNLOCK(&priv->lock);
+
+    if (!list_empty(&queue))
+        barrier_dequeue_all(this, &queue);
+
+out:
+    return ret;
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    ret = xlator_mem_acct_init(this, gf_barrier_mt_end + 1);
+    if (ret)
+        gf_log(this->name, GF_LOG_ERROR,
+               "Memory accounting "
+               "initialization failed.");
+
+    return ret;
+}
+
+int
+init(xlator_t *this)
+{
+    int ret = -1;
+    barrier_priv_t *priv = NULL;
+    uint32_t timeout = {
+        0,
+    };
+
+    if (!this->children || this->children->next) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "'barrier' not configured with exactly one child");
+        goto out;
+    }
+
+    if (!this->parents)
+        gf_log(this->name, GF_LOG_WARNING, "dangling volume. check volfile ");
+
+    priv = GF_CALLOC(1, sizeof(*priv), gf_barrier_mt_priv_t);
+    if (!priv)
+        goto out;
+
+    LOCK_INIT(&priv->lock);
+
+    GF_OPTION_INIT("barrier", priv->barrier_enabled, bool, out);
+    GF_OPTION_INIT("barrier-timeout", timeout, time, out);
+    priv->timeout.tv_sec = timeout;
+
+    INIT_LIST_HEAD(&priv->queue);
+
+    if (priv->barrier_enabled) {
+        ret = __barrier_enable(this, priv);
+        if (ret == -1)
+            goto out;
+    }
+
+    this->private = priv;
+    ret = 0;
+out:
+    if (ret && priv)
+        GF_FREE(priv);
+
+    return ret;
+}
+
+void
+fini(xlator_t *this)
+{
+    barrier_priv_t *priv = NULL;
+    struct list_head queue = {
+        0,
+    };
+
+    priv = this->private;
+    if (!priv)
+        goto out;
+
+    INIT_LIST_HEAD(&queue);
+
+    gf_log(this->name, GF_LOG_INFO,
+           "Disabling barriering and dequeuing "
+           "all the queued fops");
+    LOCK(&priv->lock);
+    {
+        __barrier_disable(this, &queue);
+    }
+    UNLOCK(&priv->lock);
+
+    if (!list_empty(&queue))
+        barrier_dequeue_all(this, &queue);
+
+    this->private = NULL;
+
+    LOCK_DESTROY(&priv->lock);
+    GF_FREE(priv);
+out:
+    return;
+}
+
+static void
+barrier_dump_stub(call_stub_t *stub, char *prefix)
+{
+    char key[GF_DUMP_MAX_BUF_LEN] = {
+        0,
+    };
+
+    gf_proc_dump_build_key(key, prefix, "fop");
+    gf_proc_dump_write(key, "%s", gf_fop_list[stub->fop]);
+
+    if (stub->frame->local) {
+        gf_proc_dump_build_key(key, prefix, "gfid");
+        gf_proc_dump_write(key, "%s",
+                           uuid_utoa(*(uuid_t *)(stub->frame->local)));
+    }
+    if (stub->args.loc.path) {
+        gf_proc_dump_build_key(key, prefix, "path");
+        gf_proc_dump_write(key, "%s", stub->args.loc.path);
+    }
+    if (stub->args.loc.name) {
+        gf_proc_dump_build_key(key, prefix, "name");
+        gf_proc_dump_write(key, "%s", stub->args.loc.name);
+    }
+
+    return;
+}
+
+static void
+__barrier_dump_queue(barrier_priv_t *priv)
+{
+    call_stub_t *stub = NULL;
+    char key[GF_DUMP_MAX_BUF_LEN] = {
+        0,
+    };
+    int i = 0;
+
+    GF_VALIDATE_OR_GOTO("barrier", priv, out);
+
+    list_for_each_entry(stub, &priv->queue, list)
+    {
+        snprintf(key, sizeof(key), "stub.%d", i++);
+        gf_proc_dump_add_section("%s", key);
+        barrier_dump_stub(stub, key);
+    }
+
+out:
+    return;
+}
+
+int
+barrier_dump_priv(xlator_t *this)
+{
+    int ret = -1;
+    char key[GF_DUMP_MAX_BUF_LEN] = {
+        0,
+    };
+    barrier_priv_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("barrier", this, out);
+
+    priv = this->private;
+    if (!priv)
+        return 0;
+
+    gf_proc_dump_build_key(key, "xlator.features.barrier", "priv");
+    gf_proc_dump_add_section("%s", key);
+    gf_proc_dump_build_key(key, "barrier", "enabled");
+
+    LOCK(&priv->lock);
+    {
+        gf_proc_dump_write(key, "%d", priv->barrier_enabled);
+        gf_proc_dump_build_key(key, "barrier", "timeout");
+        gf_proc_dump_write(key, "%ld", priv->timeout.tv_sec);
+        if (priv->barrier_enabled) {
+            gf_proc_dump_build_key(key, "barrier", "queue_size");
+            gf_proc_dump_write(key, "%d", priv->queue_size);
+            __barrier_dump_queue(priv);
+        }
+    }
+    UNLOCK(&priv->lock);
+
+out:
+    return ret;
+}
+
+struct xlator_fops fops = {
+
+    /* Barrier Class fops */
+    .rmdir = barrier_rmdir,
+    .unlink = barrier_unlink,
+    .rename = barrier_rename,
+    .removexattr = barrier_removexattr,
+    .fremovexattr = barrier_fremovexattr,
+    .truncate = barrier_truncate,
+    .ftruncate = barrier_ftruncate,
+    .fsync = barrier_fsync,
+
+    /* Writes with only O_SYNC flag */
+    .writev = barrier_writev,
+};
+
+struct xlator_dumpops dumpops = {
+    .priv = barrier_dump_priv,
+};
+
+struct xlator_cbks cbks;
+
+struct volume_options options[] = {
+    {.key = {"barrier"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "disable",
+     .op_version = {GD_OP_VERSION_3_6_0},
+     .flags = OPT_FLAG_SETTABLE,
+     .description = "When \"enabled\", blocks acknowledgements to application "
+                    "for file operations such as rmdir, rename, unlink, "
+                    "removexattr, fremovexattr, truncate, ftruncate, "
+                    "write (with O_SYNC), fsync. It is turned \"off\" by "
+                    "default."},
+    {.key = {"barrier-timeout"},
+     .type = GF_OPTION_TYPE_TIME,
+     .default_value = BARRIER_TIMEOUT,
+     .op_version = {GD_OP_VERSION_3_6_0},
+     .flags = OPT_FLAG_SETTABLE,
+     .description = "After 'timeout' seconds since the time 'barrier' "
+                    "option was set to \"on\", acknowledgements to file "
+                    "operations are no longer blocked and previously "
+                    "blocked acknowledgements are sent to the application"},
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .dumpops = &dumpops,
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "barrier",
+    .category = GF_MAINTAINED,
+};
diff --git a/xlators/features/barrier/src/barrier.h b/xlators/features/barrier/src/barrier.h
new file mode 100644
index 00000000000..1337f311f7d
--- /dev/null
+++ b/xlators/features/barrier/src/barrier.h
@@ -0,0 +1,89 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __BARRIER_H__
+#define __BARRIER_H__
+
+#include "barrier-mem-types.h"
+#include <glusterfs/xlator.h>
+#include <glusterfs/timer.h>
+#include <glusterfs/call-stub.h>
+
+#define BARRIER_FOP_CBK(fop_name, label, frame, this, params...)               \
+    do {                                                                       \
+        barrier_priv_t *_priv = NULL;                                          \
+        call_stub_t *_stub = NULL;                                             \
+        gf_boolean_t _barrier_enabled = _gf_false;                             \
+        struct list_head queue = {                                             \
+            0,                                                                 \
+        };                                                                     \
+                                                                               \
+        INIT_LIST_HEAD(&queue);                                                \
+                                                                               \
+        _priv = this->private;                                                 \
+        GF_ASSERT(_priv);                                                      \
+                                                                               \
+        LOCK(&_priv->lock);                                                    \
+        {                                                                      \
+            if (_priv->barrier_enabled) {                                      \
+                _barrier_enabled = _priv->barrier_enabled;                     \
+                                                                               \
+                _stub = fop_##fop_name##_cbk_stub(                             \
+                    frame, barrier_##fop_name##_cbk_resume, params);           \
+                if (!_stub) {                                                  \
+                    __barrier_disable(this, &queue);                           \
+                    goto unlock;                                               \
+                }                                                              \
+                                                                               \
+                __barrier_enqueue(this, _stub);                                \
+            }                                                                  \
+        }                                                                      \
+    unlock:                                                                    \
+        UNLOCK(&_priv->lock);                                                  \
+                                                                               \
+        if (_stub)                                                             \
+            goto label;                                                        \
+                                                                               \
+        if (_barrier_enabled && !_stub) {                                      \
+            gf_log(this->name, GF_LOG_CRITICAL,                                \
+                   "Failed to barrier FOPs, disabling "                        \
+                   "barrier. FOP: %s, ERROR: %s",                              \
+                   #fop_name, strerror(ENOMEM));                               \
+            barrier_dequeue_all(this, &queue);                                 \
+        }                                                                      \
+        barrier_local_free_gfid(frame);                                        \
+        STACK_UNWIND_STRICT(fop_name, frame, params);                          \
+        goto label;                                                            \
+    } while (0)
+
+typedef struct {
+    gf_timer_t *timer;
+    gf_lock_t lock;
+    struct list_head queue;
+    struct timespec timeout;
+    uint32_t queue_size;
+    gf_boolean_t barrier_enabled;
+    char _pad[3]; /* manual padding */
+} barrier_priv_t;
+
+int
+__barrier_enable(xlator_t *this, barrier_priv_t *priv);
+void
+__barrier_enqueue(xlator_t *this, call_stub_t *stub);
+void
+__barrier_disable(xlator_t *this, struct list_head *queue);
+void
+barrier_timeout(void *data);
+void
+barrier_dequeue_all(xlator_t *this, struct list_head *queue);
+call_stub_t *
+__barrier_dequeue(xlator_t *this, struct list_head *queue);
+
+#endif
diff --git a/xlators/features/bit-rot/Makefile.am b/xlators/features/bit-rot/Makefile.am
new file mode 100644
index 00000000000..f963effea22
--- /dev/null
+++ b/xlators/features/bit-rot/Makefile.am
@@ -0,0 +1 @@
+SUBDIRS = src
\ No newline at end of file
diff --git a/xlators/features/bit-rot/src/Makefile.am b/xlators/features/bit-rot/src/Makefile.am
new file mode 100644
index 00000000000..b5e4a7d62a0
--- /dev/null
+++ b/xlators/features/bit-rot/src/Makefile.am
@@ -0,0 +1 @@
+SUBDIRS = stub bitd
diff --git a/xlators/features/bit-rot/src/bitd/Makefile.am b/xlators/features/bit-rot/src/bitd/Makefile.am
new file mode 100644
index 00000000000..6db800e6565
--- /dev/null
+++ b/xlators/features/bit-rot/src/bitd/Makefile.am
@@ -0,0 +1,23 @@
+if WITH_SERVER
+xlator_LTLIBRARIES = bit-rot.la
+endif
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+bit_rot_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src/ -I$(top_builddir)/rpc/xdr/src/ \
+	-I$(top_srcdir)/rpc/rpc-lib/src -I$(CONTRIBDIR)/timer-wheel \
+	-I$(top_srcdir)/xlators/features/bit-rot/src/stub
+
+bit_rot_la_SOURCES = bit-rot.c bit-rot-scrub.c bit-rot-ssm.c \
+		     bit-rot-scrub-status.c
+bit_rot_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la \
+	$(top_builddir)/xlators/features/changelog/lib/src/libgfchangelog.la
+
+noinst_HEADERS = bit-rot.h bit-rot-scrub.h bit-rot-bitd-messages.h bit-rot-ssm.h \
+		 bit-rot-scrub-status.h
+
+AM_CFLAGS = -Wall -DBR_RATE_LIMIT_SIGNER $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-bitd-messages.h b/xlators/features/bit-rot/src/bitd/bit-rot-bitd-messages.h
new file mode 100644
index 00000000000..5bc5103a27c
--- /dev/null
+++ b/xlators/features/bit-rot/src/bitd/bit-rot-bitd-messages.h
@@ -0,0 +1,101 @@
+/*
+ Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+ This file is part of GlusterFS.
+
+ This file is licensed to you under your choice of the GNU Lesser
+ General Public License, version 3 or any later version (LGPLv3 or
+ later), or the GNU General Public License, version 2 (GPLv2), in all
+ cases as published by the Free Software Foundation.
+ */
+
+#ifndef _BITROT_BITD_MESSAGES_H_
+#define _BITROT_BITD_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(BITROT_BITD, BRB_MSG_FD_CREATE_FAILED, BRB_MSG_READV_FAILED,
+           BRB_MSG_BLOCK_READ_FAILED, BRB_MSG_CALC_CHECKSUM_FAILED,
+           BRB_MSG_NO_MEMORY, BRB_MSG_GET_SIGN_FAILED, BRB_MSG_SET_SIGN_FAILED,
+           BRB_MSG_OP_FAILED, BRB_MSG_READ_AND_SIGN_FAILED, BRB_MSG_SIGN_FAILED,
+           BRB_MSG_GET_SUBVOL_FAILED, BRB_MSG_SET_TIMER_FAILED,
+           BRB_MSG_GET_INFO_FAILED, BRB_MSG_PATH_FAILED, BRB_MSG_MARK_BAD_FILE,
+           BRB_MSG_TRIGGER_SIGN, BRB_MSG_REGISTER_FAILED,
+           BRB_MSG_CRAWLING_START, BRB_MSG_SPAWN_FAILED,
+           BRB_MSG_INVALID_SUBVOL_CHILD, BRB_MSG_SKIP_OBJECT, BRB_MSG_NO_CHILD,
+           BRB_MSG_CHECKSUM_MISMATCH, BRB_MSG_MARK_CORRUPTED,
+           BRB_MSG_CRAWLING_FINISH, BRB_MSG_CALC_ERROR, BRB_MSG_LOOKUP_FAILED,
+           BRB_MSG_PARTIAL_VERSION_PRESENCE, BRB_MSG_MEM_ACNT_FAILED,
+           BRB_MSG_TIMER_WHEEL_UNAVAILABLE, BRB_MSG_BITROT_LOADED,
+           BRB_MSG_SCALE_DOWN_FAILED, BRB_MSG_SCALE_UP_FAILED,
+           BRB_MSG_SCALE_DOWN_SCRUBBER, BRB_MSG_SCALING_UP_SCRUBBER,
+           BRB_MSG_UNKNOWN_THROTTLE, BRB_MSG_RATE_LIMIT_INFO,
+           BRB_MSG_SCRUB_INFO, BRB_MSG_CONNECTED_TO_BRICK, BRB_MSG_BRICK_INFO,
+           BRB_MSG_SUBVOL_CONNECT_FAILED, BRB_MSG_INVALID_SUBVOL,
+           BRB_MSG_RESCHEDULE_SCRUBBER_FAILED, BRB_MSG_SCRUB_START,
+           BRB_MSG_SCRUB_FINISH, BRB_MSG_SCRUB_RUNNING,
+           BRB_MSG_SCRUB_RESCHEDULED, BRB_MSG_SCRUB_TUNABLE,
+           BRB_MSG_SCRUB_THREAD_CLEANUP, BRB_MSG_SCRUBBER_CLEANED,
+           BRB_MSG_GENERIC_SSM_INFO, BRB_MSG_ZERO_TIMEOUT_BUG,
+           BRB_MSG_BAD_OBJ_READDIR_FAIL, BRB_MSG_SSM_FAILED,
+           BRB_MSG_SCRUB_WAIT_FAILED, BRB_MSG_TRIGGER_SIGN_FAILED,
+           BRB_MSG_EVENT_UNHANDLED, BRB_MSG_COULD_NOT_SCHEDULE_SCRUB,
+           BRB_MSG_THREAD_CREATION_FAILED, BRB_MSG_MEM_POOL_ALLOC,
+           BRB_MSG_SAVING_HASH_FAILED);
+
+#define BRB_MSG_FD_CREATE_FAILED_STR "failed to create fd for the inode"
+#define BRB_MSG_READV_FAILED_STR "readv failed"
+#define BRB_MSG_BLOCK_READ_FAILED_STR "reading block failed"
+#define BRB_MSG_NO_MEMORY_STR "failed to allocate memory"
+#define BRB_MSG_CALC_CHECKSUM_FAILED_STR "calculating checksum failed"
+#define BRB_MSG_GET_SIGN_FAILED_STR "failed to get the signature"
+#define BRB_MSG_SET_SIGN_FAILED_STR "signing failed"
+#define BRB_MSG_OP_FAILED_STR "failed on object"
+#define BRB_MSG_TRIGGER_SIGN_FAILED_STR "Could not trigger signing"
+#define BRB_MSG_READ_AND_SIGN_FAILED_STR "reading and signing of object failed"
+#define BRB_MSG_SET_TIMER_FAILED_STR "Failed to allocate object expiry timer"
+#define BRB_MSG_GET_SUBVOL_FAILED_STR                                          \
+    "failed to get the subvolume for the brick"
+#define BRB_MSG_PATH_FAILED_STR "path failed"
+#define BRB_MSG_SKIP_OBJECT_STR "Entry is marked corrupted. skipping"
+#define BRB_MSG_PARTIAL_VERSION_PRESENCE_STR                                   \
+    "PArtial version xattr presence detected, ignoring"
+#define BRB_MSG_TRIGGER_SIGN_STR "Triggering signing"
+#define BRB_MSG_CRAWLING_START_STR                                             \
+    "Crawling brick, scanning for unsigned objects"
+#define BRB_MSG_CRAWLING_FINISH_STR "Completed crawling brick"
+#define BRB_MSG_REGISTER_FAILED_STR "Register to changelog failed"
+#define BRB_MSG_SPAWN_FAILED_STR "failed to spawn"
+#define BRB_MSG_CONNECTED_TO_BRICK_STR "Connected to brick"
+#define BRB_MSG_LOOKUP_FAILED_STR "lookup on root failed"
+#define BRB_MSG_GET_INFO_FAILED_STR "failed to get stub info"
+#define BRB_MSG_SCRUB_THREAD_CLEANUP_STR "Error cleaning up scanner thread"
+#define BRB_MSG_SCRUBBER_CLEANED_STR "clened up scrubber for brick"
+#define BRB_MSG_SUBVOL_CONNECT_FAILED_STR                                      \
+    "callback handler for subvolume failed"
+#define BRB_MSG_MEM_ACNT_FAILED_STR "Memory accounting init failed"
+#define BRB_MSG_EVENT_UNHANDLED_STR "Event unhandled for child"
+#define BRB_MSG_INVALID_SUBVOL_STR "Got event from invalid subvolume"
+#define BRB_MSG_RESCHEDULE_SCRUBBER_FAILED_STR                                 \
+    "on demand scrub schedule failed. Scrubber is not in pending state."
+#define BRB_MSG_COULD_NOT_SCHEDULE_SCRUB_STR                                   \
+    "Could not schedule ondemand scrubbing. Scrubbing will continue "          \
+    "according to old frequency."
+#define BRB_MSG_THREAD_CREATION_FAILED_STR "thread creation failed"
+#define BRB_MSG_RATE_LIMIT_INFO_STR "Rate Limit Info"
+#define BRB_MSG_MEM_POOL_ALLOC_STR "failed to allocate mem-pool for timer"
+#define BRB_MSG_NO_CHILD_STR "FATAL: no children"
+#define BRB_MSG_TIMER_WHEEL_UNAVAILABLE_STR "global timer wheel unavailable"
+#define BRB_MSG_BITROT_LOADED_STR "bit-rot xlator loaded"
+#define BRB_MSG_SAVING_HASH_FAILED_STR                                         \
+    "failed to allocate memory for saving hash of the object"
+#endif /* !_BITROT_BITD_MESSAGES_H_ */
diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.c b/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.c
new file mode 100644
index 00000000000..5cef2ffa5e5
--- /dev/null
+++ b/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.c
@@ -0,0 +1,78 @@
+/*
+  Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <string.h>
+#include <stdio.h>
+
+#include "bit-rot-scrub-status.h"
+
+void
+br_inc_unsigned_file_count(br_scrub_stats_t *scrub_stat)
+{
+    if (!scrub_stat)
+        return;
+
+    pthread_mutex_lock(&scrub_stat->lock);
+    {
+        scrub_stat->unsigned_files++;
+    }
+    pthread_mutex_unlock(&scrub_stat->lock);
+}
+
+void
+br_inc_scrubbed_file(br_scrub_stats_t *scrub_stat)
+{
+    if (!scrub_stat)
+        return;
+
+    pthread_mutex_lock(&scrub_stat->lock);
+    {
+        scrub_stat->scrubbed_files++;
+    }
+    pthread_mutex_unlock(&scrub_stat->lock);
+}
+
+void
+br_update_scrub_start_time(br_scrub_stats_t *scrub_stat, time_t time)
+{
+    if (!scrub_stat)
+        return;
+
+    pthread_mutex_lock(&scrub_stat->lock);
+    {
+        scrub_stat->scrub_start_time = time;
+    }
+    pthread_mutex_unlock(&scrub_stat->lock);
+}
+
+void
+br_update_scrub_finish_time(br_scrub_stats_t *scrub_stat, char *timestr,
+                            time_t time)
+{
+    int lst_size = 0;
+
+    if (!scrub_stat)
+        return;
+
+    lst_size = sizeof(scrub_stat->last_scrub_time);
+    if (strlen(timestr) >= lst_size)
+        return;
+
+    pthread_mutex_lock(&scrub_stat->lock);
+    {
+        scrub_stat->scrub_end_time = time;
+
+        scrub_stat->scrub_duration = scrub_stat->scrub_end_time -
+                                     scrub_stat->scrub_start_time;
+
+        snprintf(scrub_stat->last_scrub_time, lst_size, "%s", timestr);
+    }
+    pthread_mutex_unlock(&scrub_stat->lock);
+}
diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.h b/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.h
new file mode 100644
index 00000000000..f022aa831eb
--- /dev/null
+++ b/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.h
@@ -0,0 +1,50 @@
+/*
+   Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __BIT_ROT_SCRUB_STATUS_H__
+#define __BIT_ROT_SCRUB_STATUS_H__
+
+#include <stdint.h>
+#include <sys/time.h>
+#include <pthread.h>
+
+#include <glusterfs/common-utils.h>
+
+struct br_scrub_stats {
+    uint64_t scrubbed_files; /* Total number of scrubbed files. */
+
+    uint64_t unsigned_files; /* Total number of unsigned files. */
+
+    uint64_t scrub_duration; /* Duration of last scrub. */
+
+    char last_scrub_time[GF_TIMESTR_SIZE]; /* Last scrub completion time. */
+
+    time_t scrub_start_time; /* Scrubbing starting time. */
+
+    time_t scrub_end_time; /* Scrubbing finishing time. */
+
+    int8_t scrub_running; /* Whether scrub running or not. */
+
+    pthread_mutex_t lock;
+};
+
+typedef struct br_scrub_stats br_scrub_stats_t;
+
+void
+br_inc_unsigned_file_count(br_scrub_stats_t *scrub_stat);
+void
+br_inc_scrubbed_file(br_scrub_stats_t *scrub_stat);
+void
+br_update_scrub_start_time(br_scrub_stats_t *scrub_stat, time_t time);
+void
+br_update_scrub_finish_time(br_scrub_stats_t *scrub_stat, char *timestr,
+                            time_t time);
+
+#endif /* __BIT_ROT_SCRUB_STATUS_H__ */
diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-scrub.c b/xlators/features/bit-rot/src/bitd/bit-rot-scrub.c
new file mode 100644
index 00000000000..289dd53f610
--- /dev/null
+++ b/xlators/features/bit-rot/src/bitd/bit-rot-scrub.c
@@ -0,0 +1,2070 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <math.h>
+#include <ctype.h>
+#include <sys/uio.h>
+
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/common-utils.h>
+
+#include "bit-rot-scrub.h"
+#include <pthread.h>
+#include "bit-rot-bitd-messages.h"
+#include "bit-rot-scrub-status.h"
+#include <glusterfs/events.h>
+
+struct br_scrubbers {
+    pthread_t scrubthread;
+
+    struct list_head list;
+};
+
+struct br_fsscan_entry {
+    void *data;
+
+    loc_t parent;
+
+    gf_dirent_t *entry;
+
+    struct br_scanfs *fsscan; /* backpointer to subvolume scanner */
+
+    struct list_head list;
+};
+
+/**
+ * fetch signature extended attribute from an object's fd.
+ * NOTE: On success @xattr is not unref'd as @sign points
+ * to the dictionary value.
+ */
+static int32_t
+bitd_fetch_signature(xlator_t *this, br_child_t *child, fd_t *fd,
+                     dict_t **xattr, br_isignature_out_t **sign)
+{
+    int32_t ret = -1;
+
+    ret = syncop_fgetxattr(child->xl, fd, xattr, GLUSTERFS_GET_OBJECT_SIGNATURE,
+                           NULL, NULL);
+    if (ret < 0) {
+        br_log_object(this, "fgetxattr", fd->inode->gfid, -ret);
+        goto out;
+    }
+
+    ret = dict_get_ptr(*xattr, GLUSTERFS_GET_OBJECT_SIGNATURE, (void **)sign);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_GET_SIGN_FAILED,
+               "failed to extract signature info [GFID: %s]",
+               uuid_utoa(fd->inode->gfid));
+        goto unref_dict;
+    }
+
+    return 0;
+
+unref_dict:
+    dict_unref(*xattr);
+out:
+    return -1;
+}
+
+/**
+ * POST COMPUTE CHECK
+ *
+ * Checks to be performed before verifying calculated signature
+ * Object is skipped if:
+ *  - has stale signature
+ *  - mismatches versions caches in pre-compute check
+ */
+
+int32_t
+bitd_scrub_post_compute_check(xlator_t *this, br_child_t *child, fd_t *fd,
+                              unsigned long version,
+                              br_isignature_out_t **signature,
+                              br_scrub_stats_t *scrub_stat,
+                              gf_boolean_t skip_stat)
+{
+    int32_t ret = 0;
+    size_t signlen = 0;
+    dict_t *xattr = NULL;
+    br_isignature_out_t *signptr = NULL;
+
+    ret = bitd_fetch_signature(this, child, fd, &xattr, &signptr);
+    if (ret < 0) {
+        if (!skip_stat)
+            br_inc_unsigned_file_count(scrub_stat);
+        goto out;
+    }
+
+    /**
+     * Either the object got dirtied during the time the signature was
+     * calculated OR the version we saved during pre-compute check does
+     * not match now, implying that the object got dirtied and signed in
+     * between scrubs pre & post compute checks (checksum window).
+     *
+     * The log entry looks pretty ugly, but helps in debugging..
+     */
+    if (signptr->stale || (signptr->version != version)) {
+        if (!skip_stat)
+            br_inc_unsigned_file_count(scrub_stat);
+        gf_msg_debug(this->name, 0,
+                     "<STAGE: POST> Object [GFID: %s] "
+                     "either has a stale signature OR underwent "
+                     "signing during checksumming {Stale: %d | "
+                     "Version: %lu,%lu}",
+                     uuid_utoa(fd->inode->gfid), (signptr->stale) ? 1 : 0,
+                     version, signptr->version);
+        ret = -1;
+        goto unref_dict;
+    }
+
+    signlen = signptr->signaturelen;
+    *signature = GF_MALLOC(sizeof(br_isignature_out_t) + signlen,
+                           gf_common_mt_char);
+
+    (void)memcpy(*signature, signptr, sizeof(br_isignature_out_t) + signlen);
+
+    (*signature)->signaturelen = signlen;
+
+unref_dict:
+    dict_unref(xattr);
+out:
+    return ret;
+}
+
+static int32_t
+bitd_signature_staleness(xlator_t *this, br_child_t *child, fd_t *fd,
+                         int *stale, unsigned long *version,
+                         br_scrub_stats_t *scrub_stat, gf_boolean_t skip_stat)
+{
+    int32_t ret = -1;
+    dict_t *xattr = NULL;
+    br_isignature_out_t *signptr = NULL;
+
+    ret = bitd_fetch_signature(this, child, fd, &xattr, &signptr);
+    if (ret < 0) {
+        if (!skip_stat)
+            br_inc_unsigned_file_count(scrub_stat);
+        goto out;
+    }
+
+    /**
+     * save version for validation in post compute stage
+     * c.f. bitd_scrub_post_compute_check()
+     */
+    *stale = signptr->stale ? 1 : 0;
+    *version = signptr->version;
+
+    dict_unref(xattr);
+
+out:
+    return ret;
+}
+
+/**
+ * PRE COMPUTE CHECK
+ *
+ * Checks to be performed before initiating object signature calculation.
+ * An object is skipped if:
+ *  - it's already marked corrupted
+ *  - has stale signature
+ */
+int32_t
+bitd_scrub_pre_compute_check(xlator_t *this, br_child_t *child, fd_t *fd,
+                             unsigned long *version,
+                             br_scrub_stats_t *scrub_stat,
+                             gf_boolean_t skip_stat)
+{
+    int stale = 0;
+    int32_t ret = -1;
+
+    if (bitd_is_bad_file(this, child, NULL, fd)) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, BRB_MSG_SKIP_OBJECT,
+               "Object [GFID: %s] is marked corrupted, skipping..",
+               uuid_utoa(fd->inode->gfid));
+        goto out;
+    }
+
+    ret = bitd_signature_staleness(this, child, fd, &stale, version, scrub_stat,
+                                   skip_stat);
+    if (!ret && stale) {
+        if (!skip_stat)
+            br_inc_unsigned_file_count(scrub_stat);
+        gf_msg_debug(this->name, 0,
+                     "<STAGE: PRE> Object [GFID: %s] "
+                     "has stale signature",
+                     uuid_utoa(fd->inode->gfid));
+        ret = -1;
+    }
+
+out:
+    return ret;
+}
+
+/* static int */
+int
+bitd_compare_ckum(xlator_t *this, br_isignature_out_t *sign, unsigned char *md,
+                  inode_t *linked_inode, gf_dirent_t *entry, fd_t *fd,
+                  br_child_t *child, loc_t *loc)
+{
+    int ret = -1;
+    dict_t *xattr = NULL;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, sign, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+    GF_VALIDATE_OR_GOTO(this->name, child, out);
+    GF_VALIDATE_OR_GOTO(this->name, linked_inode, out);
+    GF_VALIDATE_OR_GOTO(this->name, md, out);
+    GF_VALIDATE_OR_GOTO(this->name, entry, out);
+
+    if (strncmp(sign->signature, (char *)md, sign->signaturelen) == 0) {
+        gf_msg_debug(this->name, 0,
+                     "%s [GFID: %s | Brick: %s] "
+                     "matches calculated checksum",
+                     loc->path, uuid_utoa(linked_inode->gfid),
+                     child->brick_path);
+        return 0;
+    }
+
+    gf_msg(this->name, GF_LOG_DEBUG, 0, BRB_MSG_CHECKSUM_MISMATCH,
+           "Object checksum mismatch: %s [GFID: %s | Brick: %s]", loc->path,
+           uuid_utoa(linked_inode->gfid), child->brick_path);
+    gf_msg(this->name, GF_LOG_ALERT, 0, BRB_MSG_CHECKSUM_MISMATCH,
+           "CORRUPTION DETECTED: Object %s {Brick: %s | GFID: %s}", loc->path,
+           child->brick_path, uuid_utoa(linked_inode->gfid));
+
+    /* Perform bad-file marking */
+    xattr = dict_new();
+    if (!xattr) {
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_set_int32(xattr, BITROT_OBJECT_BAD_KEY, _gf_true);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_MARK_BAD_FILE,
+               "Error setting bad-file marker for %s [GFID: %s | "
+               "Brick: %s]",
+               loc->path, uuid_utoa(linked_inode->gfid), child->brick_path);
+        goto dictfree;
+    }
+
+    gf_msg(this->name, GF_LOG_ALERT, 0, BRB_MSG_MARK_CORRUPTED,
+           "Marking"
+           " %s [GFID: %s | Brick: %s] as corrupted..",
+           loc->path, uuid_utoa(linked_inode->gfid), child->brick_path);
+    gf_event(EVENT_BITROT_BAD_FILE, "gfid=%s;path=%s;brick=%s",
+             uuid_utoa(linked_inode->gfid), loc->path, child->brick_path);
+    ret = syncop_fsetxattr(child->xl, fd, xattr, 0, NULL, NULL);
+    if (ret)
+        gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_MARK_BAD_FILE,
+               "Error marking object %s [GFID: %s] as corrupted", loc->path,
+               uuid_utoa(linked_inode->gfid));
+
+dictfree:
+    dict_unref(xattr);
+out:
+    return ret;
+}
+
+/**
+ * "The Scrubber"
+ *
+ * Perform signature validation for a given object with the assumption
+ * that the signature is SHA256 (because signer as of now _always_
+ * signs with SHA256).
+ */
+int
+br_scrubber_scrub_begin(xlator_t *this, struct br_fsscan_entry *fsentry)
+{
+    int32_t ret = -1;
+    fd_t *fd = NULL;
+    loc_t loc = {
+        0,
+    };
+    struct iatt iatt = {
+        0,
+    };
+    struct iatt parent_buf = {
+        0,
+    };
+    pid_t pid = 0;
+    br_child_t *child = NULL;
+    unsigned char *md = NULL;
+    inode_t *linked_inode = NULL;
+    br_isignature_out_t *sign = NULL;
+    unsigned long signedversion = 0;
+    gf_dirent_t *entry = NULL;
+    br_private_t *priv = NULL;
+    loc_t *parent = NULL;
+    gf_boolean_t skip_stat = _gf_false;
+    uuid_t shard_root_gfid = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("bit-rot", fsentry, out);
+
+    entry = fsentry->entry;
+    parent = &fsentry->parent;
+    child = fsentry->data;
+
+    priv = this->private;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", entry, out);
+    GF_VALIDATE_OR_GOTO("bit-rot", parent, out);
+    GF_VALIDATE_OR_GOTO("bit-rot", child, out);
+    GF_VALIDATE_OR_GOTO("bit-rot", priv, out);
+
+    pid = GF_CLIENT_PID_SCRUB;
+
+    ret = br_prepare_loc(this, child, parent, entry, &loc);
+    if (!ret)
+        goto out;
+
+    syncopctx_setfspid(&pid);
+
+    ret = syncop_lookup(child->xl, &loc, &iatt, &parent_buf, NULL, NULL);
+    if (ret) {
+        br_log_object_path(this, "lookup", loc.path, -ret);
+        goto out;
+    }
+
+    linked_inode = inode_link(loc.inode, parent->inode, loc.name, &iatt);
+    if (linked_inode)
+        inode_lookup(linked_inode);
+
+    gf_msg_debug(this->name, 0, "Scrubbing object %s [GFID: %s]", entry->d_name,
+                 uuid_utoa(linked_inode->gfid));
+
+    if (iatt.ia_type != IA_IFREG) {
+        gf_msg_debug(this->name, 0, "%s is not a regular file", entry->d_name);
+        ret = 0;
+        goto unref_inode;
+    }
+
+    if (IS_DHT_LINKFILE_MODE((&iatt))) {
+        gf_msg_debug(this->name, 0, "%s is a dht sticky bit file",
+                     entry->d_name);
+        ret = 0;
+        goto unref_inode;
+    }
+
+    /* skip updating scrub statistics for shard entries */
+    gf_uuid_parse(SHARD_ROOT_GFID, shard_root_gfid);
+    if (gf_uuid_compare(loc.pargfid, shard_root_gfid) == 0)
+        skip_stat = _gf_true;
+
+    /**
+     * open() an fd for subsequent operations
+     */
+    fd = fd_create(linked_inode, 0);
+    if (!fd) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_FD_CREATE_FAILED,
+               "failed to create fd for inode %s",
+               uuid_utoa(linked_inode->gfid));
+        goto unref_inode;
+    }
+
+    ret = syncop_open(child->xl, &loc, O_RDWR, fd, NULL, NULL);
+    if (ret) {
+        br_log_object(this, "open", linked_inode->gfid, -ret);
+        ret = -1;
+        goto unrefd;
+    }
+
+    fd_bind(fd);
+
+    /**
+     * perform pre compute checks before initiating checksum
+     * computation
+     *  - presence of bad object
+     *  - signature staleness
+     */
+    ret = bitd_scrub_pre_compute_check(this, child, fd, &signedversion,
+                                       &priv->scrub_stat, skip_stat);
+    if (ret)
+        goto unrefd; /* skip this object */
+
+    /* if all's good, proceed to calculate the hash */
+    md = GF_MALLOC(SHA256_DIGEST_LENGTH, gf_common_mt_char);
+    if (!md)
+        goto unrefd;
+
+    ret = br_calculate_obj_checksum(md, child, fd, &iatt);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_CALC_ERROR,
+               "error calculating hash for object [GFID: %s]",
+               uuid_utoa(fd->inode->gfid));
+        ret = -1;
+        goto free_md;
+    }
+
+    /**
+     * perform post compute checks as an object's signature may have
+     * become stale while scrubber calculated checksum.
+     */
+    ret = bitd_scrub_post_compute_check(this, child, fd, signedversion, &sign,
+                                        &priv->scrub_stat, skip_stat);
+    if (ret)
+        goto free_md;
+
+    ret = bitd_compare_ckum(this, sign, md, linked_inode, entry, fd, child,
+                            &loc);
+
+    if (!skip_stat)
+        br_inc_scrubbed_file(&priv->scrub_stat);
+
+    GF_FREE(sign); /* allocated on post-compute */
+
+    /** fd_unref() takes care of closing fd.. like syncop_close() */
+
+free_md:
+    GF_FREE(md);
+unrefd:
+    fd_unref(fd);
+unref_inode:
+    inode_unref(linked_inode);
+out:
+    loc_wipe(&loc);
+    return ret;
+}
+
+static void
+_br_lock_cleaner(void *arg)
+{
+    pthread_mutex_t *mutex = arg;
+
+    pthread_mutex_unlock(mutex);
+}
+
+static void
+wait_for_scrubbing(xlator_t *this, struct br_scanfs *fsscan)
+{
+    br_private_t *priv = NULL;
+    struct br_scrubber *fsscrub = NULL;
+
+    priv = this->private;
+    fsscrub = &priv->fsscrub;
+
+    pthread_cleanup_push(_br_lock_cleaner, &fsscan->waitlock);
+    pthread_mutex_lock(&fsscan->waitlock);
+    {
+        pthread_cleanup_push(_br_lock_cleaner, &fsscrub->mutex);
+        pthread_mutex_lock(&fsscrub->mutex);
+        {
+            list_replace_init(&fsscan->queued, &fsscan->ready);
+
+            /* wake up scrubbers */
+            pthread_cond_broadcast(&fsscrub->cond);
+        }
+        pthread_mutex_unlock(&fsscrub->mutex);
+        pthread_cleanup_pop(0);
+
+        while (fsscan->entries != 0)
+            pthread_cond_wait(&fsscan->waitcond, &fsscan->waitlock);
+    }
+    pthread_mutex_unlock(&fsscan->waitlock);
+    pthread_cleanup_pop(0);
+}
+
+static void
+_br_fsscan_inc_entry_count(struct br_scanfs *fsscan)
+{
+    fsscan->entries++;
+}
+
+static void
+_br_fsscan_dec_entry_count(struct br_scanfs *fsscan)
+{
+    if (--fsscan->entries == 0) {
+        pthread_mutex_lock(&fsscan->waitlock);
+        {
+            pthread_cond_signal(&fsscan->waitcond);
+        }
+        pthread_mutex_unlock(&fsscan->waitlock);
+    }
+}
+
+static void
+_br_fsscan_collect_entry(struct br_scanfs *fsscan,
+                         struct br_fsscan_entry *fsentry)
+{
+    list_add_tail(&fsentry->list, &fsscan->queued);
+    _br_fsscan_inc_entry_count(fsscan);
+}
+
+#define NR_ENTRIES (1 << 7) /* ..bulk scrubbing */
+
+int
+br_fsscanner_handle_entry(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent,
+                          void *data)
+{
+    int32_t ret = -1;
+    int scrub = 0;
+    br_child_t *child = NULL;
+    xlator_t *this = NULL;
+    struct br_scanfs *fsscan = NULL;
+    struct br_fsscan_entry *fsentry = NULL;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", subvol, error_return);
+    GF_VALIDATE_OR_GOTO("bit-rot", data, error_return);
+
+    child = data;
+    this = child->this;
+    fsscan = &child->fsscan;
+
+    _mask_cancellation();
+
+    fsentry = GF_CALLOC(1, sizeof(*fsentry), gf_br_mt_br_fsscan_entry_t);
+    if (!fsentry)
+        goto error_return;
+
+    {
+        fsentry->data = data;
+        fsentry->fsscan = &child->fsscan;
+
+        /* copy parent loc */
+        ret = loc_copy(&fsentry->parent, parent);
+        if (ret)
+            goto dealloc;
+
+        /* copy child entry */
+        fsentry->entry = entry_copy(entry);
+        if (!fsentry->entry)
+            goto locwipe;
+
+        INIT_LIST_HEAD(&fsentry->list);
+    }
+
+    LOCK(&fsscan->entrylock);
+    {
+        _br_fsscan_collect_entry(fsscan, fsentry);
+
+        /**
+         * need not be a equality check as entries may be pushed
+         * back onto the scanned queue when thread(s) are cleaned.
+         */
+        if (fsscan->entries >= NR_ENTRIES)
+            scrub = 1;
+    }
+    UNLOCK(&fsscan->entrylock);
+
+    _unmask_cancellation();
+
+    if (scrub)
+        wait_for_scrubbing(this, fsscan);
+
+    return 0;
+
+locwipe:
+    loc_wipe(&fsentry->parent);
+dealloc:
+    GF_FREE(fsentry);
+error_return:
+    return -1;
+}
+
+int32_t
+br_fsscan_deactivate(xlator_t *this)
+{
+    int ret = 0;
+    br_private_t *priv = NULL;
+    br_scrub_state_t nstate = 0;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    ret = gf_tw_del_timer(priv->timer_wheel, scrub_monitor->timer);
+    if (ret == 0) {
+        nstate = BR_SCRUB_STATE_STALLED;
+        gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO,
+               "Volume is under active scrubbing. Pausing scrub..");
+    } else {
+        nstate = BR_SCRUB_STATE_PAUSED;
+        gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO,
+               "Scrubber paused");
+    }
+
+    _br_monitor_set_scrub_state(scrub_monitor, nstate);
+
+    return 0;
+}
+
+static void
+br_scrubber_log_time(xlator_t *this, const char *sfx)
+{
+    char timestr[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    br_private_t *priv = NULL;
+    time_t now = 0;
+
+    now = gf_time();
+    priv = this->private;
+
+    gf_time_fmt(timestr, sizeof(timestr), now, gf_timefmt_FT);
+
+    if (strcasecmp(sfx, "started") == 0) {
+        br_update_scrub_start_time(&priv->scrub_stat, now);
+        gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_START,
+               "Scrubbing %s at %s", sfx, timestr);
+    } else {
+        br_update_scrub_finish_time(&priv->scrub_stat, timestr, now);
+        gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_FINISH,
+               "Scrubbing %s at %s", sfx, timestr);
+    }
+}
+
+static void
+br_fsscanner_log_time(xlator_t *this, br_child_t *child, const char *sfx)
+{
+    char timestr[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    time_t now = 0;
+
+    now = gf_time();
+    gf_time_fmt(timestr, sizeof(timestr), now, gf_timefmt_FT);
+
+    if (strcasecmp(sfx, "started") == 0) {
+        gf_msg_debug(this->name, 0, "Scrubbing \"%s\" %s at %s",
+                     child->brick_path, sfx, timestr);
+    } else {
+        gf_msg_debug(this->name, 0, "Scrubbing \"%s\" %s at %s",
+                     child->brick_path, sfx, timestr);
+    }
+}
+
+void
+br_child_set_scrub_state(br_child_t *child, gf_boolean_t state)
+{
+    child->active_scrubbing = state;
+}
+
+static void
+br_fsscanner_wait_until_kicked(xlator_t *this, br_child_t *child)
+{
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    pthread_cleanup_push(_br_lock_cleaner, &scrub_monitor->wakelock);
+    pthread_mutex_lock(&scrub_monitor->wakelock);
+    {
+        while (!scrub_monitor->kick)
+            pthread_cond_wait(&scrub_monitor->wakecond,
+                              &scrub_monitor->wakelock);
+
+        /* Child lock is to synchronize with disconnect events */
+        pthread_cleanup_push(_br_lock_cleaner, &child->lock);
+        pthread_mutex_lock(&child->lock);
+        {
+            scrub_monitor->active_child_count++;
+            br_child_set_scrub_state(child, _gf_true);
+        }
+        pthread_mutex_unlock(&child->lock);
+        pthread_cleanup_pop(0);
+    }
+    pthread_mutex_unlock(&scrub_monitor->wakelock);
+    pthread_cleanup_pop(0);
+}
+
+static void
+br_scrubber_entry_control(xlator_t *this)
+{
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    LOCK(&scrub_monitor->lock);
+    {
+        /* Move the state to BR_SCRUB_STATE_ACTIVE */
+        if (scrub_monitor->state == BR_SCRUB_STATE_PENDING)
+            scrub_monitor->state = BR_SCRUB_STATE_ACTIVE;
+        br_scrubber_log_time(this, "started");
+        priv->scrub_stat.scrub_running = 1;
+    }
+    UNLOCK(&scrub_monitor->lock);
+}
+
+static void
+br_scrubber_exit_control(xlator_t *this)
+{
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    LOCK(&scrub_monitor->lock);
+    {
+        br_scrubber_log_time(this, "finished");
+        priv->scrub_stat.scrub_running = 0;
+
+        if (scrub_monitor->state == BR_SCRUB_STATE_ACTIVE) {
+            (void)br_fsscan_activate(this);
+        } else {
+            UNLOCK(&scrub_monitor->lock);
+            gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO,
+                   "Volume waiting to get rescheduled..");
+            return;
+        }
+    }
+    UNLOCK(&scrub_monitor->lock);
+}
+
+static void
+br_fsscanner_entry_control(xlator_t *this, br_child_t *child)
+{
+    br_fsscanner_log_time(this, child, "started");
+}
+
+static void
+br_fsscanner_exit_control(xlator_t *this, br_child_t *child)
+{
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    if (!_br_is_child_connected(child)) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, BRB_MSG_SCRUB_INFO,
+               "Brick [%s] disconnected while scrubbing. Scrubbing "
+               "might be incomplete",
+               child->brick_path);
+    }
+
+    br_fsscanner_log_time(this, child, "finished");
+
+    pthread_cleanup_push(_br_lock_cleaner, &scrub_monitor->wakelock);
+    pthread_mutex_lock(&scrub_monitor->wakelock);
+    {
+        scrub_monitor->active_child_count--;
+        pthread_cleanup_push(_br_lock_cleaner, &child->lock);
+        pthread_mutex_lock(&child->lock);
+        {
+            br_child_set_scrub_state(child, _gf_false);
+        }
+        pthread_mutex_unlock(&child->lock);
+        pthread_cleanup_pop(0);
+
+        if (scrub_monitor->active_child_count == 0) {
+            /* The last child has finished scrubbing.
+             * Set the kick to false and  wake up other
+             * children who are waiting for the last
+             * child to complete scrubbing.
+             */
+            scrub_monitor->kick = _gf_false;
+            pthread_cond_broadcast(&scrub_monitor->wakecond);
+
+            /* Signal monitor thread waiting for the all
+             * the children to finish scrubbing.
+             */
+            pthread_cleanup_push(_br_lock_cleaner, &scrub_monitor->donelock);
+            pthread_mutex_lock(&scrub_monitor->donelock);
+            {
+                scrub_monitor->done = _gf_true;
+                pthread_cond_signal(&scrub_monitor->donecond);
+            }
+            pthread_mutex_unlock(&scrub_monitor->donelock);
+            pthread_cleanup_pop(0);
+        } else {
+            while (scrub_monitor->active_child_count)
+                pthread_cond_wait(&scrub_monitor->wakecond,
+                                  &scrub_monitor->wakelock);
+        }
+    }
+    pthread_mutex_unlock(&scrub_monitor->wakelock);
+    pthread_cleanup_pop(0);
+}
+
+void *
+br_fsscanner(void *arg)
+{
+    loc_t loc = {
+        0,
+    };
+    br_child_t *child = NULL;
+    xlator_t *this = NULL;
+    struct br_scanfs *fsscan = NULL;
+
+    child = arg;
+    this = child->this;
+    fsscan = &child->fsscan;
+
+    THIS = this;
+    loc.inode = child->table->root;
+
+    while (1) {
+        br_fsscanner_wait_until_kicked(this, child);
+        {
+            /* precursor for scrub */
+            br_fsscanner_entry_control(this, child);
+
+            /* scrub */
+            (void)syncop_ftw(child->xl, &loc, GF_CLIENT_PID_SCRUB, child,
+                             br_fsscanner_handle_entry);
+            if (!list_empty(&fsscan->queued))
+                wait_for_scrubbing(this, fsscan);
+
+            /* scrub exit criteria */
+            br_fsscanner_exit_control(this, child);
+        }
+    }
+
+    return NULL;
+}
+
+/**
+ * Keep this routine extremely simple and do not ever try to acquire
+ * child->lock here: it may lead to deadlock. Scrubber state is
+ * modified in br_fsscanner(). An intermediate state change to pause
+ * changes the scrub state to the _correct_ state by identifying a
+ * non-pending timer.
+ */
+void
+br_kickstart_scanner(struct gf_tw_timer_list *timer, void *data,
+                     unsigned long calltime)
+{
+    xlator_t *this = NULL;
+    struct br_monitor *scrub_monitor = data;
+    br_private_t *priv = NULL;
+
+    THIS = this = scrub_monitor->this;
+    priv = this->private;
+
+    /* Reset scrub statistics */
+    priv->scrub_stat.scrubbed_files = 0;
+    priv->scrub_stat.unsigned_files = 0;
+
+    /* Moves state from PENDING to ACTIVE */
+    (void)br_scrubber_entry_control(this);
+
+    /* kickstart scanning.. */
+    pthread_mutex_lock(&scrub_monitor->wakelock);
+    {
+        scrub_monitor->kick = _gf_true;
+        GF_ASSERT(scrub_monitor->active_child_count == 0);
+        pthread_cond_broadcast(&scrub_monitor->wakecond);
+    }
+    pthread_mutex_unlock(&scrub_monitor->wakelock);
+
+    return;
+}
+
+static uint32_t
+br_fsscan_calculate_delta(uint32_t times)
+{
+    return times;
+}
+
+#define BR_SCRUB_ONDEMAND (1)
+#define BR_SCRUB_MINUTE (60)
+#define BR_SCRUB_HOURLY (60 * 60)
+#define BR_SCRUB_DAILY (1 * 24 * 60 * 60)
+#define BR_SCRUB_WEEKLY (7 * 24 * 60 * 60)
+#define BR_SCRUB_BIWEEKLY (14 * 24 * 60 * 60)
+#define BR_SCRUB_MONTHLY (30 * 24 * 60 * 60)
+
+static unsigned int
+br_fsscan_calculate_timeout(scrub_freq_t freq)
+{
+    uint32_t timo = 0;
+
+    switch (freq) {
+        case BR_FSSCRUB_FREQ_MINUTE:
+            timo = br_fsscan_calculate_delta(BR_SCRUB_MINUTE);
+            break;
+        case BR_FSSCRUB_FREQ_HOURLY:
+            timo = br_fsscan_calculate_delta(BR_SCRUB_HOURLY);
+            break;
+        case BR_FSSCRUB_FREQ_DAILY:
+            timo = br_fsscan_calculate_delta(BR_SCRUB_DAILY);
+            break;
+        case BR_FSSCRUB_FREQ_WEEKLY:
+            timo = br_fsscan_calculate_delta(BR_SCRUB_WEEKLY);
+            break;
+        case BR_FSSCRUB_FREQ_BIWEEKLY:
+            timo = br_fsscan_calculate_delta(BR_SCRUB_BIWEEKLY);
+            break;
+        case BR_FSSCRUB_FREQ_MONTHLY:
+            timo = br_fsscan_calculate_delta(BR_SCRUB_MONTHLY);
+            break;
+        default:
+            timo = 0;
+    }
+
+    return timo;
+}
+
+int32_t
+br_fsscan_schedule(xlator_t *this)
+{
+    uint32_t timo = 0;
+    br_private_t *priv = NULL;
+    char timestr[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    struct br_scrubber *fsscrub = NULL;
+    struct gf_tw_timer_list *timer = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    fsscrub = &priv->fsscrub;
+    scrub_monitor = &priv->scrub_monitor;
+
+    scrub_monitor->boot = gf_time();
+
+    timo = br_fsscan_calculate_timeout(fsscrub->frequency);
+    if (timo == 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_ZERO_TIMEOUT_BUG,
+               "BUG: Zero schedule timeout");
+        goto error_return;
+    }
+
+    scrub_monitor->timer = GF_CALLOC(1, sizeof(*scrub_monitor->timer),
+                                     gf_br_stub_mt_br_scanner_freq_t);
+    if (!scrub_monitor->timer)
+        goto error_return;
+
+    timer = scrub_monitor->timer;
+    INIT_LIST_HEAD(&timer->entry);
+
+    timer->data = scrub_monitor;
+    timer->expires = timo;
+    timer->function = br_kickstart_scanner;
+
+    gf_tw_add_timer(priv->timer_wheel, timer);
+    _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_PENDING);
+
+    gf_time_fmt(timestr, sizeof(timestr), (scrub_monitor->boot + timo),
+                gf_timefmt_FT);
+    gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO,
+           "Scrubbing is "
+           "scheduled to run at %s",
+           timestr);
+
+    return 0;
+
+error_return:
+    return -1;
+}
+
+int32_t
+br_fsscan_activate(xlator_t *this)
+{
+    uint32_t timo = 0;
+    char timestr[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    time_t now = 0;
+    br_private_t *priv = NULL;
+    struct br_scrubber *fsscrub = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    fsscrub = &priv->fsscrub;
+    scrub_monitor = &priv->scrub_monitor;
+
+    now = gf_time();
+    timo = br_fsscan_calculate_timeout(fsscrub->frequency);
+    if (timo == 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_ZERO_TIMEOUT_BUG,
+               "BUG: Zero schedule timeout");
+        return -1;
+    }
+
+    pthread_mutex_lock(&scrub_monitor->donelock);
+    {
+        scrub_monitor->done = _gf_false;
+    }
+    pthread_mutex_unlock(&scrub_monitor->donelock);
+
+    gf_time_fmt(timestr, sizeof(timestr), now + timo, gf_timefmt_FT);
+    (void)gf_tw_mod_timer(priv->timer_wheel, scrub_monitor->timer, timo);
+
+    _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_PENDING);
+    gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO,
+           "Scrubbing is "
+           "rescheduled to run at %s",
+           timestr);
+
+    return 0;
+}
+
+int32_t
+br_fsscan_reschedule(xlator_t *this)
+{
+    int32_t ret = 0;
+    uint32_t timo = 0;
+    char timestr[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    time_t now = 0;
+    br_private_t *priv = NULL;
+    struct br_scrubber *fsscrub = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    fsscrub = &priv->fsscrub;
+    scrub_monitor = &priv->scrub_monitor;
+
+    if (!fsscrub->frequency_reconf)
+        return 0;
+
+    now = gf_time();
+    timo = br_fsscan_calculate_timeout(fsscrub->frequency);
+    if (timo == 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_ZERO_TIMEOUT_BUG,
+               "BUG: Zero schedule timeout");
+        return -1;
+    }
+
+    gf_time_fmt(timestr, sizeof(timestr), now + timo, gf_timefmt_FT);
+
+    pthread_mutex_lock(&scrub_monitor->donelock);
+    {
+        scrub_monitor->done = _gf_false;
+    }
+    pthread_mutex_unlock(&scrub_monitor->donelock);
+
+    ret = gf_tw_mod_timer_pending(priv->timer_wheel, scrub_monitor->timer,
+                                  timo);
+    if (ret == 0)
+        gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO,
+               "Scrubber is currently running and would be "
+               "rescheduled after completion");
+    else {
+        _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_PENDING);
+        gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO,
+               "Scrubbing rescheduled to run at %s", timestr);
+    }
+
+    return 0;
+}
+
+int32_t
+br_fsscan_ondemand(xlator_t *this)
+{
+    int32_t ret = 0;
+    uint32_t timo = 0;
+    char timestr[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    time_t now = 0;
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    now = gf_time();
+    timo = BR_SCRUB_ONDEMAND;
+    gf_time_fmt(timestr, sizeof(timestr), now + timo, gf_timefmt_FT);
+
+    pthread_mutex_lock(&scrub_monitor->donelock);
+    {
+        scrub_monitor->done = _gf_false;
+    }
+    pthread_mutex_unlock(&scrub_monitor->donelock);
+
+    ret = gf_tw_mod_timer_pending(priv->timer_wheel, scrub_monitor->timer,
+                                  timo);
+    if (ret == 0)
+        gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO,
+               "Scrubber is currently running and would be "
+               "rescheduled after completion");
+    else {
+        _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_PENDING);
+        gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO,
+               "Ondemand Scrubbing scheduled to run at %s", timestr);
+    }
+
+    return 0;
+}
+
+#define BR_SCRUB_THREAD_SCALE_LAZY 0
+#define BR_SCRUB_THREAD_SCALE_NORMAL 0.4
+#define BR_SCRUB_THREAD_SCALE_AGGRESSIVE 1.0
+
+#ifndef M_E
+#define M_E 2.718
+#endif
+
+/**
+ * This is just a simple exponential scale to a fixed value selected
+ * per throttle config. We probably need to be more smart and select
+ * the scale based on the number of processor cores too.
+ */
+static unsigned int
+br_scrubber_calc_scale(xlator_t *this, br_private_t *priv,
+                       scrub_throttle_t throttle)
+{
+    unsigned int scale = 0;
+
+    switch (throttle) {
+        case BR_SCRUB_THROTTLE_VOID:
+        case BR_SCRUB_THROTTLE_STALLED:
+            scale = 0;
+            break;
+        case BR_SCRUB_THROTTLE_LAZY:
+            scale = priv->child_count * pow(M_E, BR_SCRUB_THREAD_SCALE_LAZY);
+            break;
+        case BR_SCRUB_THROTTLE_NORMAL:
+            scale = priv->child_count * pow(M_E, BR_SCRUB_THREAD_SCALE_NORMAL);
+            break;
+        case BR_SCRUB_THROTTLE_AGGRESSIVE:
+            scale = priv->child_count *
+                    pow(M_E, BR_SCRUB_THREAD_SCALE_AGGRESSIVE);
+            break;
+        default:
+            gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_UNKNOWN_THROTTLE,
+                   "Unknown throttle %d", throttle);
+    }
+
+    return scale;
+}
+
+static br_child_t *
+_br_scrubber_get_next_child(struct br_scrubber *fsscrub)
+{
+    br_child_t *child = NULL;
+
+    child = list_first_entry(&fsscrub->scrublist, br_child_t, list);
+    list_rotate_left(&fsscrub->scrublist);
+
+    return child;
+}
+
+static void
+_br_scrubber_get_entry(br_child_t *child, struct br_fsscan_entry **fsentry)
+{
+    struct br_scanfs *fsscan = &child->fsscan;
+
+    if (list_empty(&fsscan->ready))
+        return;
+    *fsentry = list_first_entry(&fsscan->ready, struct br_fsscan_entry, list);
+    list_del_init(&(*fsentry)->list);
+}
+
+static void
+_br_scrubber_find_scrubbable_entry(struct br_scrubber *fsscrub,
+                                   struct br_fsscan_entry **fsentry)
+{
+    br_child_t *child = NULL;
+    br_child_t *firstchild = NULL;
+
+    while (1) {
+        while (list_empty(&fsscrub->scrublist))
+            pthread_cond_wait(&fsscrub->cond, &fsscrub->mutex);
+
+        firstchild = NULL;
+        for (child = _br_scrubber_get_next_child(fsscrub); child != firstchild;
+             child = _br_scrubber_get_next_child(fsscrub)) {
+            if (!firstchild)
+                firstchild = child;
+
+            _br_scrubber_get_entry(child, fsentry);
+            if (*fsentry)
+                break;
+        }
+
+        if (*fsentry)
+            break;
+
+        /* nothing to work on.. wait till available */
+        pthread_cond_wait(&fsscrub->cond, &fsscrub->mutex);
+    }
+}
+
+static void
+br_scrubber_pick_entry(struct br_scrubber *fsscrub,
+                       struct br_fsscan_entry **fsentry)
+{
+    pthread_cleanup_push(_br_lock_cleaner, &fsscrub->mutex);
+
+    pthread_mutex_lock(&fsscrub->mutex);
+    {
+        *fsentry = NULL;
+        _br_scrubber_find_scrubbable_entry(fsscrub, fsentry);
+    }
+    pthread_mutex_unlock(&fsscrub->mutex);
+
+    pthread_cleanup_pop(0);
+}
+
+struct br_scrub_entry {
+    gf_boolean_t scrubbed;
+    struct br_fsscan_entry *fsentry;
+};
+
+/**
+ * We need to be a bit careful here. These thread(s) are prone to cancellations
+ * when threads are scaled down (depending on the thottling value configured)
+ * and pausing scrub. A thread can get cancelled while it's waiting for entries
+ * in the ->pending queue or when an object is undergoing scrubbing.
+ */
+static void
+br_scrubber_entry_handle(void *arg)
+{
+    struct br_scanfs *fsscan = NULL;
+    struct br_scrub_entry *sentry = NULL;
+    struct br_fsscan_entry *fsentry = NULL;
+
+    sentry = arg;
+
+    fsentry = sentry->fsentry;
+    fsscan = fsentry->fsscan;
+
+    LOCK(&fsscan->entrylock);
+    {
+        if (sentry->scrubbed) {
+            _br_fsscan_dec_entry_count(fsscan);
+
+            /* cleanup ->entry */
+            fsentry->data = NULL;
+            fsentry->fsscan = NULL;
+            loc_wipe(&fsentry->parent);
+            gf_dirent_entry_free(fsentry->entry);
+
+            GF_FREE(sentry->fsentry);
+        } else {
+            /* (re)queue the entry again for scrub */
+            _br_fsscan_collect_entry(fsscan, sentry->fsentry);
+        }
+    }
+    UNLOCK(&fsscan->entrylock);
+}
+
+static void
+br_scrubber_scrub_entry(xlator_t *this, struct br_fsscan_entry *fsentry)
+{
+    struct br_scrub_entry sentry = {
+        0,
+    };
+
+    sentry.scrubbed = 0;
+    sentry.fsentry = fsentry;
+
+    pthread_cleanup_push(br_scrubber_entry_handle, &sentry);
+    {
+        (void)br_scrubber_scrub_begin(this, fsentry);
+        sentry.scrubbed = 1;
+    }
+    pthread_cleanup_pop(1);
+}
+
+void *
+br_scrubber_proc(void *arg)
+{
+    xlator_t *this = NULL;
+    struct br_scrubber *fsscrub = NULL;
+    struct br_fsscan_entry *fsentry = NULL;
+
+    fsscrub = arg;
+    THIS = this = fsscrub->this;
+
+    while (1) {
+        br_scrubber_pick_entry(fsscrub, &fsentry);
+        br_scrubber_scrub_entry(this, fsentry);
+        sleep(1);
+    }
+
+    return NULL;
+}
+
+static int32_t
+br_scrubber_scale_up(xlator_t *this, struct br_scrubber *fsscrub,
+                     unsigned int v1, unsigned int v2)
+{
+    int i = 0;
+    int32_t ret = -1;
+    int diff = 0;
+    struct br_scrubbers *scrub = NULL;
+
+    diff = (int)(v2 - v1);
+
+    gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCALING_UP_SCRUBBER,
+           "Scaling up scrubbers [%d => %d]", v1, v2);
+
+    for (i = 0; i < diff; i++) {
+        scrub = GF_CALLOC(diff, sizeof(*scrub), gf_br_mt_br_scrubber_t);
+        if (!scrub)
+            break;
+
+        INIT_LIST_HEAD(&scrub->list);
+        ret = gf_thread_create(&scrub->scrubthread, NULL, br_scrubber_proc,
+                               fsscrub, "brsproc");
+        if (ret)
+            break;
+
+        fsscrub->nr_scrubbers++;
+        list_add_tail(&scrub->list, &fsscrub->scrubbers);
+    }
+
+    if ((i != diff) && !scrub)
+        goto error_return;
+
+    if (i != diff) /* degraded scaling.. */
+        gf_msg(this->name, GF_LOG_WARNING, 0, BRB_MSG_SCALE_UP_FAILED,
+               "Could not fully scale up to %d scrubber(s). Spawned "
+               "%d/%d [total scrubber(s): %d]",
+               v2, i, diff, (v1 + i));
+
+    return 0;
+
+error_return:
+    return -1;
+}
+
+static int32_t
+br_scrubber_scale_down(xlator_t *this, struct br_scrubber *fsscrub,
+                       unsigned int v1, unsigned int v2)
+{
+    int i = 0;
+    int diff = 0;
+    int32_t ret = -1;
+    struct br_scrubbers *scrub = NULL;
+
+    diff = (int)(v1 - v2);
+
+    gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCALE_DOWN_SCRUBBER,
+           "Scaling down scrubbers [%d => %d]", v1, v2);
+
+    for (i = 0; i < diff; i++) {
+        scrub = list_first_entry(&fsscrub->scrubbers, struct br_scrubbers,
+                                 list);
+
+        list_del_init(&scrub->list);
+        ret = gf_thread_cleanup_xint(scrub->scrubthread);
+        if (ret)
+            break;
+        GF_FREE(scrub);
+
+        fsscrub->nr_scrubbers--;
+    }
+
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, BRB_MSG_SCALE_DOWN_FAILED,
+               "Could not fully scale down "
+               "to %d scrubber(s). Terminated %d/%d [total "
+               "scrubber(s): %d]",
+               v1, i, diff, (v2 - i));
+        ret = 0;
+    }
+
+    return ret;
+}
+
+static int32_t
+br_scrubber_configure(xlator_t *this, br_private_t *priv,
+                      struct br_scrubber *fsscrub, scrub_throttle_t nthrottle)
+{
+    int32_t ret = 0;
+    unsigned int v1 = 0;
+    unsigned int v2 = 0;
+
+    v1 = fsscrub->nr_scrubbers;
+    v2 = br_scrubber_calc_scale(this, priv, nthrottle);
+
+    if (v1 == v2)
+        return 0;
+
+    if (v1 > v2)
+        ret = br_scrubber_scale_down(this, fsscrub, v1, v2);
+    else
+        ret = br_scrubber_scale_up(this, fsscrub, v1, v2);
+
+    return ret;
+}
+
+static int32_t
+br_scrubber_fetch_option(xlator_t *this, char *opt, dict_t *options,
+                         char **value)
+{
+    if (options)
+        GF_OPTION_RECONF(opt, *value, options, str, error_return);
+    else
+        GF_OPTION_INIT(opt, *value, str, error_return);
+
+    return 0;
+
+error_return:
+    return -1;
+}
+
+/* internal "throttle" override */
+#define BR_SCRUB_STALLED "STALLED"
+
+/* TODO: token buket spec */
+static int32_t
+br_scrubber_handle_throttle(xlator_t *this, br_private_t *priv, dict_t *options,
+                            gf_boolean_t scrubstall)
+{
+    int32_t ret = 0;
+    char *tmp = NULL;
+    struct br_scrubber *fsscrub = NULL;
+    scrub_throttle_t nthrottle = BR_SCRUB_THROTTLE_VOID;
+
+    fsscrub = &priv->fsscrub;
+    fsscrub->throttle_reconf = _gf_false;
+
+    ret = br_scrubber_fetch_option(this, "scrub-throttle", options, &tmp);
+    if (ret)
+        goto error_return;
+
+    if (scrubstall)
+        tmp = BR_SCRUB_STALLED;
+
+    if (strcasecmp(tmp, "lazy") == 0)
+        nthrottle = BR_SCRUB_THROTTLE_LAZY;
+    else if (strcasecmp(tmp, "normal") == 0)
+        nthrottle = BR_SCRUB_THROTTLE_NORMAL;
+    else if (strcasecmp(tmp, "aggressive") == 0)
+        nthrottle = BR_SCRUB_THROTTLE_AGGRESSIVE;
+    else if (strcasecmp(tmp, BR_SCRUB_STALLED) == 0)
+        nthrottle = BR_SCRUB_THROTTLE_STALLED;
+    else
+        goto error_return;
+
+    /* on failure old throttling value is preserved */
+    ret = br_scrubber_configure(this, priv, fsscrub, nthrottle);
+    if (ret)
+        goto error_return;
+
+    if (fsscrub->throttle != nthrottle)
+        fsscrub->throttle_reconf = _gf_true;
+
+    fsscrub->throttle = nthrottle;
+    return 0;
+
+error_return:
+    return -1;
+}
+
+static int32_t
+br_scrubber_handle_stall(xlator_t *this, br_private_t *priv, dict_t *options,
+                         gf_boolean_t *scrubstall)
+{
+    int32_t ret = 0;
+    char *tmp = NULL;
+
+    ret = br_scrubber_fetch_option(this, "scrub-state", options, &tmp);
+    if (ret)
+        goto error_return;
+
+    if (strcasecmp(tmp, "pause") == 0) /* anything else is active */
+        *scrubstall = _gf_true;
+
+    return 0;
+
+error_return:
+    return -1;
+}
+
+static int32_t
+br_scrubber_handle_freq(xlator_t *this, br_private_t *priv, dict_t *options,
+                        gf_boolean_t scrubstall)
+{
+    int32_t ret = -1;
+    char *tmp = NULL;
+    scrub_freq_t frequency = BR_FSSCRUB_FREQ_HOURLY;
+    struct br_scrubber *fsscrub = NULL;
+
+    fsscrub = &priv->fsscrub;
+    fsscrub->frequency_reconf = _gf_true;
+
+    ret = br_scrubber_fetch_option(this, "scrub-freq", options, &tmp);
+    if (ret)
+        goto error_return;
+
+    if (scrubstall)
+        tmp = BR_SCRUB_STALLED;
+
+    if (strcasecmp(tmp, "hourly") == 0) {
+        frequency = BR_FSSCRUB_FREQ_HOURLY;
+    } else if (strcasecmp(tmp, "daily") == 0) {
+        frequency = BR_FSSCRUB_FREQ_DAILY;
+    } else if (strcasecmp(tmp, "weekly") == 0) {
+        frequency = BR_FSSCRUB_FREQ_WEEKLY;
+    } else if (strcasecmp(tmp, "biweekly") == 0) {
+        frequency = BR_FSSCRUB_FREQ_BIWEEKLY;
+    } else if (strcasecmp(tmp, "monthly") == 0) {
+        frequency = BR_FSSCRUB_FREQ_MONTHLY;
+    } else if (strcasecmp(tmp, "minute") == 0) {
+        frequency = BR_FSSCRUB_FREQ_MINUTE;
+    } else if (strcasecmp(tmp, BR_SCRUB_STALLED) == 0) {
+        frequency = BR_FSSCRUB_FREQ_STALLED;
+    } else
+        goto error_return;
+
+    if (fsscrub->frequency == frequency)
+        fsscrub->frequency_reconf = _gf_false;
+    else
+        fsscrub->frequency = frequency;
+
+    return 0;
+
+error_return:
+    return -1;
+}
+
+static void
+br_scrubber_log_option(xlator_t *this, br_private_t *priv,
+                       gf_boolean_t scrubstall)
+{
+    struct br_scrubber *fsscrub = &priv->fsscrub;
+    char *scrub_throttle_str[] = {
+        [BR_SCRUB_THROTTLE_LAZY] = "lazy",
+        [BR_SCRUB_THROTTLE_NORMAL] = "normal",
+        [BR_SCRUB_THROTTLE_AGGRESSIVE] = "aggressive",
+        [BR_SCRUB_THROTTLE_STALLED] = "stalled",
+    };
+
+    char *scrub_freq_str[] = {
+        [0] = "",
+        [BR_FSSCRUB_FREQ_HOURLY] = "hourly",
+        [BR_FSSCRUB_FREQ_DAILY] = "daily",
+        [BR_FSSCRUB_FREQ_WEEKLY] = "weekly",
+        [BR_FSSCRUB_FREQ_BIWEEKLY] = "biweekly",
+        [BR_FSSCRUB_FREQ_MONTHLY] = "monthly (30 days)",
+        [BR_FSSCRUB_FREQ_MINUTE] = "every minute",
+    };
+
+    if (scrubstall)
+        return; /* logged as pause */
+
+    if (fsscrub->frequency_reconf || fsscrub->throttle_reconf) {
+        if (fsscrub->throttle == BR_SCRUB_THROTTLE_VOID)
+            return;
+        gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_TUNABLE,
+               "SCRUB TUNABLES:: [Frequency: %s, Throttle: %s]",
+               scrub_freq_str[fsscrub->frequency],
+               scrub_throttle_str[fsscrub->throttle]);
+    }
+}
+
+int32_t
+br_scrubber_handle_options(xlator_t *this, br_private_t *priv, dict_t *options)
+{
+    int32_t ret = 0;
+    gf_boolean_t scrubstall = _gf_false; /* not as dangerous as it sounds */
+
+    ret = br_scrubber_handle_stall(this, priv, options, &scrubstall);
+    if (ret)
+        goto error_return;
+
+    ret = br_scrubber_handle_throttle(this, priv, options, scrubstall);
+    if (ret)
+        goto error_return;
+
+    ret = br_scrubber_handle_freq(this, priv, options, scrubstall);
+    if (ret)
+        goto error_return;
+
+    br_scrubber_log_option(this, priv, scrubstall);
+
+    return 0;
+
+error_return:
+    return -1;
+}
+
+inode_t *
+br_lookup_bad_obj_dir(xlator_t *this, br_child_t *child, uuid_t gfid)
+{
+    struct iatt statbuf = {
+        0,
+    };
+    inode_table_t *table = NULL;
+    int32_t ret = -1;
+    loc_t loc = {
+        0,
+    };
+    inode_t *linked_inode = NULL;
+    int32_t op_errno = 0;
+
+    GF_VALIDATE_OR_GOTO("bit-rot-scrubber", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+    GF_VALIDATE_OR_GOTO(this->name, child, out);
+
+    table = child->table;
+
+    loc.inode = inode_new(table);
+    if (!loc.inode) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_NO_MEMORY,
+               "failed to allocate a new inode for"
+               "bad object directory");
+        goto out;
+    }
+
+    gf_uuid_copy(loc.gfid, gfid);
+
+    ret = syncop_lookup(child->xl, &loc, &statbuf, NULL, NULL, NULL);
+    if (ret < 0) {
+        op_errno = -ret;
+        gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_LOOKUP_FAILED,
+               "failed to lookup the bad "
+               "objects directory (gfid: %s (%s))",
+               uuid_utoa(gfid), strerror(op_errno));
+        goto out;
+    }
+
+    linked_inode = inode_link(loc.inode, NULL, NULL, &statbuf);
+    if (linked_inode)
+        inode_lookup(linked_inode);
+
+out:
+    loc_wipe(&loc);
+    return linked_inode;
+}
+
+int32_t
+br_read_bad_object_dir(xlator_t *this, br_child_t *child, fd_t *fd,
+                       dict_t *dict)
+{
+    gf_dirent_t entries;
+    gf_dirent_t *entry = NULL;
+    int32_t ret = -1;
+    off_t offset = 0;
+    int32_t count = 0;
+    char key[32] = {
+        0,
+    };
+    dict_t *out_dict = NULL;
+
+    INIT_LIST_HEAD(&entries.list);
+
+    while ((ret = syncop_readdir(child->xl, fd, 131072, offset, &entries, NULL,
+                                 &out_dict))) {
+        if (ret < 0)
+            goto out;
+
+        list_for_each_entry(entry, &entries.list, list)
+        {
+            offset = entry->d_off;
+
+            snprintf(key, sizeof(key), "quarantine-%d", count);
+
+            /*
+             * ignore the dict_set errors for now. The intention is
+             * to get as many bad objects as possible instead of
+             * erroring out at the first failure.
+             */
+            ret = dict_set_dynstr_with_alloc(dict, key, entry->d_name);
+            if (!ret)
+                count++;
+
+            if (out_dict) {
+                dict_copy(out_dict, dict);
+                dict_unref(out_dict);
+                out_dict = NULL;
+            }
+        }
+
+        gf_dirent_free(&entries);
+    }
+
+    ret = count;
+    ret = dict_set_int32_sizen(dict, "count", count);
+
+out:
+    return ret;
+}
+
+int32_t
+br_get_bad_objects_from_child(xlator_t *this, dict_t *dict, br_child_t *child)
+{
+    inode_t *inode = NULL;
+    inode_table_t *table = NULL;
+    fd_t *fd = NULL;
+    int32_t ret = -1;
+    loc_t loc = {
+        0,
+    };
+    int32_t op_errno = 0;
+
+    GF_VALIDATE_OR_GOTO("bit-rot-scrubber", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+    GF_VALIDATE_OR_GOTO(this->name, child, out);
+    GF_VALIDATE_OR_GOTO(this->name, dict, out);
+
+    table = child->table;
+
+    inode = inode_find(table, BR_BAD_OBJ_CONTAINER);
+    if (!inode) {
+        inode = br_lookup_bad_obj_dir(this, child, BR_BAD_OBJ_CONTAINER);
+        if (!inode)
+            goto out;
+    }
+
+    fd = fd_create(inode, 0);
+    if (!fd) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_FD_CREATE_FAILED,
+               "fd creation for the bad "
+               "objects directory failed (gfid: %s)",
+               uuid_utoa(BR_BAD_OBJ_CONTAINER));
+        goto out;
+    }
+
+    loc.inode = inode;
+    gf_uuid_copy(loc.gfid, inode->gfid);
+
+    ret = syncop_opendir(child->xl, &loc, fd, NULL, NULL);
+    if (ret < 0) {
+        op_errno = -ret;
+        fd_unref(fd);
+        fd = NULL;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, BRB_MSG_FD_CREATE_FAILED,
+               "failed to open the bad "
+               "objects directory %s",
+               uuid_utoa(BR_BAD_OBJ_CONTAINER));
+        goto out;
+    }
+
+    fd_bind(fd);
+
+    ret = br_read_bad_object_dir(this, child, fd, dict);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_BAD_OBJ_READDIR_FAIL,
+               "readdir of the bad "
+               "objects directory (%s) failed ",
+               uuid_utoa(BR_BAD_OBJ_CONTAINER));
+        goto out;
+    }
+
+    ret = 0;
+
+out:
+    loc_wipe(&loc);
+    if (fd)
+        fd_unref(fd);
+    return ret;
+}
+
+int32_t
+br_collect_bad_objects_of_child(xlator_t *this, br_child_t *child, dict_t *dict,
+                                dict_t *child_dict, int32_t total_count)
+{
+    int32_t ret = -1;
+    int32_t count = 0;
+    char key[32] = {
+        0,
+    };
+    char main_key[32] = {
+        0,
+    };
+    int32_t j = 0;
+    int32_t tmp_count = 0;
+    char *entry = NULL;
+    char tmp[PATH_MAX] = {
+        0,
+    };
+    char *path = NULL;
+    int32_t len = 0;
+
+    ret = dict_get_int32_sizen(child_dict, "count", &count);
+    if (ret)
+        goto out;
+
+    tmp_count = total_count;
+
+    for (j = 0; j < count; j++) {
+        len = snprintf(key, sizeof(key), "quarantine-%d", j);
+        ret = dict_get_strn(child_dict, key, len, &entry);
+        if (ret)
+            continue;
+
+        ret = dict_get_str(child_dict, entry, &path);
+        len = snprintf(tmp, PATH_MAX, "%s ==> BRICK: %s\n path: %s", entry,
+                       child->brick_path, path);
+        if ((len < 0) || (len >= PATH_MAX)) {
+            continue;
+        }
+        snprintf(main_key, sizeof(main_key), "quarantine-%d", tmp_count);
+
+        ret = dict_set_dynstr_with_alloc(dict, main_key, tmp);
+        if (!ret)
+            tmp_count++;
+        path = NULL;
+    }
+
+    ret = tmp_count;
+
+out:
+    return ret;
+}
+
+int32_t
+br_collect_bad_objects_from_children(xlator_t *this, dict_t *dict)
+{
+    int32_t ret = -1;
+    dict_t *child_dict = NULL;
+    int32_t i = 0;
+    int32_t total_count = 0;
+    br_child_t *child = NULL;
+    br_private_t *priv = NULL;
+    dict_t *tmp_dict = NULL;
+
+    priv = this->private;
+    tmp_dict = dict;
+
+    for (i = 0; i < priv->child_count; i++) {
+        child = &priv->children[i];
+        GF_ASSERT(child);
+        if (!_br_is_child_connected(child))
+            continue;
+
+        child_dict = dict_new();
+        if (!child_dict) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_NO_MEMORY,
+                   "failed to allocate dict");
+            continue;
+        }
+        ret = br_get_bad_objects_from_child(this, child_dict, child);
+        /*
+         * Continue asking the remaining children for the list of
+         * bad objects even though getting the list from one of them
+         * fails.
+         */
+        if (ret) {
+            dict_unref(child_dict);
+            continue;
+        }
+
+        ret = br_collect_bad_objects_of_child(this, child, tmp_dict, child_dict,
+                                              total_count);
+        if (ret < 0) {
+            dict_unref(child_dict);
+            continue;
+        }
+
+        total_count = ret;
+        dict_unref(child_dict);
+        child_dict = NULL;
+    }
+
+    ret = dict_set_int32(tmp_dict, "total-count", total_count);
+
+    return ret;
+}
+
+int32_t
+br_get_bad_objects_list(xlator_t *this, dict_t **dict)
+{
+    int32_t ret = -1;
+    dict_t *tmp_dict = NULL;
+
+    GF_VALIDATE_OR_GOTO("bir-rot-scrubber", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, dict, out);
+
+    tmp_dict = *dict;
+    if (!tmp_dict) {
+        tmp_dict = dict_new();
+        if (!tmp_dict) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_NO_MEMORY,
+                   "failed to allocate dict");
+            goto out;
+        }
+        *dict = tmp_dict;
+    }
+
+    ret = br_collect_bad_objects_from_children(this, tmp_dict);
+
+out:
+    return ret;
+}
+
+static int
+wait_for_scrub_to_finish(xlator_t *this)
+{
+    int ret = -1;
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", scrub_monitor, out);
+    GF_VALIDATE_OR_GOTO("bit-rot", this, out);
+
+    gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO,
+           "Waiting for all children to start and finish scrub");
+
+    pthread_mutex_lock(&scrub_monitor->donelock);
+    {
+        while (!scrub_monitor->done)
+            pthread_cond_wait(&scrub_monitor->donecond,
+                              &scrub_monitor->donelock);
+    }
+    pthread_mutex_unlock(&scrub_monitor->donelock);
+    ret = 0;
+out:
+    return ret;
+}
+
+/**
+ * This function is executed in a separate thread. This is scrubber monitor
+ * thread that takes care of state machine.
+ */
+void *
+br_monitor_thread(void *arg)
+{
+    int32_t ret = 0;
+    xlator_t *this = NULL;
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    this = arg;
+    priv = this->private;
+
+    /*
+     * Since, this is the topmost xlator, THIS has to be set by bit-rot
+     * xlator itself (STACK_WIND won't help in this case). Also it has
+     * to be done for each thread that gets spawned. Otherwise, a new
+     * thread will get global_xlator's pointer when it does "THIS".
+     */
+    THIS = this;
+
+    scrub_monitor = &priv->scrub_monitor;
+
+    pthread_mutex_lock(&scrub_monitor->mutex);
+    {
+        while (!scrub_monitor->inited)
+            pthread_cond_wait(&scrub_monitor->cond, &scrub_monitor->mutex);
+    }
+    pthread_mutex_unlock(&scrub_monitor->mutex);
+
+    /* this needs to be serialized with reconfigure() */
+    pthread_mutex_lock(&priv->lock);
+    {
+        ret = br_scrub_state_machine(this, _gf_false);
+    }
+    pthread_mutex_unlock(&priv->lock);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, BRB_MSG_SSM_FAILED,
+               "Scrub state machine failed");
+        goto out;
+    }
+
+    while (1) {
+        /* Wait for all children to finish scrubbing */
+        ret = wait_for_scrub_to_finish(this);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, BRB_MSG_SCRUB_WAIT_FAILED,
+                   "Scrub wait failed");
+            goto out;
+        }
+
+        /* scrub exit criteria: Move the state to PENDING */
+        br_scrubber_exit_control(this);
+    }
+
+out:
+    return NULL;
+}
+
+static void
+br_set_scrub_state(struct br_monitor *scrub_monitor, br_scrub_state_t state)
+{
+    LOCK(&scrub_monitor->lock);
+    {
+        _br_monitor_set_scrub_state(scrub_monitor, state);
+    }
+    UNLOCK(&scrub_monitor->lock);
+}
+
+int32_t
+br_scrubber_monitor_init(xlator_t *this, br_private_t *priv)
+{
+    struct br_monitor *scrub_monitor = NULL;
+    int ret = 0;
+
+    scrub_monitor = &priv->scrub_monitor;
+
+    LOCK_INIT(&scrub_monitor->lock);
+    scrub_monitor->this = this;
+
+    scrub_monitor->inited = _gf_false;
+    pthread_mutex_init(&scrub_monitor->mutex, NULL);
+    pthread_cond_init(&scrub_monitor->cond, NULL);
+
+    scrub_monitor->kick = _gf_false;
+    scrub_monitor->active_child_count = 0;
+    pthread_mutex_init(&scrub_monitor->wakelock, NULL);
+    pthread_cond_init(&scrub_monitor->wakecond, NULL);
+
+    scrub_monitor->done = _gf_false;
+    pthread_mutex_init(&scrub_monitor->donelock, NULL);
+    pthread_cond_init(&scrub_monitor->donecond, NULL);
+
+    /* Set the state to INACTIVE */
+    br_set_scrub_state(&priv->scrub_monitor, BR_SCRUB_STATE_INACTIVE);
+
+    /* Start the monitor thread */
+    ret = gf_thread_create(&scrub_monitor->thread, NULL, br_monitor_thread,
+                           this, "brmon");
+    if (ret != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, BRB_MSG_SPAWN_FAILED,
+               "monitor thread creation failed");
+        ret = -1;
+        goto err;
+    }
+
+    return 0;
+err:
+    pthread_mutex_destroy(&scrub_monitor->mutex);
+    pthread_cond_destroy(&scrub_monitor->cond);
+
+    pthread_mutex_destroy(&scrub_monitor->wakelock);
+    pthread_cond_destroy(&scrub_monitor->wakecond);
+
+    pthread_mutex_destroy(&scrub_monitor->donelock);
+    pthread_cond_destroy(&scrub_monitor->donecond);
+
+    LOCK_DESTROY(&scrub_monitor->lock);
+
+    return ret;
+}
+
+int32_t
+br_scrubber_init(xlator_t *this, br_private_t *priv)
+{
+    struct br_scrubber *fsscrub = NULL;
+    int ret = 0;
+
+    priv->tbf = tbf_init(NULL, 0);
+    if (!priv->tbf)
+        return -1;
+
+    ret = br_scrubber_monitor_init(this, priv);
+    if (ret)
+        return -1;
+
+    fsscrub = &priv->fsscrub;
+
+    fsscrub->this = this;
+    fsscrub->throttle = BR_SCRUB_THROTTLE_VOID;
+
+    pthread_mutex_init(&fsscrub->mutex, NULL);
+    pthread_cond_init(&fsscrub->cond, NULL);
+
+    fsscrub->nr_scrubbers = 0;
+    INIT_LIST_HEAD(&fsscrub->scrubbers);
+    INIT_LIST_HEAD(&fsscrub->scrublist);
+
+    return 0;
+}
diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-scrub.h b/xlators/features/bit-rot/src/bitd/bit-rot-scrub.h
new file mode 100644
index 00000000000..4e5f67bc021
--- /dev/null
+++ b/xlators/features/bit-rot/src/bitd/bit-rot-scrub.h
@@ -0,0 +1,46 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __BIT_ROT_SCRUB_H__
+#define __BIT_ROT_SCRUB_H__
+
+#include <glusterfs/xlator.h>
+#include "bit-rot.h"
+
+void *
+br_fsscanner(void *);
+
+int32_t
+br_fsscan_schedule(xlator_t *);
+int32_t
+br_fsscan_reschedule(xlator_t *);
+int32_t
+br_fsscan_activate(xlator_t *);
+int32_t
+br_fsscan_deactivate(xlator_t *);
+int32_t
+br_fsscan_ondemand(xlator_t *);
+
+int32_t
+br_scrubber_handle_options(xlator_t *, br_private_t *, dict_t *);
+
+int32_t
+br_scrubber_monitor_init(xlator_t *, br_private_t *);
+
+int32_t
+br_scrubber_init(xlator_t *, br_private_t *);
+
+int32_t
+br_collect_bad_objects_from_children(xlator_t *this, dict_t *dict);
+
+void
+br_child_set_scrub_state(br_child_t *, gf_boolean_t);
+
+#endif /* __BIT_ROT_SCRUB_H__ */
diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-ssm.c b/xlators/features/bit-rot/src/bitd/bit-rot-ssm.c
new file mode 100644
index 00000000000..753e31a3b23
--- /dev/null
+++ b/xlators/features/bit-rot/src/bitd/bit-rot-ssm.c
@@ -0,0 +1,124 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include "bit-rot-ssm.h"
+#include "bit-rot-scrub.h"
+#include "bit-rot-bitd-messages.h"
+
+int
+br_scrub_ssm_noop(xlator_t *this)
+{
+    return 0;
+}
+
+int
+br_scrub_ssm_state_pause(xlator_t *this)
+{
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_GENERIC_SSM_INFO,
+           "Scrubber paused");
+    _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_PAUSED);
+    return 0;
+}
+
+int
+br_scrub_ssm_state_ipause(xlator_t *this)
+{
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_GENERIC_SSM_INFO,
+           "Scrubber paused");
+    _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_IPAUSED);
+    return 0;
+}
+
+int
+br_scrub_ssm_state_active(xlator_t *this)
+{
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    if (scrub_monitor->done) {
+        (void)br_fsscan_activate(this);
+    } else {
+        gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_GENERIC_SSM_INFO,
+               "Scrubbing resumed");
+        _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_ACTIVE);
+    }
+
+    return 0;
+}
+
+int
+br_scrub_ssm_state_stall(xlator_t *this)
+{
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_GENERIC_SSM_INFO,
+           "Volume is under active scrubbing. Pausing scrub..");
+    _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_STALLED);
+    return 0;
+}
+
+static br_scrub_ssm_call *br_scrub_ssm[BR_SCRUB_MAXSTATES][BR_SCRUB_MAXEVENTS] =
+    {
+        /* INACTIVE */
+        {br_fsscan_schedule, br_scrub_ssm_state_ipause, br_scrub_ssm_noop},
+        /* PENDING  */
+        {br_fsscan_reschedule, br_fsscan_deactivate, br_fsscan_ondemand},
+        /* ACTIVE   */
+        {br_scrub_ssm_noop, br_scrub_ssm_state_stall, br_scrub_ssm_noop},
+        /* PAUSED   */
+        {br_fsscan_activate, br_scrub_ssm_noop, br_scrub_ssm_noop},
+        /* IPAUSED  */
+        {br_fsscan_schedule, br_scrub_ssm_noop, br_scrub_ssm_noop},
+        /* STALLED  */
+        {br_scrub_ssm_state_active, br_scrub_ssm_noop, br_scrub_ssm_noop},
+};
+
+int32_t
+br_scrub_state_machine(xlator_t *this, gf_boolean_t scrub_ondemand)
+{
+    br_private_t *priv = NULL;
+    br_scrub_ssm_call *call = NULL;
+    struct br_scrubber *fsscrub = NULL;
+    br_scrub_state_t currstate = 0;
+    br_scrub_event_t event = 0;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    fsscrub = &priv->fsscrub;
+    scrub_monitor = &priv->scrub_monitor;
+
+    currstate = scrub_monitor->state;
+    if (scrub_ondemand)
+        event = BR_SCRUB_EVENT_ONDEMAND;
+    else
+        event = _br_child_get_scrub_event(fsscrub);
+
+    call = br_scrub_ssm[currstate][event];
+    return call(this);
+}
diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-ssm.h b/xlators/features/bit-rot/src/bitd/bit-rot-ssm.h
new file mode 100644
index 00000000000..37b45a42eac
--- /dev/null
+++ b/xlators/features/bit-rot/src/bitd/bit-rot-ssm.h
@@ -0,0 +1,38 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __BIT_ROT_SSM_H__
+#define __BIT_ROT_SSM_H__
+
+#include <glusterfs/xlator.h>
+
+typedef enum br_scrub_state {
+    BR_SCRUB_STATE_INACTIVE = 0,
+    BR_SCRUB_STATE_PENDING,
+    BR_SCRUB_STATE_ACTIVE,
+    BR_SCRUB_STATE_PAUSED,
+    BR_SCRUB_STATE_IPAUSED,
+    BR_SCRUB_STATE_STALLED,
+    BR_SCRUB_MAXSTATES,
+} br_scrub_state_t;
+
+typedef enum br_scrub_event {
+    BR_SCRUB_EVENT_SCHEDULE = 0,
+    BR_SCRUB_EVENT_PAUSE,
+    BR_SCRUB_EVENT_ONDEMAND,
+    BR_SCRUB_MAXEVENTS,
+} br_scrub_event_t;
+
+struct br_monitor;
+
+int32_t
+br_scrub_state_machine(xlator_t *, gf_boolean_t);
+
+#endif /* __BIT_ROT_SSM_H__ */
diff --git a/xlators/features/bit-rot/src/bitd/bit-rot.c b/xlators/features/bit-rot/src/bitd/bit-rot.c
new file mode 100644
index 00000000000..a2f1c343a1d
--- /dev/null
+++ b/xlators/features/bit-rot/src/bitd/bit-rot.c
@@ -0,0 +1,2232 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <ctype.h>
+
+#include <glusterfs/logging.h>
+#include <glusterfs/compat-errno.h>
+
+#include "bit-rot.h"
+#include "bit-rot-scrub.h"
+#include <pthread.h>
+#include "bit-rot-bitd-messages.h"
+
+#define BR_HASH_CALC_READ_SIZE (128 * 1024)
+
+typedef int32_t(br_child_handler)(xlator_t *, br_child_t *);
+
+struct br_child_event {
+    xlator_t *this;
+
+    br_child_t *child;
+
+    br_child_handler *call;
+
+    struct list_head list;
+};
+
+static int
+br_find_child_index(xlator_t *this, xlator_t *child)
+{
+    br_private_t *priv = NULL;
+    int i = -1;
+    int index = -1;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+    GF_VALIDATE_OR_GOTO(this->name, child, out);
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (child == priv->children[i].xl) {
+            index = i;
+            break;
+        }
+    }
+
+out:
+    return index;
+}
+
+br_child_t *
+br_get_child_from_brick_path(xlator_t *this, char *brick_path)
+{
+    br_private_t *priv = NULL;
+    br_child_t *child = NULL;
+    br_child_t *tmp = NULL;
+    int i = 0;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+    GF_VALIDATE_OR_GOTO(this->name, brick_path, out);
+
+    priv = this->private;
+
+    pthread_mutex_lock(&priv->lock);
+    {
+        for (i = 0; i < priv->child_count; i++) {
+            tmp = &priv->children[i];
+            if (!strcmp(tmp->brick_path, brick_path)) {
+                child = tmp;
+                break;
+            }
+        }
+    }
+    pthread_mutex_unlock(&priv->lock);
+
+out:
+    return child;
+}
+
+/**
+ * probably we'll encapsulate brick inside our own structure when
+ * needed -- later.
+ */
+void *
+br_brick_init(void *xl, struct gf_brick_spec *brick)
+{
+    return brick;
+}
+
+/**
+ * and cleanup things here when allocated br_brick_init().
+ */
+void
+br_brick_fini(void *xl, char *brick, void *data)
+{
+    return;
+}
+
+/**
+ * TODO: Signature can contain null terminators which causes bitrot
+ * stub to store truncated hash as it depends on string length of
+ * the hash.
+ *
+ * FIX: Send the string length as part of the signature struct and
+ *      change stub to handle this change.
+ */
+static br_isignature_t *
+br_prepare_signature(const unsigned char *sign, unsigned long hashlen,
+                     int8_t hashtype, br_object_t *object)
+{
+    br_isignature_t *signature = NULL;
+
+    /* TODO: use mem-pool */
+    signature = GF_CALLOC(1, signature_size(hashlen + 1),
+                          gf_br_stub_mt_signature_t);
+    if (!signature)
+        return NULL;
+
+    /* object version */
+    signature->signedversion = object->signedversion;
+
+    /* signature length & type */
+    signature->signaturelen = hashlen;
+    signature->signaturetype = hashtype;
+
+    /* signature itself */
+    memcpy(signature->signature, (char *)sign, hashlen);
+    signature->signature[hashlen + 1] = '\0';
+
+    return signature;
+}
+
+gf_boolean_t
+bitd_is_bad_file(xlator_t *this, br_child_t *child, loc_t *loc, fd_t *fd)
+{
+    int32_t ret = -1;
+    dict_t *xattr = NULL;
+    inode_t *inode = NULL;
+    gf_boolean_t bad_file = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", this, out);
+
+    inode = (loc) ? loc->inode : fd->inode;
+
+    if (fd)
+        ret = syncop_fgetxattr(child->xl, fd, &xattr, BITROT_OBJECT_BAD_KEY,
+                               NULL, NULL);
+    else if (loc)
+        ret = syncop_getxattr(child->xl, loc, &xattr, BITROT_OBJECT_BAD_KEY,
+                              NULL, NULL);
+
+    if (!ret) {
+        gf_msg_debug(this->name, 0, "[GFID: %s] is marked corrupted",
+                     uuid_utoa(inode->gfid));
+        bad_file = _gf_true;
+    }
+
+    if (xattr)
+        dict_unref(xattr);
+
+out:
+    return bad_file;
+}
+
+/**
+ * Do a lookup on the gfid present within the object.
+ */
+static int32_t
+br_object_lookup(xlator_t *this, br_object_t *object, struct iatt *iatt,
+                 inode_t **linked_inode)
+{
+    int ret = -EINVAL;
+    loc_t loc = {
+        0,
+    };
+    inode_t *inode = NULL;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, object, out);
+
+    inode = inode_find(object->child->table, object->gfid);
+
+    if (inode)
+        loc.inode = inode;
+    else
+        loc.inode = inode_new(object->child->table);
+
+    if (!loc.inode) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    gf_uuid_copy(loc.gfid, object->gfid);
+
+    ret = syncop_lookup(object->child->xl, &loc, iatt, NULL, NULL, NULL);
+    if (ret < 0)
+        goto out;
+
+    /*
+     * The file might have been deleted by the application
+     * after getting the event, but before doing a lookup.
+     * So use linked_inode after inode_link is done.
+     */
+    *linked_inode = inode_link(loc.inode, NULL, NULL, iatt);
+    if (*linked_inode)
+        inode_lookup(*linked_inode);
+
+out:
+    loc_wipe(&loc);
+    return ret;
+}
+
+/**
+ * open the object with O_RDONLY flags and return the fd. How to let brick
+ * know that open is being done by bitd because syncop framework does not allow
+ * passing xdata -- may be use frame->root->pid itself.
+ */
+static int32_t
+br_object_open(xlator_t *this, br_object_t *object, inode_t *inode,
+               fd_t **openfd)
+{
+    int32_t ret = -1;
+    fd_t *fd = NULL;
+    loc_t loc = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("bit-rot", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, object, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+
+    ret = -EINVAL;
+    fd = fd_create(inode, 0);
+    if (!fd) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_FD_CREATE_FAILED,
+                "gfid=%s", uuid_utoa(inode->gfid), NULL);
+        goto out;
+    }
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+
+    ret = syncop_open(object->child->xl, &loc, O_RDONLY, fd, NULL, NULL);
+    if (ret) {
+        br_log_object(this, "open", inode->gfid, -ret);
+        fd_unref(fd);
+        fd = NULL;
+    } else {
+        fd_bind(fd);
+        *openfd = fd;
+    }
+
+    loc_wipe(&loc);
+
+out:
+    return ret;
+}
+
+/**
+ * read 128k block from the object @object from the offset @offset
+ * and return the buffer.
+ */
+static int32_t
+br_object_read_block_and_sign(xlator_t *this, fd_t *fd, br_child_t *child,
+                              off_t offset, size_t size, SHA256_CTX *sha256)
+{
+    int32_t ret = -1;
+    tbf_t *tbf = NULL;
+    struct iovec *iovec = NULL;
+    struct iobref *iobref = NULL;
+    br_private_t *priv = NULL;
+    int count = 0;
+    int i = 0;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd->inode, out);
+    GF_VALIDATE_OR_GOTO(this->name, child, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    priv = this->private;
+
+    GF_VALIDATE_OR_GOTO(this->name, priv->tbf, out);
+    tbf = priv->tbf;
+
+    ret = syncop_readv(child->xl, fd, size, offset, 0, &iovec, &count, &iobref,
+                       NULL, NULL, NULL);
+
+    if (ret < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, BRB_MSG_READV_FAILED,
+                "gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+        ret = -1;
+        goto out;
+    }
+
+    if (ret == 0)
+        goto out;
+
+    for (i = 0; i < count; i++) {
+        TBF_THROTTLE_BEGIN(tbf, TBF_OP_HASH, iovec[i].iov_len);
+        {
+            SHA256_Update(sha256, (const unsigned char *)(iovec[i].iov_base),
+                          iovec[i].iov_len);
+        }
+        TBF_THROTTLE_BEGIN(tbf, TBF_OP_HASH, iovec[i].iov_len);
+    }
+
+out:
+    if (iovec)
+        GF_FREE(iovec);
+
+    if (iobref)
+        iobref_unref(iobref);
+
+    return ret;
+}
+
+int32_t
+br_calculate_obj_checksum(unsigned char *md, br_child_t *child, fd_t *fd,
+                          struct iatt *iatt)
+{
+    int32_t ret = -1;
+    off_t offset = 0;
+    size_t block = BR_HASH_CALC_READ_SIZE;
+    xlator_t *this = NULL;
+
+    SHA256_CTX sha256;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", child, out);
+    GF_VALIDATE_OR_GOTO("bit-rot", iatt, out);
+    GF_VALIDATE_OR_GOTO("bit-rot", fd, out);
+
+    this = child->this;
+
+    SHA256_Init(&sha256);
+
+    while (1) {
+        ret = br_object_read_block_and_sign(this, fd, child, offset, block,
+                                            &sha256);
+        if (ret < 0) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_BLOCK_READ_FAILED,
+                    "offset=%" PRIu64, offset, "object-gfid=%s",
+                    uuid_utoa(fd->inode->gfid), NULL);
+            break;
+        }
+
+        if (ret == 0)
+            break;
+
+        offset += ret;
+    }
+
+    if (ret == 0)
+        SHA256_Final(md, &sha256);
+
+out:
+    return ret;
+}
+
+static int32_t
+br_object_checksum(unsigned char *md, br_object_t *object, fd_t *fd,
+                   struct iatt *iatt)
+{
+    return br_calculate_obj_checksum(md, object->child, fd, iatt);
+}
+
+static int32_t
+br_object_read_sign(inode_t *linked_inode, fd_t *fd, br_object_t *object,
+                    struct iatt *iatt)
+{
+    int32_t ret = -1;
+    xlator_t *this = NULL;
+    dict_t *xattr = NULL;
+    unsigned char *md = NULL;
+    br_isignature_t *sign = NULL;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", object, out);
+    GF_VALIDATE_OR_GOTO("bit-rot", linked_inode, out);
+    GF_VALIDATE_OR_GOTO("bit-rot", fd, out);
+
+    this = object->this;
+
+    md = GF_MALLOC(SHA256_DIGEST_LENGTH, gf_common_mt_char);
+    if (!md) {
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_SAVING_HASH_FAILED,
+                "object-gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+        goto out;
+    }
+
+    ret = br_object_checksum(md, object, fd, iatt);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_CALC_CHECKSUM_FAILED,
+                "object-gfid=%s", uuid_utoa(linked_inode->gfid), NULL);
+        goto free_signature;
+    }
+
+    sign = br_prepare_signature(md, SHA256_DIGEST_LENGTH,
+                                BR_SIGNATURE_TYPE_SHA256, object);
+    if (!sign) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_GET_SIGN_FAILED,
+                "object-gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+        goto free_signature;
+    }
+
+    xattr = dict_for_key_value(GLUSTERFS_SET_OBJECT_SIGNATURE, (void *)sign,
+                               signature_size(SHA256_DIGEST_LENGTH), _gf_true);
+
+    if (!xattr) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_SET_SIGN_FAILED,
+                "dict-allocation object-gfid=%s", uuid_utoa(fd->inode->gfid),
+                NULL);
+        goto free_isign;
+    }
+
+    ret = syncop_fsetxattr(object->child->xl, fd, xattr, 0, NULL, NULL);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_SET_SIGN_FAILED,
+                "fsetxattr object-gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+        goto unref_dict;
+    }
+
+    ret = 0;
+
+unref_dict:
+    dict_unref(xattr);
+free_isign:
+    GF_FREE(sign);
+free_signature:
+    GF_FREE(md);
+out:
+    return ret;
+}
+
+static int
+br_object_sign_softerror(int32_t op_errno)
+{
+    return ((op_errno == ENOENT) || (op_errno == ESTALE) ||
+            (op_errno == ENODATA));
+}
+
+void
+br_log_object(xlator_t *this, char *op, uuid_t gfid, int32_t op_errno)
+{
+    int softerror = br_object_sign_softerror(op_errno);
+    if (softerror) {
+        gf_msg_debug(this->name, 0,
+                     "%s() failed on object %s "
+                     "[reason: %s]",
+                     op, uuid_utoa(gfid), strerror(op_errno));
+    } else {
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno, BRB_MSG_OP_FAILED, "op=%s",
+                op, "gfid=%s", uuid_utoa(gfid), NULL);
+    }
+}
+
+void
+br_log_object_path(xlator_t *this, char *op, const char *path, int32_t op_errno)
+{
+    int softerror = br_object_sign_softerror(op_errno);
+    if (softerror) {
+        gf_msg_debug(this->name, 0,
+                     "%s() failed on object %s "
+                     "[reason: %s]",
+                     op, path, strerror(op_errno));
+    } else {
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno, BRB_MSG_OP_FAILED, "op=%s",
+                op, "path=%s", path, NULL);
+    }
+}
+
+static void
+br_trigger_sign(xlator_t *this, br_child_t *child, inode_t *linked_inode,
+                loc_t *loc, gf_boolean_t need_reopen)
+{
+    fd_t *fd = NULL;
+    int32_t ret = -1;
+    uint32_t val = 0;
+    dict_t *dict = NULL;
+    pid_t pid = GF_CLIENT_PID_BITD;
+
+    syncopctx_setfspid(&pid);
+
+    val = (need_reopen == _gf_true) ? BR_OBJECT_REOPEN : BR_OBJECT_RESIGN;
+
+    dict = dict_new();
+    if (!dict)
+        goto out;
+
+    ret = dict_set_uint32(dict, BR_REOPEN_SIGN_HINT_KEY, val);
+    if (ret)
+        goto cleanup_dict;
+
+    ret = -1;
+    fd = fd_create(linked_inode, 0);
+    if (!fd) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_FD_CREATE_FAILED,
+                "gfid=%s", uuid_utoa(linked_inode->gfid), NULL);
+        goto cleanup_dict;
+    }
+
+    ret = syncop_open(child->xl, loc, O_RDWR, fd, NULL, NULL);

path: root/libglusterfs/src/dict.c

/*
  Copyright (c) 2006-2010 Gluster, Inc. <http://www.gluster.com>
  This file is part of GlusterFS.

  GlusterFS is free software; you can redistribute it and/or modify
  it under the terms of the GNU Affero General Public License as published
  by the Free Software Foundation; either version 3 of the License,
  or (at your option) any later version.

  GlusterFS is distributed in the hope that it will be useful, but
  WITHOUT ANY WARRANTY; without even the implied warranty of
  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  Affero General Public License for more details.

  You should have received a copy of the GNU Affero General Public License
  along with this program.  If not, see
  <http://www.gnu.org/licenses/>.
*/

#include <unistd.h>
#include <string.h>
#include <stdlib.h>
#include <stdio.h>
#include <inttypes.h>

#ifndef _CONFIG_H
#define _CONFIG_H
#include "config.h"
#endif

#include "glusterfs.h"
#include "common-utils.h"
#include "dict.h"
#include "hashfn.h"
#include "logging.h"
#include "compat.h"
#include "byte-order.h"

data_pair_t *
get_new_data_pair ()
{
        data_pair_t *data_pair_ptr = NULL;

        data_pair_ptr = (data_pair_t *) GF_CALLOC (1, sizeof (data_pair_t),
                                                   gf_common_mt_data_pair_t);
        return data_pair_ptr;
}

data_t *
get_new_data ()
{
        data_t *data = NULL;

        data = (data_t *) GF_CALLOC (1, sizeof (data_t), gf_common_mt_data_t);
        if (!data) {
                return NULL;
        }

        LOCK_INIT (&data->lock);
        return data;
}

dict_t *
get_new_dict_full (int size_hint)
{
        dict_t *dict = GF_CALLOC (1, sizeof (dict_t), gf_common_mt_dict_t);

        if (!dict) {
                return NULL;
        }

        dict->hash_size = size_hint;
        dict->members = GF_CALLOC (size_hint, sizeof (data_pair_t *),
                                   gf_common_mt_data_pair_t);

        if (!dict->members) {
                GF_FREE (dict);
                return NULL;
        }

        LOCK_INIT (&dict->lock);

        return dict;
}

dict_t *
get_new_dict (void)
{
        return get_new_dict_full (1);
}

dict_t *
dict_new (void)
{
        dict_t *dict = NULL;

        dict = get_new_dict_full(1);

        if (dict)
                dict_ref (dict);

        return dict;
}


int32_t
is_data_equal (data_t *one,
               data_t *two)
{
        /* LOG-TODO */
        if (!one || !two || !one->data || !two->data)
                return 1;

        if (one == two)
                return 1;

        if (one->len != two->len)
                return 0;

        if (one->data == two->data)
                return 1;

        if (memcmp (one->data, two->data, one->len) == 0)
                return 1;

        return 0;
}

void
data_destroy (data_t *data)
{
        if (data) {
                LOCK_DESTROY (&data->lock);

                if (!data->is_static) {
                        if (data->data) {
                                if (data->is_stdalloc)
                                        free (data->data);
                                else
                                        GF_FREE (data->data);
                        }
                        if (data->vec)
                                GF_FREE (data->vec);
                }

                data->len = 0xbabababa;
                if (!data->is_const)
                        GF_FREE (data);
        }
}

data_t *
data_copy (data_t *old)
{
        if (!old) {
                gf_log_callingfn ("dict", GF_LOG_WARNING,
                                  "old is NULL");
                return NULL;
        }

        data_t *newdata = (data_t *) GF_CALLOC (1, sizeof (*newdata),
                                                gf_common_mt_data_t);

        if (!newdata) {
                return NULL;
        }

        if (old) {
                newdata->len = old->len;
                if (old->data) {
                        newdata->data = memdup (old->data, old->len);
                        if (!newdata->data)
                                goto err_out;
                }
                if (old->vec) {
                        newdata->vec = memdup (old->vec, old->len * (sizeof (void *) +
                                                                     sizeof (size_t)));
                        if (!newdata->vec)
                                goto err_out;
                }
        }

        LOCK_INIT (&newdata->lock);
        return newdata;

err_out:

        if (newdata->data)
                FREE (newdata->data);
        if (newdata->vec)
                FREE (newdata->vec);
        GF_FREE (newdata);

        return NULL;
}

static data_pair_t *
_dict_lookup (dict_t *this, char *key)
{
        if (!this || !key) {
                gf_log_callingfn ("dict", GF_LOG_WARNING,
                                  "!this || !key (%s)", key);
                return NULL;
        }

        int hashval = SuperFastHash (key, strlen (key)) % this->hash_size;
        data_pair_t *pair;

        for (pair = this->members[hashval]; pair != NULL; pair = pair->hash_next) {
                if (pair->key && !strcmp (pair->key, key))
                        return pair;
        }

        return NULL;
}


static int32_t
_dict_set (dict_t *this,
           char *key,
           data_t *value)
{
        int hashval;
        data_pair_t *pair;
        char key_free = 0;
        int tmp = 0;
        int ret = 0;

        if (!key) {
                ret = gf_asprintf (&key, "ref:%p", value);
                if (-1 == ret) {
                        gf_log ("dict", GF_LOG_WARNING, "asprintf failed %s", key);
                        return -1;
                }
                key_free = 1;
        }

        tmp = SuperFastHash (key, strlen (key));
        hashval = (tmp % this->hash_size);
        pair = _dict_lookup (this, key);

        if (pair) {
                data_t *unref_data = pair->value;
                pair->value = data_ref (value);
                data_unref (unref_data);
                if (key_free)
                        GF_FREE (key);
                /* Indicates duplicate key */
                return 0;
        }
        pair = (data_pair_t *) GF_CALLOC (1, sizeof (*pair),
                                          gf_common_mt_data_pair_t);
        if (!pair) {
                return -1;
        }

        pair->key = (char *) GF_CALLOC (1, strlen (key) + 1,
                                        gf_common_mt_char);
        if (!pair->key) {
                GF_FREE (pair);

                if (key_free)
                        GF_FREE (key);
                return -1;
        }

        strcpy (pair->key, key);
        pair->value = data_ref (value);

        pair->hash_next = this->members[hashval];
        this->members[hashval] = pair;

        pair->next = this->members_list;
        pair->prev = NULL;
        if (this->members_list)
                this->members_list->prev = pair;
        this->members_list = pair;
        this->count++;

        if (key_free)
                GF_FREE (key);
        return 0;
}

int32_t
dict_set (dict_t *this,
          char *key,
          data_t *value)
{
        int32_t ret;

        if (!this || !value) {
                gf_log_callingfn ("dict", GF_LOG_WARNING,
                                  "!this || !value for key=%s", key);
                return -1;
        }

        LOCK (&this->lock);

        ret = _dict_set (this, key, value);

        UNLOCK (&this->lock);

        return ret;
}


data_t *
dict_get (dict_t *this, char *key)
{
        data_pair_t *pair;

        if (!this || !key) {
                gf_log_callingfn ("dict", GF_LOG_INFO,
                                  "!this || key=%s", (key) ? key : "()");
                return NULL;
        }

        LOCK (&this->lock);

        pair = _dict_lookup (this, key);

        UNLOCK (&this->lock);

        if (pair)
                return pair->value;

        return NULL;
}

void
dict_del (dict_t *this, char *key)
{
        if (!this || !key) {
                gf_log_callingfn ("dict", GF_LOG_WARNING,
                                  "!this || key=%s", key);
                return;
        }

        LOCK (&this->lock);

        int hashval = SuperFastHash (key, strlen (key)) % this->hash_size;
        data_pair_t *pair = this->members[hashval];
        data_pair_t *prev = NULL;

        while (pair) {
                if (strcmp (pair->key, key) == 0) {
                        if (prev)
                                prev->hash_next = pair->hash_next;
                        else
                                this->members[hashval] = pair->hash_next;

                        data_unref (pair->value);

                        if (pair->prev)
                                pair->prev->next = pair->next;
                        else
                                this->members_list = pair->next;

                        if (pair->next)
                                pair->next->prev = pair->prev;

                        GF_FREE (pair->key);
                        GF_FREE (pair);
                        this->count--;
                        break;
                }

                prev = pair;
                pair = pair->hash_next;
        }

        UNLOCK (&this->lock);

        return;
}

void
dict_destroy (dict_t *this)
{
        if (!this) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "dict is NULL");
                return;
        }

        data_pair_t *pair = this->members_list;
        data_pair_t *prev = this->members_list;

        LOCK_DESTROY (&this->lock);

        while (prev) {
                pair = pair->next;
                data_unref (prev->value);
                GF_FREE (prev->key);
                GF_FREE (prev);
                prev = pair;
        }

        GF_FREE (this->members);

        if (this->extra_free)
                GF_FREE (this->extra_free);
        if (this->extra_stdfree)
                free (this->extra_stdfree);

        if (!this->is_static)
                GF_FREE (this);

        return;
}

void
dict_unref (dict_t *this)
{
        int32_t ref;

        if (!this) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "dict is NULL");
                return;
        }

        LOCK (&this->lock);

        this->refcount--;
        ref = this->refcount;

        UNLOCK (&this->lock);

        if (!ref)
                dict_destroy (this);
}

dict_t *
dict_ref (dict_t *this)
{
        if (!this) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "dict is NULL");
                return NULL;
        }

        LOCK (&this->lock);

        this->refcount++;

        UNLOCK (&this->lock);

        return this;
}

void
data_unref (data_t *this)
{
        int32_t ref;

        if (!this) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "dict is NULL");
                return;
        }

        LOCK (&this->lock);

        this->refcount--;
        ref = this->refcount;

        UNLOCK (&this->lock);

        if (!ref)
                data_destroy (this);
}

data_t *
data_ref (data_t *this)
{
        if (!this) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "dict is NULL");
                return NULL;
        }

        LOCK (&this->lock);

        this->refcount++;

        UNLOCK (&this->lock);

        return this;
}

/*
  Serialization format:
  ----
  Count:8
  Key_len:8:Value_len:8
  Key
  Value
  .
  .
  .
*/

int32_t
dict_serialized_length_old (dict_t *this)
{

        if (!this) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "dict is NULL");
                return -1;
        }

        int32_t len = 9; /* count + \n */
        int32_t count = this->count;
        data_pair_t *pair = this->members_list;

        while (count) {
                len += 18;
                len += strlen (pair->key) + 1;
                if (pair->value->vec) {
                        int i;
                        for (i=0; i<pair->value->len; i++) {
                                len += pair->value->vec[i].iov_len;
                        }
                } else {
                        len += pair->value->len;
                }
                pair = pair->next;
                count--;
        }

        return len;
}

int32_t
dict_serialize_old (dict_t *this, char *buf)
{
        if (!this || !buf) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "dict is NULL");
                return -1;
        }

        data_pair_t *pair = this->members_list;
        int32_t count = this->count;
        uint64_t dcount = this->count;

        // FIXME: magic numbers

        sprintf (buf, "%08"PRIx64"\n", dcount);
        buf += 9;
        while (count) {
                uint64_t keylen = strlen (pair->key) + 1;
                uint64_t vallen = pair->value->len;

                sprintf (buf, "%08"PRIx64":%08"PRIx64"\n", keylen, vallen);
                buf += 18;
                memcpy (buf, pair->key, keylen);
                buf += keylen;
                memcpy (buf, pair->value->data, pair->value->len);
                buf += pair->value->len;
                pair = pair->next;
                count--;
        }
        return (0);
}


dict_t *
dict_unserialize_old (char *buf, int32_t size, dict_t **fill)
{
        int32_t ret = 0;
        int32_t cnt = 0;

        if (!buf || !fill || !(*fill)) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "buf is NULL");
                return NULL;
        }

        uint64_t count;
        ret = sscanf (buf, "%"SCNx64"\n", &count);
        (*fill)->count = 0;

        if (!ret){
                gf_log ("dict", GF_LOG_ERROR, "sscanf on buf failed");
                goto err;
        }
        buf += 9;

        if (count == 0) {
                gf_log ("dict", GF_LOG_ERROR, "count == 0");
                goto err;
        }

        for (cnt = 0; cnt < count; cnt++) {
                data_t *value = NULL;
                char *key = NULL;
                uint64_t key_len, value_len;

                ret = sscanf (buf, "%"SCNx64":%"SCNx64"\n", &key_len, &value_len);
                if (ret != 2) {
                        gf_log ("dict", GF_LOG_ERROR,
                                "sscanf for key_len and value_len failed");
                        goto err;
                }
                buf += 18;

                key = buf;
                buf += key_len;

                value = get_new_data ();
                value->len = value_len;
                value->data = buf;
                value->is_static = 1;
                buf += value_len;

                dict_set (*fill, key, value);
        }

        goto ret;

err:
        GF_FREE (*fill);
        *fill = NULL;

ret:
        return *fill;
}


int32_t
dict_iovec_len (dict_t *this)
{
        if (!this) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "dict is NULL");
                return -1;
        }

        int32_t len = 0;
        data_pair_t *pair = this->members_list;

        len++; /* initial header */
        while (pair) {
                len++; /* pair header */
                len++; /* key */

                if (pair->value->vec)
                        len += pair->value->len;
                else
                        len++;
                pair = pair->next;
        }

        return len;
}

int32_t
dict_to_iovec (dict_t *this,
               struct iovec *vec,
               int32_t count)
{
        if (!this || !vec) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "dict is NULL");
                return -1;
        }

        int32_t i = 0;
        data_pair_t *pair = this->members_list;

        vec[0].iov_len = 9;
        if (vec[0].iov_base)
                sprintf (vec[0].iov_base,
                         "%08"PRIx64"\n",
                         (int64_t)this->count);
        i++;

        while (pair) {
                int64_t keylen = strlen (pair->key) + 1;
                int64_t vallen = 0;

                if (pair->value->vec) {
                        int i;

                        for (i=0; i<pair->value->len; i++) {
                                vallen += pair->value->vec[i].iov_len;
                        }
                } else {
                        vallen = pair->value->len;
                }

                vec[i].iov_len = 18;
                if (vec[i].iov_base)
                        sprintf (vec[i].iov_base,
                                 "%08"PRIx64":%08"PRIx64"\n",
                                 keylen,
                                 vallen);
                i++;

                vec[i].iov_len = keylen;
                vec[i].iov_base = pair->key;
                i++;

                if (pair->value->vec) {
                        int k;

                        for (k=0; k<pair->value->len; k++) {
                                vec[i].iov_len = pair->value->vec[k].iov_len;
                                vec[i].iov_base = pair->value->vec[k].iov_base;
                                i++;
                        }
                } else {
                        vec[i].iov_len = pair->value->len;
                        vec[i].iov_base = pair->value->data;
                        i++;
                }

                pair = pair->next;
        }

        return 0;
}

data_t *
int_to_data (int64_t value)
{
        int     ret = 0;
        data_t *data = get_new_data ();

        if (!data) {
                return NULL;
        }

        ret = gf_asprintf (&data->data, "%"PRId64, value);
        if (-1 == ret) {
                gf_log ("dict", GF_LOG_DEBUG, "asprintf failed");
                return NULL;
        }
        data->len = strlen (data->data) + 1;

        return data;
}

data_t *
data_from_int64 (int64_t value)
{
        int     ret = 0;
        data_t *data = get_new_data ();

        if (!data) {
                return NULL;
        }
        ret = gf_asprintf (&data->data, "%"PRId64, value);
        if (-1 == ret) {
                gf_log ("dict", GF_LOG_DEBUG, "asprintf failed");
                return NULL;
        }
        data->len = strlen (data->data) + 1;

        return data;
}

data_t *
data_from_int32 (int32_t value)
{
        int     ret = 0;
        data_t *data = get_new_data ();

        if (!data) {
                return NULL;
        }
        ret = gf_asprintf (&data->data, "%"PRId32, value);
        if (-1 == ret) {
                gf_log ("dict", GF_LOG_DEBUG, "asprintf failed");
                return NULL;
        }

        data->len = strlen (data->data) + 1;

        return data;
}

data_t *
data_from_int16 (int16_t value)
{
        int     ret = 0;
        data_t *data = get_new_data ();

        if (!data) {
                return NULL;
        }
        ret = gf_asprintf (&data->data, "%"PRId16, value);
        if (-1 == ret) {
                gf_log ("dict", GF_LOG_DEBUG, "asprintf failed");
                return NULL;
        }

        data->len = strlen (data->data) + 1;

        return data;
}

data_t *
data_from_int8 (int8_t value)
{
        int     ret = 0;
        data_t *data = get_new_data ();

        if (!data) {
                return NULL;
        }
        ret = gf_asprintf (&data->data, "%d", value);
        if (-1 == ret) {
                gf_log ("dict", GF_LOG_DEBUG, "asprintf failed");
                return NULL;
        }

        data->len = strlen (data->data) + 1;

        return data;
}

data_t *
data_from_uint64 (uint64_t value)
{
        int     ret = 0;
        data_t *data = get_new_data ();

        if (!data) {
                return NULL;
        }
        ret = gf_asprintf (&data->data, "%"PRIu64, value);
        if (-1 == ret) {
                gf_log ("dict", GF_LOG_DEBUG, "asprintf failed");
                return NULL;
        }

        data->len = strlen (data->data) + 1;

        return data;
}

static data_t *
data_from_double (double value)
{
        data_t *data = NULL;
        int     ret  = 0;

        data = get_new_data ();

        if (!data) {
                return NULL;
        }

        ret = gf_asprintf (&data->data, "%f", value);
        if (ret == -1) {
                return NULL;
        }
        data->len = strlen (data->data) + 1;

        return data;
}


data_t *
data_from_uint32 (uint32_t value)
{
        int     ret = 0;
        data_t *data = get_new_data ();

        if (!data) {
                return NULL;
        }
        ret = gf_asprintf (&data->data, "%"PRIu32, value);
        if (-1 == ret) {
                gf_log ("dict", GF_LOG_DEBUG, "asprintf failed");
                return NULL;
        }

        data->len = strlen (data->data) + 1;

        return data;
}


data_t *
data_from_uint16 (uint16_t value)
{
        int     ret = 0;
        data_t *data = get_new_data ();

        if (!data) {
                return NULL;
        }
        ret = gf_asprintf (&data->data, "%"PRIu16, value);
        if (-1 == ret) {
                return NULL;
        }

        data->len = strlen (data->data) + 1;

        return data;
}


data_t *
data_from_ptr (void *value)
{
        if (!value) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "value is NULL");
                return NULL;
        }

        data_t *data = get_new_data ();

        if (!data) {
                return NULL;
        }

        data->data = value;
        return data;
}

data_t *
data_from_static_ptr (void *value)
{
/*
  this is valid to set 0 as value..

  if (!value) {
  gf_log ("dict", GF_LOG_CRITICAL,
  "@value=%p", value);
  return NULL;
  }
*/
        data_t *data = get_new_data ();

        if (!data) {
                return NULL;
        }

        data->is_static = 1;
        data->data = value;

        return data;
}

data_t *
str_to_data (char *value)
{
        if (!value) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "value is NULL");
                return NULL;
        }
        data_t *data = get_new_data ();

        if (!data) {
                return NULL;
        }
        data->len = strlen (value) + 1;

        data->data = value;
        data->is_static = 1;

        return data;
}

data_t *
data_from_dynstr (char *value)
{
        if (!value) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "value is NULL");
                return NULL;
        }

        data_t *data = get_new_data ();

        data->len = strlen (value) + 1;
        data->data = value;

        return data;
}

data_t *
data_from_dynmstr (char *value)
{
        if (!value) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "value is NULL");
                return NULL;
        }

        data_t *data = get_new_data ();

        data->len = strlen (value) + 1;
        data->data = value;
        data->is_stdalloc = 1;

        return data;
}

data_t *
data_from_dynptr (void *value, int32_t len)
{
        data_t *data = get_new_data ();

        if (!data)
                return NULL;

        data->len = len;
        data->data = value;

        return data;
}

data_t *
bin_to_data (void *value, int32_t len)
{
        if (!value) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "value is NULL");
                return NULL;
        }

        data_t *data = get_new_data ();

        if (!data)
                return NULL;

        data->is_static = 1;
        data->len = len;
        data->data = value;

        return data;
}

int64_t
data_to_int64 (data_t *data)
{
        if (!data) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "data is NULL");
                return -1;
        }

        char *str = alloca (data->len + 1);
        if (!str)
                return -1;

        memcpy (str, data->data, data->len);
        str[data->len] = '\0';
        return (int64_t) strtoull (str, NULL, 0);
}

int32_t
data_to_int32 (data_t *data)
{
        if (!data) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "data is NULL");
                return -1;
        }

        char *str = alloca (data->len + 1);
        if (!str)
                return -1;

        memcpy (str, data->data, data->len);
        str[data->len] = '\0';

        return strtoul (str, NULL, 0);
}

int16_t
data_to_int16 (data_t *data)
{
        if (!data) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "data is NULL");
                return -1;
        }

        char *str = alloca (data->len + 1);
        if (!str)
                return -1;

        memcpy (str, data->data, data->len);
        str[data->len] = '\0';

        return strtol (str, NULL, 0);
}


int8_t
data_to_int8 (data_t *data)
{
        if (!data) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "data is NULL");
                return -1;
        }

        char *str = alloca (data->len + 1);
        if (!str)
                return -1;

        memcpy (str, data->data, data->len);
        str[data->len] = '\0';

        return (int8_t)strtol (str, NULL, 0);
}


uint64_t
data_to_uint64 (data_t *data)
{
        if (!data)
                return -1;
        char *str = alloca (data->len + 1);
        if (!str)
                return -1;

        memcpy (str, data->data, data->len);
        str[data->len] = '\0';

        return strtoll (str, NULL, 0);
}

uint32_t
data_to_uint32 (data_t *data)
{
        if (!data)
                return -1;

        char *str = alloca (data->len + 1);
        if (!str)
                return -1;

        memcpy (str, data->data, data->len);
        str[data->len] = '\0';

        return strtol (str, NULL, 0);
}

uint16_t
data_to_uint16 (data_t *data)
{
        if (!data)
                return -1;

        char *str = alloca (data->len + 1);
        if (!str)
                return -1;

        memcpy (str, data->data, data->len);
        str[data->len] = '\0';

        return strtol (str, NULL, 0);
}

char *
data_to_str (data_t *data)
{
        if (!data) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "data is NULL");
                return NULL;
        }
        return data->data;
}

void *
data_to_ptr (data_t *data)
{
        if (!data) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "data is NULL");
                return NULL;
        }
        return data->data;
}

void *
data_to_bin (data_t *data)
{
        if (!data) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "data is NULL");
                return NULL;
        }
        return data->data;
}

void
dict_foreach (dict_t *dict,
              void (*fn)(dict_t *this,
                         char *key,
                         data_t *value,
                         void *data),
              void *data)
{
        if (!dict) {
                gf_log_callingfn ("dict", GF_LOG_WARNING,
                                  "dict is NULL");
                return;
        }

        data_pair_t *pairs = dict->members_list;
        data_pair_t *next = NULL;

        while (pairs) {
                next = pairs->next;
                fn (dict, pairs->key, pairs->value, data);
                pairs = next;
        }
}


static void
_copy (dict_t *unused,
       char *key,
       data_t *value,
       void *newdict)
{
        dict_set ((dict_t *)newdict, key, (value));
}


dict_t *
dict_copy (dict_t *dict,
           dict_t *new)
{
        if (!dict) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "dict is NULL");
                return NULL;
        }

        if (!new)
                new = get_new_dict_full (dict->hash_size);

        dict_foreach (dict, _copy, new);

        return new;
}

dict_t *
dict_copy_with_ref (dict_t *dict,
                    dict_t *new)
{
        dict_t *local_new = NULL;

        GF_VALIDATE_OR_GOTO("dict", dict, fail);

        if (new == NULL) {
                local_new = dict_new ();
                GF_VALIDATE_OR_GOTO("dict", local_new, fail);
                new = local_new;
        }

        dict_foreach (dict, _copy, new);
fail:
        return new;
}

/*
 * !!!!!!! CLEANED UP CODE !!!!!!!
 */

/**
 * Common cleaned up interface:
 *
 * Return value:  0   success
 *               -val error, val = errno
 */


static int
dict_get_with_ref (dict_t *this, char *key, data_t **data)
{
        data_pair_t * pair = NULL;
        int           ret  = -ENOENT;

        if (!this || !key || !data) {
                gf_log_callingfn ("dict", GF_LOG_WARNING,
                                  "dict OR key (%s) is NULL", key);
                ret = -EINVAL;
                goto err;
        }

        LOCK (&this->lock);
        {
                pair = _dict_lookup (this, key);
        }
        UNLOCK (&this->lock);

        if (pair) {
                ret = 0;
                *data = data_ref (pair->value);
        }

err:
        return ret;
}

static int
_data_to_ptr (data_t *data, void **val)
{
        int ret = 0;

        if (!data) {
                ret = -EINVAL;
                goto err;
        }

        *val = data->data;
err:
        return ret;
}


static int
_data_to_int8 (data_t *data, int8_t *val)
{
        int    ret = 0;
        char * str = NULL;

        if (!data || !val) {
                ret = -EINVAL;
                goto err;
        }

        str = alloca (data->len + 1);
        if (!str) {
                ret = -ENOMEM;
                goto err;
        }
        memcpy (str, data->data, data->len);
        str[data->len] = '\0';

        errno = 0;
        *val = strtol (str, NULL, 0);
        if (errno != 0)
                ret = -errno;

err:
        return ret;
}

static int
_data_to_int16 (data_t *data, int16_t *val)
{
        int    ret = 0;
        char * str = NULL;

        if (!data || !val) {
                ret = -EINVAL;
                goto err;
        }

        str = alloca (data->len + 1);
        if (!str) {
                ret = -ENOMEM;
                goto err;
        }
        memcpy (str, data->data, data->len);
        str[data->len] = '\0';

        errno = 0;
        *val = strtol (str, NULL, 0);
        if (errno != 0)
                ret = -errno;

err:
        return ret;
}

static int
_data_to_int32 (data_t *data, int32_t *val)
{
        int    ret = 0;
        char * str = NULL;

        if (!data || !val) {
                ret = -EINVAL;
                goto err;
        }

        str = alloca (data->len + 1);
        if (!str) {
                ret = -ENOMEM;
                goto err;
        }
        memcpy (str, data->data, data->len);
        str[data->len] = '\0';

        errno = 0;
        *val = strtol (str, NULL, 0);
        if (errno != 0)
                ret = -errno;

err:
        return ret;
}

static int
_data_to_int64 (data_t *data, int64_t *val)
{
        int    ret = 0;
        char * str = NULL;

        if (!data || !val) {
                ret = -EINVAL;
                goto err;
        }

        str = alloca (data->len + 1);
        if (!str) {
                ret = -ENOMEM;
                goto err;
        }
        memcpy (str, data->data, data->len);
        str[data->len] = '\0';

        errno = 0;
        *val = strtoll (str, NULL, 0);
        if (errno != 0)
                ret = -errno;

err:
        return ret;
}

static int
_data_to_uint16 (data_t *data, uint16_t *val)
{
        int    ret = 0;
        char * str = NULL;

        if (!data || !val) {
                ret = -EINVAL;
                goto err;
        }

        str = alloca (data->len + 1);
        if (!str) {
                ret = -ENOMEM;
                goto err;
        }
        memcpy (str, data->data, data->len);
        str[data->len] = '\0';

        errno = 0;
        *val = strtoul (str, NULL, 0);
        if (errno != 0)
                ret = -errno;

err:
        return ret;
}

static int
_data_to_uint32 (data_t *data, uint32_t *val)
{
        int    ret = 0;
        char * str = NULL;

        if (!data || !val) {
                ret = -EINVAL;
                goto err;
        }

        str = alloca (data->len + 1);
        if (!str) {
                ret = -ENOMEM;
                goto err;
        }
        memcpy (str, data->data, data->len);
        str[data->len] = '\0';

        errno = 0;
        *val = strtoul (str, NULL, 0);
        if (errno != 0)
                ret = -errno;

err:
        return ret;
}

static int
_data_to_uint64 (data_t *data, uint64_t *val)
{
        int    ret = 0;
        char * str = NULL;

        if (!data || !val) {
                ret = -EINVAL;
                goto err;
        }

        str = alloca (data->len + 1);
        if (!str) {
                ret = -ENOMEM;
                goto err;
        }
        memcpy (str, data->data, data->len);
        str[data->len] = '\0';

        errno = 0;
        *val = strtoull (str, NULL, 0);
        if (errno != 0)
                ret = -errno;

err:
        return ret;
}

static int
_data_to_double (data_t *data, double *val)
{
        int    ret = 0;
        char * str = NULL;

        if (!data || !val) {
                ret = -EINVAL;
                goto err;
        }

        str = alloca (data->len + 1);
        if (!str) {
                ret = -ENOMEM;
                goto err;
        }
        memcpy (str, data->data, data->len);
        str[data->len] = '\0';

        errno = 0;
        *val = strtod (str, NULL);
        if (errno != 0)
                ret = -errno;

err:
        return ret;
}

int
dict_get_int8 (dict_t *this, char *key, int8_t *val)
{
        data_t * data = NULL;
        int      ret  = 0;

        if (!this || !key || !val) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_get_with_ref (this, key, &data);
        if (ret != 0) {
                goto err;
        }

        ret = _data_to_int8 (data, val);

err:
        if (data)
                data_unref (data);
        return ret;
}


int
dict_set_int8 (dict_t *this, char *key, int8_t val)
{
        data_t * data = NULL;
        int      ret  = 0;

        data = data_from_int8 (val);
        if (!data) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_set (this, key, data);

err:
        return ret;
}

int
dict_get_int16 (dict_t *this, char *key, int16_t *val)
{
        data_t * data = NULL;
        int      ret  = 0;

        if (!this || !key || !val) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_get_with_ref (this, key, &data);
        if (ret != 0) {
                goto err;
        }

        ret = _data_to_int16 (data, val);

err:
        if (data)
                data_unref (data);
        return ret;
}


int
dict_set_int16 (dict_t *this, char *key, int16_t val)
{
        data_t * data = NULL;
        int      ret  = 0;

        data = data_from_int16 (val);
        if (!data) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_set (this, key, data);

err:
        return ret;
}

int
dict_get_int32 (dict_t *this, char *key, int32_t *val)
{
        data_t * data = NULL;
        int      ret  = 0;

        if (!this || !key || !val) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_get_with_ref (this, key, &data);
        if (ret != 0) {
                goto err;
        }

        ret = _data_to_int32 (data, val);

err:
        if (data)
                data_unref (data);
        return ret;
}


int
dict_set_int32 (dict_t *this, char *key, int32_t val)
{
        data_t * data = NULL;
        int      ret  = 0;

        data = data_from_int32 (val);
        if (!data) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_set (this, key, data);

err:
        return ret;
}

int
dict_get_int64 (dict_t *this, char *key, int64_t *val)
{
        data_t * data = NULL;
        int      ret  = 0;

        if (!this || !key || !val) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_get_with_ref (this, key, &data);
        if (ret != 0) {
                goto err;
        }

        ret = _data_to_int64 (data, val);

err:
        if (data)
                data_unref (data);
        return ret;
}


int
dict_set_int64 (dict_t *this, char *key, int64_t val)
{
        data_t * data = NULL;
        int      ret  = 0;

        data = data_from_int64 (val);
        if (!data) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_set (this, key, data);

err:
        return ret;
}

int
dict_get_uint16 (dict_t *this, char *key, uint16_t *val)
{
        data_t * data = NULL;
        int      ret  = 0;

        if (!this || !key || !val) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_get_with_ref (this, key, &data);
        if (ret != 0) {
                goto err;
        }

        ret = _data_to_uint16 (data, val);

err:
        if (data)
                data_unref (data);
        return ret;
}


int
dict_set_uint16 (dict_t *this, char *key, uint16_t val)
{
        data_t * data = NULL;
        int      ret  = 0;

        data = data_from_uint16 (val);
        if (!data) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_set (this, key, data);

err:
        return ret;
}

int
dict_get_uint32 (dict_t *this, char *key, uint32_t *val)
{
        data_t * data = NULL;
        int      ret  = 0;

        if (!this || !key || !val) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_get_with_ref (this, key, &data);
        if (ret != 0) {
                goto err;
        }

        ret = _data_to_uint32 (data, val);

err:
        if (data)
                data_unref (data);
        return ret;
}



int
dict_set_uint32 (dict_t *this, char *key, uint32_t val)
{
        data_t * data = NULL;
        int      ret  = 0;

        data = data_from_uint32 (val);
        if (!data) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_set (this, key, data);

err:
        return ret;
}

int
dict_get_uint64 (dict_t *this, char *key, uint64_t *val)
{
        data_t * data = NULL;
        int      ret  = 0;

        if (!this || !key || !val) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_get_with_ref (this, key, &data);
        if (ret != 0) {
                goto err;
        }

        ret = _data_to_uint64 (data, val);

err:
        if (data)
                data_unref (data);
        return ret;
}


int
dict_set_uint64 (dict_t *this, char *key, uint64_t val)
{
        data_t * data = NULL;
        int      ret  = 0;

        data = data_from_uint64 (val);
        if (!data) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_set (this, key, data);

err:
        return ret;
}

int
dict_get_double (dict_t *this, char *key, double *val)
{
        data_t *data = NULL;
        int     ret  = 0;

        if (!this || !key || !val) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_get_with_ref (this, key, &data);
        if (ret != 0) {
                goto err;
        }

        ret = _data_to_double (data, val);

err:
        if (data)
                data_unref (data);
        return ret;
}

int
dict_set_double (dict_t *this, char *key, double val)
{
        data_t * data = NULL;
        int      ret  = 0;

        data = data_from_double (val);
        if (!data) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_set (this, key, data);

err:
        return ret;
}

int
dict_set_static_ptr (dict_t *this, char *key, void *ptr)
{
        data_t * data = NULL;
        int      ret  = 0;

        data = data_from_static_ptr (ptr);
        if (!data) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_set (this, key, data);

err:
        return ret;
}

int
dict_set_dynptr (dict_t *this, char *key, void *ptr, size_t len)
{
        data_t * data = NULL;
        int      ret  = 0;

        data = data_from_dynptr (ptr, len);
        if (!data) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_set (this, key, data);

err:
        return ret;
}

int
dict_get_ptr (dict_t *this, char *key, void **ptr)
{
        data_t * data = NULL;
        int      ret  = 0;

        if (!this || !key || !ptr) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_get_with_ref (this, key, &data);
        if (ret != 0) {
                goto err;
        }

        ret = _data_to_ptr (data, ptr);
        if (ret != 0) {
                goto err;
        }

err:
        if (data)
                data_unref (data);

        return ret;
}

int
dict_set_ptr (dict_t *this, char *key, void *ptr)
{
        data_t * data = NULL;
        int      ret  = 0;

        data = data_from_ptr (ptr);
        if (!data) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_set (this, key, data);

err:
        return ret;
}


int
dict_get_str (dict_t *this, char *key, char **str)
{
        data_t * data = NULL;
        int      ret  = -EINVAL;

        if (!this || !key || !str) {
                goto err;
        }

        ret = dict_get_with_ref (this, key, &data);
        if (ret < 0) {
                goto err;
        }

        if (!data || !data->data) {
                goto err;
        }
        *str = data->data;

err:
        if (data)
                data_unref (data);

        return ret;
}

int
dict_set_str (dict_t *this, char *key, char *str)
{
        data_t * data = NULL;
        int      ret  = 0;

        data = str_to_data (str);
        if (!data) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_set (this, key, data);

err:
        return ret;
}

int
dict_set_dynstr (dict_t *this, char *key, char *str)
{
        data_t * data = NULL;
        int      ret  = 0;

        data = data_from_dynstr (str);
        if (!data) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_set (this, key, data);

err:
        return ret;
}

/*
  for malloced strings we should do a free instead of GF_FREE
*/
int
dict_set_dynmstr (dict_t *this, char *key, char *str)
{
        data_t * data = NULL;
        int      ret  = 0;

        data = data_from_dynmstr (str);
        if (!data) {
                ret = -EINVAL;
                goto err;
        }

        ret = dict_set (this, key, data);

err:
        return ret;
}


int
dict_get_bin (dict_t *this, char *key, void **bin)
{
        data_t * data = NULL;
        int      ret  = -EINVAL;

        if (!this || !key || !bin) {
                goto err;
        }

        ret = dict_get_with_ref (this, key, &data);
        if (ret < 0) {
                goto err;
        }

        if (!data || !data->data) {
                goto err;
        }
        *bin = data->data;

err:
        if (data)
                data_unref (data);

        return ret;
}


int
dict_set_bin (dict_t *this, char *key, void *ptr, size_t size)
{
        data_t * data = NULL;
        int      ret  = 0;

        if (!ptr || (size < 0)) {
                ret = -EINVAL;
                goto err;
        }

        data = bin_to_data (ptr, size);
        if (!data) {
                ret = -EINVAL;
                goto err;
        }

        data->data = ptr;
        data->len  = size;
        data->is_static = 0;

        ret = dict_set (this, key, data);

err:
        return ret;
}


int
dict_set_static_bin (dict_t *this, char *key, void *ptr, size_t size)
{
        data_t * data = NULL;
        int      ret  = 0;

        if (!ptr || (size < 0)) {
                ret = -EINVAL;
                goto err;
        }

        data = bin_to_data (ptr, size);
        if (!data) {
                ret = -EINVAL;
                goto err;
        }

        data->data = ptr;
        data->len  = size;
        data->is_static = 1;

        ret = dict_set (this, key, data);

err:
        return ret;
}


/**
 * dict_get_str_boolean - get a boolean value based on string representation.
 *
 * @this        : dictionary
 * @key         : dictionary key queried
 * @default_val : default value if key not found
 *
 * @return      : @default_val if key not found
 *              : boolean interpretation of @this[@key] if it makes sense
 *                (ie., "on", "true", "enable" ...)
 *              : -1 if error occurs or @this[@key] doesn't make sens as
 *                  boolean
 *
 *   So if you query a boolean option, then via @default_val you can choose
 *   between following patterns:
 *
 *   - fall back to _gf_false if @key is not set  [@default_val = 0]
 *   - fall back to _gf_true if @key is not set   [@default_val = 1]
 *   - regard as failure if @key is not set       [@default_val = -1]
 *   - handle specially (not as error) if @key is not set
 *                                                [@default_val = anything else]
 */

int
dict_get_str_boolean (dict_t *this, char *key, int default_val)
{
        data_t       *data = NULL;
        gf_boolean_t  boo = _gf_false;
        int           ret  = 0;

        ret = dict_get_with_ref (this, key, &data);
        if (ret < 0) {
                if (ret == -ENOENT)
                        ret = default_val;
                else
                        ret = -1;
                goto err;
        }

        GF_ASSERT (data);

        if (!data->data) {
                ret = -1;
                goto err;
        }

        ret = gf_string2boolean (data->data, &boo);
        if (ret == -1)
                goto err;

        ret = boo;

err:
        if (data)
                data_unref (data);

        return ret;
}


/**
 * Serialization format:
 *  -------- --------  --------  ----------- -------------
 * |  count | key len | val len | key     \0| value
 *  ---------------------------------------- -------------
 *     4        4         4       <key len>   <value len>
 */

#define DICT_HDR_LEN               4
#define DICT_DATA_HDR_KEY_LEN      4
#define DICT_DATA_HDR_VAL_LEN      4

/**
 * _dict_serialized_length - return the length of serialized dict. This
 *                           procedure has to be called with this->lock held.
 *
 * @this  : dict to be serialized
 * @return: success: len
 *        : failure: -errno
 */

int
_dict_serialized_length (dict_t *this)
{
        int ret            = -EINVAL;
        int count          = 0;
        int len            = 0;
        int i              = 0;
        data_pair_t * pair = NULL;

        len = DICT_HDR_LEN;
        count = this->count;

        if (count < 0) {
                gf_log ("dict", GF_LOG_ERROR, "count (%d) < 0!", count);
                goto out;
        }

        pair = this->members_list;

        while (count) {
                if (!pair) {
                        gf_log ("dict", GF_LOG_ERROR,
                                "less than count data pairs found!");
                        goto out;
                }

                len += DICT_DATA_HDR_KEY_LEN + DICT_DATA_HDR_VAL_LEN;

                if (!pair->key) {
                        gf_log ("dict", GF_LOG_ERROR, "pair->key is null!");
                        goto out;
                }

                len += strlen (pair->key) + 1  /* for '\0' */;

                if (!pair->value) {
                        gf_log ("dict", GF_LOG_ERROR,
                                "pair->value is null!");
                        goto out;
                }

                if (pair->value->vec) {
                        for (i = 0; i < pair->value->len; i++) {
                                if (pair->value->vec[i].iov_len < 0) {
                                        gf_log ("dict", GF_LOG_ERROR,
                                                "iov_len (%"GF_PRI_SIZET") < 0!",
                                                pair->value->vec[i].iov_len);
                                        goto out;
                                }

                                len += pair->value->vec[i].iov_len;
                        }
                } else {
                        if (pair->value->len < 0) {
                                gf_log ("dict", GF_LOG_ERROR,
                                        "value->len (%d) < 0",
                                        pair->value->len);
                                goto out;
                        }

                        len += pair->value->len;
                }

                pair = pair->next;
                count--;
        }

        ret = len;
out:
        return ret;
}

/**
 * _dict_serialize - serialize a dictionary into a buffer. This procedure has
 *                   to be called with this->lock held.
 *
 * @this: dict to serialize
 * @buf:  buffer to serialize into. This must be
 *        atleast dict_serialized_length (this) large
 *
 * @return: success: 0
 *          failure: -errno
 */

int
_dict_serialize (dict_t *this, char *buf)
{
        int           ret     = -1;
        data_pair_t * pair    = NULL;
        int32_t       count   = 0;
        int32_t       keylen  = 0;
        int32_t       vallen  = 0;
        int32_t       netword = 0;


        if (!buf) {
                gf_log ("dict", GF_LOG_ERROR,
                        "buf is null!");
                goto out;
        }


        count = this->count;
        if (count < 0) {
                gf_log ("dict", GF_LOG_ERROR, "count (%d) < 0!", count);
                goto out;
        }

        netword = hton32 (count);
        memcpy (buf, &netword, sizeof(netword));
        buf += DICT_HDR_LEN;
        pair = this->members_list;

        while (count) {
                if (!pair) {
                        gf_log ("dict", GF_LOG_ERROR,
                                "less than count data pairs found!");
                        goto out;
                }

                if (!pair->key) {
                        gf_log ("dict", GF_LOG_ERROR,
                                "pair->key is null!");
                        goto out;
                }

                keylen  = strlen (pair->key);
                netword = hton32 (keylen);
                memcpy (buf, &netword, sizeof(netword));
                buf += DICT_DATA_HDR_KEY_LEN;

                if (!pair->value) {
                        gf_log ("dict", GF_LOG_ERROR,
                                "pair->value is null!");
                        goto out;
                }

                vallen  = pair->value->len;
                netword = hton32 (vallen);
                memcpy (buf, &netword, sizeof(netword));
                buf += DICT_DATA_HDR_VAL_LEN;

                memcpy (buf, pair->key, keylen);
                buf += keylen;
                *buf++ = '\0';

                if (!pair->value->data) {
                        gf_log ("dict", GF_LOG_ERROR,
                                "pair->value->data is null!");
                        goto out;
                }
                memcpy (buf, pair->value->data, vallen);
                buf += vallen;

                pair = pair->next;
                count--;
        }

        ret = 0;
out:
        return ret;
}


/**
 * dict_serialized_length - return the length of serialized dict
 *
 * @this:   dict to be serialized
 * @return: success: len
 *        : failure: -errno
 */

int
dict_serialized_length (dict_t *this)
{
        int ret            = -EINVAL;

        if (!this) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "dict is null!");
                goto out;
        }

        LOCK (&this->lock);
        {
                ret = _dict_serialized_length (this);
        }
        UNLOCK (&this->lock);

out:
        return ret;
}

/**
 * dict_serialize - serialize a dictionary into a buffer
 *
 * @this: dict to serialize
 * @buf:  buffer to serialize into. This must be
 *        atleast dict_serialized_length (this) large
 *
 * @return: success: 0
 *          failure: -errno
 */

int
dict_serialize (dict_t *this, char *buf)
{
        int           ret    = -1;

        if (!this || !buf) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "dict is null!");
                goto out;
        }

        LOCK (&this->lock);
        {
                ret = _dict_serialize (this, buf);
        }
        UNLOCK (&this->lock);
out:
        return ret;
}


/**
 * dict_unserialize - unserialize a buffer into a dict
 *
 * @buf:  buf containing serialized dict
 * @size: size of the @buf
 * @fill: dict to fill in
 *
 * @return: success: 0
 *          failure: -errno
 */

int32_t
dict_unserialize (char *orig_buf, int32_t size, dict_t **fill)
{
        char   *buf = NULL;
        int     ret   = -1;
        int32_t count = 0;
        int     i     = 0;

        data_t * value   = NULL;
        char   * key     = NULL;
        int32_t  keylen  = 0;
        int32_t  vallen  = 0;
        int32_t  hostord = 0;

        buf = orig_buf;

        if (!buf) {
                gf_log_callingfn ("dict", GF_LOG_WARNING, "buf is null!");
                goto out;
        }

        if (size == 0) {
                gf_log_callingfn ("dict", GF_LOG_ERROR,
                        "size is 0!");
                goto out;
        }

        if (!fill) {
                gf_log_callingfn ("dict", GF_LOG_ERROR,
                        "fill is null!");
                goto out;
        }

        if (!*fill) {
                gf_log_callingfn ("dict", GF_LOG_ERROR,
                        "*fill is null!");
                goto out;
        }

        if ((buf + DICT_HDR_LEN) > (orig_buf + size)) {
                gf_log_callingfn ("dict", GF_LOG_ERROR,
                                  "undersized buffer passed. "
                                  "available (%lu) < required (%lu)",
                                  (long)(orig_buf + size),
                                  (long)(buf + DICT_HDR_LEN));
                goto out;
        }

        memcpy (&hostord, buf, sizeof(hostord));
        count = ntoh32 (hostord);
        buf += DICT_HDR_LEN;

        if (count < 0) {
                gf_log ("dict", GF_LOG_ERROR,
                        "count (%d) <= 0", count);
                goto out;
        }

        /* count will be set by the dict_set's below */
        (*fill)->count = 0;

        for (i = 0; i < count; i++) {
                if ((buf + DICT_DATA_HDR_KEY_LEN) > (orig_buf + size)) {
                        gf_log_callingfn ("dict", GF_LOG_ERROR,
                                          "undersized buffer passed. "
                                          "available (%lu) < required (%lu)",
                                          (long)(orig_buf + size),
                                          (long)(buf + DICT_DATA_HDR_KEY_LEN));
                        goto out;
                }
                memcpy (&hostord, buf, sizeof(hostord));
                keylen = ntoh32 (hostord);
                buf += DICT_DATA_HDR_KEY_LEN;

                if ((buf + DICT_DATA_HDR_VAL_LEN) > (orig_buf + size)) {
                        gf_log_callingfn ("dict", GF_LOG_ERROR,
                                          "undersized buffer passed. "
                                          "available (%lu) < required (%lu)",
                                          (long)(orig_buf + size),
                                          (long)(buf + DICT_DATA_HDR_VAL_LEN));
                        goto out;
                }
                memcpy (&hostord, buf, sizeof(hostord));
                vallen = ntoh32 (hostord);
                buf += DICT_DATA_HDR_VAL_LEN;

                if ((buf + keylen) > (orig_buf + size)) {
                        gf_log_callingfn ("dict", GF_LOG_ERROR,
                                          "undersized buffer passed. "
                                          "available (%lu) < required (%lu)",
                                          (long)(orig_buf + size),
                                          (long)(buf + keylen));
                        goto out;
                }
                key = buf;
                buf += keylen + 1;  /* for '\0' */

                if ((buf + vallen) > (orig_buf + size)) {
                        gf_log_callingfn ("dict", GF_LOG_ERROR,
                                          "undersized buffer passed. "
                                          "available (%lu) < required (%lu)",
                                          (long)(orig_buf + size),
                                          (long)(buf + vallen));
                }
                value = get_new_data ();
                value->len  = vallen;
                value->data = memdup (buf, vallen);
                value->is_static = 0;
                buf += vallen;

                dict_set (*fill, key, value);
        }

        ret = 0;
out:
        return ret;
}


/**
 * dict_allocate_and_serialize - serialize a dictionary into an allocated buffer
 *
 * @this: dict to serialize
 * @buf:  pointer to pointer to character. The allocated buffer is stored in
 *        this pointer. The buffer has to be freed by the caller.
 *
 * @return: success: 0
 *          failure: -errno
 */

int32_t
dict_allocate_and_serialize (dict_t *this, char **buf, size_t *length)
{
        int           ret    = -EINVAL;
        ssize_t       len = 0;

        if (!this || !buf) {
                gf_log_callingfn ("dict", GF_LOG_DEBUG,
                                  "dict OR buf is NULL");
                goto out;
        }

        LOCK (&this->lock);
        {
                len = _dict_serialized_length (this);
                if (len < 0) {
                        ret = len;
                        goto unlock;
                }

                *buf = GF_CALLOC (1, len, gf_common_mt_char);
                if (*buf == NULL) {
                        ret = -ENOMEM;
                        goto unlock;
                }

                ret = _dict_serialize (this, *buf);
                if (ret < 0) {
                        GF_FREE (*buf);
                        *buf = NULL;
                        goto unlock;
                }

                if (length != NULL) {
                        *length = len;
                }
        }
unlock:
        UNLOCK (&this->lock);
out:
        return ret;
}